Datasets:

bzantium

/

LongPerplexity

like 0

the study of spectra of graphs associated with finitely generated groups , such as cayley graphs or schreier graphs ( natural analogues of cayley graphs associated to not necessarily free group actions ) has a long history and is related to many problems in modern mathematics . still , very little is known about the dependence of the spectrum of the laplacian on the choice of generators in the group and on the weight on these generators . in the recent paper @xcite we addressed the issue of dependence on the weights on generators in the example of grigorchuk s group of intermediate growth . more specifically , we determined the spectral type of the laplacian on the schreier graphs describing the action of grigorchuk s group on the boundary of the infinite binary tree and showed that it is different in the isotropic and anisotropic cases . in fact , the spectrum is shown to be a cantor set of lebesgue measure zero in the anisotropic case , whereas , as has been known for a long time , it consists of one or two intervals in the isotropic case . moreover , we showed almost surely ( with respect to a natural measure on the boundary of the tree ) the absence of eigenvalues for the laplacians in question . our investigation in @xcite provides ( and relies on ) a surprising link between discrete schrdinger operators with aperiodic order and the substitutional dynamics arising from a presentation of the group by generators and relators . the purpose of the present paper is two - fold . firstly , we want to survey the spectral theoretic results of @xcite . secondly , we want to put these results in wider perspective by discussing the background on aperiodic order in dimension one . in this context we present a discussion of subshifts and aperiodic order in section [ section - subshifts ] and of schrdinger operators arising in models of ( dis)ordered solids in section [ schroedinger ] . we also continue our study of the substitution that is instrumental for the results in @xcite . it first appeared in 1985 in the presentation of grigorchuk s group by generators and relators found by lysenok @xcite ( such infinite recursive presentations are now called l - presentations ) . a remarkable fact is that the same substitution also describes basic dynamical properties of the group . here we review its combinatorial properties and carry out a detailed study of the factor map to its maximal equicontinuous factor , which is the binary odometer . this factor map is proven to be one - to - one everywhere except on three orbits . this in turn can be linked to the phenomenon of pure point diffraction which is at the core of aperiodic order . all these discussions concerning the substitution are contained in section [ the - substitution ] . the necessary background from graphs and dynamical systems is discussed in section [ section - background - graphs - etc ] and basics on grigorchuk s group and its schreier graphs can be found in section [ grigorchuk ] . the connection between schrdinger operators and laplacians on schreier graphs revealed in @xcite ( and reviewed in sections [ connecting ] and [ section - spectral ] ) can also be used to show that the kesten - von - neuman trace and the integrated density of states agree . this seems to be somewhat folklore . we provide a proof in section [ ids - schreier ] . our approach can certainly be carried out for various further families of groups generalizing grigorchuk s group . in particular , our results fully extend to a larger family of self - similar groups acting on the infinite binary tree considered by sunic in @xcite , as well as to similar families of self - similar groups acting on regular rooted trees of arbitrary degrees that also have linear schreier graphs for their action on the boundary of the tree . moreover , each of these self - similar groups belongs to an uncountable family of groups parametrized by sequences in a certain finite alphabet , in the same way as grigorchuk s group belongs to the family of groups @xmath0 constructed by the first named author in @xcite . the schreier graphs of the groups in the same family look the same , but their labelling by generators , and thus their spectra , depend on the sequence @xmath1 . the associated subshifts are also different , in particular , they do nt have to come from a substitution in the case when the corresponding group is not self - similar . this more general setup will be considered in a separate paper . * acknowledgments . * r. g. was partially supported by the nsf grant dms-1207669 , by erc ag compasp and by the nsa grant h98230 - 15 - 1 - 0328 . the authors acknowledge support of the swiss national science foundation . part of this research was carried out while d.l . and r. g. were visiting the department of mathematics of the university of geneva . the hospitality of the department is gratefully acknowledged . the authors also thank fabien durand and ian putnam for most enlightening discussions concerning the material gathered in section [ section - recoding ] , yaroslav vorobets for allowing them to use his figures @xmath2 and @xmath3 and olga klimecki for help in preparing figure @xmath4 . finally , the authors would like to thank the anonymous referee for a very careful reading of the manuscript and several helpful suggestions . long range aperiodic order ( or aperiodic order for short ) denotes an intermediate regime of order between periodicity and randomness . it has received a lot of attention over the last thirty years or so , see e.g. the article collections and monographs @xcite . this interest in aperiodic order has various reasons . on the one hand it is due to the many remarkable and previously unknown features and phenomena arising from aperiodic order in various branches of mathematics . on the other hand , it is also due to the relevance of aperiodic order in physics and chemistry . indeed , aperiodic order provides a mathematical theory for the description of a new type of matter discovered in 1982 by shechtman via diffraction experiments @xcite . these experiments showed sharp peaks in the diffraction pattern indicating long range order and at the same time a five fold symmetry indicating absence of periodicity . the discovery of solids combining both long range order with aperiodicity came as a complete surprise to physicists and chemists and shechtman was honored with a nobel prize in 2011 . by now , the solids in question are known as quasicrystals . in one dimension aperiodic order is commonly modeled by subshifts of low complexity . in higher dimensions it is modeled by dynamical systems consisting of point sets with suitable regularity features ( which are known as delone dynamical systems ) . here , we present the necessary notation in order to deal with the one - dimensional situation . * when we speak about aperiodic order subsequently , this will always mean that we have a subshift with suitable minimality features at our disposal . * let a finite set @xmath5 be given . we call @xmath5 the _ alphabet _ and refer to its elements as _ letters_. we will consider the set @xmath6 of finite words ( including the empty word ) over the alphabet @xmath5 , viewed as a free monoid ( with the multiplication given by concatenation of words and the empty word representing the identity ) . elements of @xmath6 will often be written as @xmath7 with @xmath8 . the length of a word is the number of its letters . it will be denoted by @xmath9 . the empty word has length zero . then , @xmath10 denotes the set of functions from @xmath11 to @xmath5 . we think of the elements of @xmath10 as of bi - infinite words over the alphabet @xmath5 . the set @xmath12 denotes the set of functions from @xmath13 to @xmath5 . we think of its elements of one - sided infinite words over @xmath5 . they will often be denoted by @xmath14 . if @xmath15 are finite words and @xmath16 satisfies @xmath17 we write @xmath18 and say that @xmath19 _ denotes the position of the origin_. we equip @xmath5 with the discrete topology and @xmath20 with the product topology . by the tychonoff theorem , @xmath10 is then compact . in fact , it is homeomorphic to the cantor set . a pair @xmath21 is called a _ subshift _ over @xmath5 if @xmath22 is a closed subset of @xmath20 which is invariant under the _ shift transformation _ @xmath23 if there exists a natural number @xmath24 with @xmath25 for all @xmath26 then @xmath21 is called _ periodic _ otherwise it is called _ non - periodic_. whenever @xmath1 is a word over @xmath5 ( finite or infinite , indexed by @xmath13 or by @xmath11 ) we define @xmath27 by convention , the set of finite subwords includes the empty word . every subshift @xmath28 comes naturally with the set @xmath29 of associated finite words given by @xmath30 a word @xmath31 is said to _ occur with bounded gaps _ if there exists an @xmath32 such that every @xmath33 with @xmath34 contains a copy of @xmath35 . as is well known ( and not hard to see ) @xmath28 is minimal if and only if every @xmath36 occurs with bounded gaps . for proofs and further discussion we refer to standard textbooks such as @xcite . we will be concerned here with the following strengthening of the bounded gaps condition . it concerns the case that @xmath37 can be chosen as @xmath38 with fixed @xmath39 ( independent of @xmath35 ) . a subshift @xmath21 is called _ linearly repetitive _ ( lr ) , if there exists a constant @xmath40 such that any word @xmath41 occurs in any word @xmath42 of length at least @xmath38 . this notion has been discussed under various names by various people . in particular it was studied by durand , host and skau @xcite in the setting of subshifts ( under the name linearly recurrent ) . for delone dynamical systems it was brought forward at about the same time by lagarias and pleasants @xcite under the name linearly repetitive. it has also featured in the work of boris solomyak @xcite ( under the name uniformly repetitive ) . it was also already discussed in an unpublished work of boshernitzan in the 90s . that work also contains a characterization in terms of positivity of weights . a corresponding result for delone systems was recently given in @xcite . durand @xcite gives a characterization of such subshifts in terms of primitive @xmath43-adic systems and shows the following ( which was already known to boshernitzan ) . [ theorem - unique - ergodicity ] let @xmath21 be a linearly repetitive subshift . then , the subshift is uniquely ergodic . in fact , linear repetitivity implies a strong form of subadditive ergodic theorem @xcite . validity of such a result together with the fundamental work of kotani @xcite is at the core of the approach to cantor spectrum of lebesgue measure zero developed in @xcite . our considerations on cantor spectrum rely on an extension of that approach worked out in @xcite . in this section , we present ( parts of ) the spectral theory of discrete schrdinger operators associated to minimal dynamical systems . schrdinger operators occupy a prominent position in the theory of aperiodic order . indeed , they arise in the quantum mechanical description of conductance properties of quasicrystals and exhibit quite interesting mathematical properties . in fact , already the first two papers on them written by physicists suggest that the corresponding spectral measures are purely singular continuous and the spectrum is a cantor set of lebesgue measure zero @xcite . by now these features as well as other conductance - related properties known as anomalous transport have been thoroughly studied in a variety of models by various authors , see the survey articles @xcite . the phenomenon that the underlying spectrum is a cantor set of lebesgue measure zero is usually referred to as _ cantor spectrum of lebesgue measure zero _ and this is how we will refer to it subsequently . the laplacians on schreier graphs discussed later will turn out to be unitarily equivalent to certain such schrdinger operators . subsequently , we first discuss basic mathematical features of schrdinger operators associated to dynamical systems in the next section [ basic - schroedinger ] , then turn to absolutely continuous spectrum and the spectrum as a set for subshifts in section [ spectrum - set - ac ] and finally give some background from physics in section [ aperiodic - order - schroedinger ] . in this section we review some basic theory of discrete schrdinger operators ( or rather jacobi matrices ) associated to minimal topological dynamical systems . this framework is slightly more general than the framework of minimal subshifts . the results we discuss include constancy of the spectrum and uniform convergence of the so - called integrated density of states . all these results are well known . whenever @xmath44 is a homeomorphism of the compact space @xmath22 we will refer to @xmath21 as a _ topological dynamical system_. later we will meet an even more general definition of dynamical system . to continuous functions @xmath45 we then associate a family of _ discrete operators _ @xmath46 . specifically , for each @xmath47 , @xmath48 is a bounded selfadjoint operator from @xmath49 to @xmath49 acting via @xmath50 for @xmath51 and @xmath52 . in the case @xmath53 the above operators are known as _ discrete schrdinger operators_. for general @xmath54 the name _ jacobi matrices _ is often used in the literature . here , we will deal with the case @xmath55 but still mostly refer to the arising operators as schrdinger operators . as the operator @xmath48 is selfadjoint , the operator @xmath56 is bijective with continuous inverse @xmath57 for any @xmath58 . moreover , for any @xmath59 there exists a unique positive borel measure @xmath60 on @xmath61 with @xmath62 for any @xmath58 . this measure if finite and assigns the value @xmath63 to the set @xmath61 . for fixed @xmath64 , the measures @xmath60 , @xmath65 , are called the _ spectral measures _ of @xmath48 . the spectrum of @xmath48 is defined as @xmath66 it is the smallest set containing the support of any @xmath60 ( see e.g. @xcite ) . the spectrum is said to be _ purely absolutely continuous _ if the spectral measures for all @xmath67 are absolutely continuous with respect to lebesgue measure . the spectrum is said to be _ purely singular continuous _ if all spectral measures are both continuous ( i.e. do not have discrete parts ) and singular with respect to lebesgue measure . the following result is well - known . it can be found in various places , see e.g. @xcite . recall that @xmath21 is called _ minimal _ if @xmath68 is dense in @xmath22 for any @xmath47 . [ constancy]let @xmath21 be minimal and @xmath69 continuous . then , there exists a closed subset @xmath70 such that the spectrum @xmath71 of @xmath48 equals @xmath72 for any @xmath64 . we will refer to the set @xmath72 in the previous theorem as the _ spectrum of the schrdinger operator associated to @xmath21 ( and ( f , g))_. the spectrum @xmath72 is a one of the main objects of interest in our study . before turning to a finer analysis of the spectrum we will introduce a further quantity of interest the so called integrated density of states . in order to do so , we will assume that the underlying dynamical system @xmath21 is not only minimal but also _ uniquely ergodic _ i.e. possesses a unique invariant probability measure , which we call @xmath73 . then , we can associate to the family @xmath74 the positive measure @xmath75 on @xmath61 defined via @xmath76 ( for @xmath77 any continuous function on @xmath61 with compact support ) . here , @xmath78 is just the characteristic function of @xmath79 . this measure @xmath75 is called the _ integrated density of states_. let @xmath80 , n(e):=\int_{(-\infty , e ] } d k,\ ] ] be the distribution function of @xmath75 . there is a direct relation between the measure @xmath75 and the spectrum of the @xmath48 . [ theorem - support - and - atomfree ] let @xmath21 be minimal and uniquely ergodic . then , the set @xmath72 is the support of the measure @xmath75 . if the function @xmath54 does not vanish anywhere then @xmath75 does not have atoms ( i.e. it assigns the value zero to any set containing only one element ) . this is rather standard in the theory of random operators . specific variants of it can be found in many places . in particular , the first statement of the theorem can be found in @xcite . in the case of @xmath54 which do not vanish anywhere the statements of the theorem are contained in section 5 of @xcite . given the constancy of the spectrum , theorem [ constancy ] , the statements are also very special cases of the results of section 5 of @xcite . the key ingredient for the absence of atoms is amenability of the underlying group @xmath11 . the statement on the support does not even need this property . as is well known , it is possible to calculate @xmath75 via an approximation procedure . this will be discussed next . for @xmath81 let @xmath82 be the canonical inclusion and let @xmath83 be the adjoint of @xmath84 . thus , @xmath85 is the canonical projection . define for @xmath64 then @xmath86 we will be concerned with the spectral theory of these operators . let the measure @xmath87 on @xmath61 be defined as @xmath88 ( for any continuous @xmath77 on @xmath61 with compact support ) and let @xmath89,\ : n_\omega^{n } ( e):=\int_{(\infty , e ] } d k_\omega^{n } ( x),\ ] ] be its distribution function . let @xmath90 be the eigenvalues of @xmath91 counted with multiplicity . then , straightforward linear algebra ( diagonalization of @xmath91 and independence of the trace of the chosen orthonormal basis ) shows that @xmath92 holds for any continuous @xmath77 on @xmath61 with compact support and that the distribution functions of the measures @xmath93 are given by @xmath94 where @xmath95 denotes the cardinality of a set . in this sense , @xmath87 is just an averaged eigenvalue counting . [ theorem - convergence - ids ] let @xmath21 be minimal and uniquely ergodic . then , for any continuous @xmath77 on @xmath61 with compact support and any @xmath96 there exists an @xmath97 with @xmath98 for all @xmath64 and all @xmath52 with @xmath99 . it is well - known that the measures @xmath100 converge weakly toward @xmath75 for @xmath101 for almost every @xmath64 . this can be found in many places , see e.g. lemma 5.12 in @xcite . ( that lemma assumes that @xmath54 does not vanish anywhere but its proof does not use this assumption . ) a key step in the proof is the use of the birkhoff ergodic theorem . the desired statement now follows by replacing the birkhoff ergodic theorem with the uniform ergodic theorem ( oxtoby theorem ) valid for uniquely ergodic systems @xcite . the operators @xmath91 are sometimes thought of as arising out of the @xmath48 by some form of dirichlet boundary condition. the previous result is stable under taking different boundary conditions. in fact , even a more general statement is true as we will discuss next . ( the more general statement will even save us from saying what we mean by boundary condition . ) let for any @xmath102 and @xmath103 be a selfadjoint operator @xmath104 on @xmath105 be given . then , the statement of the theorem essentially continues to hold if the operators @xmath91 are replaced by the operators @xmath106 provided the rank of the @xmath107 is not too big . here , the rank of an operator @xmath39 on a finite dimensional space , denoted by @xmath108 , is just the dimension of the range of @xmath39 . in order to be more specific , we introduce the measure @xmath109 on @xmath61 defined as @xmath110 ( for any continuous @xmath77 on @xmath61 with compact support ) and its distribution function given by @xmath111,\ ; \widetilde{n}_\omega^{n } ( e):= \frac{1}{n } \sharp\{\mbox{eigenvalues of $ \widetilde{h}_\omega^{n}$ not exceeding $ e$ } \}.\ ] ] [ perturbation - ids ] consider the situation just described . let @xmath64 be given with @xmath112 then , for any continuous @xmath77 on @xmath61 with compact support and any @xmath96 there exists an @xmath97 with @xmath113 for all @xmath99 . a consequence of the min - max principle , see e.g. theorem 4.3.6 in @xcite shows @xmath114 independent of @xmath115 . this directly gives the desired statement . the theorem allows one to obtain an inclusion formula for the spectrum @xmath72 . denote the spectrum of the operator @xmath116 by @xmath117 . by construction , @xmath117 is just the support of the measure @xmath109 . [ cor - inclusion ] assume the situation of the previous theorem . then , the inclusion @xmath118 holds for all @xmath119 . the corollary is somewhat unsatisfactory in that it only gives an inclusion . in certain cases more is known . this is further discussed in section [ ids - equal - kvns - trace ] . for a general result on how to construct approximations whose spectra converge with respect to the hausdorff distance we refer the reader to @xcite . in this section we consider schrdinger operators associated to locally constant functions on minimal subhifts . here , a function @xmath120 on a subshift @xmath22 over a finite alphabet is _ locally constant _ if there exists an @xmath121 such that the value of @xmath122 only depends on the word @xmath123 . a key distinction in our considerations will then be whether @xmath124 is periodic or not . the overall structure of the spectrum in the periodic case is well - known . this can be found in many references , see e.g. the monograph @xcite . [ theorem - periodic - abstract ] let @xmath21 be a minimal subshift and @xmath69 locally constant with @xmath125 for all @xmath26 . if @xmath124 is periodic ( with period @xmath126 ) , then the spectra @xmath72 of the associated schrdinger operators consist of finitely many ( and not more than @xmath126 ) closed intervals of positive length and all spectral measures are absolutely continuous with respect to lebesgue measure . note that periodicity of @xmath124 may have its origin in both properties of @xmath21 and properties of @xmath124 . for example periodicity always occurs if @xmath127 irrespective of the nature of @xmath21 . also , periodicity occurs for arbitrary @xmath128 if @xmath21 is periodic ( i.e. there exists a natural number @xmath129 with @xmath130 for all @xmath64 ) . the previous theorem gives rather complete information on @xmath72 in the periodic case . in order to deal with the non - periodic case , we will need a further assumption on @xmath21 . this condition is linear repetitivity . [ theorem - aperiodic - abstract ] let @xmath21 be a linearly repetitive subshift and @xmath69 locally constant with @xmath125 for all @xmath26 . if @xmath124 is non - periodic , then there exists a cantor set @xmath72 of lebesgue measure zero in @xmath61 such that @xmath131 for all @xmath103 . the above theorem was first proven in @xcite in the case @xmath53 . this result was then extended in @xcite from linearly repetitive subshifts to arbitrary subshifts satisfying a certain condition known as boshernitzan condition ( b ) ( again for the case @xmath132 ) . in the form stated above it can be inferred from the recent work @xcite , corollary 4 . this corollary treats the even more general situation , where condition ( b ) is satisfied . condition ( b ) was introduced by boshernitzan as a sufficient condition for unique ergodicity @xcite ( see @xcite for an alternative approach as well ) . in our context , we do not actually need its definition here . it suffices to know that linear repetitivity implies ( b ) ( see e.g. @xcite ) . the previous result deals with the appearance of @xmath72 as a set . it also gives some information on the spectral type . [ corollary - aperiodic - abstract ] assume the situation of the previous theorem . then , no spectral measure can be absolutely continuous with respect to the lebesgue measure . schrdinger operators with aperiodic order can be considered within the context of random schrdinger operators . indeed , they arise in quantum mechanical treatment of solids . as this may be revealing we briefly present this context in this section . further discussion and references can be found e.g. in the textbooks @xcite . consider a subshift @xmath21 over the finite alphabet @xmath5 . assume without loss of generality that @xmath5 is a subset of the real numbers . let a @xmath44-invariant probability measure @xmath133 on @xmath22 be given . to these data we can associate the family @xmath134 of bounded selfadjoint operators on @xmath49 acting via @xmath135 such operators are ( slightly special ) cases of the operators considered in the previous section . they arise in the quantum mechanical treatment of disordered solids . the solid in question is modeled by the sequence @xmath26 . the operator @xmath48 then describes the behavior of one electron under the influence of this @xmath1 . more specifically , if the state of the electron is @xmath136 at time @xmath137 then the time evolution is governed by the _ schrdinger equation _ @xmath138 this equation has a unique solution given by @xmath139 the behavior of this solution is then linked to the spectral properties of @xmath48 . the two basic pieces of philosophy underlying the investigations are now the following : * increasing regularity of the spectral measures increases the conductance properties of the solid in question . * the more disordered the subshift is the more singular the spectral measures are . of course , this has to be taken with ( more than ) a grain of salt . in particular , precise meaning has to be given to what is meant by _ regularity and singularity of the spectral measures _ and _ conductance properties _ and _ disorder in the subshift_. a large part of the theory is then devoted to giving precise sense to these concepts and then prove ( or disprove ) specific formulations of the mentioned two pieces of philosophy . regarding the first point of the philosophy we mention @xcite as basic references for proofs of lower bounds on transport via quantitative continuity of the spectral measures . as for the second point of the philosophy , the two - in some sense - most extremal cases are given by periodic subshifts representing the maximally ordered case on the one hand and the bernoulli subshift ( with uniform measure ) on the other hand representing the maximally disordered case . the periodic situation can be thought of as one with maximal order . as discussed in the previous section the spectral measures are all absolutely continuous ( hence not at all singular ) and the spectrum consists of non - trivial intervals . these intervals are known in the physics literature as ( conductance ) bands . the bernoulli subshift can be thought of as having maximal disorder . in this case all spectral measures turn out to be pure point measures and the spectrum is pure point spectrum with the eigenvalues densely filling suitable intervals . we refer to the monographs @xcite for details and further references . the subshifts considered in the previous section are characterized by some intermediate form of disorder . they are not periodic . however , they are still minimal and uniquely ergodic and have very low complexity . so they are close to the periodic situation ( or rather well approximable by periodic models with bigger and bigger periods ) . accordingly , one can expect the following spectral features of the associated schrdinger operators : * absence of absolutely continuous spectral measures ( due to the presence of disorder i.e. the lack of periodicity ) . * absence of point spectrum ( due to the closeness to the periodic case ) . * cantor spectrum of lebesgue measure zero ( as a consequence of approximation by periodic models with bigger and bigger periods and , hence , more and more gaps ) . indeed , a large part of the theory for schrdinger operators with aperiodic order is devoted to proving these features ( as well as more subtle properties ) for specific classes of models . further details and references can be found in the surveys @xcite . here , we emphasize that also our discussion of the operators associated to a certain substitution generated subshift below will be focused on establishing the above features . in this section we study the two - sided subshift @xmath143 induced by a particular substitution @xmath140 on @xmath144 with @xmath145 the one - sided subshift induced by this substitution had already been studied by vorobets @xcite . some of our results can be seen as providing the two - sided counterparts to his investigations . the key ingredient in the investigations of @xcite is that the arising one - sided sequences can be considered as toeplitz sequences . this is equally true in our case of two - sided sequences . thus , it seems very likely that one could also base our analysis of the corresponding features on the connection to toeplitz sequenes . here , we will present a different approach based the @xmath146-decomposition and @xmath146-partition introduced in @xcite . the subshift @xmath142 will be of crucial importance for us as it will turn out that the schrdinger operators associated to it are unitarily equivalent to the laplacians on the schreier graphs of the grigorchuk s group @xmath147 . while we do not use it in the sequel we would like to highlight that the substitution in question has already earlier appeared in the study of grigorchuk s group @xmath147 . indeed , it is ( a version of ) the substitution used by lysenok @xcite for getting a presentation of grigorchuk s group @xmath147 . more specifically , @xcite gives that @xmath148 where @xmath149 is the substitution on @xmath150 obtained from @xmath140 by replacing @xmath151 by @xmath152 , @xmath153 by @xmath154 and @xmath155 by @xmath156 . let the alphabet @xmath144 be given and let @xmath140 be the substitution mentioned above mapping @xmath157 , @xmath158 , @xmath159 , @xmath160 . let @xmath141 be the associated set of words given by @xmath161 then , the following three properties obviously hold true : * the letter @xmath162 is a prefix of @xmath163 for any @xmath164 . * the lengths @xmath165 converge to @xmath166 for @xmath167 . * any letter of @xmath5 occurs in @xmath168 for some @xmath146 . by the first two properties @xmath163 is a prefix of @xmath169 for any @xmath170 . thus , there exists a unique one - sided infinite word @xmath171 such that @xmath163 is a prefix of @xmath171 for any @xmath170 . this @xmath171 is then a fixed point of @xmath140 i.e. @xmath172 . we will refer to it as _ the fixed point of the substitution @xmath140_. clearly , @xmath171 is then a fixed point of @xmath173 as well for any natural number @xmath146 . by the third property we then have @xmath174 we can now associate to @xmath140 the subshift @xmath175 note that every other letter of @xmath171 is an @xmath162 ( as can easily be seen ) . thus , @xmath162 occurs in @xmath171 with bounded gaps . this implies that any word of @xmath141 occurs with bounded gaps ( as the word is a subword of @xmath163 and @xmath171 is a fixed point of @xmath173 ) . for this reason @xmath142 is minimal and @xmath176 holds for any @xmath177 . we can then apply theorem 1 of @xcite to obtain the following . [ theorem - tau - linear - repetitive ] the subshift @xmath142 is linearly repetitive . in particular , @xmath142 is uniquely ergodic and minimal . it is well - known that subshifts associated to primitive substitutions are linearly repetitive ( see e.g. @xcite ) . theorem 1 of @xcite shows that linear repetitivity in fact holds for subshifts associated to any substitution provided minimality holds . unique ergodicity is then a direct consequence of linear repetitivity due to theorem [ theorem - unique - ergodicity ] . our further considerations will be based on a more careful study of the @xmath163 . we set @xmath178 a direct calculation gives @xmath179 thus , the following _ recursion formula _ for the @xmath180 @xmath181 with @xmath182 is valid[recursion ] . this recursion formula is a very powerful tool . this will become clear in the subsequent sections . here we first note that it implies @xmath183 for all @xmath184 . we will now use it to present a formula for the occurrences of the @xmath185 in @xmath171 and to introduce some special elements in @xmath186 . consider the fixed point @xmath187 of @xmath188 on @xmath10 . then the following holds . * the letter @xmath162 occurs exactly at the positions @xmath189 , @xmath190 ( i.e. at the odd positions ) . * the letter @xmath151 occurs exactly at the positions of the form @xmath191 , @xmath192 ( i.e. at the positions of the form @xmath193 with @xmath194 an odd integer and @xmath164 arbitrary ) . * the letter @xmath153 occurs exactly at the positions of the form @xmath195 , @xmath192 ( i.e. at the positions of the form @xmath196 with @xmath194 an odd integer and @xmath164 arbitrary ) . * the letter @xmath155 occurs exactly at the positions of the form @xmath197 , @xmath192 ( i.e. at the positions of the form @xmath198 with @xmath194 an odd integer and @xmath164 arbitrary ) . we first note that the given sets of positions are pairwise disjoint and cover @xmath13 . thus , it suffices to show that the mentioned letters occur at these positions . the statement for @xmath162 is clear . the statements for @xmath185 can all be proven similarly . thus , we only discuss the statement for @xmath151 . repeated application of ( rf ) shows that @xmath199 with @xmath200 . moreover , ( rf ) implies @xmath201 combining these formula we see that @xmath151 must occur at all positions of the form @xmath202 this finishes the proof . the previous proposition shows that @xmath171 is a _ toeplitz sequence _ ( with periods of the form @xmath203 for @xmath204 ) . as mentioned already the analysis of the one - sided subshift in @xcite is based on this property . we now head further to use ( rf ) to introduce some special two - sided sequences . as is not hard to see from ( rf ) , for any @xmath164 and any single letter @xmath205 the word @xmath206 occurs in @xmath171 . thus , for any @xmath205 there exists a unique element @xmath207 such that @xmath208 holds for all natural numbers @xmath146 , where the @xmath19 denotes the position of the origin . the elements @xmath209 will play an important role in our subsequent analysis . they clearly have the property that they agree on @xmath13 with @xmath171 . indeed , they can be shown to be exactly those elements of @xmath186 which agree with @xmath171 on @xmath13 ( see below ) . here , we already note that these three sequences are different . thus , @xmath186 contains different sequences , which agree on @xmath13 . hence , @xmath22 is not periodic . we finish this section by noting a certain reflection invariance of our system . recall that a non - empty word @xmath210 with @xmath8 is called a _ palindrome _ if @xmath211 . an easy induction using ( rf ) shows that for any @xmath212 the word @xmath180 is a palindrome . it starts and ends with @xmath213 for any @xmath190 with @xmath214 . as each @xmath180 is a palindrome and any word belonging to @xmath141 is a subword of some @xmath180 we immediately infer that @xmath141 is closed under reflections in the sense that the following proposition holds . [ prop - palindrome ] whenever @xmath215 with @xmath8 belongs to @xmath141 then so does @xmath216 . as a direct consequence of the definitions we obtain that for any @xmath170 the word @xmath171 has a ( unique ) decomposition as @xmath217 with @xmath218 . clearly , this decomposition can be thought of as a way of writing @xmath171 with letters from the alphabet @xmath219 . moreover , setting @xmath220 we have @xmath221 for any @xmath222 . this way of writing @xmath171 will be called the _ @xmath146-decomposition of @xmath171_. it turns out that an analogous decomposition can actually be given for any element @xmath223 . this will be discussed in this section . specifically , we will discuss next that each @xmath224 admits for each @xmath170 a unique decomposition of the form @xmath225 with * @xmath226 for all @xmath227 , * the origin @xmath228 belongs to @xmath229 . such a decomposition will be referred to as _ @xmath146-decomposition _ of @xmath1 . a short moment s thought reveals that if such a decomposition exists at all , then it is uniquely determined by the position of any of the @xmath230 s in @xmath1 . moreover , the positions of the @xmath230 s are given by @xmath231 with @xmath232 . thus , the positions are given by an element of @xmath233 . this suggests the following definition . for @xmath212 we call an element @xmath234 an @xmath235 of @xmath223 if for any @xmath236 both * @xmath237 and * @xmath238 hold . clearly , for each @xmath239 , existence ( uniqueness ) of an @xmath146-partition is equivalent to existence ( uniqueness ) of an @xmath146-decomposition . in this sense these two concepts are equivalent . it is not apparent that such an @xmath146-partition exists at all . here is our corresponding result . [ theorem - n - partition ] let @xmath170 be given . then any @xmath223 admits a unique @xmath146-partition @xmath240 and the map @xmath241 is continuous and equivariant ( i.e. @xmath242 ) . based on n - partitions ( and n - decompositions ) and the previous theorem one can then study the dynamical system @xmath142 as well as various questions on the structure of @xmath141 . this is the content of the next sections . in this section we use n - partitions to study the structure of the dynamical system @xmath142 . for any @xmath243 we can consider the cyclic group @xmath244 together with the map @xmath245 , called _ addition map _ , which sends @xmath246 to @xmath247 . then , @xmath248 is a periodic minimal dynamical system . moreover , there are natural maps @xmath249 for any @xmath250 . these maps allow one to construct the topological abelian group @xmath251 as the inverse limit of the system @xmath252 , @xmath243 . specifically , the elements of @xmath251 are sequences @xmath253 with @xmath254 and @xmath255 for all @xmath243 . this group is called the _ group of dyadic integers_. the addition maps @xmath245 are compatible with the inverse limit and lift to a map @xmath256 on @xmath251 ( which is just addition by @xmath4 on each member of the sequence in question ) . in this way , we obtain a dynamical system @xmath257 . it is known as the _ binary odometer_. as @xmath256 is just addition one can think of this system as an adding machine. it is well known that this dynamical system is minimal . now theorem [ theorem - n - partition ] can be rephrased as saying that the dynamical system @xmath258 is a factor of @xmath142 via the factor map @xmath259 . clearly , the factor maps @xmath259 are compatible with the natural canonical projections @xmath260 in the sense that @xmath261 holds for all @xmath262 . thus , we can combine the @xmath259 for all @xmath263 to get a factor map @xmath264 we first use this to study continuous eigenvalues . let @xmath265 be a dynamical system ( i.e. @xmath266 is a compact space and @xmath267 is a homeomorphism ) . denote the unit circle in @xmath268 by @xmath269 . then , @xmath270 is called a _ continuous eigenvalue _ of the dynamical system @xmath265 if there exists a continuous not everywhere vanishing function @xmath54 with values in @xmath268 on @xmath266 satisfying @xmath271 for all @xmath272 . such a function is called a _ continuous eigenfunction_. then , any continuous eigenvalue belongs to @xmath269 and the continuous eigenvalues form a group under multiplication whenever the underlying dynamical system is minimal . indeed , by minimality any continuous eigenfunction has constant ( non - vanishing ) modulus . then , the product of two eigenfunctions is an eigenfunction to the product of the corresponding eigenvalues . the complex conjugate of an eigenfunction is an eigenfunction to the inverse of the corresponding eigenvalue and the constant function is an eigenfunction to the eigenvalue @xmath4 . moreover , it is not hard to see that minimality implies that the multiplicity of each continuous eigenvalue is one ( i.e. for any two continuous eigenfunctions @xmath128 to the same eigenvalue there exists a complex number c with @xmath273 ) . let now @xmath274 be the group of continuous eigenvalues of @xmath275 . this is just the subgroup of @xmath269 given by @xmath276 . then , clearly the groups @xmath274 and @xmath275 are dual to each other via @xmath277 the dual maps to the canonical projections @xmath278 are then the canonical embeddings @xmath279 thus , @xmath274 is a subgroup of @xmath280 . let @xmath281 be the group arising as the union of the @xmath282 , i.e. @xmath283 this group is often denoted as @xmath284 . equip it with the discrete topology and denote its pontryagin dual @xmath285 by @xmath286 . [ prop - canonical - isom ] the group @xmath286 is canonically isomorphic to the group @xmath251 via @xmath287 by construction @xmath251 comes about as inverse limit of the @xmath252 , @xmath262 . then , the dual group of @xmath251 arises as the direct limit of the dual system @xmath288 . this limit is just the group @xmath281 . dualising once more we obtain that @xmath286 is indeed canonically isomorphic to the dual of @xmath251 . to obtain the actual formula we can proceed as follows : define @xmath289 . then , each @xmath290 is a complex primitive @xmath291-th root of @xmath4 with @xmath292 . now , consider an arbitrary element @xmath293 i.e. a character @xmath294 . then , @xmath295 is completely determined by its values on the @xmath290 , @xmath296 . moreover , for each @xmath146 we have @xmath297 for a unique @xmath298 as @xmath299 it is not hard to see that @xmath300 goes to @xmath301 under the natural surjection @xmath302 . thus , to each character @xmath303 there corresponds a sequence @xmath304 with @xmath305 and @xmath255 for all @xmath263 ( and vice versa ) . the set of such sequences is exactly @xmath251 . as any eigenvalue belongs to @xmath269 , there is a canonical embedding of groups @xmath306 . in fact , as things are set up here this embedding is just inclusion of subsets of @xmath268 . by duality , this gives rise to a group homomorphism @xmath307 with dense range . this homomorphism induces then an action @xmath308 of @xmath11 on @xmath286 via @xmath309 it is not hard to see that this @xmath308 corresponds to @xmath256 if @xmath286 is identified with @xmath251 according to proposition [ prop - canonical - isom ] . disentangling definitions , we also infer that the action is given by @xmath310 ( for @xmath311 and @xmath312 ) . we denote the arising dynamical system as @xmath313 . it is isomorphic to the binary odometer . by its very construction it is what is called a _ rotation on a compact abelian group _ ( viz the action @xmath308 comes about by multiplication with @xmath314 , where @xmath315 is a group homomorphism ) . thus , by standard theory ( see e.g. @xcite ) its group of continuous eigenvalues is exactly given by the dual of @xmath286 i.e. by @xmath281 . now , obviously any eigenfunction of @xmath275 gives immediately rise to an eigenfunction of @xmath186 for the same eigenvalue ( by composing with the factor map ) . as the factor map is continuous , we obtain in this way continuous eigenfunctions to each of the elements from @xmath281 . minimality easily shows that ( up to an overall scaling ) each of these eigenfunctions is unique . thus , we obtain a family of continuous eigenfunctions . at this point it is not clear that all continuous eigenvalues of @xmath142 belongs to @xmath281 . however , this ( and more ) will be shown later . we can use the preceding considerations to introduce a closed equivalence relation @xmath316 on @xmath186 via @xmath317 then , @xmath318 is a compact topological space when equipped with the quotient topology . clearly , @xmath319 if and only if @xmath320 . thus , the relation @xmath316 is compatible with the shift operation . hence , the quotient @xmath318 becomes a dynamical system with the operation @xmath321 induced by the shift . fix now an @xmath322 . as discussed above continuous eigenfunctions do not vanish anywhere and the multiplicity of each continuous eigenvalue is one . thus , for each @xmath312 there exists a unique eigenfunction @xmath323 to @xmath75 on @xmath186 with @xmath324 . then , the arising system of eigenfunctions will have the property that @xmath325 for all @xmath326 . thus , any @xmath224 will give rise to an element of @xmath327 via @xmath328 even more is true and the following result holds . it is well - known and can be found in various places in the literature . recent discussions are given in @xcite . [ conjugate ] the dynamical systems @xmath329 and @xmath313 are conjugate via the map @xmath330 \mapsto ( k\mapsto f_k ( \omega)).\ ] ] in particular , the eigenvalues of @xmath331 are exactly given by the elements of @xmath281 . we now further investigate @xmath316 and provide a characterization of @xmath319 . here , we will again use the special words @xmath332 introduced above . [ prop - characterization - approx ] for @xmath333 the relation @xmath319 holds if and only if one of the following two properties hold : * @xmath334 . * there exist @xmath335 with @xmath336 and @xmath337 with @xmath338 and @xmath339 . let @xmath340 with @xmath341 and @xmath342 be given . by definition of @xmath316 and the above construction of the eigenfunctions of @xmath142 , we then have that @xmath343 for all @xmath212 . call this quantity @xmath259 . in the remaining part of the proof we will identify such a @xmath259 with its unique representative in @xmath344 . as @xmath341 we infer that one of the sequences @xmath345 or @xmath346 must be bounded . ( otherwise , @xmath1 and @xmath347 would agree on larger and larger pieces around the origin and then had to be equal . ) assume without loss of generality that @xmath348 is bounded . by restricting attention to a subsequence we can then assume without loss of generality that @xmath349 for all @xmath146 . after shifting the sequences by @xmath350 to the left we can then assume without loss of generality that @xmath351 for all @xmath146 . by definition of @xmath350 , there exist then letters @xmath352 with latexmath:[\[\omega = ... s| p^{(n)} ... \;\:\mbox{and}\;\ : \omega { ' } = ... s { ' } denotes the position of the origin . this gives , by definition of the @xmath146-partition that in fact @xmath354 and @xmath355 . as @xmath356 we infer that @xmath357 . this finishes the proof . the previous result shows that the factor map from @xmath186 to @xmath358 is one - to - one except on three orbits . this has strong consequences as will be discussed next . as @xmath142 is uniquely ergodic , there exists a unique @xmath44-invariant probability measure @xmath73 on @xmath186 . the operation @xmath44 then induces a unitary operator @xmath359 on the associated @xmath360-space via @xmath361 an element @xmath362 ( with @xmath363 ) is called a _ measurable eigenfunction to @xmath364 _ if @xmath365 . the subshift is said to have _ pure point spectrum _ if there exists an orthonormal basis of measurable eigenfunctions . from the two previous results we immediately infer the following . the dynamical system @xmath142 has pure point spectrum and any measurable eigenvalue is a continuous eigenvalue and belongs to @xmath281 . the dynamical system @xmath366 has pure point spectrum with all eigenvalues being continuous and belonging to @xmath281 as it is a shift on the compact abelian group @xmath286 which is the pontryagin dual of @xmath281 , @xcite . as the dynamical system @xmath318 is conjugate to @xmath313 due to lemma [ conjugate ] it has also pure point spectrum with all eigenvalues being continuous and belonging to @xmath281 . now , proposition [ prop - characterization - approx ] shows that factor map from @xmath186 to @xmath358 is one - to - one except on three countable orbits . this implies that in terms of measures the associated @xmath360-spaces are isomorphic . this easily gives the desired result . the occurrence of pure point dynamical spectrum is a key feature in the investigation of aperiodic order . in fact , while there is no axiomatic framework for aperiodic order a distinctive feature is ( pure ) point diffraction . now , pure point diffraction has been shown to be equivalent to pure point dynamical spectrum . for the case of subshifts at hand this can be inferred ( after some work ) from @xcite . a more general result ( dealing with uniquely ergodic delone systems ) was then given in @xcite . the result can even further be generalized to measure dynamical systems and even processes @xcite . the previous theorem implies that @xmath313 is exactly the _ maximal equicontinuous factor _ of @xmath142 ( see e.g. @xcite for definition ) . indeed , one of the many equivalent ways to describe this factor is as the dual group of the group of continuous eigenvalues . a recent discussion of this and various related facts can be found in @xcite . henceforth , we will denote the maximal equicontinuous factor of @xmath142 by @xmath367 and the corresponding factor map by @xmath368 . then , our findings so far provide the following theorem . [ main - max - equicontinuous - factor ] the three dynamical systems @xmath369 , @xmath370 , @xmath313 and @xmath367 are topologically conjugate . the factor map @xmath371 is one - to - one in all points except on the images of the points of the orbits of @xmath332 . in these points it is three - to - one . * minimal systems with the property that their factor map to the maximal equicontinuous factor is one - to - one in at least one point are known as _ almost automorphic systems _ ( see e.g. @xcite for further details ) . their study has attracted a lot of attention . as the previous result shows , @xmath142 is an almost automorphic system . in fact , as @xmath186 is uncountable , the factor map is one - to - one in almost every point with respect to the unique invariant probability measure @xmath73 on @xmath186 . this is sometimes expressed as _ the factor map from @xmath142 to its maximal equicontinuous factor is almost - everywhere one - to - one_. * there is an alternative description of the relation @xmath316 for almost automorphic systems . more specifically , define the _ proximality relation _ @xmath372 by @xmath373 where @xmath156 is any metric on @xmath186 inducing the topology . ( due to compactness of @xmath186 the relation is indeed independent of the chosen metric . ) note that the proximality relation can be though of as describing asymptotic agreement . then , for almost automorphic systems the proximality relation @xmath372 and the relation @xmath316 agree . this can be found in the book of auslander @xcite . a recent discussion is given in @xcite . in fact , this result is even more general in that one does not need almost automorphy but only a weaker condition called _ coincidence rank one_. we refrain from further discussion of this concept and refer the reader to e.g. @xcite for further investigation . we just note that in our situation equality of @xmath372 and @xmath316 and the characterization of @xmath316 in proposition [ prop - characterization - approx ] imply that sequences which are proximal ( i.e. asymptotically equal ) are in fact equal everywhere up to one position . * in @xcite vorobets shows that the one - dimensional subshift associated to @xmath140 has the binary odometer @xmath313 as a factor with the factor map being @xmath374 in all points except three orbits . he uses this to conclude pure point spectrum and unique ergodicity . our corresponding results above for the two - sided subshift can therefore be seen as analogues to his results . however , our approach is different : it is based on @xmath146-partition whereas his approach is based on toeplitz sequences . in this section we have a closer look at the structure of @xmath141 . the main focus will be on occurrences of three blocks and the index of words ( also known as critical exponent ) . we will use @xmath146-partitions in our study in a spirit similar to @xcite . we start by investigating occurrences of almost four blocks . an easy inspection of @xmath171 gives the following lemma . [ one - three - plus - block ] the word @xmath375 belongs to @xmath141 . the previous result deals with occurrence of a cube ( and even a bit more ) of the special word @xmath376 . as @xmath141 is invariant under @xmath140 this then yields the occurrence of many more cubes . this can be used to exclude eigenvalues for schrdinger operators via the so - called gordon argument . this is discussed in @xcite for the case at hand . such an application of the gordon arguments for subshifts coming from substitution goes back to @xcite , see @xcite for a survey as well . here , we turn next to showing that there are no fourth powers occurring in @xmath141 . let us recall from the considerations on @xmath146-partitions in section [ section - n - partition ] that there exists a sequence @xmath377 such that the fixed point @xmath171 of @xmath378 can be written as @xmath217 with @xmath379 for any @xmath380 . this way of writing @xmath171 is referred to as the @xmath146-decomposition of @xmath171 . call the sequence @xmath381 the _ @xmath146-th derived sequence of @xmath171_. note that for any natural number @xmath146 the combinatorial properties of the sequence @xmath382 are exactly the same as the combinatorial properties of the sequence @xmath383 as @xmath173 is injective on @xmath384 and @xmath385 holds . @xcite[isolation ] in the derived sequence @xmath386 the letters @xmath153 and @xmath155 always occur isolated preceded and followed by an @xmath151 . the letter @xmath151 always occurs either isolated ( i.e. preceded and followed by elements of @xmath387 ) or in the form @xmath388 . in particular , there is no occurrence of @xmath389 . the analogue statements hold for any natural number @xmath146 for the sequence @xmath382 ( with @xmath185 replaced by @xmath390 and @xmath391 ) . in terms of information the derived sequences @xmath382 are as useful as the original sequence . we will base our subsequent investigations on the relatively simple properties of the derived sequence @xmath383 given in the preceding proposition . more information should be obtainable from a more detailed study of the derived sequences . the @xmath146-decomposition of @xmath171 gives a way of writing @xmath171 as a concatenation of the words @xmath180 and elements from @xmath384 . for example @xmath171 can be written as @xmath392 where we have put brackets around the @xmath393 . still , @xmath171 can contain further occurrences of @xmath394 as indicated in the preceding formula . more generally , it is not true that a @xmath180 occurring somewhere in @xmath171 is in fact one of the words @xmath180 appearing in the @xmath146-decomposition of @xmath171 . however , it turns out that whenever @xmath206 occurs in @xmath171 then both of its @xmath180 actually stem from the @xmath146-partition . in this sense , there is some form of alignment . this is the content of the next proposition @xcite . [ prop - alignment ] consider a natural number @xmath146 and @xmath395 . if @xmath396 occurs in @xmath171 at the position @xmath397 ( i.e. @xmath398 holds ) , then @xmath397 is of the form @xmath399 for some @xmath400 . this means that if @xmath401 occurs somewhere in @xmath171 then both of its words @xmath180 actually agree with blocks @xmath180 appearing in the @xmath146-decomposition @xmath402 if @xmath403 is a finite word in @xmath141 and @xmath35 is a prefix of @xmath403 and @xmath126 is a natural number we define the _ index of the word @xmath403 in @xmath404 _ by @xmath405 and denote it by @xmath406 . we then define the _ index of the word @xmath403 _ by @xmath407 as our subshift is minimal and aperiodic the index of every word can easily be seen to be finite . \(a ) for every @xmath408 the inequality @xmath409 holds . in particular , @xmath171 does not contain a fourth power . \(b ) we have @xmath410 _ remark . _ the supremum over all the indices is sometimes known as the critical exponent . a proof can be found in @xcite . here , we only sketch the idea . by lemma [ one - three - plus - block ] the word @xmath411 ( with @xmath412 ) belongs to @xmath141 . for each @xmath263 we then have @xmath413 and we infer that the index must be at least @xmath3 . thus , it suffices to show that @xmath171 does not contain a fourth power . to do so , it suffices to consider occurrences of powers of words @xmath403 in @xmath171 . for short words the statement can easily be checked . consider now the case @xmath414 for some @xmath262 . assume that @xmath415 occurs in @xmath141 . then , the proposition on alignment gives that the length of @xmath403 is given by @xmath416 . this then easily implies the desired statement . the proof of the theorem shows that the length of any word @xmath408 whose cube @xmath415 also belongs to @xmath141 is given by @xmath291 for some @xmath263 . in this section we present a result on the word complexity of @xmath141 . a detailed proof can be found in @xcite . we define the _ word complexity _ of the subshift @xmath142 as @xmath417 recall that a word @xmath418 is called _ right special _ if the set of its extensions @xmath419 has more than one element . \(a ) for any @xmath420 and @xmath421 with @xmath422 we have @xmath423 \(b ) the complexity function @xmath424 satisfies @xmath425 and then for any @xmath420 and @xmath421 with @xmath422 @xmath426 \(c ) consider @xmath420 and @xmath421 with @xmath422 . * if @xmath427 , then there exist exactly two right special words of length @xmath428 . these are given by the suffix of @xmath180 of length @xmath428 ( which can be extended by @xmath185 ) and the suffix of @xmath429 of length @xmath428 ( which can be extended by @xmath430 and by @xmath431 ) . * if @xmath432 , then there exists exactly one right special words of length @xmath428 . this is given by the suffix of @xmath180 of length @xmath428 ( which can be extended by @xmath185 ) . the proof relies on a detailed investigation of the @xmath146-partition of @xmath171 . this allows one to directly determine all words of length @xmath433 in @xmath141 and this gives @xmath434 for all @xmath250 . at the same time the study of the @xmath146-partition allows one to obtain a lower bound on the difference @xmath435 for all @xmath436 . this in turn gives a lower bound on @xmath424 . combining the lower bound and the precise values we obtain the statements of the theorem . as @xmath141 is closed under reflections by proposition [ prop - palindrome ] the above statements about right special words easily translate on corresponding statements about left special words . ( here , a word @xmath418 is called _ left special _ if the set @xmath437 has more than one element . ) this shows in particular that the words @xmath180 , @xmath250 , are both right special and left special ( and are the only words with this property ) . in this section we present an automaton that generates the fixed point @xmath171 of the substitution @xmath140 . this is well in line with general theory on how to exhibit fixed points of substitutions by automata , see e.g. the monograph @xcite to which we also refer for background on automata . there are numerous applications of automatic sequences in group theory . for a recent example and a list of further reference we refer the reader to @xcite . ] consider the automaton @xmath5 from figure [ automaton ] . it is an automaton over the alphabet @xmath438 with four states @xmath439 , which are labeled by @xmath440 respectively . then , the infinite sequence @xmath441 generated by the automaton with initial state @xmath442 is defined as follows : write @xmath250 in its binary expansion as @xmath443 with @xmath444 and @xmath445 , @xmath446 . consider now the path @xmath447 in the automaton starting in @xmath442 and following the sequence @xmath448 . then , @xmath449 is defined to be the label of the state where this path ends . the fixed point @xmath171 of @xmath140 agrees with @xmath450 ( where the fixed point is considered as a map from @xmath451 to @xmath452 ) . this theorem is an immediate consequence of the next proposition . to state the proposition we will need some further pieces of notation . for each @xmath250 and each state @xmath453 of the automaton we define @xmath454 to be the word over @xmath452 of length @xmath455 obtained in the following way : let @xmath456 be the list of all words of length @xmath146 over @xmath438 in lexicographic order ( where @xmath457 ) . consider now for each @xmath458 the path in the automaton starting at @xmath453 and following the word @xmath459 . then , the @xmath75-th letter of @xmath460 is defined to be the label of the state where this path ends . define the letters @xmath182 and recall the recursion formula @xmath461 with @xmath462 . for each natural number @xmath146 and each @xmath463 we have @xmath464 this is proven by induction . the case @xmath465 follows by inspection . assume now that the statement is true for some @xmath466 and consider @xmath467 . then the lexicographic ordering of the words of length @xmath468 over @xmath438 is given by @xmath469 where @xmath470 is the lexicographic ordering of the words of length @xmath467 over @xmath438 . thus , from the rules of the automaton we obtain for each @xmath463 @xmath471 where we set @xmath472 . from our assumption for @xmath146 and the recursion we then find @xmath473 for each @xmath474 and this is the desired statement . the substitution @xmath140 arises naturally in the study of grigorchuk groups @xmath147 and its schreier graphs ( see below ) . from the point of view of subshifts it has the ( slight ) disadvantage of not being primitive . it turns out that it is possible to find a primitive substitution @xmath475 with the same fixed point - and hence the same subshift - as @xmath140 . this can then be used to obtain alternative proofs for the ( proven above ) linear repetitivity and pure discreteness of the spectrum . this is discussed at the end of this section . the material presented here was pointed out to us by fabien durand @xcite . consider the substitution @xmath476 on the alphabet @xmath452 with @xmath477 as any letter of the alphabet is contained in @xmath478 for any letter @xmath479 , this is a primitive substitution . to relate it to @xmath140 we use ( again ) for @xmath250 the letters @xmath480 as well as the recursion formula @xmath481 with @xmath462 . for any natural number @xmath146 the equality @xmath482 holds . in particular , the fixed point @xmath171 of @xmath140 agrees with the fixed point of @xmath476 . this is shown by induction . the cases @xmath483 are easily checked by direct inspection . assume now that the statement is true for all integers up to some @xmath484 . then , we can compute @xmath485 by @xmath486 we then find @xmath487 from our assumption on @xmath146 we then infer @xmath488 successive application of the recursion formula and the fact that @xmath489 then give @xmath490 this is the desired statement . the subshift @xmath491 generated by the primitive substitution @xmath476 agrees with the subshift @xmath142 . as it is well known , see e.g. @xcite , that subshifts associated to primitive substitutions are linearly repetitive , an immediate consequence of the previous corollary is that @xmath142 is linearly repetitive . also , as @xmath476 has constant length ( i.e. the length of @xmath492 is the same for any letter @xmath493 ) and @xmath494 starts with @xmath162 for any letter @xmath493 we can apply a result of dekking @xcite to obtain purely discrete spectrum . in this section we recall some basic notions and concepts from the theory of graphs , dynamical systems , and schreier graphs . in the next section we will meet all these abstract concepts in the context of a particular group . here we first recall some terminology from the theory of graphs and introduce the topological space of ( isomorphism classes of ) rooted labeled graphs . let @xmath495 be a finite non - empty set together with an involution @xmath496 . graph with edges labeled by @xmath495 _ is a pair @xmath497 consisting of a set @xmath498 and a set @xmath499 such that @xmath500 belongs to @xmath501 if @xmath502 belongs to @xmath501 . the elements of @xmath498 are called vertices and the elements of @xmath501 are called edges . whenever @xmath503 is an edge , then @xmath154 is called the _ label _ , @xmath504 the _ origin _ and @xmath505 the _ terminal vertex _ of the edge . we say that _ there is an edge from the vertex @xmath35 the the vertex @xmath403 with label @xmath154 _ if @xmath500 belongs to @xmath501 . an edge @xmath506 is said to _ emanate _ from the vertex @xmath35 if @xmath507 . the number of edges emanating from a vertex is called the _ degree _ of the vertex . an edge of the form @xmath508 is called a _ loop _ at @xmath35 ( with label @xmath154 ) . we will a need the _ combinatorial distance _ on a graph given as follows . each vertex has distance @xmath509 to itself . the distance between different vertices @xmath35 and @xmath403 is one if and only if there exists a label @xmath154 such that @xmath500 belongs to @xmath501 . more generally the distance between different vertices @xmath35 and @xmath403 is then defined inductively as the smallest natural number @xmath146 such that there exists a vertex @xmath510 with distance @xmath511 to @xmath35 and distance one to @xmath403 . if no such @xmath146 exists the combinatorial distance is defined to be @xmath166 . the graph is called _ connected _ if the distance between any two if its vertices is finite . likewise the _ connected component _ of a vertex is the set of all vertices with finite distance to it . a _ ray _ in an infinite graph is an infinite sequence @xmath512 of pairwise different vertices with distance one between consecutive vertices . two rays are equivalent if there exists a third ray containing infinitely many vertices of each of the rays . an equivalence class of rays is called an _ end _ of the graph . for example finite graphs have @xmath509 ends , the cayley graphs of @xmath11 , @xmath513 and @xmath514 ( the free group on two letters ) have @xmath515 and infinitely many ends , respectively . in general the cayley graph of a group may have @xmath516 or infinitely many ends . a _ rooted graph _ is a pair consisting of a graph and a vertex belonging to the vertex set of the graph . this vertex is then called the _ root_. two rooted graphs @xmath517 and @xmath518 labeled by the same set @xmath495 are called isomorphic if there exists a bijective map @xmath519 from the vertices of @xmath520 to the vertices of @xmath521 taking @xmath522 to @xmath523 such that the vertices @xmath151 and @xmath153 in @xmath520 are connected by an edge of color @xmath154 if and only if their images in @xmath524 are connected by an edge of color @xmath154 . in this case we write @xmath525 . let us now consider the set @xmath526 of isomorphism classes of connected rooted graphs labeled with elements from @xmath495 that we endow with the following natural metric . the distance between the isomorphism classes of the two rooted graphs @xmath527 and @xmath528 is then defined as @xmath529,[(y_2,v_2)]):=\inf\left\{\frac{1}{r+1 } : b_{y_1}(v_1,r ) \cong b_{y_2}(v_2,r ) \right\}\ ] ] where @xmath530 is the ( labeled ) ball of radius @xmath531 centered in @xmath35 in the combinatorial metric on @xmath266 . if we only consider graphs of uniformly bounded degree ( as we will in this paper ) , the space @xmath532 is compact . we now turn to dynamical systems . whenever the group @xmath533 acts on the compact space @xmath266 via the continuous map @xmath534 we call @xmath535 a _ dynamical system_. we will mostly suppress the @xmath536 in the notation . in particular , we will write @xmath537 instead of @xmath538 and we will write @xmath539 for @xmath540 ( with @xmath272 and @xmath541 ) . if @xmath533 is the infinite cyclic group @xmath11 , then the action of @xmath533 on @xmath266 is determined by @xmath542 and we then just write @xmath543 instead of @xmath544 ( and this is well in line with the notation used in the previous sections ) . whenever a dynamical system @xmath537 and @xmath272 is given we call @xmath545 the _ orbit of @xmath153 ( under h)_. the dynamical system @xmath537 is _ minimal _ if every orbit is dense in @xmath266 . the dynamical system @xmath537 is called _ uniquely ergodic _ if there exists exactly one @xmath533-invariant probability measure on @xmath266 . the dynamical system @xmath537 is called a _ factor _ of the dynamical system @xmath546 if there exist a continuous surjective map @xmath547 with @xmath548 for all @xmath541 and @xmath549 . this map @xmath550 is then called _ factor map_. the dynamical system @xmath537 is then also referred to as _ extension _ of the dynamical system @xmath546 . we will deal with graphs arising from dynamical systems . more specifically , consider a group @xmath533 generated by a symmetric finite set @xmath43 and assume that @xmath533 acts on the compact space @xmath266 . then any point @xmath272 gives rise to the _ orbital schreier graph of @xmath153 _ denoted by @xmath551 . this is a rooted graph labeled by @xmath43 , which is equipped with the involution @xmath552 . the set of vertices of @xmath551 is given by the points in the orbit of @xmath153 . the root is given by @xmath153 and rhere is an edge from @xmath35 to @xmath403 with label @xmath553 if @xmath554 . note that by the required symmetry of @xmath43 we then have also an edge from @xmath403 to @xmath35 with label @xmath555 ( as is needed according to our definition of a labeled graph ) . as an example we note the cayley graph of a group @xmath147 with generating set @xmath43 . this cayley graph ( with the neutral element of @xmath147 as the root ) is the orbital schreier graph of @xmath147 corresponding to the action of @xmath147 on itself via left multiplication . if the elements of @xmath43 happen to all be involutions then there will be an edge of label @xmath479 from @xmath35 to @xmath403 if and only if there is an edge of label @xmath479 from @xmath403 to @xmath35 . this is the situation we will encounter in the next section . in this section we introduce the main object of our interest : the first group of intermediate growth @xmath147 and the laplacians on the associated schreier graphs . the group @xmath147 is the first group with intermediate word growth and was introduced by the first author in @xcite . by now it is generally known as grigorchuk s group @xmath147 and this is how we will refer to it . it is generated by four involutions @xmath556 . with notation to be introduced presently , the group @xmath147 can be viewed as a group acting by automorphisms on the full infinite binary tree @xmath557 . the action extends by continuity to an action by homeomorphisms on the boundary @xmath558 . the action of @xmath147 gives rise to schreier graphs the schreier graphs arising from actions of automorphism groups of infinite regular rooted trees have attracted substantial attention in recent years @xcite . of particular interest are so - called self - similar groups ( as @xmath147 ) whose action reflect the self - similar structure of the tree . their schreier graphs also have self - similarity features . some are closely related to julia sets @xcite , others are fractal sets close to e.g. the sierpinski gasket or the apollonian gasket @xcite . the spectra of the laplacians on such graphs have been described in some cases @xcite . the spectrum can be a union of intervals @xcite , a cantor set @xcite , or a union of a cantor set together with an infinite set of isolated points that accumulate to it , as in the case of the so - called hanoi tower group @xcite . these investigations all use the method introduced in @xcite . here we study the spectra of the laplacians on the schreier graphs by a different method via the connection to aperiodic order . this allows us to determine the spectral type of the laplacians for arbitrary choice of weights on the generators of the group . we thus obtain one of the rare examples where we understand how the spectrum of the laplacian depends on the weights . in general this dependence is very poorly understood , as well as the dependence of the spectrum on the generating set in the group . let us mention here a recent result of grabowski and virag who show in @xcite that there is a weighted laplacian on the lamplighter group that has singular continous spectrum , whereas the unweighted laplacian on the same generators is known to have pure point spectrum @xcite . let us denote by @xmath559 , @xmath560 with @xmath561 , the _ rooted regular tree of degree _ @xmath453 . the vertex set of @xmath559 is given by @xmath562 , i.e. the set of all words over the alphabet @xmath563 . the root of @xmath559 is the empty word . there is an edge between @xmath35 and @xmath403 whenever @xmath564 or @xmath565 holds for some @xmath566 . the words @xmath567 constitute the _ @xmath146-the level _ of the tree . ( in the tree , they are at combinatorial distance exactly @xmath146 from the root . ) the boundary @xmath568 of @xmath559 consisting of infinite geodesic rays in @xmath559 emanating from the root ( i.e. infinite paths starting in the root all of whose edges are pairwise different ) can then be identified with the set @xmath569 of one - sided infinite words over @xmath570 . as mentioned above , the set @xmath571 is equipped with the product topology and is thus a compact space homeomorphic to the cantor set . any automorphism of @xmath559 necessarily preserves the root ( which is the only vertex with degree @xmath453 ) and maps paths starting in the root to paths starting in the root . this readily implies that any automorphisms group action on @xmath559 is level preserving , i.e. maps words of length @xmath146 to words of length @xmath146 . any such action then extends to an action of the same group by homeomorphisms on the boundary @xmath568 . a regular rooted tree is a self - similar object . indeed , the subtree rooted at an arbitrary vertex of the tree is isomorphic to the whole tree @xmath559 . the full group of automorphisms inherits this self - similarity property in the following sense : any automorphism of @xmath559 is completely determined by the permutation it induces on the @xmath453 branches growing from the root ( an element of @xmath572 ) and the collection of @xmath453 automorphisms @xmath573 which coincide with the restrictions of @xmath574 on the corresponding branches . however , if one is interested in a subgroup @xmath575 and wants it to be self - similar , one has to impose the condition that all the restrictions @xmath573 are again elements of the same group @xmath533 , so that every @xmath576 can be represented as @xmath577 where @xmath536 belongs to @xmath578 and describes the action of @xmath574 on the first level of @xmath559 and @xmath579 is the restriction of @xmath574 on the full subtree of @xmath559 rooted at the vertex @xmath580 of the first level of @xmath559 . this leads to the following definition . [ defiselfsimilar ] a group @xmath533 of automorphisms of @xmath559 is _ self - similar _ if , for all @xmath581 , there exist @xmath582 such that @xmath583 for all finite words @xmath403 over the alphabet @xmath563 . we refer the interested reader to @xcite for more information about self - similar groups . we now turn our attention to one particular example of a self - similar group that will be the central object of our study , the grigorchuk group @xmath147 . it is generated by four automorphisms @xmath556 of the rooted binary tree @xmath584 as follows : * @xmath585 , @xmath586 * @xmath587 , @xmath588 * @xmath589 , @xmath590 * @xmath591 , @xmath592 for an arbitrary word @xmath403 over @xmath438 . these automorphisms can also be expressed in the self - similar form , as above : @xmath593 where @xmath506 and @xmath594 are , respectively , the trivial and the non - trivial permutations in the group @xmath595 . observe that all the generators are involutions and that @xmath596 commute and constitute a group isomorphic to the klein group @xmath597 . let as also mention that there are many more relations and the group is not finitely presented . for our subsequent discussion it will be important that @xmath147 acts _ transitively _ on each level , i.e. for arbitrary words @xmath598 over @xmath438 with the same length there exists a @xmath599 with @xmath600 . the action of @xmath147 on the set @xmath602 of vertices of the rooted binary tree and on its boundary @xmath603 induces on this set the structure of a graph labeled with @xmath604 . specifically , the vertex set of this graph is given by @xmath605 and there is an edge with label @xmath606 and origin @xmath35 and terminal vertex @xmath403 if and only if @xmath607 holds . note that the set of arising edges has indeed the symmetry property required in our definition of a labeled graph as any @xmath608 is an involution . for the first three levels of the tree the resulting graphs are shown in figure [ schreier - finite ] . the connected components of this graph correspond to the orbits of the group action . thus , the rooted graphs consisting of such a connected component together with a root are orbital schreier graphs ( in the sense defined above ) . there are two kind of such connected components viz finite and infinite components . the finite ones correspond to the levels of the tree ( recall that the action of @xmath147 on the levels of the tree is transitive ) . we will refer to them simply as _ schreier graphs_. the infinite ones correspond to the orbits of the action on the boundary . the corresponding rooted graphs will be referred to as _ orbital schreier graphs_. we will need the isomorphism classes of schreier graphs and introduce therefore the map @xmath609,\ ] ] where @xmath610 is the connected component of @xmath35 . the graphs @xmath611 and @xmath610 coincide ( as non - rooted graphs ) whenever @xmath35 and @xmath403 are in the same orbit of the action of @xmath147 . in particular , as @xmath147 acts transitively on each level of the tree , for @xmath263 we can therefore define @xmath612 which coincides with @xmath611 for all @xmath613 with @xmath614 . in general , the graph @xmath615 has a linear shape ; it has @xmath616 simple edges , all labeled by @xmath162 , and @xmath617 cycles of length 2 whose edges are labeled by @xmath618 . it is regular of degree 4 , with one loop at each edge . the loop contributes 1 to the degree of the vertex because all generators are elements of order 2 , and the labeling of the loop is uniquely determined by the labeling of the other edges around the vertex , as edges around one vertex are labeled by @xmath150 . the graphs @xmath619 , @xmath620 corresponding to the action of @xmath147 on the boundary are infinite and have either two ends or one end . the graph @xmath621 corresponding to the orbit of the rightmost infinite ray , is one - ended ( see figure [ one - ended ] ) , and so are then all graphs in the same orbit . . ] all the other graphs @xmath622 , are two - ended . they are all isomorphic as unlabeled graphs . in @xcite , vorobets studied the closure @xmath623 of the space of schreier graphs in the space of isomorphism classes of rooted labeled graphs . we recall some of his results next . he showed that the one - ended graphs are exactly the isolated points of this closure @xmath623 , and that the other points in @xmath624 are two ended graphs . this suggests to consider the compact set @xmath625 the group @xmath147 acts on @xmath626 by changing the root of the graph . the arising dynamical system is denoted as @xmath601 . it is minimal and uniquely ergodic and its invariant probability measure will be denoted as @xmath627 . a precise description of @xmath626 can be given as follows . the space @xmath626 consists of all isomorphism classes of two - ended rooted schreier graphs @xmath628 and of three additional countable families of isomorphism classes of two - ended graphs . these families are obtained by gluing two copies of the one - ended graph @xmath629 , at the root in three possible ways corresponding to choosing a pair @xmath630 , @xmath631 or @xmath632 and then choosing an arbitrary vertex of the arising graph as the root . one of these three possibilities is shown in figure [ connecting - two - copies ] . there , the chosen pair is @xmath632 and to avoid confusion with other edges with the same labels , labels at the gluing point are denoted with a prime . ] the decomposition of @xmath626 into isomorphism classes of the @xmath633 and the three families mentioned above gives immediately rise to a factor map @xmath634 which is one - to - one except in a countable set of points , where it is three - to - one . in fact , the inverse map @xmath635 exists on the complement @xmath636 of the orbit of the point @xmath637 and agrees there with @xmath638 . under this factor map @xmath639 the unique @xmath147-invariant probability measure @xmath627 on @xmath626 is mapped to the uniform bernoulli measure @xmath133 on @xmath640 . whenever a group @xmath533 acts on a measure space @xmath641 by measure preserving transformations , one obtains the _ koopman representation _ @xmath642 of @xmath533 on @xmath643 via @xmath644 ( for @xmath645 ) . any @xmath646 is then a unitary operator ( as the action is measure preserving ) . in our situation when @xmath647 , we have moreover that for @xmath608 the unitary operator @xmath648 is its own inverse ( as @xmath479 is an involution ) and hence must be selfadjoint . in particular , for any set of parameters @xmath649 we obtain a selfadjoint operator @xmath650 consider first an arbitrary @xmath620 . then , there is an action of @xmath147 on the ( countable ) vertex set of @xmath651 of @xmath619 . specifically , @xmath652 maps the vertex @xmath653 to the vertex @xmath654 , which is the unique vertex of @xmath655 connected to @xmath35 by an edge of label @xmath479 . clearly , this action preserves the counting measure on @xmath651 . thus , we obtain a representation @xmath656 of @xmath147 on @xmath657 . an operator @xmath658 defined by @xmath659 with @xmath620 and @xmath649 will be called _ ( weighted ) laplacian of the schreier graph _ @xmath619 . * it is possible to understand @xmath651 as @xmath660 , where @xmath661 is the stabilizer of @xmath475 in the action of @xmath147 on @xmath558 , and then @xmath656 is the quasi - regular representation @xmath662 associated to @xmath660 . * if @xmath663 are positive with @xmath664 then it is possible to interpret the operators @xmath665 as the markov operators of a random walk on the graph @xmath619 . in the general case , the operator @xmath666 can still be seen as the natural weighted adjacency matrix or laplacian associated to the the graph @xmath619 . we can also equip @xmath667 with the uniform bernoulli measure @xmath133 and consider the koopman representation @xmath668 of @xmath147 on @xmath669 given via @xmath670 this is a unitary representation of @xmath147 and any @xmath671 , @xmath672 , is a unitary selfadjoint involution . for @xmath673 we then obtain the operator @xmath674 via @xmath675 the following is a crucial result on the spectral theory of the above operators . [ constancy - bg ] for any given set of parameters @xmath649 the spectrum of @xmath666 does not depend on @xmath676 and coincides with the spectrum of @xmath677 . of course , there is the question what the spectrum is in terms of the parameters @xmath663 . a complete answer was given in @xcite in the case @xmath678 . the spectrum then consists of two points or one or two intervals , and an explicit description of the spectrum can be given in terms of the parameters@xmath679 and @xmath680 . in fact , the case @xmath680 is the case of periodic schrdinger type operators and can easily be treated by classical means ( floquet decomposition ) . it can also be treated by the method suggested in @xcite . for related results on the kesten - von - neummann - serre trace we refer to @xcite . below we will come to a main result of @xcite which treats the case , where @xmath680 does not hold . we will see that in this case , the spectrum is a cantor set of lebesgue measure zero . we will now provide a map @xmath681 from @xmath186 to ( isomorphism classes of ) labeled rooted graphs with labels belonging to the alphabet @xmath150 . in order to get a better understanding of this map it will be useful to think of the letters @xmath185 as encoding the pairs @xmath682 respectively . roughly speaking the map will convert @xmath224 into a graph with vertex set @xmath11 and root @xmath4 and edges between @xmath146 and @xmath467 with labels from @xmath150 according to the value of @xmath683 . we now provide the precise * definition of the labeled rooted graph @xmath684 * associated to @xmath239 as follows : _ vertices : _ the set of vertices is @xmath11 . _ root : _ the number @xmath4 is chosen as the root . _ edges : _ there are edges between vertices @xmath685 if an only if @xmath686 . specifically , edges are assigned between @xmath146 and @xmath467 and from @xmath146 to itself and from @xmath467 to itself in the following way : * if @xmath687 , then there is an edge between @xmath146 and @xmath467 and this edge carries the label @xmath162 . * if @xmath688 , then there are two edges between @xmath146 and @xmath467 ; one carries the label @xmath154 and the other carries the label @xmath152 . moreover , there is an additional edge from @xmath146 to itself labeled with @xmath156 and an additional edge from @xmath467 to itself labeled with @xmath156 . * if @xmath689 , then there are two edges between @xmath146 and @xmath467 ; one carries the label @xmath154 and the other carries the label @xmath156 . moreover , there is an additional edge from @xmath146 to itself labeled with @xmath152 and an additional edge from @xmath467 to itself labeled with @xmath152 . * if @xmath690 , then there are two edges between @xmath146 and @xmath467 ; one carries the label @xmath152 and the other carries the label @xmath156 . moreover , there is an additional edge from @xmath146 to itself labeled with @xmath154 and an additional edge from @xmath467 to itself labeled with @xmath154 . we note that in this way any vertex has for each label @xmath150 exactly one edge of this color emanating from it . we also note that the arising graphs have a linear structure in a natural sense . let @xmath691 be the metric space of isomorphism classes of connected rooted graphs labeled with elements from @xmath150 . then , @xmath681 gives rise to a map @xmath692 via @xmath693,\ ] ] where @xmath694 $ ] is the equivalence class of @xmath695 in the space of isomorphism classes of the rooted connected labeled graphs . recall that @xmath626 denotes the closure of @xmath696 in @xmath697 without its isolated points ( see section [ section - schreier ] ) . then , it turns out that the image of @xmath186 under @xmath692 is exactly @xmath626 . in fact , much more is true and this is our main result on the connection of the subshift @xmath142 and the dynamical system @xmath601 . define the maps @xmath698 from @xmath186 into itself by * @xmath699 if @xmath700 and @xmath701 if @xmath702 . * @xmath703 if @xmath704 , @xmath705 if @xmath706 and @xmath707 in all other cases . * @xmath708 if @xmath709 , @xmath710 if @xmath711 and @xmath712 in all other cases . * @xmath713 if @xmath714 , @xmath715 if @xmath716 and @xmath717 in all other cases . then , clearly , @xmath698 are homeomorphisms and involutions . let @xmath533 be the group generated by @xmath698 ( within the group of homeomorphisms of @xmath186 ) . [ main - connection ] the following statements hold : \(a ) the group @xmath147 is isomorphic to the group @xmath533 via @xmath718 with @xmath719 , @xmath720 , @xmath721 and @xmath722 . in particular , there is a well defined action @xmath536 of @xmath147 on @xmath186 given by @xmath723 for @xmath599 and @xmath724 and via this action we obtain a dynamical system @xmath725 . \(b ) the dynamical system @xmath601 is a factor of the dynamical system @xmath725 with factor map @xmath726 which is two - to - one . \(c ) for every @xmath177 the orbits @xmath727 and @xmath728 coincide . \(d ) the dynamical system @xmath725 is uniquely ergodic and the unique @xmath44-invariant probability measure on @xmath186 coincides with the unique @xmath147-invariant probability measure on @xmath186 . the theorem provides a factor map @xmath729 . in section [ section - schreier ] we have already encountered the factor map @xmath730 . putting together these factor maps provide a tower of extensions of the dynamical system @xmath731 via @xmath732 the remainder of this section is devoted to providing additional perspective on these two factor maps by discussing their respective merits . we will see that @xmath639 resolves the non - typical points and that @xmath733 allows one to embed the group @xmath147 into a topological full group . we will need a bit of notation first . let @xmath537 be a dynamical system . let @xmath576 be given . then , an @xmath272 is called _ @xmath574-typical _ if either @xmath734 or @xmath574 acts trivially in some neighborhood of @xmath153 . if @xmath272 is @xmath574-typical for any @xmath735 it is called _ typical_. typical points have many claims to be indeed typical . for instance , if @xmath533 is countable the set of typical points has a meager complement @xcite . moreover , the following is shown in @xcite . let @xmath537 be a minimal dynamical system and @xmath533 finitely generated . then , the schreier graphs of typical points are locally isomorphic ( as labeled graphs ) . moreover , if @xmath736 is typical and @xmath272 is not typical , then any ball around a vertex of @xmath737 is isomorphic to a ball around some vertex in @xmath551 . in order to get a further understanding of the typical points we will next discuss another characterization of typical points . define the _ stabilizer _ of @xmath153 as @xmath738 and the _ neighborhood stabilizer _ of @xmath153 as @xmath739 then , @xmath740 is a normal subgroup of @xmath741 and the quotient @xmath742 is called the _ group of germs _ ( at @xmath153 ) . let @xmath537 be a dynamical system . then , the following assertions are equivalent for @xmath272 . * the point @xmath153 is typical . * the group of germs @xmath743 is trivial . ( i)@xmath744 ( ii ) : if @xmath745 is not trivial , there exits a @xmath746 . hence , @xmath153 is not typical . ( ii)@xmath744 ( i ) : if @xmath153 is not typical then there exists a @xmath599 with @xmath747 but @xmath574 is not the identity on a neighborhood of @xmath153 . thus , @xmath743 is not trivial ( as it contains the class of @xmath574 ) . the points @xmath153 with non - trivial @xmath745 are sometimes called _ singular_. we will also need the concept , going back to @xcite , of the ( topological ) _ full group _ @xmath748 $ ] of a homeomorphism @xmath749 of the cantor set @xmath266 . this is the group of those homeomorphisms of @xmath266 that at any point coincide locally with powers of t. it is a countable group with remarkable properties . for example it is amenable if @xmath44 is minimal @xcite , its commutator subgroup is simple @xcite and it is finitely generated if and only if @xmath44 is a minimal subshift over a finite alphabet @xcite . after these preparations we can now came back to the situation at hand . the dynamical system @xmath731 is minimal and uniquely ergodic . however , it does have non - typical points . in fact , a point in @xmath750 is non - typical if and only if it belongs to @xmath751 ( see @xcite ) . the dynamical system @xmath601 is minimal and uniquely ergodic ( as discussed in @xcite ) . moreover , all its points are typical ( as can easily be seen ) . thus , from this point of view a key merit of @xmath601 is to provide an extension of of @xmath731 which resolves the non - typical points . the dynamical system @xmath725 is minimal and uniquely ergodic and all its points are typical ( as can easily be seen ) . moreover , as shown in the previous theorem , the dynamical system @xmath725 has the additional feature that there exists a homeomorphism @xmath44 on @xmath186 such that the orbits of @xmath44 agree with the orbits of @xmath147 . this in turn can be seen to imply that @xmath147 is a subgroup of the topological full group @xmath752 $ ] of @xmath44 . in fact , @xmath147 embeds into the topological full group of @xmath753 , as the action of generators @xmath556 on @xmath754 can be represented locally as the action by @xmath755 and @xmath756 , so @xmath147 embeds into @xmath748 $ ] . [ connection - bon ] in this context we also mention a recent article of matte bon @xcite showing that the group @xmath147 ( and other groups of intermediate growth introduced by the first author in @xcite ) embed into the topological full group @xmath757 $ ] of a minimal subshift @xmath639 over a finite alphabet . while his approach is different from ours it leads to the the same subshift for the group @xmath147 . we finish this section with the question whether there exists a minimal homeomorphism @xmath43 on @xmath626 such that @xmath147 is a subgroup of @xmath758 $ ] . in this section we discuss spectral theory of the operators @xmath666 ( defined in section [ operators ] ) for arbitrary values of the parameters @xmath663 . we do so by using the results of the previous section to transfer the problem of spectral theory of the operators @xmath666 , @xmath760 , to the field of the spectral theory of discrete schrdinger operators with aperiodic order , @xmath761 . we consider the subshift @xmath142 . in order to define the schrdinger operators we will define specific functions @xmath128 on @xmath186 depending on four real parameters @xmath663 . given these parameters we set @xmath762 and define @xmath763 and @xmath764 for a given @xmath765 and these @xmath128 we let @xmath48 , @xmath239 , be the associated operators . @xcite[unitary - equivalence ] let @xmath766 be given . let @xmath767 be arbitrary . then , there exists an @xmath1 in @xmath186 such that @xmath48 is unitarily equivalent to @xmath666 . as a consequence of the previous proposition we can translate results on the spectral theory of schrdinger operators associated to @xmath142 into results on the @xmath665 . recall that the spectrum of the @xmath666 does not depend on @xmath768 ( due to theorem [ constancy - bg ] ) . define @xmath769 [ main - laplacian - cantor ] let @xmath770 be given and let @xmath771 be the spectrum of the associated family of laplacians @xmath666 , @xmath772 . then , the following holds : \(a ) if @xmath773 holds then @xmath72 consists of one or two closed non - trivial intervals and all spectral measures are absolutely continuous . \(b ) if @xmath680 does not hold then @xmath72 is a cantor set of lebesgue measure zero and no spectral measure is absolutely continuous . note that @xmath142 is linearly repetitive due to theorem [ theorem - tau - linear - repetitive ] . now , the theorem is a direct consequence of the preceding proposition and theorem [ theorem - periodic - abstract ] and theorem [ theorem - aperiodic - abstract ] . the case @xmath680 has already be treated in @xcite and an explicit description of the spectrum ( in terms of the value of @xmath774 ) has been given there . we can also use the above considerations to translate results on absence of eigenvalues from the @xmath74 to the @xmath665 . recall that @xmath133 denotes the uniform bernoulli measure on @xmath775 . [ main - laplacian - absence - eigenvalues ] let @xmath770 be given and assume that @xmath680 does not hold . \(a ) for @xmath133-almost every @xmath620 the operator @xmath665 does not have eigenvalues . \(b ) for any @xmath776 the operator @xmath777 does not have eigenvalues . the considerations of this section are concerned with the case @xmath778 . for @xmath779 the operators in question can be decomposed as a sum of finitely many different finite dimensional operators each appearing with infinite multiplicity . thus , the spectrum is pure point with finitely many eigenvalues each with infinite multiplicity . in this section we show that the kesten - von - neumann - serre spectral measure actually agrees with the integrated density of states . this result is folklore and certainly known to experts . also , a recent discussion of a somewhat more general setting can also be found in @xcite . here , we give a direct reasoning for the case at hand based on our approach . there is one more representation studied in the context of spectral approximation . this representation comes from the action of @xmath147 on the @xmath146-th level of the tree . this action clearly preserves the counting measure on the finite set of vertices of the @xmath146-th level . it hence gives rise to a representation @xmath780 of @xmath147 on the finite dimensional @xmath781 . for @xmath673 we then set @xmath782 to each such @xmath783 we associate the _ spectral distribution _ which is the measure @xmath784 on @xmath61 given by @xmath785 where the sum runs over eigenvalues @xmath501 of @xmath783 counted with multiplicities . in the case @xmath678 it is shown in @xcite that the measures @xmath786 , @xmath263 , converge weakly and the limiting measure @xmath787 is called the _ kesten - von - neumann - serre spectral measure _ there . next we will show that the limiting measure exists for any values of the parameters @xmath663 and coincides with the so - called _ integrated density of states _ of the associated schrdinger operators . [ ids - equal - kvns - trace ] let @xmath788 be given . for @xmath263 , let @xmath783 be the corresponding operator on @xmath789 and @xmath786 its spectral distribution . then , the sequence of measures @xmath790 converges weakly towards the integrated density of states @xmath75 of @xmath74 . in particular , the integrated density of states of the @xmath74 agrees with the kesten - von - neumann - serre spectral measure . the graph arising from restricting @xmath791 to the vertex set @xmath792 $ ] and the graph @xmath789 differ at most in @xmath793 loops at the ends ( as is clear from the definitions ) . thus , a simple variant of the argument in proposition [ unitary - equivalence ] shows that the restriction of the operator @xmath794 to @xmath792 $ ] is a perturbation of @xmath783 of rank at most @xmath793 . now , corollary [ perturbation - ids ] gives the convergence of the @xmath795 towards the integrated density of states of @xmath74 . * denote the @xmath475-independent spectrum of the operators @xmath796 by @xmath797 and the spectrum of @xmath798 by @xmath799 . clearly , the support of @xmath786 is given by @xmath799 . by theorem [ theorem - support - and - atomfree ] support of the integrated density of states @xmath75 is given by @xmath797 . thus , the previous theorem gives in a certain and very weak sense the convergence of the @xmath799 towards @xmath797 . in particular , there is an inclusion of spectra as shown in corollary [ cor - inclusion ] . + in the case at hand , however , convergence of the @xmath800 towards @xmath797 holds in a much stronger sense . more precisely , the results of @xcite give the following : * * @xmath801 for all @xmath263 . * * @xmath802 . + these are rather remarkable features and not at all true for approximation of the integrated density of states of schrdinger operators in other cases . + let us note , however , that there are recent results on approximation of spectra of minimal subshifts with respect to the hausdorff distance by suitable periodic approximations @xcite . * it is also worth pointing out that the approximation of the spectra is done from below i.e. via unions . this is in contrast to other approximation schemes used in the investigations of schrdinger operators , where the approximation is done from above i.e. via intersections . * the feature presented in the preceding two points also explain that the spectrum of @xmath677 agrees with @xmath803 . the reason is simply that the representation @xmath668 decomposes as a sum of the representations @xmath260 ( as shown in @xcite ) . accordingly , @xmath677 is a sum of the finite dimensional operators @xmath783 , @xmath243 , and its spectrum is then given as the closure of the union of the spectra of the @xmath798 . l. bartholdi and r. i. grigorchuk : _ on the spectrum of hecke type operators related to some fractal groups _ , steklova * 231 * ( 2000 ) , din . gruppy , 545 ; translation in _ proc . steklov inst . _ ( 2000 ) , no . 4 ( * 231 * ) , 141 . l. bartholdi , r. i. grigorchuk and v. nekrashevych : _ from fractal groups to fractal sets _ in : fractals in graz ( p. grabner and w. woess editors ) , trends in mathematics , birkuser verlag , basel , 2003 , 25118 . r. i. grigorchuk : _ some problems of the dynamics of group actions on rooted trees ( russian ) _ , tr . mat . steklova * 273 * ( 2011 ) , sovremennye problemy matematiki , 72191 ; translation in proc . * 273 * ( 2011 ) , 64175 . r. i. grigorchuk , ya . s. krylyuk : _ the spectral measure of the markov operator related to 3-generated 2-group of intermediate growth and its jacobi parameters _ , algebra discrete math . * 13 * ( 2012 ) , 237272 .

index coding , introduced in @xcite , considers the problem of efficiently broadcasting a number of messages available at a source , to receivers that already possess some prior knowledge of the messages . in the index coding framework , the source is allowed to encode the messages ( and thereby use the channel efficiently ) while satisfying the receiver demands . the general class of _ groupcast _ index coding problems consists of @xmath0 messages generated at a source , where each message is demanded by at least one receiver . index coding problems where each receiver demands a unique message are called _ single unicast index coding _ problems and are the most widely studied class . although index coding continues to be open in general , several researchers have made inroads into characterising the _ _ rate of index coding _ _ and presenting achievable schemes . the landmark paper @xcite famously connected the scalar linear index coding problem to finding a quantity called _ minrank _ associated with the _ side - information graph _ related to the given single unicast index coding problem . upper and lower bounds on the rate for single unicast index coding have been presented via graph theoretic ideas like clique cover , chromatic number @xcite , local chromatic number @xcite , fractional clique covering and hyperclique covering @xcite , and recently , the ` generalized interlinked cycle cover ' @xcite . many of these papers naturally lead to constructions of ( scalar and vector ) linear index codes . linear codes however are not always found to be optimal @xcite . random coding approaches to index coding were studied in @xcite . bounds on the rate of groupcast index coding were presented in @xcite . interference alignment , well known as a powerful tool to study degrees of freedom in wireless interference networks , was employed to the linear index coding problem in @xcite , by modelling the unavailable side - information as interference . the idea is to assign precoding matrices to the message vectors such that all receivers can decode even in the presence of interference ( thereby requiring some degree of linear independence between the precoding matrices , and hence reducing the rate ) . however , at the same time the interference at the receivers must be as ` aligned ' as possible ( in order to reduce the amount of linear independence required , i.e. to increase the rate ) . this technique was further explored in @xcite and several classes of index coding instances with certain feasible rates were identified based on the properties of the interference seen by the receivers . this work builds primarily upon the results in @xcite . in @xcite , a necessary and sufficient condition for the feasibility of _ _ rate half _ _ ( some finite field ) symbols transmitted through the channel , one @xmath3 symbol of each message is conveyed . ] in a groupcast index coding problem was established based on the properties satisfied by two graphs obtained from the interference structure of the problem , called the _ conflict graph _ and the _ alignment graph_. also , a sufficient ( but not necessary ) condition on the structure of these graphs was given for rate @xmath2 feasibility . prior work @xcite also gives a necessary ( but not sufficient ) condition for rate @xmath2 feasibility based on the interference structure of the problem . the relevant details of the prior work are discussed elaborately in the forthcoming sections of this paper . * firstly , we revise the definition of the conflict graph given in @xcite . the conflict graph definition in @xcite does not capture the interference structure of the index coding problem sufficiently . in order to rectify this , we define the _ conflict hypergraph _ , which is shown to capture the interference structure sufficiently . ( subsection [ conflicthypergraphsubsec ] ) * our main result shows that rate @xmath2 is achievable in a given index coding problem under certain conditions on the topology of the alignment graph and conflict hypergraph of the index coding problem . the sufficient condition which we present for rate @xmath2 feasibility is looser than the sufficient condition shown in @xcite . therefore , this class of index coding problems is bigger than the previously known class of rate @xmath2 feasible problems . the achievability of rate @xmath2 in such problems is shown by presenting a construction of an index code by random generation of precoding vectors over a large field . ( subsection [ rateonethirdfeasibleproblems ] ) * towards obtaining our main result , we also obtain a necessary condition for rate @xmath2 feasibility , which is stricter than the prior condition from @xcite ( subsection [ necessaryconditionrateonethirds ] ) . our feasibility conditions can thus be seen as being midway between those of @xcite and @xcite . _ notations : _ throughout the paper , we use the following notations . let @xmath4 $ ] denote @xmath5 . for a set of vectors @xmath6 , @xmath7 denotes their span . for a vector space @xmath8 , @xmath9 denotes its dimension . an arbitrary finite field is denoted by @xmath3 . a vector from the @xmath10-dimensional vector space @xmath11 is said to be picked _ at random _ if it is selected according to the uniform distribution on @xmath11 . formally , the index coding problem ( over some field @xmath12 ) consists of a broadcast channel which can carry symbols from @xmath3 , along with the following . * a set of @xmath13 receivers * a source which has messages @xmath14\}$ ] , each of which is modelled as a vector over @xmath3 . * for each receiver @xmath15 , a set @xmath16 denoting the set of messages demanded by the receiver @xmath15 . * for each receiver @xmath15 , a set @xmath17 denoting the set of side - information messages available at the @xmath18 receiver . this general class of index coding problems is known as _ groupcast _ index coding problems . an index code of symmetric rate @xmath1 for a given index coding problem consists of an encoding function @xmath19 for some @xmath20 mapping the @xmath0 @xmath21-length message vectors ( @xmath22 ) to some @xmath23-length codeword which is broadcast through the channel , as well as decoding functions @xmath24 at the receivers @xmath25,$ ] mapping the received codeword and the side - information messages to the demanded messages @xmath26 , i.e. , @xmath27.\ ] ] we could in general have different rates for different messages , but in this paper we restrict our attention to symmetric rates . therefore any rate referred to in this paper is the symmetric rate . for a given index coding problem , a rate @xmath1 is said to be _ achievable _ if there exists an index code of rate @xmath1 , and the index coding problem is said to be _ rate _ @xmath1 _ feasible_. if a rate @xmath28 is achievable , the associated index code is a _ scalar index code of length _ @xmath23 . if the encoding and decoding functions are linear , then we have a _ linear index code_. if we have a linear index code of rate @xmath1 , then we can represent the encoding function as follows . @xmath29 where each @xmath30 is a @xmath31 matrix with elements from @xmath12 . in the case of scalar linear index coding , we have @xmath32 . finding a scalar linear index code of length @xmath23 ( i.e. , with a feasible rate @xmath33 ) is equivalent to finding an assignment of these @xmath23-length vectors @xmath30s to the @xmath0 messages such that the receivers can all decode their demanded messages , i.e. , @xmath34.\ ] ] in the case of a scalar linear index code , the encoding function @xmath35 fixes the vectors assigned to the messages . however not every vector assignment is a valid index coding encoding function , as the decoding functions may not exist . in a number of proofs in this paper , we start with some vector assignment and show that it leads to an encoding function of a valid index code . therefore , in such proofs we refer to the initial vector assignment as a valid encoding function @xmath35 by abusing the notation . we restrict our attention to scalar linear index codes for the rest of this paper . however we believe that our results can be extended to vector linear index codes as well . for some receiver @xmath15 and for some message @xmath36 , let @xmath37 denote the set of messages ( except @xmath38 ) not available at the receiver @xmath15 . the sets @xmath39 are called the _ interfering sets at receiver _ @xmath15 . if receiver @xmath15 does not demand message @xmath38 , then we define @xmath40 . if a message @xmath41 is not available at a receiver @xmath15 demanding at least one message @xmath42 , then @xmath41 is said to _ interfere at receiver _ @xmath15 , and @xmath41 and @xmath38 are said to be _ in conflict_. for a set of vertices @xmath43 , let @xmath44 denote the vector space spanned by the vectors assigned to the messages in @xmath6 , under the specified encoding function @xmath35 . if @xmath45 , we define @xmath44 as the zero vector . for a given assignment of vectors to the messages ( or equivalently , for a given encoding function @xmath35 ) , we say that _ conflicts within a subset _ @xmath46 _ are resolved _ , if @xmath47,\end{aligned}\ ] ] where @xmath48 is the vector assigned to @xmath38 under the encoding function @xmath35 . if ( [ resolvedconflicts ] ) holds for @xmath49 then _ all the conflicts _ in the given index coding problem _ are said to be resolved_. we now state a simple lemma , rephrased from @xcite , which is easily proved . [ lemmasuccessdecoding ] for any encoding function @xmath35 , successful decoding at the receivers is possible if and only if all the conflicts are resolved . let @xmath48 be the vector assigned to @xmath38 under the encoding function @xmath35 . _ if part : _ at all receivers @xmath15 , consider that we have @xmath50 $ ] . what is received at a receiver @xmath15 is the codeword @xmath51 , from which it wants to obtain @xmath38 . as receiver @xmath15 can always subtract the contribution from the side - information messages from @xmath51 , it only remains to be shown that @xmath52 can be decoded from @xmath53 where the equality is because @xmath54 because of the assumption that @xmath55 , receiver @xmath15 can get @xmath38 from ( [ eqn1 ] ) . _ only if part : _ consider now that there is some receiver @xmath15 and some message @xmath56 such that @xmath57 clearly , solving for @xmath38 from ( [ eqn1 ] ) does not lead to an unique solution . hence decoding fails . this concludes the proof . by lemma [ lemmasuccessdecoding ] , it should also be clear that if there is an assignment of @xmath23-length vectors @xmath30s to the messages @xmath41s such that the condition in lemma [ lemmasuccessdecoding ] is satisfied , then these vectors naturally define an index code of length @xmath23 for the given index coding problem . in @xcite , the authors defined the notions of _ alignment graph _ and _ conflict graph _ whose properties were used to characterise index coding problems for which rate @xmath58 is feasible . both of these graphs have the same vertex set , which is the set of messages @xmath59 . in the alignment graph , the vertices @xmath41 and @xmath52 are connected by an edge ( called an _ alignment edge _ , shown in our figures by a solid edge ) when the messages @xmath41 and @xmath52 are not available at a receiver demanding a message other than @xmath41 and @xmath52 . a connected component of the alignment graph is called an _ alignment set_. it is easy to see that the alignment sets define a partition of the alignment graph . also , the messages in @xmath60 for all messages @xmath56 at all receivers @xmath15 are fully connected in the alignment graph . in the conflict graph , @xmath41 and @xmath52 are connected by an edge ( called an _ conflict edge _ , shown by a dotted edge ) if @xmath41 is not available at a receiver demanding @xmath52 , or @xmath52 is not available at a receiver demanding @xmath41 . the following example illustrates that the conflict graph definition does not capture the _ directionality _ of the conflicts . [ conflictissueexm ] consider two single unicast index coding problems with four messages . in the first problem , the interfering sets are as follows @xmath61 . in the second problem , @xmath62 all other interfering sets corresponding to the receivers are empty . these two problems have the same alignment and conflict graphs ( see fig.[fig : conflictgraphsame ] ) . in particular , the conflict graphs are the same because the definition of the conflict graph does not model the directionality of the absent side - information messages . note that they both have different solutions . we leave it to the reader to check that the first problem is rate @xmath58 feasible , while the second problem is not rate @xmath58 feasible , but has a rate @xmath2 solution . to overcome this issue , we define the _ conflict hypergraph _ as follows . the conflict hypergraph is an undirected hypergraph with vertex set @xmath59 ( the set of messages ) , and its hyperedge set defined as follows . * for any receiver @xmath15 demanding any message @xmath38 , @xmath38 and @xmath63 are connected by a hyperedge , which is denoted by @xmath64 . for example , the two problems presented in example [ conflictissueexm ] are now represented using different conflict hypergraphs in fig . [ fig : conflicthyp1 ] and fig . [ fig : conflicthyp2 ] ( the alignment graphs remains the same ) . note that even though this definition for the conflict hypergraph does not explicitly contain direction , the directionality of interference seen by any receiver is modelled correctly whenever the number of interfering messages is more than one . this is sufficiently general as we see in the following lemma . suppose two index coding problems , denoted by @xmath65 and @xmath66 are modelled by the same conflict hypergraph . then any index coding solution for @xmath65 is an index coding solution for @xmath67 . let @xmath68 be the encoding function of the given index code for @xmath65 . let @xmath48 be the vector assigned to @xmath38 . by lemma [ lemmasuccessdecoding ] , we must have that for @xmath50,~\forall j\in[1:t]$ ] in @xmath65 . now assume that the same index code is used for @xmath67 . for any receiver @xmath15 and message @xmath69 with @xmath70 in @xmath67 , the conflict hyperedge @xmath71 present in the conflict hypergraph of @xmath67 is present also in that of @xmath65 as both @xmath65 and @xmath67 have the same conflict hypergraph . in other words , message @xmath72 is an interfering set of a receiver @xmath15 in @xmath65 also and hence we must have @xmath73 , which means that @xmath69 is recoverable at @xmath15 in @xmath67 also . the only case left to check is when @xmath74 in @xmath67 for some receiver @xmath15 and message @xmath69 . let us assume that @xmath75 because @xmath65 and @xmath67 share the same conflict hypergraphs , this conflict ( in some direction ) is present in @xmath67 also . by assignment @xmath35 , we must have that @xmath30 and @xmath48 are linearly independent , which ensures that this conflict is resolved in @xmath67 also , irrespective of its directionality . towards showing a necessary and sufficient condition for rate @xmath58 feasibility , the following definition for _ internal conflicts _ was given in @xcite . a conflict between two messages within an alignment set is called an _ internal conflict_. the following theorem was proved in @xcite on rate @xmath58 feasible index coding problems . [ ratehalfthm ] an index coding problem is rate @xmath58 feasible if and only if there are no internal conflicts . the following lemma plays a crucial role in our proof of theorem [ ratehalfthm ] , which is included for the sake of completeness as it is not available in its complete form in prior literature . we will also see in subsection [ rateonethirdfeasibleproblems ] that it is central to proving our main theorem which deals with the feasibility of rate @xmath2 . [ intersectingkdimspaces ] let @xmath76 be @xmath77 sets of vectors , such that @xmath78 $ ] . then the space spanned by @xmath79 has dimension @xmath80 if and only if each @xmath81 spans a vector space of dimension @xmath80 . _ if part : _ suppose each @xmath81 spans a vector space of dimension @xmath80 , and @xmath78.$ ] clearly , all the vector spaces @xmath82s are exactly equal , and thus @xmath83 _ only if part : _ suppose @xmath84 and @xmath85.\ ] ] by ( [ eqn3 ] ) , @xmath86 . if some @xmath87 , then @xmath88 , and we have a contradiction . thus , we must have that @xmath89 . we now use lemma [ intersectingkdimspaces ] and lemma [ lemmasuccessdecoding ] to prove theorem [ ratehalfthm ] . corresponding to any vertex @xmath38 in the alignment graph , let @xmath90 denote the alignment set it belongs to ( this is unique as the alignment sets partition the alignment graph ) . we first note that in any ( scalar linear ) index coding scheme for the given problem , all the vertices must be assigned some non - zero vectors ( zero vector can not be assigned to any message as this means that the message can not be decoded by any receiver ) . _ if part : _ suppose that there are no internal conflicts . we assume a large field @xmath12 . for each alignment set , we independently generate a random @xmath91 vector over @xmath3 and assign it to the vertices of the alignment set . because of random generation , we can assume that any assigned vector is non - zero and any two assigned vectors are linearly independent with high probability ( whp ) . let @xmath35 denote the associated encoding function and @xmath48 denote the vector assigned to vertex @xmath38 . since there are no internal conflicts , we only have to check conflicts between alignment sets . for any vertex @xmath38 , the set @xmath63 for any receiver @xmath15 which demands @xmath38 belongs to a @xmath92 different and @xmath93 unique alignment set than @xmath90 ( @xmath92 is because there are no internal conflicts , @xmath93 is because all the messages in @xmath63 must be in the same alignment set ) . since any two alignment sets get independent vectors ( whp ) , we have that @xmath55 , and the same argument is true for all receivers @xmath15 and all messages @xmath38 . hence this assignment of vectors ensures successful decoding by lemma [ lemmasuccessdecoding ] . _ only if part : _ suppose that there is some internal conflict ( represented in the conflict graph as an edge between node @xmath94 and node @xmath56 ) in an alignment set @xmath90 . because @xmath95 are part of the same alignment set ( connected component ) @xmath90 , there lies a path from @xmath94 to @xmath56 , given by an ordered set @xmath96 , such that every adjacent pair of elements ( @xmath97 , etc . ) belong to the interfering set of some receiver . in some assignment corresponding to a rate @xmath58 solution , let @xmath98 be the non - zero vectors assigned to the vertices @xmath96 . we define the sets , @xmath99 $ ] and @xmath100 . and @xmath56 are in conflict , they must be assigned linearly independent vectors . this requires that at least one of the sets @xmath101 has to be two dimensional.,width=211 ] suppose some @xmath102 for some @xmath103 . then a receiver @xmath15 ( at which the messages corresponding to @xmath104 are unavailable ) ` sees ' an interfering space of dimension @xmath105 . thus we need to assign a vector linearly independent from @xmath104 ( which itself has dimension @xmath105 ) to the corresponding demanded message of receiver @xmath15 . this can be possible only if the assigned vectors are of length at least @xmath106 , i.e. the rate can be at most @xmath2 . therefore , for a rate @xmath58 index coding assignment , all @xmath104 should spanning a space of dimension @xmath107 . thus we have @xmath108 and @xmath109.$ ] by lemma [ intersectingkdimspaces ] , we should thus have @xmath110 however @xmath94 and @xmath56 are in conflict , which means that they should be assigned linearly independent vectors , i.e. , @xmath111 which means that @xmath112 . thus there is a contradiction and thus any internal conflicts forces the rate to be less than @xmath58 . this concludes the proof . from section [ sec3 ] , the following is clear . * if there are no conflicts ( not even conflicts between two alignment sets ) in the alignment graph , rate @xmath107 is achievable ( this is the case when any receiver demanding a message has all the other messages as side - information ) . * for rate @xmath107 infeasible index coding problems , rate @xmath58 is feasible if and only if there are no internal conflicts . towards obtaining our main result , which characterises a class of index coding problems which are rate @xmath2 feasible , we first give a prior known necessary condition for feasibility of rate @xmath2 . suppose there are @xmath113 messages @xmath114 such that message @xmath115 is demanded by some receiver ( say receiver @xmath116 ) and @xmath117 following @xcite , we call such a set of messages @xmath114 as an _ acyclic subset of messages of size _ @xmath113 . the following theorem can be obtained from the results in @xcite . [ thmmais ] an index coding problem which is rate @xmath2 feasible can not have an acyclic subset of messages of size @xmath113 . assigned to the four messages must be linearly independent.,width=192 ] the following example however shows that theorem [ thmmais ] is not a sufficient condition for rate @xmath2 feasibility . [ ex : onethirdinfeasible ] consider a single unicast index coding problem with six messages . the interfering sets are as follows : @xmath118 . the alignment graph and the conflict hypergraph corresponding to the problem are given in fig . [ fig : infeasible ] . it can be seen that the conflict hypergraph of this problem does not have an acyclic subset of messages of size @xmath113 . if we assume that the problem is rate @xmath2 feasible , then it has to be necessarily true that both @xmath119 and @xmath120 are two dimensional . we leave it to the reader to check that @xmath121 must also be two dimensional . however , the conflict between the messages @xmath122 and @xmath123 are resolved only if @xmath121 is three dimensional . thus , the problem is rate @xmath2 infeasible . in @xcite , the following theorem about the achievability of rate @xmath2 was proved ( this is a special case of corollary 9 of @xcite ) . [ thmnocyclesforks ] consider a rate @xmath58 infeasible index coding problem with no acyclic subset of size @xmath124 if none of its alignment sets have both forks ( a _ fork _ is a vertex connected by three or more edges ) and cycles , then the index coding problem is rate @xmath2 feasible . as mentioned in @xcite , the condition that there are no alignment sets with both forks and cycles of length @xmath106 means that there is no interfering set @xmath63 at any receiver of size @xmath125 , as such a set would mean that there is both a cycle and fork within an alignment set ( since the messages in @xmath63 are fully connected in the alignment set ) . therefore theorem [ thmnocyclesforks ] characterises a rather limited class of index coding problems which are @xmath2 feasible . example [ ex : onethirdfeasible ] shows an index coding problem which does not satisfy the conditions of theorem [ thmnocyclesforks ] but is rate @xmath2 feasible . [ ex : onethirdfeasible ] consider a single unicast index coding problem with six messages . the interfering sets of the problem are as follows : @xmath126 . the alignment graph and the conflict hypergraph corresponding to the problem are given in fig . [ fig : feasible ] . it can be seen that the alignment graph of this problem has both forks and a cycle . consider @xmath106 linearly independent @xmath127 vectors @xmath128 . we note that the following assignment of vectors resolves all conflicts : ( i ) vector @xmath129 to messages @xmath130 and @xmath123 , ( ii ) vector @xmath131 to messages @xmath132 and @xmath133 , ( iii ) vector @xmath134 to messages @xmath135 and @xmath136 . thus , the problem is rate @xmath2 feasible . in this subsection , we develop a new framework for studying the rate @xmath2 feasibility of groupcast index coding problems . towards this , we define the notions of a triangular interfering set and a type-2 alignment set . a subset @xmath137 of size three of the set of messages @xmath138 is said to be a _ triangular interfering set _ if all the messages in @xmath137 interfere simultaneously at some receiver , and at least two of the messages in @xmath137 are in conflict . two distinct triangular interfering sets @xmath139 and @xmath140 are said to be _ adjacent _ if they ` meet ' at a conflicting edge , i.e. , @xmath141 such that @xmath41 and @xmath52 are in conflict . [ type2align ] two triangular interfering sets @xmath139 and @xmath140 are said to be _ connected _ if there exists a _ ( i.e. , a sequence ) of adjacent triangular interfering sets starting from @xmath139 and ending at @xmath140 . type-2 alignment set _ is a maximal set of triangular interfering sets which are connected to each other . examples to illustrate the above definitions are shown in fig . [ fig : triangleinterfset ] . note that the maximality in definition [ type2align ] means that we can not add another triangular interfering set to a type-2 alignment set and still maintain connectivity ( as in definition [ type2align ] ) . by definition , every type-2 alignment set must be a subset of a ( regular ) alignment set , and there could be many type-2 alignment sets within any alignment set . given an index coding problem , we can identify type-2 alignment sets as follows . within any alignment set , we identify a triangular interfering set of messages ( if there is no such set then there is no type-2 alignment set inside that alignment set ) . then we repeatedly add triangular interfering sets which are adjacent to the existing connected triangular interfering sets . when we can no longer add such adjacent triangular interfering sets , then we have our type-2 alignment set . we now prove a necessary condition for rate @xmath2 feasibility based on the vectors assigned to type-2 alignment sets . this theorem is another application of the key lemma , lemma [ intersectingkdimspaces ] . [ type2sets2dim ] in any rate @xmath2 solution to a given index coding problem , all the messages in any type-2 alignment set must be assigned vectors from a vector space of dimension two . let @xmath35 be the encoding function of a rate @xmath2 solution . consider a type-2 alignment set @xmath142 with triangular interfering sets @xmath143.$ ] suppose for any triangular interfering set @xmath144 of @xmath142 , we have @xmath145 . since all the vertices in @xmath144 interfere at some receiver ( say , a receiver which requests some other message @xmath52 ) , the message @xmath52 must be assigned a vector which is linearly independent from those assigned to the messages in @xmath144 , and thus we need at least @xmath113 linearly independent vectors , and hence the rate has to be @xmath146 . thus no triangular interfering set @xmath144 of @xmath142 has @xmath145 . however , any triangular interfering set @xmath144 of @xmath142 _ must _ have @xmath147 , as @xmath144 has a conflict . suppose @xmath148 . consider three messages @xmath149 and @xmath150 in @xmath151 that have been assigned three linearly independent vectors , belonging to some three triangular interfering sets ( not necessarily different ) , @xmath152 and @xmath153 respectively . we have already shown that we can not have @xmath154 . so at least two of the three triangular interfering sets are different . suppose all three sets @xmath152 and @xmath153 are different . because the three triangular interfering sets are within the same type-2 alignment set , it must be the case that there exists a path consisting of adjacent triangular interfering sets starting from @xmath155 through @xmath156 and upto @xmath157 let @xmath77 be the number of triangular interfering sets on this path ( counted as we go along the path ; repetitions are counted separately ) . for @xmath158 $ ] , let @xmath159 denote the set of @xmath106 vectors assigned to the @xmath160 triangular interfering set in this path . [ fig : type2assignment ] illustrates this scenario for the type-2 alignment set example shown in fig . [ fig : type2ex ] . . to determine the dimension of the span of these messages , consider the following path of triangular interfering sets : @xmath161 and @xmath162 are the sets of @xmath106 vectors assigned to @xmath160 triangular interfering set.,width=249 ] by the previous arguments , we have that @xmath163 also , @xmath164,$ ] as the @xmath160 and the @xmath165 triangular interfering sets are adjacent by construction of the path . therefore , by lemma [ intersectingkdimspaces ] , it must be the case that @xmath166 . however , the vector assigned to the message @xmath167 belongs to @xmath168 for @xmath169 , and according to our assumption the vectors assigned to these three messages are linearly independent vectors . thus there is a contradiction , which means that we can not have three messages in three different triangular interfering sets which have been assigned linearly independent vectors . a similar claim can be proved if the three messages come from two different triangular interfering sets . thus , no three messages in a type-2 alignment set can be assigned linearly independent vectors . in other words , any type-2 alignment set @xmath142 in a rate @xmath2 solution must have @xmath170 . theorem [ type2sets2dim ] is stricter than theorem [ thmmais ] , as theorem [ thmmais ] applies only to an acyclic subset of messages of size @xmath171 which basically is equivalent to a triangular interfering set . theorem [ type2sets2dim ] on the other hand considers a ` connected component ' of such triangular interfering sets , and is therefore more strict . we leave it to the reader to verify that the problem in example [ ex : onethirdinfeasible ] , while ` passing ' the condition of theorem [ thmmais ] , ` fails ' the condition of theorem [ type2sets2dim ] . theorem [ type2sets2dim ] prescribes that type-2 alignment sets must be ` two - dimensional ' in a rate @xmath2 code . in this subsection , we give a necessary and sufficient condition for achieving this two - dimensionality . for this purpose , we require the notion of a restricted index coding problem . let @xmath172 denote an index coding problem with message set @xmath59 . for some @xmath173 , a @xmath142__-restricted index coding problem _ _ is defined as the index coding problem @xmath174 consisting of * the messages @xmath142 . * the subset @xmath175 ( of size @xmath176 ) of the receivers of @xmath172 which demand messages in @xmath142 . * for each @xmath177 the demand sets @xmath178 and the side - information sets @xmath179 are restricted within @xmath142 , i.e. , @xmath180 the alignment graph and the alignment sets of the restricted index coding problem @xmath174 are called the @xmath142-_restricted alignment graph _ and @xmath142-_restricted alignment sets _ respectively . a @xmath142-_restricted internal conflict _ is a conflict between any two messages within a restricted alignment set of @xmath142 . the proof of the following theorem is a direct application of theorem [ ratehalfthm ] , and hence is skipped . [ restrictedratehalfthm ] the restricted index coding problem @xmath174 is rate @xmath58 feasible if and only if there are no @xmath142-restricted internal conflicts . [ 3cross1restrictedcorr ] for a given index coding problem @xmath172 , there exists an assignment of @xmath181 vectors from a two dimensional vector space ( over a large enough field @xmath3 ) to the messages in a subset @xmath142 such that the conflicts within @xmath142 are resolved , if and only if the restricted index coding problem @xmath174 is rate @xmath58 feasible . we first recall that conflicts within @xmath142 are said to be resolved if ( [ resolvedconflicts ] ) is satisfied . _ if part : _ the proof for the if - part follows the achievability scheme shown in the proof of theorem [ ratehalfthm ] , with the difference that we assign to the messages in @xmath142 randomly generated @xmath181 vectors ( not @xmath91 vectors as in theorem [ ratehalfthm ] ) from a two dimensional space over a suitably large field @xmath3 . because @xmath174 is rate @xmath58 feasible , such an assignment resolves the conflicts within @xmath142 . _ only if part : _ suppose there is an assignment of @xmath181 vectors from a two dimensional vector space to messages in @xmath142 such that all conflicts within @xmath142 are resolved . then we can always obtain a @xmath182 matrix @xmath6 such that premultiplying all the vectors assigned to @xmath142 with @xmath6 gives us a rate @xmath58 ( length 2 ) index coding solution for @xmath174 . the following theorem gives a necessary and sufficient condition for assigning vectors from a two dimensional space to the type-2 alignment sets ( i.e. , for satisfying the necessary condition of theorem [ type2sets2dim ] ) . [ norestrconflictstype2 ] let @xmath142 be a type-2 alignment set of the given index coding problem @xmath172 . if @xmath172 is rate @xmath2 feasible , then @xmath174 must be rate @xmath58 feasible which holds if and only if there are no @xmath142-restricted internal conflicts . the proof follows by combining the claims of theorem [ type2sets2dim ] , corollary [ 3cross1restrictedcorr ] and theorem [ restrictedratehalfthm ] . we now prove the main result of this paper , which connects all the previously proved results and widens the class of index coding problems for which rate @xmath2 is achievable . because of the framework we have developed in the previous subsections , the proof of this theorem is simpler than theorem [ thmnocyclesforks ] , while also subsuming that result . [ thm : main ] a rate @xmath58 infeasible index coding problem @xmath172 is rate @xmath2 feasible if every alignment set of @xmath172 satisfies _ either _ of the following conditions . 1 . it does not have both forks and cycles . it is a type-2 alignment set with no restricted internal conflicts . we first make some observations before giving the achievable index coding scheme . _ observation 1 : _ consider an alignment set @xmath183 of @xmath184 which does not contain both forks and cycles and is also not a triangular interfering set . we claim that there is no triangular interfering set within @xmath183 . this is because if there is a triangular interfering set ( say @xmath139 ) then there should be at least one more message which is not included within @xmath139 . however this would mean that there is both a fork and a cycle within @xmath183 , which is not allowed . _ observation 2 : _ now suppose there are three messages in @xmath183 such that all of them interfere with a particular receiver . then there must necessarily be no conflicts in - between the three messages . this is because if there were conflicts in between the three messages , then these three messages will form a triangular interfering set . thus no three messages from @xmath183 having at least one conflict in - between themselves can interfere at the same receiver . _ observation 3 : _ by a similar argument as in observation @xmath107 , if @xmath183 has three messages interfering at a receiver , then it can not have any other message than these three messages ( because if it did , then we would have both cycles and forks within @xmath183 . _ observation 4 : _ consider an alignment set @xmath185 of @xmath184 which does not contain both forks and cycles , but is also a triangular interfering set . then , by definition @xmath185 must be a type-2 alignment set . for the sake of this proof , we consider an alignment set such as @xmath186 as being under the class of type-2 alignment sets . we thus have three kinds of alignment sets in @xmath172 . 1 . alignment sets which have no three messages interfering at any receiver . alignment sets which consists only of three messages , all three interfering at some receiver , without any conflicts in - between . ( these three messages may interfere at other receivers also , but at least one common receiver where they all interfere exists ) . alignment sets which are also type-2 alignment sets without restricted internal conflicts . are shown . in the type-2 alignment set , the receivers at which the triangular interfering sets interfere are not explicitly shown in the figure . since there should not be any restricted internal conflict within a type-2 alignment set , no triangular interfering set interferes at a receiver inside the same type-2 alignment set.,width=249 ] we now give the achievability index coding scheme by assigning vectors independently to each alignment set of @xmath172 , and the technique of assignment , which follows , depends on the type of the alignment set . _ alignment set which has no three messages interfering at any receiver : _ for each message in such an alignment set , we assign an independently generated random @xmath181 vector ( over a large field @xmath3 ) . _ alignment set which consists only of three messages interfering at any receiver without any conflicts in - between : _ we randomly generate a @xmath127 vector and assign it to all the messages in such an alignment set . _ alignment set which is a type-2 alignment set without restricted internal conflicts : _ let @xmath142 be the type-2 alignment set under concern . for each @xmath142-restricted alignment set , we assign an independently generated random @xmath181 vector from a two dimensional space . note that this resolves all the conflicts between the messages within @xmath142 , by theorem [ restrictedratehalfthm ] and corollary [ 3cross1restrictedcorr ] . all the messages fall under one of these alignment sets , and hence all of the messages have been assigned vectors at this point . let @xmath35 denote encoding function corresponding to this assignment and @xmath48 denote the vector assigned to message @xmath38 . we now show that this assignment resolves all the conflicts in @xmath172 . consider a receiver @xmath15 which requests a message @xmath38 . we proceed on a case by case basis , depending on the size of @xmath63 . for each case we check whether the condition , @xmath55 ( whp ) , is met . we call this condition as the _ no conflict condition _ for the sake of this proof . _ case 1 : _ @xmath187 there are two cases here , either @xmath38 and @xmath63 are in the same alignment set or they are in different alignment sets . if @xmath38 and @xmath63 are in different alignment sets , then @xmath38 and the messages @xmath63 have been assigned independently and randomly generated @xmath181 vectors . thus the no conflict condition is met . now , @xmath38 and @xmath63 are in the same alignment set . this can only be an alignment set where no three messages interfere at any receiver , or a type-2 alignment set . in the former case , the no conflict condition is met as any three messages in such an alignment set are assigned independent vectors ( whp ) . now if it is a type-2 alignment set , the conflict(s ) between @xmath38 and @xmath63 is(are ) within that type-2 alignment set . because there are no restricted internal conflicts in any type-2 alignment set , it must be the case that @xmath38 and @xmath63 are in different restricted alignment sets . by our scheme , such conflicts should therefore be resolved . hence , the no conflict condition is met in this case too . _ case 2 : _ @xmath188 then we have two cases . the first case is that no two messages in @xmath63 are in conflict . this means that @xmath63 must be an alignment set which consists only of three messages without any conflicts in - between , and @xmath38 necessarily belongs to another alignment set . by our scheme , @xmath189 is an one - dimensional space generated by a random vector which is independently generated from the vector assigned to @xmath38 . hence , the no conflict condition is met . finally we consider the case when @xmath190 and at least one conflict exists within @xmath63 . then @xmath63 must be within some type-2 alignment set ( say @xmath142 ) . again we have two sub - cases here , i.e. , @xmath38 and @xmath63 are within the same ( type-2 ) alignment set , or @xmath38 is in a different alignment set than @xmath63 which is within a type-2 alignment set . in the former case , the conflicts between @xmath38 and @xmath63 are within the type-2 alignment scheme , which is resolved by our scheme ( by the same arguments as in the last subcase of case 1 ) . in the latter case , we must have that @xmath189 lies within a two dimensional space ( as it is within a type-2 alignment set ) which is generated independently from the vector assigned to @xmath38 . hence again the no conflict condition is met . by the previous arguments , the no conflict condition is met for any receiver @xmath15 and any demand @xmath38 at @xmath15 . thus , all the conflicts in @xmath172 are resolved . this proves the theorem . in this work , we presented a class of index coding problems for which rate @xmath2 is feasible . this class of problems is larger than what was previously known . we believe that the framework developed in this work in order to obtain our results can be leveraged to settle the rate @xmath2 feasibility completely . in particular , we conjecture that the the necessary condition for rate @xmath2 feasibility of theorem [ norestrconflictstype2 ] is also sufficient . developing conditions for feasibility of rates of the form @xmath191 would also be an interesting area of future study . the connection to topological interference management problems follows from @xcite . from @xcite , it is known that the length of scalar linear index codes for the single unicast index coding problem is known to be equal to _ minrank _ of the side - information graph . thus , our results also imply a class of graphs whose minrank is equal to @xmath106 . this is a promising result as the general minrank problem is known to be np hard @xcite . 160 y. birk and t. kol , coding on demand by an informed source ( iscod ) for efficient broadcast of different supplemental data to caching clients , ieee transactions on information theory , vol . 52 , no . 6 , june , 2006 , pp . 2825 - 2830 . a. blasiak , r. kleinberg , and e. lubetzky , broadcasting with side information : bounding and approximating the broadcast rate , ieee transactions on information theory , vol . 2013 , pp . 5811 - 5823 .

the study of ultra high energy cosmic rays ( uhecr ) and neutrinos ( uhe@xmath0s ) is currently a high priority in astroparticle physics with many experimental efforts being dedicated to these two related areas of research . uhecrs are routinely detected through the extensive atmospheric showers ( eas ) they produce when interacting in the atmosphere . despite the recent advances in the measurement of the flux of uhecrs @xcite , their primary composition remains unknown @xcite , and this is one of the main obstacles to extract precise conclusions on their origin . there are strong reasons to believe that uhe@xmath0s should be produced in the interactions of uhecrs with the material surrounding the sources , and/or in their propagation through the observed cosmic microwave background radiation @xcite . however their detection has not yet been achieved . efforts are being made to improve the experimental situation in both fields . the challenge is to instrument sufficiently large target volumes to compensate for the low cross section and low fluxes for uhe@xmath0 detection , and to find measurements that provide large aperture at the highest energies and help to constrain the primary composition of uhecrs . the radio technique is being explored in both fields . as early as in 1962 , g. askaryan proposed to detect uhecrs and uhe@xmath0s by observing the coherent radio pulse from the excess of electrons in a shower developing in a dense , dielectric and nonabsorptive to radiowaves medium @xcite . soon after , pulses were observed in coincidence with air shower arrays @xcite . the emission is coherent in wavelengths which are large compared to the characteristic size of the electric charge and current distributions associated to the induced showers . the technique has been receiving a lot of attention in the field of astroparticle physics in the last decade because of the relatively low cost of the antennas needed for the detection systems . also , coherence implies that the pulse energy scales with the square of the primary energy which favours long range detection , a requirement to achieve large areas and volumes both for uhecr and uhe@xmath0 detection . indeed , quadratic scaling has been confirmed in accelerator experiments @xcite leading to very strong pulses associated to uhe energy showers . a large number of initiatives have been made or are currently in development or planning stages . in dense media neutrino - induced showers are less than a meter in width and full coherence is expected to be mantained up to the ghz range ( see for instance @xcite ) . a variety of past and present experiments search for pulses at those frequencies produced by neutrinos in ice @xcite , the moon regolith @xcite or salt domes @xcite . the largest and most promising experiments are in planning stages looking at ice @xcite . electromagnetic pulses emitted in the mhz - ghz frequency range by eas induced by uhecr are being measured in the hope of using this complementary information to constrain its composition and/or to develop new cost effective detection systems @xcite . in addition the anita balloon flown antennas , initially devised to search for neutrinos , has recorded coherent pulses up to the ghz range @xcite consistent with eas induced by uhecrs and there are plans for future developments @xcite . the success of the radio technique requires an accurate and computationally efficient calculation of the radio emission properties of uhe showers . it is necessary to perform this in an efficient way since the number of particles in a shower at eev energies is @xmath1 . simulation techniques for evaluating pulses in dense media have been used for more than 20 years @xcite . the shower is simulated to obtain , for every particle track , the information needed to calculate its contribution to the electric field . the formula used for this purpose stems from an approximate solution of maxwell s equations in the fraunhofer limit @xcite adapted for simulation purposes in @xcite . the electric field due to a short charged particle track , assumed to be travelling at a constant speed , can be approximated by two terms which correspond to the start and end of the track . the resulting electric field depends on the particle speed and the angle between the line of sight from the track to the observer @xcite . to calculate the emission in a shower , tracks are chosen so that the particle velocity can be approximated to be constant , and all track contributions are added taking into account interference effects . although the approximate formula is sufficient for many practical applications , its range of validity is limited in frequency and position of the observer with respect to the track . it is possible to extend its range of validity by subdividing each track in sub - intervals . the resulting algorithm ( often referred to as zhs algorithm " ) is easy to implement and fast enough for the simulation of particle showers . this is convenient since it has allowed the simulation of pulses in the fresnel region for neutrino detection @xcite and in measurements of eas @xcite . although the range of applicability of the algorithm is enhanced when used in this way , it is not obvious that the sum of the subcontributions correclty accounts for all the radiation in regions close to the emission source . the object of this article is to study the conditions for the zhs algorithm to be valid and to establish its accuracy . for this purpose we obtain in section ii an exact solution to the problem of a charged particle instantaneously accelerating to a constant speed and stopping abruptly after a discrete time interval @xcite . this allows us to establish the precise conditions necessary to turn this solution into the basic formula of the zhs algorithm . in section iii we compare the exact solutions for an infinite and a finite track to the result of the zhs algorithm , and stress the compatible interpretation of the radiation regime in terms of cherenkov radiation for both cases . in section iv we compare the exact solution of the single track problem in nearby regions with the results of applying the zhs algorithm with track subdivisions to test its validity and accuracy . section v is devoted to discussing the zhs algorithm in relation to other approximations made to calculate the pulses emitted from showers . section vi presents the summary and conclusions of our work . let us assume an electron ejected from an atom at time @xmath2 that travels at constant speed @xmath3 through a medium along a finite track until it is absorbed by another atom at time @xmath4 . neglecting the movement of the atoms , we can model the electric current associated to the electron as , @xmath5 where @xmath6 is the charge of a positron , @xmath7 is its position and @xmath8 an arbitrary reference position . the step @xmath9-functions account for the fact that the electron only moves in the time interval ( @xmath10 , @xmath4 ) . in a dielectric medium with permitivity @xmath11 and magnetic susceptibility @xmath12 , maxwell s equations for the vector potential in the frequency domain @xmath13 can be written as @xcite : @xmath14 = -\mu\mathbf{j}(\mathbf{x},\omega ) , \end{split } \label{eq : avec}\ ] ] where @xmath15 is the fourier transform of the scalar potential and we use the following convention for the fourier - transform of a function @xmath16 : @xmath17 . in principle @xmath11 and @xmath12 can depend on frequency and our results below would be equally valid , but we drop the explicit dependence of @xmath11 and @xmath12 with @xmath18 for simplicity . we use the lorenz gauge condition which implies that eq . ( [ eq : avec ] ) for the vector potential becomes : @xmath19 with @xmath20 and @xmath21 the refractive index of the medium . the lorenz gauge condition in the frequency domain @xcite : @xmath22 implies that the scalar potential for non - zero frequencies is entirely determined by the divergence of the vector and only the vector potential is needed to calculate the field . the solution of the helmholtz equation in eq . ( [ eq : helmholtz ] ) is standard physics and can be obtained , using green s method , as an integral over source positions @xcite : @xmath23 where from now on , @xmath24 will denote the position of the observer and @xmath25 the position of the source . for simplicity we rewrite the current in eq . ( [ eq : current ] ) as : @xmath26 with @xmath27 ; @xmath28 and @xmath29 , where we assume without loss of generality that the electron travels along the @xmath30-axis which is parallel to @xmath31 . the equations that follow below are clearly valid for any continuous and differentiable functions @xmath32 and @xmath33 . they are also valid for the case considered in eq . ( [ eq : current ] ) which can be obtained as a limiting case of suitable continuous and differentiable functions . transforming this definition of the current to the frequency domain and substituting into eq . ( [ eq : avec_w ] ) for the vector potential we obtain : @xmath34 in the lorenz gauge , the only non - zero component of the vector potential for a charge moving along @xmath30 is the @xmath30-component . the scalar potential @xmath35 is obtained through eq . ( [ eq : gauge ] ) . for this purpose we need to obtain @xmath36 : @xmath37 \frac{(z - z')}{|\mathbf{x } - \mathbf{x'}| } \nonumber\end{aligned}\ ] ] the electric field @xmath38 can be obtained from the scalar and vector potentials @xcite as : @xmath39 due to the cylindrical symmetry of a charged particle track we calculate only the radial ( @xmath40 ) and @xmath30 ( @xmath41 ) components of the field . since @xmath42 only has @xmath30-component , then the radial component of the field is given by : @xmath43 where @xmath44 is the radial coordinate , and @xmath45 denotes @xmath46 . the @xmath47component of the field is given by : @xmath48 with @xmath49 denoting @xmath50 . performing the derivatives in eqs . ( [ field_rho ] ) and ( [ field_z ] ) we obtain : @xmath51 \end{split } \label{eq : rhofield}\ ] ] and , @xmath52 ~+~ \\ & i\omega~\frac{\mu}{4\pi}~qv~\int_{t_1}^{t_2 } \mathrm{d}t'~e^{i\omega t'}~\frac{e^{ikr}}{r } \end{split } \label{eq : zfield}\ ] ] where @xmath53 and @xmath54 are both functions of the source time @xmath55 which is defined as , @xmath56 and @xmath57 eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) provide an exact solution for the electric field of a finite track . in general they do not have an analytical form and a numerical integration needs to be performed to obtain the field . for the calculations in this paper , we have divided the integration interval and applied simpson s rule , increasing the number of divisions until the integral converged . under certain conditions , however , we can give analytical approximations for relevant physical situations . in the following we show that the basic expression used in the zhs algorithm is a particular case of eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) under certain approximations . a simple expression for the approximate calculation of the electric field from a single charged particle track moving at constant speed was found in @xcite . in this section we derive the expression used for the zhs algorithm from the exact solution and compare the electric field as obtained in both the exact calculation and with the zhs expression . this allows us to establish under which circumstances the formula gives a good account of the electric field . the zhs algorithm can be obtained from eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) if the following set of conditions are fulfilled : 1 . the observer is in the far field " zone i.e. @xmath58 2 . the fraunhofer approximation holds . this can be stated as a condition for the phase factor to be approximated as : @xmath59 \label{eq : fraunhofer}\ ] ] where @xmath60 is the distance from the observation point to a reference point along the track where the particle is located at a reference time @xmath61 , and @xmath62 is the angle between the particle track and the direction from the reference point to the observer . this approximation holds provided the parameter @xmath63 with @xmath64 defined as : @xmath65 ^ 2}{r}\sin^2\theta , \label{eq : cond3}\ ] ] this condition should be fulfilled at any time @xmath66 from @xmath10 to @xmath4 . a more commonly used and nearly equivalent form of this condition @xcite is : @xmath67 where @xmath68 is the length of the track . this condition is necessary to ensure that the second and higher order terms for the phases @xmath69 in eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) have no relevance , even when the sum of the leading and first order terms in the taylor expansion of the phases is zero , as it occurs for observation at the cherenkov angle @xcite defined as @xmath70 with @xmath71 . 3 . finally the distance to the observer appearing in the denominators of several terms of eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) must be approximated as : @xmath72 over the length @xmath73 of the track , where @xmath60 is the distance to a reference point along the track ( in the zhs algorithm the mid - point of the track is selected ) . the error when making this approximation is of order @xmath74 . in particular , the condition in eq . ( [ eq : cond1 ] ) implies : @xmath75 with these approximations the radial component of the field in eq . ( [ eq : rhofield ] ) becomes : @xmath76 where we have used : @xmath77 and @xmath78 eq . ( [ eq : rhozhs ] ) can be easily integrated yielding : @xmath79 . \end{split}\ ] ] if we make @xmath80 this becomes the expression for the radial field as used in the zhs algorithm @xcite except for a factor 2 due to the fourier transform convention used in @xcite . similarly applying the approximations in eqs . ( [ eq : fraunhofer ] ) , ( [ eq : cond4 ] ) and ( [ eq : b ] ) to eq . ( [ eq : zfield ] ) for the @xmath30-component of the field , and using that @xmath81 , it is straightforward to show that : @xmath82 which can be cast as : @xmath83 performing the integral , and taking @xmath80 , the zhs formula is recovered : @xmath84~. \end{split}\ ] ] to calculate the electric field of the pulse emitted from a current distribution , such as that produced in a high energy shower , the zhs algorithm uses the zhs expressions in eqs . ( [ eq : rhofield_zhs ] ) and ( [ eq : zfield_zhs ] ) to calculate the emission from all the charged particle tracks . the final result is obtained adding up all the contributions . the value of @xmath60 used for the phase factor in each particle track is the distance from the first point of the track to the observation point . this definition is consistent with the convention to account for the phase change between emission arising from the start and end points of the track . the actual value of @xmath60 used for the denominator is the distance between the midpoint of the track and the observer which is for all practical purposes the same value when the approximation is valid . the algorithm used in alternative simulation programs is similar but different in these technical details @xcite . naturally it is possible to divide any charged particle track into arbitrarily small subtracks in order to make the computation more accurate . this will extend the range of validity of the approximation . it is interesting to note that it makes absolutely no difference to subdivide the track of a uniformly moving charge when observed in the fraunhofer limit because the term associated to the end of one subtrack cancels the term associated to the beginning of the next subtrack . this is because in this limit the differences in @xmath60 between adjacent subtracks are arbitrarily small . however in practical situations this cancelation is not exact if @xmath60 is allowed to change from a track to the next one according to geometry . in the original zhs program @xcite the tracks were not subdivided but soon it was realized that a more accurate result was obtained by subdividing all tracks at every point there was a discrete interaction in the simulation program @xcite . this was the main modification that was ever made to the original zhs monte carlo @xcite and has been effective since then . with this subdivision the distribution of tracks for a shower in ice has a peak at about @xmath85 mm . the exact electric field for a charged particle track derived in eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) must account for every single feature of the electric field , since no approximations have been made . in particular , it must reproduce cherenkov radiation which is the only radiation emitted by a charged particle moving at constant speed , when @xmath86 in the limit of an infinite track . an analytical solution for the electric field produced by such particle can be obtained and it is given in @xcite(chapter 4 ) . the radial @xmath87 and @xmath47components of the fields given in @xcite converted to si units , and using the fourier transform convention adopted in this work , can be written as : @xmath88 @xmath89 where @xmath87 is the radial coordinate , @xmath90 and @xmath91 are the modified bessel functions of the second kind , and @xmath92 is a function that can take two different values depending on the magnitude of the particle speed , @xmath93 or @xmath86 ( subluminal or superluminal regime ) . in our convention : @xmath94 @xmath95 if the particle travels below the speed of light in the medium , the argument @xmath96 of the bessel functions is real and the particle does not radiate as shown in @xcite . on the contrary , if the speed of the particle is larger than the speed of light , then @xmath96 is imaginary and the particle radiates . the latter case corresponds to pure cherenkov radiation @xcite . with the help of the asymptotic forms for the bessel functions , we can obtain the limits of eqs . ( [ eq : rhofield_inf ] ) and ( [ eq : zfield_inf ] ) when @xmath97 i.e. for small distances to the track compared to the radiation wavelength . conversely we can also obtain the limits when @xmath98 i.e. for large distances compared to the wavelength . if @xmath97 the @xmath91 bessel function dominates over the @xmath90 and only the radial component of the field @xmath40 matters . in this case , @xmath99 and a @xmath100 dependence with distances is obtained , as well as no dependence with frequency . if @xmath98 and keeping in mind that when @xmath86 the argument @xmath101 is imaginary the fields can be written as : @xmath102 @xmath103 in this case the field is proportional to @xmath104 . this is in agreement with @xcite where the same behavior is deduced using simple arguments of energy conservation through a cylindrical surface surrounding the track . in fig . [ fig : inf_vs_finzhs ] the fourier components of the modulus of the electric field for an infinite track as obtained from eqs . ( [ eq : rhofield_inf ] ) and ( [ eq : zfield_inf ] ) are shown , for different frequencies , as a function of @xmath87 , the radial distance to the track . the particle speed is @xmath105 travelling in homogeneous ice with refractive index @xmath106 . under these circumstances the quantity @xmath107 can be approximated as : latexmath:[\[\label{eq : u } at large distances to the track when @xmath98 the fields shown in fig . [ fig : inf_vs_finzhs ] scale with distance as @xmath109 and with frequency as @xmath110 , in agreement with the aysmptotic field components in eqs . ( [ eq : limit_rho_urho_large ] ) and ( [ eq : limit_z_urho_large ] ) . this behavior takes place when @xmath111 , 1 and 10 m for frequencies @xmath112 ghz , 100 mhz and 10 mhz respectively , in agreement with eq . ( [ eq : u ] ) , as can be seen in fig . [ fig : inf_vs_finzhs ] . as the distance to the track decreases and the condition @xmath97 starts to be valid , the field behaves as @xmath100 and does not depend on frequency as expected from eq . ( [ eq : limit_urho_small ] ) . as can be clearly seen in fig . [ fig : inf_vs_finzhs ] , the transition from the @xmath100 behaviour to @xmath104 occurs at a distance that depends on frequency because @xmath107 involves the frequency ( eq . [ eq : u ] ) . for instance at distances @xmath113 m the condition @xmath97 applies for both @xmath114 and 10 mhz . the fourier component of the field scales with @xmath100 and has the same value for the two frequencies as seen in fig . [ fig : inf_vs_finzhs ] , while this is not the case for a frequency of @xmath112 ghz . in fig . [ fig : inf_vs_fin ] the modulus of the field for a charged particle in an infinite track - as obtained from eqs . ( [ eq : rhofield_inf ] ) and ( [ eq : zfield_inf ] ) - is shown as a function of frequency for an observer at a fixed radial distance . at large enough frequencies so that the condition @xmath115 applies , the field scales as @xmath110 as expected from eqs . ( [ eq : limit_rho_urho_large ] ) and ( [ eq : limit_z_urho_large ] ) , while it is constant with frequency for small enough frequencies so that @xmath116 as predicted from the asymptotic eq . ( [ eq : limit_urho_small ] ) . more quantitatively , since the observer in fig . [ fig : inf_vs_fin ] is located at @xmath117 m , the field should behave as @xmath110 for @xmath118 mhz ( applying eq . ( [ eq : u ] ) ) . this is approximately the case as can be seen in fig . [ fig : inf_vs_fin ] . the result of the exact calculation for a track of length @xmath119 km is also shown in figs . [ fig : inf_vs_finzhs ] and [ fig : inf_vs_fin ] . the agreement between both calculations is excellent because the finite track is long compared to observation distance . this confirms that eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) must also account for cherenkov radiation . as can be appreciated in fig . [ fig : inf_vs_fin ] , the exact results for finite and infinite tracks differ for wavelengths larger than the length of the track - frequencies typically below @xmath120 with @xmath121 which gives @xmath122 mhz . we have numerically evaluated the exact expressions for the @xmath30 and @xmath87 components of the electric field in eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) at different frequencies and observer distances , for a single tracks of different lenghts , and for the tracks constituting a shower in ice as obtained in full simulations performed with the zhs monte carlo code @xcite . in this section we present the results of this comparison . the applicability of the zhs expressions in eqs . ( [ eq : rhofield_zhs ] ) and ( [ eq : zfield_zhs ] ) relies on the conditions @xmath123 in section [ sec : zhs ] . ( [ eq : cond3 ] ) and ( [ eq : cond4 ] ) are easily fulfilled for any wavenumber @xmath124 simply by dividing the particle track in a sufficiently large number of subtracks . once this is guaranteed , the zhs expression is applied to every sub - track and the electric field is obtained adding the corresponding contributions . the validity of this procedure will be numerically confirmed below when comparing the exact calculation with the electric field obtained using the zhs algorithm in the manner just described . however the condition @xmath125 , eq . ( [ eq : cond1 ] ) , does not depend on the size of the track and can not be enforced by applying the procedure outlined above . as a consequence @xmath125 is an intrinsic limit to the range of observing frequencies and distances in which the zhs algorithm gives accurate results as will be shown in the following . in fig . [ fig : single_track_1 mm ] we compare the fourier components of the electric field modulus for a single particle track as obtained with the exact calculation , eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) , to that obtained with the zhs algorithm . the length of the track is chosen to be small @xmath126 m ( close to the peak value of the distribution of track lengths in the standard zhs code ) . to test the validity of the zhs algorithm , we have calculated the spectra for observers at different distances ( @xmath60 ) measured with respect to the center of the track and placed at the cherenkov angle . the condition @xmath127 in ice with refractive index @xmath106 can be cast as : @xmath128 for observers at distances @xmath129 and @xmath130 m from the particle track , the condition in eq . ( [ eq : kr_ice ] ) is fulfilled as long as @xmath131 mhz and 1 ghz respectively . the zhs algorithm is expected to reproduce the results of the exact calculation in this range . this can be seen in fig . [ fig : single_track_1 mm ] . the relative difference between the zhs and exact calculations is less than @xmath132 at the frontier of the validity range in the explored frequency and distance space . the accuracy can be however orders of magnitude better . if @xmath133 is enforced for instance the corresponding relative difference is below @xmath134 . this condition is satisfied for @xmath135 m and @xmath136 mhz what corresponds to a range of frequencies and distances typically encountered in experiments that search for neutrino induced radio transients . what is striking is that these conclusions also apply in the `` near field '' provided the track is subdivided in sufficiently small sub - tracks . this can be seen in fig . [ fig : single_track_1 m ] comparing the exact and zhs results for a @xmath119 m track . the same range of validity is obtained because conditions ( [ eq : fraunhofer ] ) , ( [ eq : cond3 ] ) and ( [ eq : cond4 ] ) are guaranteed by reducing the length of the sub - tracks . for fixed frequencies the condition in eq . ( [ eq : kr_ice ] ) only applies at sufficiently large distances @xmath60 to the track . this is illustrated in fig . [ fig : single_track_distance ] for a track of length @xmath119 m. the fourier components at frequencies of @xmath137 , 100 mhz and 1 ghz are in agreement with the exact calculation respectively at @xmath138 , 1 and 0.1 m as expected . since the typical distance between antennas in experiments such as the askaryan radio array ( ara ) @xcite is @xmath139 m , we expect the results of the zhs algorithm obtained through the procedure outlined above , to be accurate enough in most practical situations . it has been questioned whether the zhs algorithm reproduces cherenkov radiation from a single charged particle track @xcite . in figs . [ fig : inf_vs_finzhs ] and [ fig : inf_vs_fin ] the algorithm is shown to be in very good agreement with the exact solution for a 1.2 km track as long as @xmath127 as explained above . the emission from such track is in turn practically equivalent to the cherenkov emission from an infinite track ( also displayed ) and it follows that the zhs calculation must account for cherenkov radiation . as can be seen in fig . [ fig : single_track_1 mm ] ( solid lines ) the exact solution for the modulus of the electric field scales linearly with frequency provided that @xmath140 . as a result the electric field can be approximated with the zhs expressions , eqs . ( [ eq : rhofield_zhs ] ) and ( [ eq : zfield_zhs ] ) . for an observer close to the cherenkov angle as in fig . [ fig : single_track_1 mm ] , the factor @xmath141 and the term in brackets in eqs . ( [ eq : rhofield_zhs ] ) and ( [ eq : zfield_zhs ] ) can be approximated by : @xmath142 \approx \\ & \frac{1 + i(\omega-\mathbf{k}\cdot\mathbf{v})t_2 - 1 - i(\omega-\mathbf{k}\cdot\mathbf{v})t_1}{i(\omega-\mathbf{k}\cdot\mathbf{v } ) } = t_2 - t_1 \end{split}\ ] ] making the dependence of the field with @xmath18 apparent . clearly the field can not grow indefinetely with frequency . this apparent ultraviolet divergence " is only an artifact of considering the unrealistic medium in which the permittivity @xmath143 is constant with frequency . in a physical medium absorption at high frequencies will tame the growth with frequency of the electric field . when @xmath144 the field behaves with frequency as @xmath145 as can be also seen in fig . [ fig : single_track_1 mm ] . in the model of a charged particle at rest for @xmath146 , moving with a speed @xmath147 between @xmath2 and @xmath148 , and becoming again at rest for @xmath149 , the coulomb field dominates at small distances to the track and/or low frequencies . the field can be modeled as : @xmath150 for @xmath151 and @xmath152 for @xmath153 , where @xmath154 and @xmath155 are respectively the distances from the charge to the observer at times @xmath10 and @xmath4 , and @xmath156 and @xmath157 denote the instants of time at which the coulomb field arrives at the observer . the fourier transform of a heaviside function at non - zero frequency is proportional to @xmath145 , and the two coulomb fields interfere coherently at low frequencies , explaining the frequency dependence of the electric field . naturally the zhs algorithm does not reproduce this behavior which is not associated to radiation . the growth of the field at low frequencies is an artifact of not accounting for screening of the field by the atoms in the medium . in fig . [ fig : single_track_distance ] the dependence of the fourier components of the field modulus with distance is shown for several frequencies . at sufficiently large distances to the track the electric field behaves as @xmath158 for all frequencies . this is the radiation zone , the field is expected to behave as @xmath158 as explained in conventional radiation theory @xcite . if the observer is placed at very small distances compared to the length of the track , the situation resembles that of an infinite track . in this case the discussion in section [ sec : cherenkov ] applies . the field behaves as @xmath159 at distances much smaller than the length of the track provided the frequency is high enough to satisfy @xmath115 - see eq . ( [ eq : u ] ) . this can be seen in the curve of the fourier component at @xmath112 ghz for @xmath160 m. at small distances and sufficiently low frequencies , when @xmath116 , the field becomes proportional to @xmath158 and independent of frequency - see eq . ( [ eq : zfield_inf ] ) . this feature can be appreciated in fig . [ fig : single_track_distance ] at distances below 0.1 m for the calculations at 10 and 100 mhz . the zhs algorithm reproduces the calculation provided the @xmath127 condition is satisfied as could be expected , reproducing both the @xmath158 and @xmath159 behaviors . it is possible to test the zhs algorithm in a more realistic situation . in this section we compare the fourier components of the electric field predicted by the zhs algorithm with those obtained using the exact calculation in a full simulation of electromagnetic showers . for this purpose we have applied the exact solutions of the field of a track given in eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) in the zhs monte carlo code @xcite for the simulation of electron and photon - induced showers in ice . the zhs monte carlo calculates the start and end points of small sub - tracks of all charged particles ( electrons and positrons ) in an electromagnetic shower down to a kinetic energy threshold of @xmath161 kev . with these we can calculate the exact electric field produced by each single sub - track and add the fields up accounting for interference between different tracks . since eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) are only valid for a charged particle travelling along the @xmath30 axis ( parallel to the shower axis ) , we perform the necessary rotations of eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) to obtain the field for a particle track moving along an arbitrary direction . simultaneously with the exact calculation , we also obtain the field as predicted by the zhs algorithm for exactly the same shower ( i.e. the same set of tracks and sub - tracks ) . as explained above the subdivisions are such that the conditions in eqs . ( [ eq : fraunhofer ] ) , ( [ eq : cond3 ] ) and ( [ eq : cond4 ] ) are fulfilled for all the sub - tracks in the shower . the result is qualitatively the same as in the case of single tracks . as long as the condition @xmath127 is fulfilled , the zhs algorithm gives an accurate prediction for the fourier components of the electric field with a difference of less than a few percent relative to those obtained with the exact calculation . as can be seen in fig . [ fig : zhs_10tev ] this occurs for distances to the shower axis as small as @xmath162 m and frequencies above @xmath163 mhz , well in the distance and frequency ranges relevant for experiments looking for particle shower induced radio pulses in dense media @xcite . we stress here that the accuracy reported above refers to the approximation of using the zhs formula applied to the standard subdivision of tracks in the zhs code , instead of the exact expression for the radiation emitted by the same particle sub - tracks . by comparing the results obtained in the fraunhofer limit with the standard subdivision of tracks to those obtained with a much finer subdivison , it was determined that the accuracy of the zhs code is @xmath164 at frequencies @xmath165 ghz , improving significantly at lower frequencies . we do not further address this uncertainty in this paper , nor the uncertainty due to the shower simulation itself . it is also worth remarking that in terms of computing time the exact calculation is roughly a factor @xmath165 slower than the calcuation performed with the zhs algorithm . since the zhs algorithm can only be applied in a limited range of frequencies , an accurate representation of the electric field in the time - domain can not be obtained with an inverse fourier transform of eqs . ( [ eq : rhofield_zhs ] ) and ( [ eq : zfield_zhs ] ) . the low frequency components which do not satisfy the condition @xmath127 are not accurately described by the zhs algorithm as shown before . also at very high frequencies the number of steps in which the tracks have to be divided in order to fulfill eqs . ( [ eq : fraunhofer ] ) , ( [ eq : cond3 ] ) and ( [ eq : cond4 ] ) can become prohibitively large from the computational point of view what can compromise the calculation of the pulses below the 100 picosecond scale . in practice however these techniques , although not exact , provide fast and accurate calculations in the region of interest to uhe neutrino detection . for experiments with typical time resolutions of the order of 1 ns , and which are only sensitivity to frequencies from @xmath166 mhz to few ghz , the zhs algorithm has been shown to give a very accurate representation of the fourier components of the electric field ( fig . [ fig : zhs_10tev ] ) . several calculations of the field emitted in showers developing in dense media can be found in the literature . in @xcite the finite diference time domain method is used for calculating the field of a pancake - like shower with a gaussian longitudinal development and gaussian radial profile in the time - domain which is then transformed to the frequency - domain . in @xcite using the saddle - point approximation , an analytic equation for the calculation of the electric field of a charge distribution exhibiting a longitudinal profile with a well - pronounced maximum is derived . the result is factorized into an integral accounting for the longitudinal variation of the charge and a form factor that accounts for the lateral spread of the shower , a procedure revisited in @xcite for realistic showers . assuming a gaussian longitudinal and lateral development for the charge distribution both results were directly compared and turned out to be in good overall agreement as shown in @xcite . minor differences could be attributed to the form factors used . with the exact calculation of the electric field performed in this work , the field due to a gaussian profile can also be obtained and compared to the calculations mentioned before . the electric current for a shower with a gaussian profile is given by eq . ( [ eq : current ] ) with the following replacements : @xmath167 and @xmath168 the shower develops in the longitudinal direction parallel to the @xmath169 coordinate ( shower axis ) , and radially along the @xmath170 and @xmath171 coordinates . @xmath172 is a normalization constant . @xmath173 characterizes the width of the shower along shower axis and @xmath174 the corresponding lateral width . after substituting this current in eq . ( [ eq : vectorpotential ] ) , and following the same steps as in section [ sec : exact ] , the expression for the field is the same as in eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) with the following change : @xmath175 eqs . ( [ eq : rhofield ] ) and ( [ eq : zfield ] ) after the changes in eq . ( [ eq : field_gaussian ] ) can be solved numerically . the zhs algorithm can also be applied to this situation as long as the condition @xmath125 for all distances @xmath176 to the shower . the procedure consists on slicing the volume occupied by the bulk of the shower in small cubes and approximating each as a track with constant charge given by the gaussian distributions in eq . ( [ eq : current_gaussian ] ) . comparison of the result of the exact calculation in this work ( or the zhs algorithm ) with the saddle - point calculation in @xcite requires knowing the form factor @xmath177 for a gaussian profile . @xmath177 is defined in @xcite as : @xmath178 with @xmath179 with @xmath180 the radial position of the observer . also , @xmath181 with @xmath182 . @xmath60 is the distance from the maximum of the shower to the observer . the function @xmath183 represents a normalized charge density of the travelling pancake . assuming a gaussian for @xmath183 of the form : @xmath184 and substituting @xmath183 into eq . ( [ eq : form_factor ] ) , the form factor reads , @xmath185 setting @xmath186 , @xmath187 m , @xmath188 m and @xmath189 m with the refractive index of ice @xmath106 , the electric fields for a gaussian charge profile were calculated using the saddle - point approach as described in @xcite , and compared to the exact formula given in this work . the results are shown in fig . [ fig : comparison ] . since @xmath60 is large the condition @xmath127 is satisfied for frequencies above @xmath190 mhz ( see eq . [ eq : kr_ice ] ) and the agreement between the zhs and the exact calculation presented in this work is very good ( not shown in fig . [ fig : comparison ] for clarity ) . the saddle - point approach is also in very good agreement with both the exact and zhs calculations . the results of the field for a gaussian charge profile are also in good agreement with those obtained in @xcite using the finite difference time domain method . we have obtained the results of an exact calculation of the electric field produced by a charged particle that accelerates instantaneously moves at constant speed and intantaneously deccelerates . the results are used to obtain the approximate expression used in the zhs algorithm to calculate the radio emission from showers in dense media using shower simulations . this allows the precise determination of the conditions necessary for this approximation to be valid , namely , the observer must be in the far field zone @xmath127 and the frauhofer approximation must apply , @xmath191 for each track . the exact and simulated results are compared for tracks in a variety of circumstances to illustrate both the behavior of the fields and the validity of the approximation made in different frequency and distance ranges . we have shown that the range of validity of the expressions can be greatly enlarged by subdividing long charged particle tracks in smaller sub - tracks and adding up the conributions , as it is done in the zhs algorithm . nevertheless the far field condition @xmath127 has been shown to be an intrinsic limit of the zhs algorithm . the zhs algorithm has been tested for long tracks in regions where the fraunhofer approximation does not hold comparing it to the exact solution . the results clearly indicate that the zhs algorithm reproduces the exact behavior provided the far field condition is satisfied and the length of the sub - tracks is small enough for the fraunhofer approximation to be valid . by comparing the emission with the exact emission from an infinite track it is shown that the emission calculated with the zhs algorithm does indeed contain what is conventionally described as cherenkov radiation . the precision of the zhs algorithm is shown to be below the @xmath192 level provided that @xmath193 , corresponding to @xmath135 m and @xmath194 mhz . by enforcing @xmath133 the precision improves to better than @xmath195 , what corresponds to @xmath135 m and @xmath196 mhz , which are conditions met in most experimental arrangements trying to detect radio pulses form neutrinos interacting in dense media . the zhs algorithm is tested for completitude when applied to a shower simulation . this is done comparing the result of the zhs algorithm to that obtained when the zhs expression is replaced by the exact calculation for every charged particle sub - track for exactly the same shower . the results indeed confirm that the accuracies reported for individual sub - tracks are approximately mantained in the final zhs result . finally , in order to compare to alternative calculations the results are compared to the saddle point approximation . this approximation has been used to test solutions using the method of finite differences to solve maxwell s equations directly in the time domain ( fdtd ) using a simplified shower front based on gaussian distributions in the shower plane and in time . the comparison of the saddle point approximation to both the exact solution and the zhs algorithm give compatible results confirming that both approaches reproduce the radiation emitted from showers . the results presented here in summary confirm that the zhs algorithm can be used to describe most practical applications to detect pulses emitted from high energy showers produced in dense media by neutrinos . the approach only begins to show significant discrepancies when the observer is at distances comparable to the lateral dimensions of the shower ( @xmath197 1 m in ice ) . since the typical distance between antennas in experiments such as the askaryan radio array ( ara ) @xcite is @xmath139 m , we expect the results to be accurate enough in most practical situations . the results obtained are also of interest for the application of the zhs algorithm to calculate pulses from eas . indeed the application of the method of track subdivision has been applied in zhaires @xcite , a recent code developed for this purpose , by making track subdivisions that are forced to satisfy the fraunhofer condition . j.a - m , w.r.c . , d.g .- f . and e.z . thank xunta de galicia ( incite09 206 336 pr ) and consellera de educacin ( grupos de referencia competitivos consolider xunta de galicia 2006/51 ) ; ministerio de educacin , cultura y deporte ( fpa 2010 - 18410 , fpa2012 - 39489 and consolider cpan - ingenio 2010 ) ; aspera ( pri - pimasp-2011 - 1154 ) and feder funds , spain . we thank cesga ( centro de supercomputacin de galicia ) for computing resources . part of this research was carried out at the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration . [ hires collaboration ] phys . * 100 * , 101101 ( 2008 ) j. abraham _ [ pierre auger collaboration ] , phys . * 101 * , 061101 ( 2008 ) ; j. abraham _ et al . _ [ pierre auger collaboration ] , physics letters b * 685 * , 239 ( 2010 ) . j. abraham _ [ pierre auger collaboration ] , phys . lett . * 104 * , 091101 ( 2010 ) r.u . [ hires collaboration ] phys . lett . * 104 * , 161101 ( 2010 ) k. kotera , a. olinto , jcap * 10 * , 013 ( 2010 ) . askaryan , _ soviet physics _ jetp * 14,2 * 441443 ( 1962 ) ; * 48 * 988990 ( 1965 ) . jelley , nuovo cimento x * 46 * , 649 ( 1966 ) . h. r. allan , in : j. g. wilson , s. a. wouthuysen ( eds . ) , progress in elementary particle and cosmic ray physics , north holland , ( 1971 ) , p. 169 . d. saltzberg _ et al . _ , phys . rev . lett . * 86 * , 2802 ( 2001 ) . p. miocinovic _ et al . d * 74 * , 043002 ( 2006 ) . d * 72 * , 023002 ( 2005 ) . * 99 * , 171101 ( 2007 ) . e. zas , f. halzen , t. stanev , phys . d * 45 * , 362 ( 1992 ) . i. kravchenko _ _ , astropart . * 19 * , 15 ( 2003 ) ; i. kravchenko _ et al . _ [ rice collaboration ] , phys . d * 85 * , ( 2012 ) 062004 . gorham _ et al . _ [ anita collaboration ] , astropart . phys . * 32 * , ( 2009 ) 10 ; p.w . gorham _ et al . * 103 * , 051103 ( 2009 ) ; p.w . gorham _ et al . _ [ anita collaboration ] , phys . d * 82 * , ( 2010 ) 022004 ; phys . d * 85 * , ( 2012 ) 049901(e ) . t. hankins _ et al . _ mon . not . royal astron . soc . * 283 * , 1027 ( 1996 ) . gorham _ et al . lett . * 93 * , 041101 ( 2004 ) . o. scholten _ astropart . * 26 * , 219 ( 2006 ) . a.r . beresnyak _ _ astronomy reports * 49 * 2 , 127 ( 2005 ) . james _ et al . d * 81 * , 042003 ( 2010 ) . jaeger , r.l . mutel , k.g . gayley , astropart . phys . * 34 * , 293 ( 2010 ) . p. allison _ [ ara collaboration ] astropart . phys . * 35 * , ( 2012 ) 457 s.w . barwick et al . j. phys . 60 , 276 - 286 ( 2007 ) . s. fliescher for the pierre auger collaboration , nucl . a * 662 * , s124 ( 2012 ) d. ardouin _ et al . _ , astropart . phys . * 31 * , 192 ( 2009 ) . s. hoover _ et al . _ [ anita collaboration ] , phys . * 105 * , 151101 ( 2010 ) . gorham , f.e . baginski , p. allison , k.m . liewer , c. miki , b. hill , g.s . varner , astropart . 35 * 242 , ( 2011 ) f. halzen , e. zas , t. stanev , phys . b * 257 * , 432 ( 1991 ) . j. alvarez - muiz , e. zas , phys . b * 411 * , 218 ( 1997 ) . j. alvarez - muiz , e. zas , phys . b * 434 * , 396 ( 1998 ) . j. alvarez - muiz , r.a . vzquez and e. zas , phys . d * 61 * , 023001 ( 1999 ) . j. alvarez - muiz , r.a . vzquez , e. zas , phys . d * 62 * , 063001 ( 2000 ) . s. razzaque _ et al . d * 65 * , 103002 ( 2002 ) ; j. alvarez - muiz , e. marqus , r.a . vzquez , e. zas , phys . d * 67 * , 101303 ( 2003 ) . s. razzaque _ et al . d * 69 * , 047101 ( 2004 ) . s. hussain , d.w . mckay , phys . d * 70 * , 103003 ( 2004 ) . j. alvarez - muiz , e. marqus , r.a . vzquez , e. zas , phys . d * 74 * , 023007 ( 2006 ) . j. alvarez - muiz , c.w . james , r.j . protheroe and e. zas , astropart . 32 * , 100 ( 2009 ) . j. alvarez - muiz , a. romero - wolf , e. zas , phys . d * 81 * , 123009 ( 2010 ) . j. alvarez - muiz , w. rodrigues , m. tueros and e. zas , astroparticle physics * 35 * , 287 ( 2012 ) j. alvarez - muiz , a. romero - wolf , e. zas , phys . d * 84 * , 103003 ( 2011 ) j. alvarez - muiz , w. rodrigues , m. tueros and e. zas , astroparticle physics * 35 * , 325 ( 2012 ) i. tamm , journal of physics * i * , no . 5 - 6 , 439 ( 1939 ) j.d . jackson , _ classical electrodynamics 3rd ed . _ wiley , new york , ( 1998 ) . james , h. falcke , t. huege , and m. ludwig phys . rev . e * 84 * , 056602 ( 2011 ) r.v . buniy , j.p . ralston , phys . d * 65 * , 016003 ( 2002 ) . hu , c .- c . chen , p. chen , astroparticle physics * 35 * , 421 ( 2012 ) . j. alvarez - muiz , g. parente , e. zas , in procs . of the 24@xmath198 icrc , rome ( italy ) , vol . 1 , 1023 ( 1995 ) . afanasiev , _ vavilov - cherenkov and synchrotron radiation : foundations and applications _ , kluwer academic publishers , ( 2004 )

one of the methods for studying extensive air showers of high energies ( @xmath1 ev ) is to register their images in the optical ( mainly fluorescence ) light . this can be done by observing showers from the side in order to avoid the more intense cherenkov light emitted rougly in the shower direction . the observations are made by a number of optical telescopes , each containing a large mirror and a camera with a matrix of photomultipliers ( pmts ) placed at the focus of the optical system ( hires @xcite , the pierre auger observatory @xcite , the telescope array @xcite ) , so that photons arriving from a given direction on the sky are focused on a particular pmt ( pixel ) . photon arrival time can also be measured if the time structure of the pmt signals is recorded . a cosmic ray induced shower produces at a given time a light spot on the camera which moves across it as the shower develops in the atmosphere so that succeeding pmts are being hit . + it would have been ideal if the light producing a shower image had contained the fluorescence photons only . this is because there exists experimental evidence that the number of fluorescence photons induced by a charged electron in the atmosphere is proportional to the energy lost by it for ionisation @xcite . as practically all primary particle energy is eventually used for ionisation , this energy can be determined by measuring the fluorescence light emitted along the shower track in the atmosphere . + however , there are several problems in deriving the flux of the fluorescence light emitted by a shower from that arriving at the detector . firstly , the arriving light contains not only the fluorescence but also cherenkov photons . if the viewing angle ( the angle between the line of sight and the shower direction ) is large ( say , @xmath2 ) then it is mainly the cherenkov light _ scattered _ in the atmosphere region just passed by the shower and observed by the detector ( this light , before being scattered , travels roughly along the directions of the shower particles ) . typically its fraction at the detector is about @xmath3 of the fluorescence flux . for smaller viewing angles it is the cherenkov light _ produced _ at the observed part of the shower that may dominate even the fluorescence signal . the contribution of the cherenkov light , which has to be subtracted from the total signal , has been extensively studied @xcite . + the subject of this paper is another phenomenon , affecting shower images , most commonly called the multiple scattering ( ms ) of light . photons produced at the observed shower element , whatever their origin ( fluorescence or cherenkov ) , may undergo scattering in the atmosphere on their way from the shower to the detector , causing an attenuation of the light flux arriving at the detector and a smearing of the image . this scattering may take place on the air molecules ( rayleigh scattering ) or on larger transparent particles , aerosols ( mie scattering ) . most of the scattered photons change their directions , so that they no longer arrive at the pixel registering the not scattered ( direct ) light . moreover , they arrive later having longer path lengths to pass . on the other hand , photons emitted by the shower at earlier times and scattered somewhere , may fall in the field of view of the pixels just registering the direct photons emitted at a later time . the net effect is that the scattered light forms its own instantaneous image superimposed on that in the direct light . + our aim is to calculate the shower images in the multiply scattered light , so that this effect could be allowed for ( subtracted ) when determining the shower primary energy from the pmt signals . this problem was already studied by roberts @xcite , by us in several , short conference contributions @xcite and more recently by pkala et al @xcite . the approach of the other authors was based on monte carlo simulations of photons emitted by a shower . photons were followed up to 5 - 6 scatterings and their arrival directions and time were registered by the detector . many shower simulations were needed to obtain the ms images for various distances , heights , viewing angles of the observed shower parts . finally , a phenomenological parametrisation of the number of ms photons was made as a function of the parameters found as relevant . + in contrast to roberts and pkala et al this approach is based on an analytical treatment . the main idea is to consider the arriving ms light as a sum of the photons scattered only once ( the first generation ) , of those scattered two times ( the second generation ) and so on and calculate separately the angular and temporal distributions for each generation . + we start ( section 2 ) with a consideration of the simplest situation when a point source of isotropic light flashes for a very short time in an uniform medium . we derive analytical expressions for the angular and temporal distributions of the first and next generations of light arriving at a particular distance from the source . + as our aim is to apply our results to cosmic ray showers we need to consider a non - uniform medium like the atmosphere . assuming an exponential distribution of the gas density and similarly for aerosols we show that an effective scattering length between any two points in the atmosphere can be easily calculated analytically . signals of the first two generations arriving at a particular detector within a given angle @xmath4 to the direction to the source are found as a function of time ( section 3 ) . + using these it is straightforward to derive the corresponding distributions if the source moves across the atmosphere , integrating the point source distributions over changing distance and time of light emission . in section 4 we consider a moving light source , modelling a distant shower . we calculate angular distributions of ms light arriving at a detector at the same time as the direct ( not scattered ) photons emitted by the shower . this particular approach is quite natural because the data from optical detectors consist of the recorded signals by the camera pmts within short time intervals @xmath5 so that one needs to know how much of the ms light has to be subtracted from the main , direct signal . the method for calculating images in the ms light _ simultaneous _ with images in the direct light is relatively simple ( for the first and the second generations ) as it is based on the geometry of the scattered photons in the particular generation . it does not require time - consuming monte - carlo simulations that were done for various shower - observer geometries . making some approximations we derived analytical formula for a shower image produced by the first generation ( section 4.2 ) . our analytical derivations allowed us to choose easily the variables on which and how to parametrise the ms signals . they also made us to realise that the dependence of it on the viewing angle was different for rayleigh and mie scatterings . thus , we introduced a new , simple parametrisation of the fraction of the scattered photons arriving at the telescope within a given viewing cone , depending on the viewing angle of the shower ( what has not been done before ) , separately for the two scatterings . for the same reason we also parametrised separately the second generation ( section 4.3 ) . + a discussion of the results and the implication of the ms effect on the derivation of shower parameters is given in section 5 . the last section ( 6 ) contains a summary and conclusions . at any fixed time a distant shower can be regarded as a point source emitting isotropically fluorescence light ( about cherenkov light see later ) . as explained above , we treat the light scattered in the medium as a sum of consecutive generations consisting of photons scattered only once , twice , and so on , on their way from the source to the observation point . of a sphere with radius @xmath6 . two rays shown are scattered at points @xmath7 and @xmath8 , correspondingly , arriving at the surface of the sphere with radius @xmath6 at angles @xmath9 and @xmath10.,width=288 ] let us consider the first generation , consisting of photons scattered once only . we shall calculate the flux of these photons , @xmath11 , at a distance @xmath6 from the source , such that @xmath12 is the number of photons scattered only once , arriving at time @xmath13 after the flash , within a solid angle @xmath14 at the surface @xmath15 ( perpendicular to the arrival direction ) located at a distance @xmath6 . to do this we shall calculate first the number of photons crossing the sphere of radius @xmath6 ( from inside ) at an angle ( @xmath16 ) with respect to the normal , at time ( @xmath17 ) ; see fig . the average number of photons , per one photon emitted , interacting at a distance ( @xmath18 ) from the source and scattered at an angle @xmath19 within @xmath20 , equals @xmath21 where @xmath22 is the mean scattering path length , @xmath23 is the probability that , once the scattering has occured , the scattering angle is @xmath19 within . to each pair of variables @xmath24 there corresponds another pair @xmath25 related to the former by @xmath26 and @xmath27 where @xmath28 and @xmath29 is the speed of light . + the jacobian of the transformation gives @xmath30 thus , we obtain @xmath31 finally , the number of photons arriving at a unit surface at an angle @xmath32 ( all azimuths ) at time @xmath13 equals @xmath33 and @xmath34 for the rayleigh scattering we have @xmath35 expressing @xmath19 as a function of @xmath36 and @xmath37 ( eq . 2 ) we obtain @xmath38\ ] ] where @xmath39 . thus , the flux @xmath40 of the first generation , defined above , equals @xmath41 where @xmath42 is the mean free path length for the rayleigh scattering . however , @xmath22 in the exponent depends on all the scattering processes active . in general it is determined by @xmath43 for @xmath44 processes . thus , if both molecular and aerosol scatterings are active but one wants to calculate the flux of photons scattered by the rayleigh process only , @xmath22 in the exponent equals @xmath45 but in the denominator one has @xmath42 . + from eq . 10 one can find the number of photons @xmath46 arriving per unit time at a unit surface within a given angle @xmath4 , as a function of time . + we have @xmath47 the integral can be found analytically , giving the result @xmath48\cdot\frac{y_2^{i+1}-y_1^{i+1}}{i+1}+ \nonumber \\ & & + \big [ ( \tau^2 + 1)a_0-a_{-1 } \big ] ln \frac{y_2}{y_1}-a_2 \frac{y_2 ^ 3-y_1 ^ 3}{3 } \bigg)\end{aligned}\ ] ] where @xmath49 , @xmath50 , @xmath51 , @xmath52 , @xmath53 , @xmath54 , @xmath55 . + we have also calculated analytically a similar distribution @xmath56 if the scattering was isotropic . i.e. if @xmath57 ( appendix a ) . + one can also find analytically the angular distribution @xmath58 of the arriving light ( integrated over time ) , but for small angles only ( appendix b ) . the result is @xmath59 \bigg ) \nonumber\\ \,\end{aligned}\ ] ] where @xmath60 , @xmath61 , if there is no mie scattering and @xmath62 is the euler constant . + the ratio of all photons arriving within a small angle @xmath36 to those not scattered @xmath63 , equals @xmath64 where terms @xmath65 have been neglected . in the next paragraph we shall consider the scattering of light emitted by showers developing in the real atmosphere , i.e. with the density depending on height . in addition to the rayleigh process one has to take into account the mie scattering occurring on particles ( aerosols ) larger that the light wavelength . the mie angular distribution is concentrated at rather small angles , in contrast to the rayleigh case . moreover , in the deeper parts of the atmosphere the mean free path length for the mie scattering may be comparable to that for rayleigh , so that it is necessary to calculate the distribution of the light scattered by the mie process only . + as before , we start with a simpler case - a uniform medium . the angular distribution of light scattered on particles with sizes larger than the light wavelength depends on the distribution of the sizes and is not a well known function . roberts @xcite adopts a function of the form @xmath66 here , however , we prefer an expression allowing us to perform some integrations analytically . most crucial is to have the number of numerical integrations for the second generation as few as possible . we shall see that to find @xmath67 for the rayleigh scattering ( section 2.3 ) there is only one integration ( over @xmath68 ) to be done numerically since the form of @xmath69 enables one to integrate analytically over @xmath70 and @xmath71 ( eq . 25 and 26 ) . thus , we adopt the following form for the mie angular distribution : @xmath72 where @xmath73 , @xmath74 , @xmath75 . + this function is normalised as follows @xmath76 it describes quite reasonably the distribution used by roberts . + from eq . 6 and 7 we have @xmath77}{2\pi \lambda_m r^2(\tau^2 - 2\tau\,cos\theta+1)}\ ] ] where @xmath78 if @xmath79 + and @xmath80 if @xmath81 + since @xmath82 we obtain @xmath83\ ] ] where @xmath84 , and @xmath85 is determined as before . in principle , it is possible to find analytically the number of photons @xmath86 arriving within an angle @xmath4 after time @xmath87 per unit time . however , each of the nine integrals @xmath88 contains many terms itself , so that an analytical dependence on @xmath4 and/or @xmath37 would be practically lost . thus , we have found @xmath89 by integrating the flux ( eq . 21 ) numerically ( similarly to eq . + finally , the total flux of the first generation is the sum of the two fluxes arising from the two active mechanisms of the scattering @xmath90 and the detector is the surface of the sphere with radius @xmath6 . the picture shows the last scattering.,width=288 ] these are the photons scattered exactly two times . we shall consider first a general case when there are more than one scattering processes ( as rayleigh and mie ) . let us call the process of the first scattering as @xmath91 and this of the second one as @xmath92 . both @xmath91 and @xmath92 can be either rayleigh or mie . we denote their mean scattering path lengths by @xmath93 and @xmath94 , and the angular distribution functions of the scattering by @xmath95 and @xmath96 , correspondingly . the light source flashes isotropically at the centre of a sphere with radius @xmath6 at time @xmath97 ( fig . 2 ) . as before we want to calculate the number of photons crossing the surface of the sphere from inside at a given angle @xmath36 , at time @xmath87 , per unit time . + let us consider the photons scattered for the second time at a distance @xmath98 from the source . the number of photons incident on a small surface @xmath99 at an angle @xmath71 ( within @xmath100 ) at time @xmath101 and scattered within a spherical shell of thickness @xmath102 by an angle @xmath103 ( within @xmath104 ) equals @xmath105 where @xmath106 . as now both processes are active the meaning of @xmath22 in the factor @xmath107 in the expression for @xmath108 is the effective mean path length for both processes . + the direction of the scattered photons is at an angle @xmath19 to the radius of the sphere and @xmath109 , where @xmath110 is the azimuth of the photons scattered for the second time . for any given direction @xmath111 before and @xmath112 after the second scattering , the scattering angle @xmath103 fulfils the relation @xmath113 . the only function depending on the azimuth angle @xmath70 of the incident photons is @xmath114 . denoting @xmath115 and integrating ( 25 ) over @xmath71 we obtain for the number of photons incident on @xmath99 and scattered within @xmath102 into the solid angle @xmath20 the following expression @xmath116 where the function @xmath117 is defined by the integral in the l.h.s . of ( 26 ) . + the pair of fixed variables @xmath98 and @xmath19 defines another pair @xmath68 and @xmath36 , where @xmath68 is the photon path length after the second scattering . the jacobian of the transformation gives the relation @xmath118 putting @xmath119 , the contribution of photons scattered for the second time at a distance @xmath120 to arrive at an angle @xmath32 at the sphere with radius @xmath6 equals @xmath121\cdot } \nonumber \\ & & \cdot 4\pi r^2\cdot 2\pi sin\theta\,|cos\theta| \ , \ , dx'd\theta dt\end{aligned}\ ] ] the factor @xmath122 multiplied by @xmath123 in the expression for @xmath108 gives @xmath124 , independent of @xmath68 . integration over @xmath68 gives the total number of the above photons @xmath125 where @xmath126 . + the maximum value of @xmath68 results from fixing time @xmath87 . we have that @xmath127 thus , the flux of the second generation @xmath128 equals @xmath129 and , integrated over @xmath36 for @xmath130 , gives @xmath131 . with the rayleigh and mie processes active we must take into account all four cases @xmath132 or @xmath133 and @xmath134 or @xmath133 . finally , the flux of the combined second generation photons is a sum of all specific fluxes @xmath135 some of the integrals defined in this section can be found as analytical functions ( appendix c ) . it is of some importance when calculating higher generations ( see the next section ) . any next generation of the scattered photons can be calculated in the same way as the second one has been found from the previous one ( the first ) . to calculate the flux @xmath136 of the @xmath137 generation , given @xmath138 we proceed as before when calculating the second generation from the first one ( eq . the number of photons , incident on @xmath99 at an angle @xmath139 within @xmath140 at time @xmath141 and scattered along @xmath142 into @xmath143 equals : @xmath144 if there are two scattering processes , @xmath145 , then @xmath146 is the scattering probability by an angle @xmath147 by any process per unit distance per unit solid angle . the rest of the derivation of @xmath148 is the same as in the previous section . however , for each next generation the number of numerical integrations increases , unless values of @xmath149 are stored as a 3-dimension matrix . thus , it is convenient to find analytical solutions of the integrals @xmath150 and/or @xmath151 , if possible . fig . 3a shows the number of photons arriving at the detector within an angle @xmath152 per unit area per unit @xmath28 as a function of @xmath153 . the upper curves refer to the first generation , the lower - to the second one . we also compare here the time distributions obtained for the rayleigh with those for the isotropic scattering . the distance detector- source equals to one scattering length ( @xmath154 ) . + first of all we notice that for short times ( @xmath155 ) the first generation dominates over the second one , and ( as we can guess ) over the higher ones . it can be seen from the formulae for the isotropic scattering ( appendix a ) that the ratio @xmath156 for any given time should be proportional to @xmath157 , so that the importance of the second ( and the higher ) generation will be bigger for larger @xmath158 . + we can also see that the number of photons arriving within an opening angle @xmath4 reaches the dependence @xmath159 only at later times . this reflects the fact that the initial angular distribution of light is steep and becomes almost flat at times @xmath160 or so . when comparing the rayleigh curves with the isotropic ones one can see that the latter are slightly flatter for shorter times , as might be expected but become parallel to the former for later times . + next , we consider a situation when there are two scattering processes , rayleigh and mie with quite different angular distributions @xmath161 ( as discussed before ) . we adopt @xmath162 and @xmath163 for simplicity . + the result for the first and the second generation depending on time is shown in fig . 3b . there is now more light at earlier times than in the previous case ( fig . 3a ) due to the strong mie scattering in the forward directions . however , the flux of the first generation decreases about 3 times quicker over the considered time region . although the ratio of the second to the first generation is practically the same at @xmath164 in both cases @xmath165 . the importance of the second one is reached at later times when the mie scattering is present . , shown by numbers , looking from the detector ( at @xmath166 , @xmath167 ) to a point on the line , for @xmath168 and @xmath169 at the ground and the scale heights @xmath170 and @xmath171.,height=326 ] now we shall study the situation when a point light source flashes in a non - uniform medium , such as the atmosphere . we assume that the atmosphere is composed of two sorts of matter , molecules and aerosols , each having its density decreasing with height exponentially with a different scale heights , @xmath172 - for molecules and @xmath173 for aerosols , and having the corresponding mean path lengths for scattering at the ground @xmath174 and @xmath175 . it is not difficult to derive that the effective mean free path for a scattering for light travelling between two arbitrary points @xmath176 and @xmath177 equals @xmath178 where @xmath179 and @xmath180 are the heights of points @xmath176 and @xmath177 above the level , where the rayleigh and mie scattering path lengths are correspondingly @xmath174 and @xmath175 . if the source ( point @xmath176 ) is at a distance @xmath6 from the detector ( point @xmath181 on the ground ) then the ratio @xmath158 , of @xmath6 to the mean free path length along @xmath182 equals @xmath183 \nonumber\end{aligned}\ ] ] it can be seen that increasing the distance @xmath6 to infinity ( keeping @xmath184 constant ) the ratio @xmath158 reaches its maximum finite value @xmath185 this situation is illustrated in fig . a vertical cross - section of the atmosphere is shown . detector is at @xmath186 and lines represent constant values of @xmath158 corresponding to the straight path from the detector to the point on the line . we have adopted the following values : @xmath187 , @xmath188 and @xmath189 . these values describe approximately the atmospheric conditions at the pierre auger observatory @xcite . + from fig . 4 one can also deduce that relevant values of @xmath158 , if light sources are at distances @xmath190 ( as extensive air showers seen by auger ) are @xmath191 . thus , this will be the region of our interest . , detector at @xmath181 . scattering takes place at @xmath177.,width=288 ] after time @xmath192 . the corresponding numbers are equal to @xmath193 . light source is at @xmath176 , detector at @xmath181 . , width=288 ] as the medium is non - uniform , we can not use the idea of a sphere to be crossed by the scattered photons , as in section 2 . now their flux will depend on the zenith angle @xmath184 of the source and on the azimuth angle @xmath110 around the direction towards it . 5 shows a trajectory @xmath194 of a first generation photon scattered at @xmath177 . we want to calculate the angular distribution of the first generation as a function of time , for a fixed @xmath6 and @xmath184 , @xmath195 , crossing a unit area perpendicular to the direction towards the source . + we notice that for a fixed arrival direction of photons @xmath196 and time @xmath87 , the scattering point @xmath177 is uniquely determined . to arrive at the detector at angles @xmath197 within @xmath198 after time @xmath13 , photons have to cross the surface @xmath199 ( shaded in the figure ) and be scattered along a path length @xmath142 by the angle @xmath19 determined by eq.2 . the number of such photons equals @xmath200 where @xmath201 , @xmath202 , @xmath203 is the angle between the normal to the surface @xmath199 and the direction of the incident photons @xmath204 , @xmath205 is the solid angle determined by the unit area at @xmath181 and the scattering point @xmath177 ( @xmath206 ) . there is no need to calculate @xmath203 because @xmath207 , so that it cancels out . it can be shown that @xmath208 inserting this into ( 40 ) we obtain @xmath209 one can see that this formula is practically the same as ( 7 ) for the uniform medium , the only difference being in the scattering path lengths depending not only on distances but also on the geometry . + the height of the scattering point @xmath177 necessary to calculate @xmath210 and @xmath211 equals : @xmath212 we calculate numerically the time distributions of light @xmath213 , arriving at the detector at angles smaller than @xmath4 for different zenith angles of the source . the results , in the form of the ratio : @xmath214 are presented in figs . 7 , 8 and 9 , where @xmath215 are the distributions obtained in the previous section for a uniform medium . + in real atmosphere to that in uniform medium @xmath215 as a function of time ( @xmath216 ) for the rayleigh scattering only for various values of zenith angle of the source . solid lines - @xmath217 , dashed lines - @xmath218 , @xmath219.,height=288 ] . , height=288 ] . each group of lines is for @xmath220 , and @xmath221 from bottom to top ( at @xmath164 ) . , height=288 ] to compare the light flux obtained for the real atmosphere with that for a uniform medium we adopt the same value of @xmath158 and the same distance from the source to detector for both cases . to understand the effect of a purely exponential atmosphere we start with considering only the rayleigh scattering ( fig . 7 ) , neglecting mie ( @xmath222 ) . understanding the behaviour of the curves in this figure is easier with the help of fig . 6 . each ellipse is a cross - section of a rotational ellipsoid with the symmetry axis determined by point @xmath181 - the detector and point @xmath176 - the light source . these are the focal points of all the ellipses . photons arriving at the detector at @xmath181 after some fixed time @xmath223 ( in units of the distance pd ) must have been scattered on the surface of an ellipsoid with the eccentricity @xmath224 equal @xmath225 the ellipses refer to @xmath226 and @xmath227 keeping the right proportions . the detector field of view cuts only a part of the ellipsoid surface where the photons registered after time @xmath228 must have been scattered . we notice that all ratios @xmath229 in fig . 7 are smaller than 1 and decrease with time ( although those for @xmath217 are practically constant ) and ratios for @xmath218 are larger than those for @xmath217 . all this becomes clear when inspecting fig . 6 and the corresponding scattering sites . for example - the constancy of @xmath230 ) for times @xmath231 reflects the fact that the scattering takes place very close to point @xmath176 during all this time and starts to move away from it ( higher in the atmosphere ) only for larger times i.e @xmath232 . since the scattering path length in the uniform medium is chosen equal to the effective path length in the atmosphere @xmath233 ( eq . 36 ) , we have that @xmath234 and the scattering probability at @xmath176 is smaller in the exponential atmosphere . + let us consider now a more realistic atmosphere with both processes , rayleigh and mie at work . fig . 8 shows time dependence of the ratio @xmath235 for @xmath217 and @xmath236 and zenith angles let us take a closer look at the case @xmath217 and @xmath238 ( upper solid curve ) . it may seem strange that the ratio @xmath229 increases since in the case of the rayleigh scattering only it decreases , although very slowly . we have checked that a similar slow decrease takes place if the scattering is only mie . the behaviour of the curves in this figure would be hard to understand without the presentation of the scattering sites in fig . it can be seen that for @xmath193 smaller than a few @xmath239 the scattering sites are close to the source at @xmath176 . thus , we may approximate the ratio @xmath229 as follows @xmath240 where @xmath241 are some effective angular distributions of photons scattered by rayleigh ( mie ) on the cut surface . as time increases , none of the @xmath242 changes much . however , the typical scattering angles increase , what affects much more @xmath243 than @xmath244 , so that @xmath245 decreases . since @xmath246 the numerator decreases by a smaller factor than the denominator so that the ratio @xmath229 increases . it can be seen from fig . 6 that for @xmath247 the scattering angles of the registered photons do not change much ( the denominator stays constant ) but now @xmath248 and @xmath249 have to be substituted by @xmath250 and @xmath251 , where @xmath177 is an effective scattering point with growing height . as @xmath252 , the ratio @xmath253 decreases and so does @xmath229 . + a different behaviour of @xmath254 can be explained also with the help of fig . at first @xmath229 decreases ( the smaller @xmath184 - the stronger decrease ) because the detector field of view cuts out a growing part of the deep atmosphere where the scattering is strong in the real atmosphere . at @xmath255 @xmath229 starts to increase for the same reason as just described in the case @xmath217 . it must finally decrease since the scattering takes places further and further behind the source , where @xmath256 and @xmath257 are growing in the real atmosphere . 9 we show @xmath258 for @xmath259 for several intermediate values of @xmath4 , and for two values @xmath260 and @xmath261 . note that changing @xmath262 must result in changing @xmath6 - the distance to the source . this is the main reason why the curves for @xmath263 are lower than those for @xmath260 . it can be seen from fig . 5 that for @xmath259 ( the scales on both axes are the same so in the figure the angles are correctly represented ) the distance @xmath6 is much shorter for @xmath260 than for @xmath263 , implying that the corresponding heights of the source differ considerably . inspecting fig . 6 we can estimate that for @xmath264 and @xmath217 the curves for @xmath263 should be down with respect to those for @xmath260 by a factor @xmath265 where @xmath266 and @xmath267 are positions of the source referring to @xmath260 and 1 respectively , with the meaning of all @xmath242 as defined before . in our example @xmath268 , @xmath269 , @xmath270 , @xmath271 so that the above factor equals @xmath272 whereas the exact ratio from fig . 9 @xmath273 . to the first generation @xmath274 as a function of time ( @xmath216 ) from a flash ( at @xmath97 ) of a point source at zenith angle @xmath238 in the real atmosphere . fluxes are integrated within @xmath217 ( solid lines ) and @xmath218 ( dashed lines ) . three curves for each @xmath4 refer to @xmath275 ( from bottom to top ) . dotted line refers to a uniform medium with rayleigh scattering only , for @xmath263 , @xmath217.,height=288 ] we proceed like in the case of the first generation ( fig . 5 ) , but now point @xmath177 refers to the second scattering . photons scattered only once arrive at the surface @xmath199 from all directions according to @xmath276 , where @xmath277 . thus , the number of photons @xmath278 incident on @xmath199 and scattered for the second time towards the detector ( to arrive there within @xmath279 after time @xmath13 equals @xmath280 where the integration has to be done over full solid angle @xmath281 , @xmath282 is the path length for the second scattering to occur ( @xmath283 ) and @xmath284 is the angle of the second scattering . now the pair of variables , @xmath36 and @xmath87 , does not determine uniquely the positions of the second scattering , since the times @xmath285 elapsed from photon emission to their arrival at @xmath177 ( or strictly speaking , at @xmath199 ) have some distribution . however , @xmath285 can not be smaller than @xmath286 , thus @xmath287 . expressing @xmath288 as a function of @xmath36 , @xmath87 and @xmath6 only we obtain @xmath289 and the distribution of the second generation equals @xmath290 finally , the number of photons within an angle @xmath4 equals @xmath291 the above integrals have been calculated numerically . + the ratio @xmath292 as a function of @xmath293 in the real atmosphere , where the source is at @xmath259 at the distance @xmath294 corresponding to @xmath263 , within two opening angles of the detector @xmath217 and @xmath295 , is presented in fig . 10 . for comparison we have also drawn there the same ratio for the uniform medium , for @xmath263 with rayleigh scattering active only ( following from fig . 3a for @xmath217 ) . first we see that the contribution of the second generation increases with time as it should be expected . next , it is considerably larger for the uniform medium than for the real atmosphere , which may not be so obvious at first sight . in fact , at @xmath296 the curve for the uniform medium is @xmath297 times higher than that for the real atmosphere . this factor equals ( for simplicity we neglect the operators @xmath298 ) @xmath299 the numerator @xmath300 as explained in the previous paragraph . the denominator refers to photons scattered exactly two times . for small @xmath193 the scatterings must take place close to the source or along the field of view ( @xmath217 in our example ) . thus we should have that @xmath301 our calculations show that @xmath302 for @xmath303 few @xmath239 so that @xmath304 , in agreement with the exact calculations . so far we have been interested in the light signals from a flash of a stationary point source in a detector at a given distance , within a given field of view , as a function of time . + now , we shall consider a moving point emitting light on its way through the atmosphere . this is a reasonable model of a distant cosmic ray shower exciting atmosphere what results in emitting isotropic fluorescence light . moreover , the cherenkov light produced by @xmath305 of shower electrons @xcite , propagating alongside the particles will be scattered to the sides by the rayleigh and mie processes in an anisotropic way . thus , in principle , a shower is a moving source , emitting light anisotropically . our aim is to find an instantaneous optical image of the shower , produced by the scattered light at _ the same time _ as the shower image in the direct ( not scattered ) light . + the image obtained for a small integration time @xmath5 is a radially symmetric light spot , corresponding ( roughly ) to the emission distribution at the shower lateral cross - section @xcite . it is produced mainly by the direct light . but at the same time as the direct light , there arrive also photons produced by shower particles at an earlier stage of shower evolution and scattered in the atmosphere . as shower parameters follow straightforwardly from the amount of the direct light , it is important to calculate this effect and to take it into account when deriving the flux of the direct light . + let us assume that we know the geometry and the timing of the shower . if the source is at point @xmath306 at time @xmath87 and emits @xmath307 photons per unit length , the telescope will record the following number of the direct photons at time @xmath308 , in the time bin @xmath5 : @xmath309 where @xmath310 is the angle between the shower direction and the direction from point @xmath306 to the detector @xmath181 , @xmath311 is the effective mean free path for attenuation along the distance @xmath6 , @xmath91 is the diaphragm area determining the collection solid angle and @xmath312 is the angular distribution of the emitted light . + in the case of isotropic fluorescence light @xmath313 and @xmath307 equals to the number of photons produced per unit length ( @xmath314 , per one electron ) . in the case of the scattered cherenkov photons , or those from a laser , @xmath315 for the rayleigh scattering ( or a corresponding distribution for mie ) and @xmath316 , where @xmath317 is the mean free path ( in length units ) for the process in consideration at point @xmath306 and @xmath318 is the number of cherenkov photons at the observed shower level propagating ( approximately ) in the direction of the shower . the number of the scattered cherenkov photons to the fluorescence ones is typically @xmath319 , and a lot of the former are scattered by rayleigh , what is not far from isotropic . thus , with a good approximation one can treat a shower as a moving isotropic light source . this is what has been adopted in this paper : @xmath320 . + simultaneously with the direct light some photons produced earlier ( above point @xmath306 ) will also arrive at the telescope . however , as we already mentioned , they must have been scattered on their way to the telescope . our earlier calculations @xcite showed that if the distance to the shower is not very much longer than the mean scattering path , then the main contribution to the scattered light is due to the photons scattered only once ( the first generation ) . here , we will calculate the fraction of the total light consisting of the first and the second generations . let us consider photons produced at point @xmath176 , along a shower path element @xmath282 ( fig . the first generation photons arriving at the detector @xmath181 at the same time as the direct photons produced at @xmath306 must have been scattered on the surface of the rotational ellipsoid with the focal points at @xmath181 and @xmath176 , with the @xmath321 line being the rotational symmetry axis of the ellipsoid . + as we have already mentioned , the main interest of the shower experiments is to measure the direct light . thus , we are interested in the light arriving from the direction of point @xmath306 and the region around it . in practice ( as it is in the pierre augere experiment ) the light signal arriving simultaneously with the direct light from point @xmath306 is measured within a certain angle @xmath4 around the line @xmath322 . the cone with the opening angle @xmath4 cuts out on the surface of the ellipsoid a region of our interest ( the shaded surface ) . it is at this part of the ellipsoid that the photons produced at @xmath176 have to be scattered to arrive simultaneously with the direct light produced at @xmath306 , within an angle smaller than @xmath4 with respect to the latter . thus , to find the contribution of the first generation to the direct light one has to integrate the number of the scattered photons over the cutout surface , and then integrate the result over the distance @xmath323 ( the upper part of the shower ) . + , scattered once and arriving at detector @xmath181 within its field of view simultaneously with direct photons produced at @xmath306 . @xmath324.,height=288 ] and scattered on the shaded surface ( inside field of view of detector @xmath181 ) of the ellipsoid arrives at @xmath181 simultaneously with the direct light produced at @xmath306 . , width=288 ] the proportions in fig . 11 have not been preserved . for example the actual distance from point @xmath176 to the detector at @xmath181 is much longer than @xmath325 ( although in principle the integration over @xmath326 should go far up the shower , the distances contributing to the total scattered light are rather close to point @xmath306 ) . also the angle @xmath4 does not need to be larger than a few degrees , but on the figure it is much larger for the sake of clarity . + let us first find the contribution to the first generation from photons produced by the shower at point @xmath176 along a path length element @xmath282 , assuming that @xmath176 lies outside the detector field of view . [ [ a - theta_pzeta - fig.-11 ] ] _ a ) _ @xmath327 _ ( fig . 11 ) _ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath328\end{aligned}\ ] ] where @xmath329 , @xmath330 , @xmath331 , @xmath36 is the arrival angle ( with respect to the direction to the source at @xmath176 ) , and @xmath332 . + the azimuth angle @xmath110 is measured in the plane perpendicular to the axis @xmath182 and @xmath333 refer to points on the shower - detector plane ( fig . 11 plane ) . the length @xmath334 can be found from the triangle @xmath335 : @xmath336 and @xmath337 . [ [ b - theta_pzeta ] ] _ b ) _ @xmath338 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ( fig . + it is clear from the figure that now the integration limits of @xmath36 and @xmath110 are different so that one has to add to the integral in case a ) the following term @xmath339\nonumber\end{aligned}\ ] ] and the lower limit of @xmath36 in the term a ) changes sign , so that @xmath340 . to find the total flux of the first generation arriving at the same time as the direct light from point @xmath306 one has to integrate over @xmath326 the contribution from case b ) from 0 to @xmath341 , ( where @xmath342 ) , and add to it the contribution from a ) integrated from @xmath343 to @xmath344 . ( it turns out that contribution from points with @xmath345 are negligible ) . + if there are two scattering processes the first generation means the sum of the number of photons scattered by rayleigh and those by mie and then @xmath346 , and in the attenuation exponent @xmath347 and @xmath348 are the effective mean free paths for both processes . by making some approximations we have found an analytical solution of the shower image in the first generation light scattered by the rayleigh process . + we shall treat separately light produced within the detector field of view @xmath350 and that outside it , @xmath351 , starting with the former ( fig . as the scattering points lie close to point @xmath306 we assume that @xmath352 . for the same reason we assume that the exponential factor describing light attenuation equals @xmath353 . our main assumption , however , consists in integrating over the symmetric ( with respect to axis @xmath182 ) part of the ellipsoid so that we could solve the integrals over @xmath354 and finally @xmath326 . expressing scattering angle @xmath19 as a function of @xmath36 and @xmath355 ( eq . 2 ) we obtain that the contribution from a shower path element @xmath282 equals @xmath356 where @xmath357 , @xmath358 . + we adopt @xmath359 being the mean of the limiting values of @xmath36 in the exact integration . + in the air shower experiments it is the direct light which should be measured so that the viewing angle @xmath4 is small ( a few degrees ) and so are angles @xmath36 . this leads to @xmath360 . taking all this into account the result of the integration over @xmath36 is @xmath361\frac{dl}{r'^2}\nonumber\end{aligned}\ ] ] where @xmath362 . from geometrical considerations it can be derived that @xmath363d\tau'\simeq } \\ & & \simeq\frac{d\tau'}{r(1-cos\delta ) } = \frac{\zeta dx}{r(1-cos\delta ) } \nonumber\end{aligned}\ ] ] where we have dropped the second term @xmath364 . + to obtain @xmath365 it remains to integrate ( 59 ) over @xmath366 ( including ( 60 ) from @xmath367 to @xmath368 , corresponding to point @xmath176 lying on the edge of the viewing cone ( @xmath369 ) . again , assuming that @xmath370 , it can be derived that @xmath371 ( for @xmath372 ) . the result of the integration is the following @xmath373 where @xmath374 . + must be scattered for the second time ( point @xmath375 ) within detector @xmath181 field of view . first scattering ( @xmath267 ) must occur on the surface of the ellipsoid with focal points at @xmath176 and @xmath375 . direct light from @xmath306 arrives at @xmath181 at the same time : @xmath376.,height=288 ] to calculate the contribution from the outer part of the shower i.e. outside the field of view we keep the previous approximate assumptions . however , since now the scattering surface looks differently ( fig . 11 ) we make another simplifications . we assume that all scattering angles equal @xmath310 so that @xmath377 . next , as @xmath4 is small we assume that all @xmath378 . integration over @xmath379 , the solid angle of the field of view , is then reduced to multiplying the integrand for @xmath378 by @xmath380 . thus we have @xmath381 it can be derived that the denominator in the above expression equals @xmath382 assuming needs to allow for light attenuation along @xmath326 in the integration over @xmath326 . the correction to @xmath383 is a few percent , being negligible for the total @xmath384 . ] that @xmath385 , we obtain @xmath386 since @xmath387 we have @xmath388 and @xmath389 expressing the number of photons @xmath390 as the ratio to that of the direct light ( eq . 54 for @xmath391 ) we finally obtain @xmath392\end{aligned}\ ] ] where @xmath374 ( in radians ) . terms @xmath393 and of higher order have been neglected . we can see that , with the approximations adopted , the ratio @xmath394 is proportional to @xmath395 reflecting the fact ( as we shall see below ) that most scatterings take place close to point @xmath306 since @xmath396 is proportional to the shower path segment seen by the detector . for @xmath397 and the dependence on @xmath310 separates from that on other parameters . it becomes also obvious that it must be different for different angular distributions @xmath161 , of photons at scattering . as @xmath398 is quite different from @xmath399 which is peaked in forward directions , we expect another dependence on @xmath310 for the latter . + while deriving the analytical formula ( 66 ) we have also assumed that the length of the shower segment cut out by the detector field of view is small when compared to the distance @xmath6 . for @xmath370 it is fulfilled for almost all @xmath310 , apart from the case when @xmath310 is close to 0 or @xmath400 . we have calculated analytically the ratio @xmath394 for @xmath401 for @xmath402 . however , as the derivation is lengthy and ( as it will be seen later ) agrees with the exact , numerical results only at @xmath403 , we present here only the final result : @xmath404 these values are marked as stars ( for @xmath403 ) in fig . + we have not tried to calculate @xmath405 because the function @xmath399 is not well known , depending on the sizes of the aerosol particles and , anyway , our analytical approach is approximate . thus , it seems better to calculate it exactly numerically and , being led by our analytical solutions for rayleigh , find an appropriate parametrisation of the numerical results . the ratio of the first generation to the direct light arriving simultaneously is obtained by numerical integration of point contributions over @xmath326 ( shower track above point @xmath306 ) and dividing the result by @xmath408 . 14 shows the ratio for the rayleigh a ) and mie b ) scattering as a function of @xmath409 for @xmath410 and @xmath411 and for several different distances @xmath6 . it can be seen that for the rayleigh case the ratio is proportional to @xmath412 and @xmath4 , as it has been derived analytically . for mie the dependence follows a power law , with the indices depending slightly on @xmath4 . however , in each case there is practically no dependence on the distance @xmath6 itself . these results are for @xmath413 . + the dependence on @xmath310 is shown in fig . 15a , b for @xmath414 1 and @xmath415 and @xmath411 . one can see the difference of the behaviour of this dependence between rayleigh and mie . in the figure a comparison of our analytical calculations with the exact ones is also shown . the biggest difference reaches some @xmath416 at large @xmath310 . there is no normalisation there . we find this agreement quite satisfactory . + nevertheless , as the agreement is not perfect we have parametrised the exact numerical results . for the rayleigh scattering the ratio @xmath417 can be expressed as @xmath418 where @xmath419 , @xmath4 is in degrees , and @xmath420 for mie the factorisation is not as complete as for rayleigh , and our fit to the numerical results is the following : @xmath421 @xmath422 these fits are also shown in fig . 14 and 15 . it is seen that they do not deviate from the exact values by more than a few percent . + the second generation has been calculated numerically . 13 should be helpful for understanding the calculation . the thin solid line shows a path of a photon produced at @xmath176 , scattered at points @xmath267 and @xmath375 and arriving at @xmath181 . the point @xmath375 of the second scattering must lie somewhere inside the viewing cone . fixing it at a distance @xmath423 from the detector corresponds to fixing the value of the remaining photon path length @xmath424 . we have that @xmath425 where @xmath426 and @xmath427 . + in order to assure that the photons scattered twice arrive at @xmath181 at the same time as the direct light emitted at @xmath306 , point @xmath267 must lie anywhere on the surface of the ellipsoid with focal points at @xmath176 and @xmath375 . thus , to calculate the number of photons produced at @xmath176 and scattered twice one has to integrate the contributions like in ( 49 ) over the position of the first scattering ( the whole surface of the ellipsoid ) and then integrate the result over the volume of the viewing cone . finally , the integration has to be performed over the distance @xmath428 . + the ratio @xmath407 , including rayleigh and mie scatterings as a function of @xmath429 for @xmath430 and as a function of @xmath310 for @xmath431 is presented in fig various point signs refer to different distances @xmath6 . as in the case of the first generation the ratios depend on @xmath432 rather than @xmath6 itself . however , now they are proportional to the second power of it . we have parametrise the obtained numerical results as follows @xmath433 as we see the numbers of the scattered photons , @xmath435 and @xmath436 , depend on the angle @xmath4 within which they arrive at the detector . this angle has to be chosen in such a way as to encompass the total direct signal from the shower . it is obvious that for closer showers @xmath4 has to be larger and for those more distant - smaller , so that @xmath396 - meaning the lateral spread of the direct light - should remain constant . our approximations show that in this case ( @xmath437 ) @xmath438 depends only on height ( by @xmath317 ) and @xmath310 . the power indices in the dependence of @xmath406 on @xmath6 and @xmath4 are close to @xmath261 , whereas @xmath439 ( roughly ) , so that their dependence on the distance @xmath6 itself should be weak . 17 presents the contributions of the individual components of the scattered light as a function of height , for @xmath440 , @xmath430 and two ( quite different ) values of @xmath6 : 12 and 36 @xmath441 . the dependence of the sum @xmath434 on @xmath6 is very weak , indeed , the more so as the ratios @xmath442 and @xmath407 change in opposite directions . the chosen value @xmath443 corresponds to a shower lateral radius of @xmath444 , equal to @xmath445 ( molire radii ) . this may seem too large to contain the total fluorescence signal . however , deep in the atmosphere it is the shower cherenkov light what determines its lateral dimensions and it goes up to @xmath446 @xcite . + as the typical viewing angles @xmath310 are not far from @xmath447 one can draw a conclusion that the maximum contribution of the ms light is @xmath448 for @xmath449 and the adopted atmospheric parameters . , dashed lines @xmath452 . curves denoted by `` rayleigh '' and `` mie '' refer to the first generation , by `` n2 '' - to the second one . if a constant lateral dimension is chosen ( @xmath453 , @xmath430 ) the ratios do not ( practically ) depend on shower distance @xmath6 . , height=288 ] let us estimate how not allowing for the ms effect would affect a reconstructed value of a shower primary energy @xmath450 . the total signal would then be treated as the direct light , therefore the primary energy would be overestimated . to estimate how much it would be let us consider a typical ( for auger ) shower with @xmath454 ev , falling to the ground 25@xmath441 away from the light detector , at @xmath455 and azimuthal angle @xmath456 ( angle between the direction from the detector to the shower core and the projection of the shower axis on the ground ) . if the primary is a proton the mean depth of its maximum equals @xmath457 , corresponding to the height above auger level ( @xmath458 ) @xmath459 . from fig . 17 we see that the contribution of the scattered light is @xmath460 if @xmath443 , what corresponds to @xmath461 . however , in the auger telescopes the individual pixel has a diameter of @xmath462 , so that it would be necessary to choose also the neighbouring pixels around the one with the stronger signal to collect the total direct light . thus , we choose @xmath463 . within this angle the ms light grows to @xmath464 , overestimating the number of particles at @xmath451 by this value . for shower parts near the ground , where shower age equals 1.25 , this number grows to @xmath465 . to find the shower primary energy @xmath450 ( by integrating the energy deposit over depth @xmath466 ) it is necessary to extrapolate the shower for larger ages , so that the overestimation from these parts would be even larger . we estimate that not allowing for the ms effect would increase the reconstructed @xmath450 by @xmath0 . + not allowing for the ms light would also lead to an overestimation of the depth of the shower maximum @xmath451 . fitting a new cascade curve to that from the previous example with the increased values according to the ms contribution , we obtain that the overestimation equals @xmath467 . with the difference between the depths of an average proton and iron showers being @xmath468 , this is not a negligible number for drawing conclusions about the mass composition of the highest energy cosmic rays . at first sight it may seem odd that it is only the first two generations that contribute to the scattered light . one might expect that the main contribution should come from the generation with the number closest to the value @xmath469 , according to the poisson distribution ( although we have shown that the first two generations scale with @xmath470 rather than with @xmath469 ) . for @xmath471 and @xmath472 and a shower close to the ground we have @xmath473 . thus , if the distance @xmath474 then @xmath475 , so should one expect the third and the fourth generations to contribute even more than the first one ? the answer is no , since the scattered photons should fall into the detector field of view , so that it is not the path length of a scattering _ by any angle _ ( what @xmath311 means ) that counts but by some _ particular values _ of it . this decreases the probability of such a scattering and can be viewed as an increase of the ( actually ) effective mean free path length @xmath476 for the process _ we are interested in _ ( i.e. photons arriving at the detector ) . one could estimate its value from the ratio @xmath477 or @xmath478 adopting for this aim a poisson distribution of the number of contributing generations . the two obtained values differ typically by 30 - 60@xmath479 , so that an estimation of the share of the remaining generations on the basis of this uncertain @xmath476 is uncertain as well . however , let us take an example : @xmath480 , elevation angle @xmath481 , @xmath482 , what are typical conditions for registered showers in the auger observatory . the two values of the mean number of ( effective ) scatterings are @xmath483 ( from @xmath477 ) and @xmath484 ( from @xmath478 ) . assuming the larger one , one gets @xmath485 for a fraction of the third and higher generations . + however , as our earlier work for a uniform medium @xciteb ) shows , where the third generation was calculated , its share was two times smaller than that deduced from the poisson model and the ratios @xmath477 and @xmath478 , so that this estimation may be too high . thus , we estimate that the contribution of the higher number generations should be small , at least for the conditions of auger . integrated within angle @xmath4 , for shower 24 @xmath441 away from the detector , at elevation angle @xmath221 . our results ( first and second generation ) are represented by thick solid lines for @xmath486 and thin solid lines - for @xmath487 and @xmath488 ( second generation ) . parametrisation from monte - carlo works : roberts - dashed line and pkala et al - dotted line.,height=288 ] the problem of the influence of the scattered light on the shower image in the direct light was first undertaken by roberts @xcite . he treated it by monte - carlo simulations of tracks of individual photons emitted by vertical showers . such shower ( although , strictly speaking , non - existing ) enabled the author to introduce a detector in a form of a ring around the shower as its symmetry axis , increasing dramatically the number of registered photons . although the author appreciate that the singly scattered photons dominated the total ms signal , his simulations went to more than five scatters , including both rayleigh and mie . he obtained a parametrisation of the ratio of the ms signal to the total signal ( ms + direct ) in the form @xmath489 where @xmath490 is the total optical depth ( molecular and aerosol ) between the detector and the light source , @xmath19 - total scattering coefficient ( in @xmath491 ) at the source . in our terms @xmath492 , @xmath493 , where @xmath311 is determined by formula ( 36 ) and @xmath317 depends only on the height in the stable atmosphere , so that ( 74 ) would mean the following @xmath494 this parametrisation differs from those obtained in the present work although not very much . it has a weaker dependence on the angle @xmath4 , which in our case is as ( about ) @xmath495 . the smaller power index at @xmath496 ( @xmath497 in our case ) is to some extent compensated by the term @xmath498 . + another work , dealing with the present problem was that of pkala et al @xcite . the calculation method was monte - carlo as well . this paper did consider inclined showers so that the cylindrical symmetry , as in the previous case did not apply . instead large `` packages '' of photons were assumed to behave in the same way , according to simulated points and directions of their scattering . after extensive simulations the authors arrived at a parametrisation of the ratio @xmath133 of the ms signal to the direct one as follows @xmath499 with @xmath500 . for a one component exponential atmosphere with the scale height @xmath501 we would have @xmath502 indeed , it is not difficult to check that an effective scale height of the two - component atmosphere used by the authors for @xmath503 is about @xmath504 what approximately equals @xmath501 . thus , their parametrisation would be as @xmath505 it is not far from our derivation for the first generation , where the term @xmath506 does not appear . a comparison of the ratio of the total ms signal to the direct one as a function of @xmath4 , obtained in this study with that of roberts and pkala et al is presented in fig . the agreement is quite good for @xmath507 but for larger @xmath4 the other authors seem to have underestimated a little the effect . previous studies of the influence of the scattering of photons on their way from the shower to the detector on the shower image used monte - carlo simulations . the effect has been called `` the multiple scattering '' since it was not clear how many times the photons arriving simultaneously with the direct light actually scatter . although roberts @xcite appreciated the importance of photons scattered one time only he went on with simulating till several scatterings . + the main idea of this study was to divide the scattered light into separate generations : the first generation being photons scattered exactly one time , the second one two times , and so on . this enabled us to find analytically the angular distributions of photons as a function of time for the first generation which turned out to be the most important one . we also showed that any next generation can be found by some numerical integrations of the previous one . these results were obtained for a point source in a uniform medium as well as in the actual atmosphere ( although more numerical integrations were needed in the latter case ) . by considering separately the first generation it was easy to understand the differences between the two cases . the ratio of the second to the first generation is very small for small times after the light emission which will turn out to be relevant when considering the scattered light from extensive air showers what is a final aim of this study . + in particular , this aim was to calculate the angular distributions of scattered photons arriving simultaneously with the direct ( not scattered ) light emitted by an extensive air shower and find relevant parameters determining these distributions . the solutions for a point source were useful here since a shower could be treated as a series of consecutive ( in space and time ) point sources . for the first and the second generations ( the most important ) we have explained where actually the scattering points took place , so that we were able to integrate over all of them . + we have derived an analytical expression ( although approximate ) for the angular distribution of the first generation photons , scattered by the rayleigh process , fitting quite well the exact numerical distributions . the analytical solutions have led us to a proper choice of the independent variables for a final parametrisation of the numerical results . for example the number of photons in the n - th generation turns out to be proportional to @xmath508 , at least for @xmath509 ( although for mie with some correction ) . knowing the positions of the scatterings we were able to show that there should be a dependence on the viewing angle @xmath310 , not taken into account in the considerations of the previous authors . moreover , this dependence should be different for the rayleigh and the mie scatterings . thus , we parametrised the first generation separately for these two processes . concerning the second generation we combined all four possibilities of the two scatterings ( rr , rm , mr and mm ) to give just one parametrisation since its contribution was rather small . + from our parametrisation it follows that for @xmath510 ( corresponding to a constant observed lateral dimension of a shower ) the contribution of the ms component to the direct light depends practically only on the height of the observed shower element and on the viewing angle ( the latter dependence being rather mild ) . a typical overestimation of the primary shower energy ( if the ms effect was ignored ) would be @xmath0 and of the depth of shower maximum @xmath467 . our analytical and numerical approach gives similar results as the previous monte - carlo work of other authors , although the parametrisation obtained in this paper is different . only for larger opening angles @xmath511 the prediction of the ms contribution obtained here is a bit larger , what would have some implications for a reconstruction of very close showers . + summarising we conclude that the ms effect is not to be neglected while reconstructing parameters of the extensive air showers . to be properly allowed for a good knowledge of the atmospheric conditions is necessary @xcite for determining values of @xmath317 , as the shower traverses the atmosphere . + [ [ acknowledgements ] ] acknowledgements + + + + + + + + + + + + + + + + this work has been financially supported by polish ministry of science and higher education , grant no n n202 200239 . it was stimulated by our participation in the pierre auger collaboration . we thank the collaboration for fruitful discussions . 00 t. abu - zayyad et . al , astroparticle physics 12 ( 1999 ) 121 - 134 the pierre auger collaboration : j. abraham et al , nucl . instrum . meth . a 620 ( 2010 ) 227 - 251 t. nonaka et al , nuclear physics b proc . suppl . 190 ( 2009 ) 26 - 31 the macfly collaboration : p. colin et al , astroparticle physics 30 ( 2009 ) 312 - 317 m. giller et al , j. phys . ( 2004 ) 97 - 105 m. giller et al , j. phys . g : nucl . part . ( 2005 ) 947 - 958 f. nerling at al , astroparticle physics 24 ( 2006 ) 421 m. giller and g. wieczorek , astroparticle physics 31 ( 2009 ) 212 - 219 m. d. roberts , j. phys . g : nucl . part . ( 2005 ) 1291 - 1301 a ) m. giller and a. miakowski , ( 2005 ) proceedings of the 29th int . cosmic ray conference , pune , india + b ) m. giller and a. miakowski , ( 2007 ) proceedings of the 30th int . cosmic ray conference , merida , mexico + c ) m. giller and a. miakowski , ( 2011 ) proceedings of the 32nd int . cosmic ray conference , beijing , china j. pkala et al , nucl . a605 ( 2009 ) 388 - 398 p. sommers , astroparticle physics 3 ( 1995 ) 349 - 360 the pierre auger collaboration : j. abraham et al , astroparticle physics 33 ( 2010 ) 108 - 129 it may be of some value to know how much the fluxes @xmath512 for the rayleigh scattering differ from those when the scattering function @xmath513 , so when the scattering is isotropic . to calculate it we proceed as in section 2 putting @xmath514 in eq . we obtain @xmath515 and @xmath516 \nonumber\end{aligned}\ ] ] the second generation can be obtained by proceeding as in section 2.2 , giving the result @xmath517 where @xmath518 ; @xmath519 and @xmath520 . to find @xmath521 one must integrate the flux over @xmath36 numerically . how does the ratio of the two generations depend on the distance ? we can represent the first generation ( a.2 ) in the form @xmath522 where @xmath523 depends on @xmath87 and @xmath4 only . + the flux of the second generation equals @xmath524 where @xmath525 , @xmath526 , @xmath527 . + we see that the integral depends on @xmath37 and @xmath36 only so that @xmath528 thus @xmath529 so that for a given time the ratio is proportional to the distance @xmath6 measured in units of @xmath22 . it can be verified that this relation holds for any scattering process in a uniform medium , not only for isotropy . the formula for the angular distribution of the first generation integrated over time , can be rewritten in the following form @xmath531f(\beta+\theta)d\beta \nonumber\end{aligned}\ ] ] where @xmath532 is the angular distribution of the scattering angle @xmath19 , normalised to @xmath261 when integrated over the full solid angle . + for the rayleigh scattering we obtain @xmath533\big[1+cos^2(\beta+\theta)\big ] d\beta \nonumber\end{aligned}\ ] ] to find the above integral for @xmath60 we divide the integration region into two parts : @xmath534= } \nonumber \\ & & = c(i_1+i_2)\end{aligned}\ ] ] where @xmath535 . + choosing @xmath536 it will be possible to find analytically the integrals @xmath537 and @xmath538 . each of them is a sum of the two integrals : @xmath539d\beta+ \\ \displaystyle + \int_{\beta_{i1}}^{\beta_{i2 } } exp\big[-k\frac{sin\beta+sin\theta}{sin(\beta+\theta)}\big]cos^2(\beta+\theta)d\beta=\\ \displaystyle = i_{i1}+i_{i2 } \end{array}\ ] ] for @xmath540 and @xmath541 , @xmath542 , @xmath543 . + using @xmath60 we have @xmath544d\beta\simeq \nonumber \\ & \simeq & \int_0^{\pi-\theta-\epsilon } exp\big[-k(1+\frac{\theta}{sin\beta}-\theta\ , ctg\beta)\big]d\beta = \nonumber \\ & = & e^{-k}\int_0^{\pi-\theta-\epsilon } exp\big(-k\,\theta \,tg\frac{\beta}{2}\big)d\beta\end{aligned}\ ] ] where we have neglected all terms @xmath545 . note that in the integration region @xmath546 so that our approximation @xmath547 is justified . for the same reason the exponential index @xmath548 so that @xmath549\simeq \nonumber \\ & & \simeq e^{-k}(\pi-\theta-\epsilon+2k\theta\,ln\frac{\epsilon}{2 } ) \nonumber\end{aligned}\ ] ] the integral @xmath550 can be found easily : @xmath551\nonumber\end{aligned}\ ] ] thus @xmath552\nonumber\end{aligned}\ ] ] when calculating @xmath538 we introduce an angle @xmath553 to take advantage of @xmath554 being small in the integration region of @xmath538 . + we have @xmath555 } { \chi(1-\frac{\chi^2}{6 } ) } \big\}d\chi\simeq \nonumber \\ & \simeq & e^{-k}\int_0^{\epsilon}exp\bigg(-\frac{2k\theta}{x}\bigg)dx = e^{-k}2k\theta\int_{\frac{2k\theta}{\epsilon}}^{\infty}\frac{e^{-y}dy}{y^2 } \nonumber \\ & \simeq & e^{-k}\big[2k\theta\big(c_{eu}-ln\frac{\epsilon}{2}+ln\,k\theta -1\big)+\epsilon\big]\end{aligned}\ ] ] where @xmath556 is the euler constant . + the last integral , @xmath557 , equals @xmath558\nonumber\end{aligned}\ ] ] the second integral is of the order of @xmath559 , so that @xmath560 and @xmath561 + 2\epsilon\big\}\nonumber \\ \,\end{aligned}\ ] ] thus @xmath562\big\}\nonumber \\ \,\end{aligned}\ ] ] and finally @xmath563\nonumber \\ \,\end{aligned}\ ] ] as should be expected , the terms depending on @xmath193 have cancelled in the final result . + the ratio of the scattered light arriving within a certain angle @xmath4 to the point source direction to that non - scattered ( direct ) equals @xmath564\big\}\nonumber\\ \,\end{aligned}\ ] ] here we present results of some analytical integrations of functions @xmath565 and @xmath566 defined in section 2.2 . for the rayleigh scattering we have @xmath567 since @xmath568 , @xmath569 ( eq . 25 ) can be easily found analytically @xmath570 \nonumber\end{aligned}\ ] ] the calculation of a corresponding function @xmath571 for the mie scattering is a little more complicated . the adopted here form of @xmath572 has been chosen in such a way as to make the integration over @xmath70 analytically possible . we have ( eq . 16 ) @xmath573 the angle of the second scattering @xmath103 changes within limits @xmath574 where @xmath575 since @xmath576 separates the two regions of @xmath103 where @xmath572 has different shapes , the solution of ( c.3 ) depends on the positions of @xmath577 on the @xmath103 axis with respect to the point @xmath576 . there are three cases to be considered + then , of course , @xmath579 as well and the function @xmath580 applies to the whole region of @xmath70 ( @xmath581 ) . thus we have @xmath582 \nonumber \\ & = & 2\big[a_1\sum_{k=0}^8c_k\int_0^{\pi } cos^k\phi'\,d\phi'+\pi b\big]\end{aligned}\ ] ] where @xmath583 , @xmath584 and @xmath585 . + for odd @xmath158 the integrals in ( c.5 ) vanish . for even @xmath586 we have @xmath587 hence @xmath588\ ] ] this case is a bit more complicated since there are both parts @xmath590 and @xmath591 involved . the value @xmath592 corresponding to @xmath576 equals @xmath593 . + we have @xmath594= \nonumber\\ & = & 2\sum_{k=0}^8c_k\big[a_1\int_0^{\phi'_*}cos^k\phi'd\phi'+\nonumber \\ & + & a_2\int_{\phi'_*}^{\pi}cos^k\phi'd\phi'\big]+2\pi b\end{aligned}\ ] ] solving the integrals we finally obtain @xmath595+\nonumber \\ & & + \frac{a_2}{2^{2n}}{2n \choose n}\pi\big\ } \sum_{n=0}^3 c_{2n+1}(a_1-a_2)g'_n(\phi'_*)+b\pi\bigg]\nonumber\\ & & \quad\end{aligned}\ ] ] where @xmath596}{2n-2i } \\ g'_n(\phi ' _ * ) & = & \frac{1}{2^{2n}}\sum_{i=0}^{n}{2n+1 \choose i}\frac{sin[(2n-2i+1)\phi'_*]}{2n-2i+1}\nonumber\end{aligned}\ ] ] then , of course @xmath597 as well and for all values of @xmath70 we have that @xmath598 , so that @xmath599 . the calculation of @xmath600 is the same as in case a ) but @xmath601 . thus @xmath602\ ] ] the next integration over @xmath71 ( eq . 26 ) defining @xmath117 can be found analytically only , @xmath603 and @xmath604 the result could also be found analytically but due to the high power of @xmath605 ( index = 8) there would be very many terms making such an effort worthless . ] for @xmath606 . the calculation is lengthily so that we will present here only the final result for the flux of the second generation @xmath607 where @xmath608\cdot\big(\sum^2_{i=-2 \atop i\neq 0 } a_i\frac{x_2^i - x_1^i}{i}+\nonumber \\ & + & a_0\ , ln\frac{x_2}{x_1}+ \frac{b}{4\tau'^2}\bigg[\sum_{i=1}^4 a_{i-2}\cdot\frac{x_2^i - x_1^i}{i}+a_{-2}ln\frac{x_2}{x_1}+\nonumber\\ & -&2(\tau'^2 + 1)\cdot \big(\sum^3_{i=-1 \atop i \neq 0}a_{i-1}\frac{x_2^i - x_1^i}{i}+a_{-1}ln\frac{x_2}{x_1 } \big ) \bigg]\bigg\ } \nonumber\end{aligned}\ ] ]

the be / x - ray binaries represent the largest subclass of high - mass x - ray binaries . about two - thirds of the identified systems fall into this category . these systems consist of , generally , a neutron star and a be star with a cool ( @xmath2 ) equatorial disc , which is geometrically thin and nearly keplerian . the orbit is wide @xmath3 and usually eccentric ( @xmath4 ) ( e.g. , @xcite ) . most of the be / x - ray binaries show only transient activity in the x - ray emission and are termed be / x - ray transients . be / x - ray transients show periodical ( type i ) outbursts , which are separated by the orbital period and have the lumiocity of @xmath5 , and giant ( type ii ) outbursts of @xmath6 with no orbital modulation . the transient nature in the x - ray activity of be / x - ray binaries is considered to result from the interaction between the accreted material from the circumsteller disc of the be star and the rotating magnetized neutron star @xcite . if the accreted material is dense enough to make the magnetospheric radius smaller than the corotation radius , the accretion on to the neutron star causes a bright x - ray emission ( the direct accretion regime ) . otherwise , the magnetospheric radius is larger than the corotation radius , and the accretion is prevented by the centrifugal inhibition . the system is then in quiescence ( the propeller regime ) . during the outbursts , the neutron stars in most be / x - ray binaries show spin - up ( e.g. , 4u0115 + 63 , gs1843 - 02 , exo2030 + 375 and xtej1946 + 26 ) ( @xcite ; @xcite ; @xcite ; @xcite ) , and some even show quasi - periodic oscillations ( qpos ) ( e.g. , a0535 + 26 and exo2030 + 375 ) ( @xcite ; @xcite ) . these observational features strongly suggest the presence of an accretion disc during the outbursts . the accretion disc is also indicated to exist between the outbursts in several systems ( e.g. , a0535 + 26 ) ( @xcite ; see also @xcite ) . on the other hand , @xcite searched the optical / infrared contribution from the accretion disc in a0535 + 26 during the x - ray quiescent phase , and found no signature of the accretion disc in propeller regime . obviously , more work is needed to determine whether the accretion disc also exists in the quiescent state . recently , @xcite studied the interaction between the be - star disc and the neutron star , using a three - dimensional ( 3d ) smoothed particle hydrodynamics ( sph ) code ( @xcite ; @xcite ) . they also qualitatively discussed the relation between the x - ray luminosity and the accretion rate using the simulated mass - capture rate by the neutron star . in their simulations , however , the neutron star was modelled by a sink particle @xcite with the size of the roche lobe . hence , no information was available how the captured material accretes on to the neutron star . moreover , no direct comparison with the observed x - ray data was possible . in this paper , we study the accretion on to the neutron star in be / x - ray binaries , using a 3d sph code . the simulation should ideally take account of all the processes at work in a be / x - ray binary , such as the disc evolution around the be star , the mass transfer from the be - star disc to the neutron star , and the accretion on to the neutron star . such a simulation , however , would require very intense computations . therefore , we confine ourselves to simulate only the accretion flow around the neutron star , with an outer boundary condition constructed by the data of the particles captured by the neutron star from a high - resolution simulation by @xcite . this situation is illustrated in fig . 1 , where the right panel taken from their simulation shows the mass transfer from the be - star disc to the neutron star at periastron , while the left panel shows the accretion disc formed around the neutron star in one of our simulations . simulations presented here were performed with a 3d sph code . the sph code is basically the same as that used by @xcite , which is based on a version originally developed by benz ( @xcite ; @xcite ) . the smoothing length is variable in time and space . the sph equation with the standard cubic - spline kernel are integrated using a second - order runge - kutta - fehlberg integrator with individual time steps for each particle @xcite , which results in an enormous computational saving when a large range of dynamical time - scales are involved . in our code , the accretion flow on to the neutron star is modelled by an ensemble of gas particles , each of which has an arbitrary and negligible mass chosen to be @xmath7 with a variable smoothing length . note that the accretion flow is nonself - gravitating in our simulations . on the other hand , the be star and the neutron star are modelled by two sink particles @xcite with corresponding masses . the gas particles which fall within a specified accretion radius are accreted by the sink particles . we assume that the neutron star has the fixed accretion radius of @xmath8 , @xmath9 being the semi - major axis of the binary . for the be star , we adopt a variable accretion radius of @xmath10 , where @xmath11 is the roche - lobe radius for a circular binary . this is because our interest is only in the flow around the neutron star . an approximate formula for @xmath11 is given by @xmath12 ( see , for example , @xcite ; @xcite ) with the mass ratio @xmath13 , where @xmath14 and @xmath15 are the masses of the neutron star and the be star , respectively , and @xmath16 is the distance between the stars . in disc simulations using a three dimensional sph code with the standard cubic spline kernel , one has an approximate relation between the shakura - sunyaev viscosity parameter @xmath17 and the linear sph artificial viscosity parameter @xmath18 as @xmath19 if the non - linear sph artifical viscosity parameter @xmath20 is set equal to zero @xcite . here @xmath21 is the sph smoothing length and @xmath22 is the scale - height of the disc . in a simulation we report in this paper , we adopt @xmath23 and @xmath24 in which case @xmath17 is variable in time and space . in the other simulations , however , we adopt a constant value of @xmath17 in order to roughly model the @xmath25 disc . in these simulations , @xmath26 was variable in time and space and @xmath27 . in our simulations , we adopt a polytropic equation of state specified by the exponent @xmath28 , in which the effects of an energy dissipation resulting from viscosity and a radiative cooling are approximately boxed . our binary configuration is the same as that of @xcite . we set the binary orbit on the @xmath29-@xmath30 plane with the semi - major axis along the @xmath29-axis . initially , the neutron star is at the periastron . it orbits about the be star with the orbital period @xmath31 and the eccentricity @xmath32 . the unit of time adopted here and throughout the paper is @xmath33 . the be disc is coplanar with the binary orbital plane . as the be star , we take a b0v star with @xmath34 , @xmath35 and @xmath36 , where @xmath37 and @xmath38 are the stellar radius and effective temprature , respectively . for the neutron star , we take @xmath39 and @xmath40 . these parameters are adopted to target 4u0115 + 63 , which has shown only type ii x - ray outbursts followed by a short series of type i x - ray outbursts @xcite . as mentioned earlier , we concentrate only on the simulations of the accretion flow on to the neutron star . this is done by constructing an outer boundary condition from the data of particles captured by the neutron star in a high - resolution simulation by @xcite , which was run for @xmath41 and finally had @xmath42 particles with @xmath43 . we first fold the data on the orbital period over @xmath44 to reduce the fluctuation noise . the phase dependence of the mass - capture rate by the neutron star obtained by this procedure is shown in fig . [ fig : mdot_cap ] . note that there is a strong phase - dependence in the mass - transfer rate . the peak comes slightly after the periastron passage and decreases by two orders of magnitude by apastron passage . we then fit the spatial and velocity distribution of the captured particles in each phase bin with separate gaussian functions . this is for running simulations with an arbitrary mass - transfer rate from the be disc . we finally model the mass - transfer process from the be disc to the neutron star by injecting gas particles at a given phase - dependent rate shown in fig . 2 with the spatial and velocity distributions specified by these gaussian functions . the injected particles are assumed to have the initial temperature of @xmath45 . as described earlier , the mass of each gas particle is arbitrary and negligible in our simulations . all the mass - related quantities discussed later , such as the disc mass and the mass - accretion rate , are normalised by using the mass - capture rate shown in fig . 2 . . the mass - capture rate is measured in units of @xmath46 , where @xmath47 is highest local density in the be - star disc normalized by @xmath48 , a typical value for be stars . the right axis shows the x - ray luminosity corresponding to the mass - capture rate with the x - ray emission efficiency @xmath49 , where @xmath50 is defined by @xmath51.,width=302,height=226 ] . , width=264 ] . , width=264 ] [ cols="<,^,^,^,^,^,^,^,^ " , ] before describing the results of our simulations , we discuss below the possibility of the formation of an accretion disc around the neutron star . we first consider the radius @xmath52 at which the material transferred from the be disc is circularized via the interaction with the previously transferred gas . we assume the circularization radius @xmath52 to be approximately given by @xmath53 , where @xmath54 is the initial specific angular momentum of the transferred material . in the upper panel of fig . [ fig : rcirc ] , we show the dependence of @xmath52 on the orbital phase , based on the data taken from a simulation by @xcite . it is noted that there is a strong phase - dependence of @xmath52 . it has a minimum of @xmath55 slightly before periastron and then rapidly increases with phase . for comparison , we also show the corotation radius , @xmath56 , for 4u0115 + 63 by the dotted line . here , @xmath56 is given by @xmath57 where @xmath58 is the spin period of the neutron star in 4u0115 + 63 . note that @xmath59 for most of the orbital phase . therefore , we expect that an accretion disc is formed around the neutron star in 4u0115 + 63 . next , we discuss whether the accretion disc formed around the neutron star is persistent or transient . this is done by comparing the viscous accretion time - scale with the orbital period . for simplicity , we assume the accretion disc to be geometrically thin and isothermal with the shakura - sunyaev viscosity parameter @xmath17 and evaluate the accretion time - scale at @xmath60 . the ratio of these two time - scales for 4u0115 + 63 is then given by @xmath61^{1/2 } \left ( \frac{t_{\rm{d}}}{10^{4}\rm{k } } \right)^{-1 } , \label{eq : tratio}\end{aligned}\ ] ] where @xmath62 is the sound speed and @xmath63 is the disc temprature . the orbital - phase dependence of @xmath64 is shown in the lower panel of fig . [ fig : rcirc ] . it is immediately observed that the viscous accretion time - scale is much longer than the orbital period for most of the orbital phase . therefore , once formed , the accretion disc around the neutron star in 4u0115 + 63 will survive over the whole orbital phase . note that this result depends little on the orbital period , because @xmath65 increases almost linearly with increasing @xmath9 . based on the procedure described in the previous section , we have performed several simulations with different parameters . details of the models are given in table [ tbl : models ] . the polytropic exponents considered are @xmath66 ( model 3 ) , 1.2 ( models 1 , 2 , 5 and 6 ) and 5/3 ( model 4 ) . we have taken the inner radius of the simulation region , @xmath67 , to be @xmath68 in all simulations but model 1 with a smaller inner radius of @xmath69 , which was run to study the effect of the inner radius on the simulation results . in models 1 - 4 and 6 , we have assumed @xmath43 , where @xmath18 varies in time and space and @xmath27 . for comparison purpose , we have also run model 5 with @xmath23 and @xmath24 , where @xmath17 varies in time and space . model 6 was run solely for clearly illustrating the initial build - up phase of an accretion disc . for this purpose , we have taken the injection rate of the sph particles in model 6 four times as high as in other models and run only for one orbital period . analysing multi - wavelength long term monitoring observations of 4u0115 + 63 , @xcite found that the be star undergoes quasi - cyclic @xmath70 activity , losing and reforming its circumsteller disc . then , the mass transfer to the neutron star should vary , depending on the state of the be disc . by the time the be disc is fully developed and becomes capable of supplying mass to the neutron star , even a large accretion disc would be exhausted . therefore , to investigate the evolution of an accretion flow around the neutron star after the be disc is fully developed , we performed 3d sph simulations based on the numerical models described above , where there is intially no gas particle around the neutron star . since all models have shown similar structure and evolution of the accretion flow , we will mainly show the results from model 1 with @xmath71 and @xmath72 . + + we consider the initial phase of accretion flow around the neutron star . as discussed in section [ sec : possibility ] , it is expected that the material transferred from the be disc near the periastron does not directly fall on to the neutron star and forms a disc - like structure with a size corresponding to its specific angular momentum . for illustration purpose , we give in fig . [ fig : model6_ss ] the snapshots of the initial accretion flow around the neutron star from model 6 , which covers @xmath73 . at @xmath74 , the be star is at periastron . each panel shows the surface density in the range of four orders of magnitude in the logarithmic scale . we note that the accreting matter first forms an elliptical ring around the neutron star , which is then relaxed to an elliptical disc via the viscous diffusion . because the viscous time - scale is longer than the orbital period , the orbits of the accreting particles are not circularized and the flow is highly eccentric during the initial phase of accretion . comparing the number of particles , @xmath75 , with the number of particles inside the effective roche radius of the neutron star , @xmath76 , we also note that most of the material transferred from the be disc stays inside the effective roche radius of the neutron star . it should be noted that these dynamical features are common to all models studied in this paper . [ fig : model6_rad ] shows the time - dependence of the radial structure of the accretion flow around the neutron star at the same times as in fig . [ fig : model6_ss ] . in each panel , the solid line , the dashed line and the dash - dotted line denote the surface density @xmath77 , the azimuthal velocity @xmath78 normalized by the keplerian velocity at the inner boundary and the radial mach number @xmath79 , respectively . in the figure , the density is integrated vertically and averaged azimuthally , while the velocity components are averaged vertically and azimuthally . from fig . [ fig : model6_rad ] we note that while the azimuthal velocity distribution gradually approaches the keplerian one , the behaviour of the radial velocity is very complex , reflecting the dynamical process of the disc formation shown in fig . [ fig : model6_ss ] . the surface density distribution also shows the initial evolution of the accretion flow described above . in the panel at @xmath80 , there appears a hump at @xmath81 caused by the infall of injected particles . outside the hump the density decreases rapidly , making a ring - like structure with a sharp outer edge . this rapid decrease in the density is due to the ram pressure of the infalling particles . as the hump moves inward , the density gradient outside the hump becomes less steep . by the next periastron passage ( @xmath82 ) , a disc - like structure is formed , the inner radius of which is at @xmath83 . in our simulations , the presence of the inner simulation boundary causes an artificial decrease in the density near the boundary , because the gas particles which pass through the boundary are removed from the simulation . as a result , the density distribution has a break near the inner boundary , as seen in fig . [ fig : model6_rad ] . we have found that our simulations are reliable for @xmath84 . + + many be / x - ray binaries with known orbital parameters have orbital eccentricities in the range of 0.3 to 0.9 . in such systems , the mass - transfer rate from the be disc is expected to have the strong phase dependence @xcite . accordingly , the structure of an accretion disc formed around the neutron star in these systems is also likely to be strongly phase dependent . in this section , we first discuss the phase - dependent structure of a developed accretion disc in model 1 and then compare that with those in other models . fig . [ fig : model1_ss ] gives a sequence of snapshots of the accretion flow around the neutron star covering one orbital period for @xmath85 . the format of the figure is the same as that of fig . [ fig : model6_ss ] , except that the surface density is shown in the range of three orders of magnitude . we note from the figure that the disc shrinks at periastron passage by a negative torque exerted by the be star and restores its radius afterwards by the viscous diffusion . we also note that the disc is significantly eccentric even after it is fully developed , although the eccentricity is much smaller than that in the initial phase of the disc formation . [ fig : model1_rad ] shows the radial disc structure at the orbital phases corresponding to fig . [ fig : model1_ss ] . the format of the figure is the same as that of fig . is noted from fig . [ fig : model1_rad ] that the disc is nearly keplerian at any phase and the accretion flow is highly subsonic except in the outermost part for a short period of time when the material transferred from the be disc falls towards the neutron star . we call this state of the accretion disc developed . we have found that these features of a developed disc hold for @xmath86 in model 1 . as in the initial phase of the disc formation , the ram pressure by the supersonic infall of matter transferred from the be disc around periastron enhances the density in the outermost part of the accretion disc , making the disc outer edge sharp ( see panels for @xmath87 in figs . [ fig : model1_ss ] and [ fig : model1_rad ] ) . afterwards , the disc gradually expands until the next periastron passage of the be star . [ fig : chardisc ] shows how hot the disc is at @xmath88 in models 1 and 3 - 5 . in each panel , the solid line , the dash - dotted line and the dashed line denote the disc temprature normalized by the initial temprature @xmath89 , the ratio of the smoothing length to the harf - thickness of the disc @xmath90 and the relative disc thickness @xmath91 , respectively . from the upper - left panel of fig . [ fig : chardisc ] , it is noted that the disc in model 1 is vertically resolved except for the innermost region of the disc ( @xmath92 ) . the disc temprature varies from @xmath93 at @xmath94 to @xmath95 at the inner boundary . the disc is geometrically thin with the relative thickness of about 0.1 - 0.2 . in this section , we describe the effects of the simulation parameters in table [ tbl : models ] on the structure of the accretion flow around the neutron star . we ran model 2 to study the effect of the inner boundary on the disc strcture . [ fig : model2 ] shows the snapshots and the radial structures of the accretion flow around the neutron star at @xmath96 ( periastron ) and @xmath97 ( apastron ) for model 2 , which has the same simulation parameters as model 1 , except for a larger inner boundary of @xmath98 . comparing fig . [ fig : model2 ] with the panels at the corresponding times in figs . [ fig : model1_ss ] and [ fig : model1_rad ] for model 1 , we note that the presence of the inner boundary artificially reduces the density in a narrow region ( @xmath99 ) near the boundary , but outside of it , the disc structure is similar for both models . in order to study the dependence of the disc structure on the polytropic exponent , we also ran models 3 ( @xmath100 ; isothermal case ) and 4 ( @xmath101 ; adiabatic case ) , for which the polytropic exponent of 1.2 for models 1 , 2 and 6 is in between . [ fig : model3 ] shows the snapshots and the radial structures of the accretion disc for model 3 ( isothermal case ) at the same times as in fig . [ fig : model2 ] . from the figure , we note that the distributions of the surface density and both of the azimuthal and radial velocity components are similar to those in the initial phase of the disc formation for @xmath102 ( see figs . [ fig : model6_ss ] and [ fig : model6_rad ] ) . consequently , in model 3 the accretion disc is still developing at @xmath103 . this is because the viscous time - scale of the isothermal disc , particularly in an inner region , is longer than those in other models ( see also fig . [ fig : rcirc ] ) . from the upper - right panel of fig . [ fig : chardisc ] , it is noted that the disc in model 3 is not vertically resolved in most of the disc region . [ fig : model4 ] shows the disc structure for model 4 ( adiabatic case ) . we note that the disc is nearly keplerian at both orbital phases and that the surface density is lower and the disc is bigger than those for model 2 with @xmath102 . this is due to a higher viscous stress in model 4 , resulting from a higher pressure . as expected , in model 4 , the accretion rate is much higher than the models with lower polytropic exponents , which we will see in a later section . from the lower - left panel of fig . [ fig : chardisc ] , it is noted that the disc in model 4 is much hotter than in other models , reaching the temprature of @xmath104 near the inner boundary . the disc is vertically resolved except for the innermost region . we also studied a model with the standard , artificial viscosity parameters . model 5 has the same simulation parameters as those in model 2 , except that it has constant articficial viscosity parameters @xmath105 and @xmath106 , for which @xmath17 varies in time and space . [ fig : model5 ] shows the disc structure for model 5 . from the figure and the lower - right panel of fig . [ fig : chardisc ] , it is immediately noted that the inner boundary affets the disc structure much more widely in this model . this is mainly because the inner disc region is not resolved vertically for this number of sph particles . since the ratio of the smoothing length to the harf - thickness of the disc @xmath90 in the inner region is much higher than unity , the shear viscosity parameter @xmath17 in this model is much higher in the inner region than that in other models . except for the above - mentioned dependence of the disc structure on the simulation parameters , the accretion discs in our models share the common features : they are nearly keplerian after developed and their structures modulate with the orbital phase . they are also significantly eccentric . we will analyse the non - axisymmentry of the accretion discs in be / x - ray binaries in a subsequent paper . + + + + + @xcite investigated the tidal / resonant truncation of circumsteller and circumbinary discs in eccentric binaries and found that a gap is always opened between the disc and the binary orbit . following their formulation , @xcite and @xcite have shown that the be disc is truncated via the resonant interaction with the neutron star , as long as @xmath107 , which has recently been confirmed by numerical simulations by @xcite . it is interesting to study whether the tidal / resonant torque by the be star also truncates the accretion disc around the neutron star . in order to have a measure of the accretion disc radius at which the disc density has a major break , we have applied a non - linear least squares fitting method to the radial distribution of the azimuthally - averaged surface density @xmath77 , adopting the following simple fitting function , @xmath108 where @xmath109 and @xmath110 are constants and @xmath111 is the radius of accretion disc . this method is the same as that adopted by @xcite . [ fig : radius ] shows the orbital - phase dependence of the accretion disc radius in model 1 . to reduce the fluctuation noise , we folded the data on the orbital period over @xmath112 . it is noted that the disc radius varies between @xmath113 and @xmath114 . the disc shirinks at periastron due to a negative torque by the be star and gradually restores its radius towards apastron due to the viscous diffusion . the rapid increase of the disc radius before periastron is due to the infall of matter from the be star . next , we examine whether the disc radius shown in fig . 13 is determined by the resonant truncation as is the case in the be disc in be / x - ray binaries . fig . 8 of @xcite gives the truncation radius of a circum - secondary disc as a function of the orbital eccentricity @xmath115 and the disc reynolds number @xmath116 for systems with the mass ratio of 0.1 , which is similar to that of be / x - ray binaries . since the accretion disc in our model is geometrically thin and nearly keplerian , the reynolds number @xmath116 is given by @xmath117 @xmath118 , where @xmath119 is the keplerian velocity . in model 1 , @xmath120 for @xmath121 . from fig . 8 of @xcite , with @xmath122 and @xmath123 , we have the truncation radius of @xmath124 , which corresponds to the 6:1 resonance radius . note that this radius is much larger than the disc radius shown in fig . [ fig : radius ] . it is likely that the interaction with the be star is significantly non - linear owing to its large mass . then , the accretion disc would be truncated well inside the 6:1 radius given above . even such a strong interaction , however , is unlikely to truncate the disc at a radius as small as that shown in fig . [ fig : radius ] . it is suggestive that the disc radius is about the same as the circularization radius @xmath65 of the matter transferred from the be disc after the periastron passage ( see fig . [ fig : rcirc ] ) . since the viscous time - scale at the circularization radius is much longer than the run time of our simulations , the matter once circularized dose not have time enough to change its radius significantly . therefore , we conclude that the disc radius shown in fig . [ fig : radius ] is determined by the specific angular momentum of the matter transferred from the be disc . it is probable , however , that the disc expands well beyond the circularization radius and is truncated by a tidal torque in a time - scale longer than the viscous one . as mentioned in section 1 , most of the be / x - ray binaries show transient x - ray activities , of which nature is considered to result from the interactions between the accreted material and the rotating magnetized neutron star @xcite . in this section , we first describe the growth of the accretion disc in mass and radius . we then examine the evolution of the mass - accretion rate obtained by our simulations , from which we finally discuss the x - ray activity expected in be / x - ray binaries . [ fig : masrad](a ) shows the evolution of the disk mass in model 1 over the period of @xmath125 . in the long term , the disc mass increases with time , because the the gas transferred from the be star accumulates in the accretion disc owing to the viscous time - scale longer than the orbital period . on the other hand , in the short term , it modulates with the orbital phase . the disk mass starts increasing , when the mass - supply from the be disc begins slightly before every periastron passage . it shows a rapid increase for a short while , followed by a gradual decrease by accretion with no mass - supply from the be disc , which lasts until the next periastron passage . at first glance , the disc mass in model 1 seems to increase secularly and approach no steady average state . on closer inspection , however , we note that the increasing rate of the disc mass gradually decrases with time after the disc is developed . this suggests an interesting possibility that the disc reachs a steady average state in a time - scale much longer than the term shown in fig . 14(a ) . in fig . [ fig : masrad](b ) , we show the evolution of the disc radius over @xmath125 . one may expect that the disc radius gradually increases with the disc mass . from the figure , however , we note that this is the case only for @xmath86 . in fact , for @xmath126 , the disc radius decreases as the disc grows . this counter - intuitive decrease in radius is related to the change in the disc eccentricity . in the early stage of disc formation , the disc / ring is highly eccentric ( see fig . [ fig : model6_ss ] ) , for which the radius @xmath111 calculated by equation ( [ eq : fitting ] ) is larger than that for a circular disc with the same disc mass . since the disc eccentricity decreases with time , so does the disk radius . in model 1 , the effect of the disc circularization is balanced with that of the disc growth at @xmath127 . after this epoch , the disc radius increases with the disc mass . thus , the evolution of the accretion flow around the neutron star in be / x - ray binaries involves a two - stage process , which consists of the initial developing stage and the later developed stage . [ fig : mdot_acc](a ) shows the evolution of the mass - accretion rate and the corresponding x - ray luminosity in model 1 for @xmath128 . here we calculated the x - ray luminosity by @xmath129 with the x - ray emission effciency @xmath49 for the neutron star . the thick and thin lines denote the mass - accretion rate in models 1 and 2 , respectively . note that the mass - accretion rate in our simulations generally has double peaks per orbit . the first peak , which occurs at the periastron , is due to the contraction of the disc by the tidal torque of the be star , except that the peaks at periastron during the initial phase of disc formation are mainly due to the direct accretion of gas with low specific angular momentum . the second peak , which lags behind the peak in the mass - transfer rate from the be disc shown in fig . [ fig : mdot_cap ] , results from the viscous accretion of matter transferred from the be disc . the first peak could be artificial since it is related to the presence of the inner simulation boundary . the orbital modulation in the mass - accretion rate is significantly different between the developing disc and the developed disc . the first peak becomes weak as the disc developes . after the disc is fully developed , the second peak is higher than the first peak . [ fig : mdot_acc](a ) also shows that , once developed , the accretion disc in 4u0115 + 63 shows the peak x - ray luminosity of @xmath130 . this level of x - ray emission enters the transition regime ( see @xcite ) and is consistent with the observed x - ray luminosity in the quiescent state . for comparison purpose , we present in fig . [ fig : mdot_234 ] the evolution of the mass - accretion rate and the corresponding x - ray luminosity in models 2 - 4 for @xmath125 . the thick and thin solid lines and the thin dotted line are for models 2 , 3 and 4 , respectively . we see from the figure that the accretion rate increases with increasing polytropic exponent . in models 2 ( @xmath131 ) and 3 ( @xmath100 ) , the accretion rate profile has the double peaks as described above . the long - term change in the peak profiles is also similar to that in model 1 . we note that the disc in model 2 entered the stage of the developed disc at @xmath132 . on the other hand , the disc in model 3 is at the developing stage even at @xmath133 , which is consistent with the disc structure shown in fig . [ fig : model3 ] . in contrast to the models with low @xmath28 , the accretin rate in model 4 has a single and periodic peak per orbit , even after the disc is developed . with @xmath134 , the pressure in the disc in model 4 is much higher than that in other models . this means that much larger viscous torque exerts on the disc in model 4 than in other models , because the viscous stress in the @xmath25-viscosity prescription is proportional to the pressure . as a result , the viscous torque dominates the tidal torque in model 4 even at periastron . this is why there is no peak at periastron and the accretion rate shows the regular orbital modulation with the amplitude much higher than in other models . it is important to note that the disc evolution and its orbital modulation is sensitive to the polytropic exponent @xmath28 . this strongly implies that the cooling effciency in the accretion disc plays an important role in the x - ray activity in be / x - ray binaries . changes in the mass flux on the magnetosphere of the neutron star determine whether the system is in the direct accretion regime or in the propeller regime . while the matter falls directly on to the neutron star in the direct accretion regime , no material accretes on to the neutron star in the propeller regime by the centrifugal inhibitation . if the magnetospheric radius @xmath135 is smaller than the corotation radius @xmath136 , the system is in the direct accretion regime , otherwise , in the propeller regime . assuming the neutron star magnetic field as the dipole - like field , the magnetospheric radius @xmath135 , obtained by equating the magnetic field pressure of the neutron star to the ram pressure of the accreting matter from the accretion disc , is written by @xmath137 ( @xcite ; @xcite ; see also @xcite ) , where @xmath138 is the magnetic moment . here , @xmath139 is a constant model parameter of order unity , depending on the physics and geometry of accretion . for spherical accretion , @xmath140 . the original model calculations by @xcite for disc accretion on to the aligned rotating magnetized neutron star obtained @xmath141 , whereas calculations by @xcite , which included an induced wind not considered by @xcite , resulted in the value of @xmath142 for disc accretion . fig . [ fig : mdot_acc](b ) shows the evolution of the magnetospheric radius @xmath135 normalized by the corotation radius @xmath136 , where @xmath143 was adopted . in the figure , the thick and thin lines are for models 1 and 2 , respectively . recall that these models have the same parameters except for the radius of the inner simulation boundary @xmath67 ; @xmath144 in model 1 , while @xmath145 in model 2 . it should be noted that the magnetospheric radii in both models evolve in a very similar way , despite the different inner radii . our result on the magnetospheric radius is robust in this sense . we also note from fig . [ fig : mdot_acc](b ) that the ratio of the magnetospheric radius to the corotation radius , @xmath146 , varies between @xmath147 and @xmath148 . consequently , for a plausible range of @xmath139 ( @xmath149 ) , the magnetospheric radius in models with @xmath150 is almost always larger than the corotation radius so that the accretion on to the neutron star is inhibited , which is consistent with the oberved x - ray luminosity of 4u0115 + 63 in quiescence ( @xcite ) . on the other hand , we can see from equation ( [ eq : magnetosphere ] ) and fig . [ fig : mdot_234 ] that model 4 with @xmath101 is consistent with the observed x - ray behaviour of 4u0115 + 63 only if @xmath151 . for @xmath152 , the neutron star in model 4 enters the direct accretion regime after every peristron passage of the be star . then , the system would show regular , periodic x - ray outbursts at a level of @xmath153 , which has not been observed for 4u0115 + 63 . for the first time , we have performed numerical simulations of accretion on to the neutron star in be / x - ray binaries , taking into account the phase - dependent mass transfer from the be disc . we have adopted the mass - transfer rate from a high - resolution simulation by @xcite for a coplanar system with a short period ( @xmath31 ) and moderate eccentricity @xmath1 , which targeted 4u0115 + 63 , one of the best studied be / x - ray binaries . we have found that a nearly keplerian , non - steady accretion disc is formed around the neutron star regardless of the parameters adopted . the formation of the accretion disc involves a two - stage process , which consists of the initial developing stage and the later developed stage . the disc structure shows a strong dependence on the orbital phase , because both the tidal potential and the mass - transfer rate change periodically . the disc is also significantly eccentric , because the gas particles from the be disc originally have elliptical orbits . in 4u0115 + 63 , type i x - ray outbursts have occurred only as a short series after occasional type ii x - ray outbursts @xcite . except for such occasions , the system is in the quiescent state ( @xmath154 ) @xcite . the peak x - ray luminosity estimated from our simulations corresponds to the transition regime between the direct accretion regime and the propeller regime and is consistent with the observed x - ray behaviour . note , however , that the discussion based on the x - ray luminosity estimated from the simulations only gives the necessary condition for the quiescent state . in order to see whether our model satisfies a more stringent condition , we also have estimated the magnetospheric radius by using a general formula with the mass - accretion rate from our simulations and compared with the corotation radius for 4u0115 + 63 . the result was that our models with @xmath150 are consistent with the observed x - ray behaviour . other models with @xmath101 and @xmath155 were ruled out , because in these models the system would show an x - ray outburst after every periastron passage , which has never been observed . we have to admit , however , that the radius of our inner simulation boundary is about 60 - 100 times as large as the corotation radius of the neutron star in 4u0115 + 63 . to make the constraint on the system parameters more reliable , we need numerical simulations with a much smaller inner boundary . it is important to note that our model is not only consistent with the quiescent state of 4u0115 + 63 . it can also give a natural explanation for a short series of type i x - ray outbursts after occasional type ii x - ray outbursts . the mass - transfer rate from the be disc in the x - ray outbursts is likely much higher than in the quiescent state . suppose that the mass - transfer rate from the be disc in 4u0115 + 63 is temporarily increased by an order of magnitude . then , the estimated x - ray luminosity in our model increases by an order of magnitude to a level of several @xmath156 , a typical x - ray luminosity of type i x - ray outbursts . at the same time , the magnetosphric radius @xmath135 decreases by a factor of about two , for which @xmath135 is almost always smaller than the corotation radius , @xmath136 , if @xmath152 ( see fig . [ fig : mdot_acc](b ) ) . this means that the system is in the direct accretion regime . thus , our model explains the type i x - ray outbursts in 4u0115 + 63 as a temporal phenomena with an enhanced mass - transfer from the be disc . be / x - ray binaries are distributed over a wide range of orbital periods @xmath157 and eccentricities @xmath158 . it is interesting to consider here the effects of orbital parameters . in this paper , we have seen that the main ingredients which determines the x - ray behaviour of a system are the viscous time - scale relative to the orbital period and the mass - transfer rate from the be disc . since the former is insensitive to the orbital period itself ( see equation ( [ eq : tratio ] ) ) , the latter is most important if the viscosity parameter and the equation of state are similar in accretion discs of be / x - ray binaries . in systems with @xmath159 , the viscous time - scale in an outer part of the accretion disc is much longer than the orbital period unless the cooling is ineffective and the gas obeys an adiabatic equation of state . in these systems , persistent accretion discs are expected to form . then , the mass - transfer rate from the be disc determines whether the system shows type i x - ray outbursts . for example , for a system with a moderate orbital eccentricity of @xmath160 and a typical density of be disc , the peak mass - transfer rate is @xmath161 , for which the system stays in quiescence . it is obvious , however , that the situation should be different in systems with much higher rate of mass - transfer . a high mass - transfer rate is expected to result from a high orbital eccentricity and/or an intrinsically dense be disc . therefore , it is not strange that many systems which have regularly shown type i x - ray outbursts have orbital eccentricities higher than that of 4u0115 + 63 studied in this paper . in a subsequent paper , we will perform numerical simulations of accretion in systems with high orbital eccentricities . above we have assumed that the viscous time - scale is much longer than the orbital period in all be / x - ray binaries . however , it is possible that an accretion disc has a relatively short viscous time - scale by some reason . one possibility is that the disc has a low cooling efficiency and becomes hot , particularly in an inner region . our model 4 with @xmath101 gives an extreme example of such discs ( see also fig . [ fig : chardisc ] ) . as shown in fig . [ fig : mdot_234 ] , in a system with a short viscous time - scale , the mass - accretion rate has a high and sharp peak shortly after every periastron passage . the resemblance between the accretion rate profile in model 4 and the x - ray light curves of regular type i x - ray outbursts from systems such as exo2030 + 375 @xcite implies that these are systems with relatively short viscous time - scales . be / x - ray binaries are an ideal group of objects for studying the physics of accretion under the circumstances which have been studied little . unlike close binaries , many be / x - ray binaries have eccentric orbits . they can also be highly inclined systems . such systems provide us a valuable oppotunity to study the effects of the periodically - changing tidal potential and mass - transfer rate and the inclination angle on the structure and evolution of the accretion flow . another important characteristic of be / x - ray binaries is that they are systems with double circumstellar discs ( the be disc and the accretion disc ) . the interaction in such systems is threefold : the interactions between the neutron star and the be disc , the be star and the accretion disc and the be disc and the accretion disc ( mainly via the mass transfer ) . much more work is desirable both theoretically and observationally in order to understand this interesting and important group of objects . we acknowlege the anonymous refree for constructive comments . we would like to thank matthew r. bate for allowing us to use his sph code . kh acknowledes james r. murray for useful comments during the short stay at the swinburne university of technology , australia . kh also thanks noboru kaneko for his continuous encouragement . the simulations reported here were performed using the facility at the hokkaido university information initiative center , japan . this work was supported in part by grant - in - aid for scientific reserch ( 13640244 ) of japan society for the promotion of science . artymowicz p & lubow s.h . , 1994 , apj , 421 , 651 bate m.r . , bonnell i.a.,price n.m . , 1995 , mnras , 285 , 33 benz w. , 1990 , in buchler j. r.,ed.,the numerical modelling of nonlinear stellar pulsations . kluwer , dordrecht , p.269 benz w. , bowers r.l . , cameron a.g.w . , press w.h . , 1990 , apj , 348 , 647 bildsten , l. , chakrabarty , d. , chiu , j. , et al . 1997 , apjs , 113 , 367 campana s. , gastaldello f. , stella l. , israel g.l . , colpi m. , pizzolato f. , orlandini m. , dal fiume d. , 2001 , apj , 561 , 924 clark j.s . , lyuty v.m . , zaitseva g.v . , larionov v.m . , larionova l.v . , finger m. , tarasov a.e . , roche p. , coe m.j . , 1999 , mnras , 302 , 167 finger m.h . , wilson r.b . , harmon b.a . , 1996 , apj , 459 , 288 finger m.h . , bildsten l. , chakrabarty d. , prince t.a . , scott d.m . , wilson c.a . , wilson r.b & zhang s.n . , 1999 , apj , 517 , 449 frank j. , king a.r . , raine d.j . , 2002 , accretion power in astrophysics , 3rd edn . cambridge univ . press , cambridge ghosh p & lamb f.k . , 1979 , apj , 234 , 296 lamb f.k . , pethick c.j & pines d. , 1973 , apj , 184 , 271 negueruela i. , okazaki a.t . , fabregat j. , coe m.j . , munari u & tomov . , 2001 , a&a , 369 , 108 negueruela i & okazaki a.t . , 2001 , a&a , 369 , 108 okazaki a.t & negueruela i. , 2001 , a&a , 377 , 161 okazaki a.t . , bate m.r . , ogilvie g.i & pringle j.e . , 2002 , mnras , 337 , 967 ostriker e.c . , shu f.h . , 1995 , 447 , 813 pringle j.e & rees m.j . , 1972 , a&a , 21 , 1 stella l. , white n.e . , rosner r. , 1986 , apj , 208 , 669 warner b. , 1995 , cataclysmic variable stars . cambridge univ . press , cambridge wilson c.a & finger m.h . , coe m.j . , laycock s. , fabregat j. , 2002 , apj , 570 , 287 wilson c.a & finger m.h . , coe m.j . , negueruela i. , 2003 , apj , 584 , 996 ziolkowski j. , 2002 , mem . ital , 73 , 1038

circular planar electrical networks were studied by curtis - ingerman - morrow @xcite and de verdire - gitler - vertigan @xcite . associated to any circular planar electrical network of order @xmath0 is its @xmath1 _ response matrix _ , and response matrices are characterized in ( * ? ? ? * theorem 4 ) as the symmetric matrices with row sums equal to zero and _ circular minors _ non - negative . furthermore , the circular minors which are strictly positive can be identified combinatorially using ( * ? ? ? * lemma 4.2 ) . this positivity condition is of particular interest to us , and the goal of this paper is to investigate the connections between electrical networks and other positivity phenomena in the literature . until now , the combinatorial properties of these response matrices have remained largely unstudied , despite their inherently combinatorial descriptions . a natural question that arises is : which sets of circular minors can be positive , while the others are zero ? it is clear ( for example , from the grassmann - plcker relations ) that one can not construct response matrices with arbitrary sets of positive circular minors . postnikov @xcite studied a similar question in the _ totally nonnegative grassmannian _ , as follows : for @xmath2 matrices @xmath3 , with @xmath4 and all @xmath5 minors nonnegative , which sets ( in fact , matroids ) of @xmath5 minors can be the set of positive minors of @xmath3 ? these sets , called _ positroids _ by knutson - lam - speyer @xcite , were found in @xcite to index many interesting combinatorial objects . two of these objects , plabic graphs and alternating strand diagrams , are highly similar to circular planar electrial networks and medial graphs , respectively , which we study in this paper . our introduction of _ electrical positroids _ is therefore a natural extension of the theory of positroids . we give a novel axiomatization of electrical positroids , motivated by the grassmann - plcker relations , and prove the following : a set @xmath6 of circular pairs is the set of positive circular minors of a response matrix if and only if @xmath6 is an electrical positroid . another point of interest is that of _ positivity tests _ for response matrices . in @xcite , fomin - zelevinsky describe various positivity tests for _ totally positive matrices _ : given an @xmath1 matrix , there exist sets of @xmath7 minors whose positivity implies the positivity of all minors . these sets of minors are described combinatorially by _ double wiring diagrams_. fomin - zelevinsky later introduced _ cluster algebras _ in @xcite , in part , to study similar positivity phenomena . in particular , their double wiring diagrams fit naturally within the realm of cluster algebras as manifestations of certain cluster algebra mutations . in a similar way , we describe sets of @xmath8 minors of an @xmath1 matrix @xmath9 whose positivity implies the positivity of all circular minors , that is , that @xmath9 is a response matrix for a top - rank ( in @xmath10 , as defined in @xcite ) electrical network . some such sets were first described by kenyon - wilson @xcite . however , these sets do not form clusters in a cluster algebra . instead , they form clusters in a _ laurent phenomenon ( lp ) algebra _ , a notion introduced by lam - pylyavskyy in @xcite . this observation leads to the last of our main theorems : there exists an lp algebra @xmath11 , isomorphic to the polynomial ring on @xmath8 generators , with an initial seed @xmath12 of diametric circular minors . @xmath12 is a positivity test for circular minors , and furthermore , all `` plcker clusters '' in @xmath11 , that is , clusters of circular minors , are positivity tests . in proving theorem [ lpalg ] , we find that @xmath11 is , in a sense , `` double - covered '' by a cluster algebra @xmath13 that behaves very much like @xmath11 when we restrict to certain types of mutations . further investigation of the clusters leads to an analogue of weak separation , as studied by oh - speyer - postnikov @xcite and scott @xcite . conjecturally , the `` plcker clusters , '' of @xmath11 correspond exactly to the maximal pairwise weakly separated sets of circular pairs . furthermore , we conjecture that these maximal pairwise weakly separated sets are related to each other by mutations corresponding to the grasmann - plcker relations . while we establish several weak forms of the conjecture , the general statement remains open . the roadmap of the paper is as follows . we begin by briefly reviewing terminology and known results in [ defs ] , referring the reader to @xcite for a more detailed exposition . in [ circpos ] , we motivate and introduce electrical positroids , and prove theorem [ circposthm ] . in [ lpsec ] , using positivity tests as a springboard , we construct @xmath11 and prove theorem [ lpalg ] , and conclude by establishing weak forms of conjecture [ wsconj ] , which relates the clusters of @xmath11 to positivity tests and our new analogue of weak separation . we recommend that the reader begin by reviewing the definitions and notational conventions from @xcite . for convenience , we re - state some of the more important definitions and results that we will use in this paper . a * circular planar graph * @xmath14 is a planar graph embedded in a disk @xmath15 . @xmath14 is allowed to have self - loops and multiple edges , and has at least one vertex on the boundary of @xmath15 - such vertices are called * boundary vertices*. a * circular planar electrical network * ( or more simply an * electrical network * is a circular planar graph @xmath14 , together with a * conductance map * @xmath16 . associated to an electrial network is its * response matrix * , which is obtained by imposing voltages at the boundary vertices and measuring the resulting boundary currents . more details are given in @xcite . two electrical networks @xmath17 are * equivalent * if they have the same response matrix . the equivalence relation is denoted by @xmath18 . recall also the * local equivalences * , described in detail in @xcite and @xcite : self - loop and spike removal , series and parallel edge replacement , and @xmath19-@xmath20 transformations . these local equivalences are sufficient to generate the equivalence relation @xmath18 : [ localeqenough ] two electrical networks are equivalent if and only if they are related by a sequence of local equivalences . as in @xcite , we will often consider only the underlying circular planar graph of an electrical network . accordingly , we use the following notion of equivalence of circular planar graphs : let @xmath21 be circular planar graphs , each with the same number of boundary vertices . then , @xmath21 are * equivalent * ( with the equivalence abusively denoted @xmath18 ) if there exist conductances @xmath22 on @xmath21 , respectively such that @xmath17 are equivalent electrical networks . alternatively , @xmath23 if @xmath24 and @xmath25 are related by a sequence of local equivalences . the central ingredient to this paper is the characterization of response matrices given in @xcite . in order to state this characterization , we recall the definitions of circular pairs and circular minors . [ circpair ] let @xmath26 and @xmath27 be disjoint ordered subsets of the boundary vertices of an electrical network @xmath28 . we say that @xmath29 is a * circular pair * if @xmath30 are in clockwise order around the circle . we will refer to @xmath31 as the * size * of the circular pair . [ flippq ] we will take @xmath29 to be the same circular pair as @xmath32 , where @xmath33 denotes the ordered set @xmath34 with its elements reversed . almost all of our definitions and statements are compatible with this convention ; most notably , by theorem [ circularminorspositive](a ) , because response matrices are positive , the circular minors @xmath35 and @xmath36 are the same . whenever there is a question as to the effect of choosing either @xmath29 or @xmath32 , we take extra care to point the possible ambiguity . [ connection ] let @xmath29 and @xmath28 be as in definition [ circpair ] . we say that there is a * connection * from @xmath34 to @xmath37 in @xmath14 if there exists a collection of vertex - disjoint paths from @xmath38 to @xmath39 in @xmath14 , and furthermore each path in the collection contains no boundary vertices other than its endpoints . we denote the set of circular pairs @xmath29 for which @xmath34 is connected to @xmath37 by @xmath40 . let @xmath29 and @xmath28 be as in definition [ circpair ] , and let @xmath9 be the response matrix . we define the * circular minor * associated to @xmath29 to be the determinant of the @xmath5 matrix @xmath35 with @xmath41 . as in @xcite , we often refer to submatrices and their determinants both as minors , interchangeably . we now have the language needed to state the characterization of response matrices : [ circularminorspositive ] let @xmath9 be an @xmath1 matrix . then : 1 . @xmath9 is the response matrix for an electrical network @xmath28 if and only if @xmath9 is symmetric with row and column sums equal zero , and each of the circular minors @xmath35 is non - negative . 2 . if @xmath9 is the response matrix for an electrical network @xmath28 , the positive circular minors @xmath35 are exactly those for which there is a connection from @xmath34 to @xmath37 . let us also mention include two more tools which we will need in what follows . [ equivalentconnections ] the circular planar graphs @xmath42 and @xmath43 are equivalent if and only if @xmath44 . it is easily checked that local equivalences preserve @xmath45 , so if @xmath46 , then @xmath44 . conversely , if @xmath44 , there exist critical ( see @xcite graphs @xmath47 with @xmath48 and @xmath49 , by ( * ? ? ? * thorme 2 ) . then , @xmath50 , so the claim follows from ( * ? ? ? * theorem 1 ) . let @xmath51 be a circular planar graph , and let @xmath52 be an edge with endpoints @xmath53 . the * deletion * of @xmath52 from @xmath51 is exactly as named ; the edge @xmath52 is removed while leaving the rest of the vertices and edges of @xmath51 unchanged . if @xmath53 are not both boundary vertex of @xmath51 , we may also perform a * contraction * of @xmath52 , which identifies all points of @xmath52 . if exactly one of @xmath53 is a boundary vertex , then the image of @xmath52 under the contraction is a boundary vertex . note that edges connecting two boundary vertices can not be contracted to either endpoint . by theorem [ circularminorspositive ] , @xmath1 response matrices are characterized in the following way : a square matrix @xmath9 is the response matrix for an electrical network @xmath28 if and only if @xmath9 is symmetric , its row and column sums are zero , and its circular minors @xmath35 are non - negative . furthermore , @xmath35 is positive if and only if there is a connection from @xmath34 to @xmath37 in @xmath14 . the sets @xmath6 of circular pairs for which there exists a response matrix @xmath9 with @xmath35 is positive if and only if @xmath54 , then , are thus our next objects of study . the case of the totally nonnegative grassmannian was studied in @xcite : for @xmath2 ( with @xmath4 ) matrices with non - negative maximal minors , the possible sets of positive maximal minors are called _ positroids _ , and are a special class of matroids . our objects will be called _ electrical positroids _ , which we first construct axiomatically , then prove are exactly those sets @xmath6 of positive circular minors in response matrices . here , we present the axioms for electrical positroids , which arise naturally from the grassmann - plcker relations . let @xmath9 be a fixed matrix , whose rows and columns are indexed by some sets @xmath55 . we write @xmath56 for the determinant of the matrix @xmath57 formed by deleting the rows corresponding to @xmath58 and @xmath59 , provided @xmath57 is square . while the meaning @xmath56 depends on the underlying sets @xmath55 , these sets will always be implicit . [ gp ] we have the following two * grassmann - plcker * relations . 1 . let @xmath9 be an @xmath1 matrix , with @xmath60 elements of its row set and @xmath61 elements of its column set . furthermore , suppose that the row @xmath62 appears above row @xmath63 and column @xmath64 appears to the left of column @xmath65 . then , @xmath66 2 . let @xmath9 be an @xmath67 matrix , with @xmath68 elements of its row set ( appearing in this order , from top to bottom ) , and let @xmath65 an element of its column set . then , @xmath69 while the grasmann - plcker relations are purely algebraic in formulation , they encode combinatorial information concerning the connections of circular pairs in a circular planar graph @xmath14 . as a simple example , consider four boundary vertices @xmath70 in clockwise order of an electrical network @xmath28 , and let @xmath71 . if @xmath9 is the response matrix of @xmath28 , then @xmath72 is the circular minor associated to the circular pair @xmath73 ; thus , @xmath57 has non - negative determinant . furthermore , the entries of @xmath57 are @xmath74 circular minors of @xmath9 , so they , too , must be non - negative . now , suppose that the left hand side of ( [ p1 ] ) is positive , that is , @xmath75 . equivalently , there are connections between @xmath63 and @xmath65 and between @xmath62 and @xmath64 in @xmath14 . then , at least one of the two terms on the right hand side must be strictly positive ; combinatorially , this means that either there are connections between @xmath63 and @xmath64 and between @xmath62 and @xmath65 , or there is a connection between @xmath76 and @xmath77 . one can derive similar combinatorial rules by assuming one of the terms on the right hand side is positive , and deducing that the left hand side must be positive as well . the first six of the electrical positroid axioms given in definition [ posaxioms ] summarize all of the information that can be extracted in this way from the grassmann - plcker relations . if @xmath78 , write @xmath79 for the ordered set formed by removing @xmath62 from @xmath34 . [ posaxioms ] a set @xmath6 of circular pairs is an * electrical positroid * if it satisfies the following eight axioms : 1 . for ordered sets @xmath80 and @xmath81 , with @xmath82 in clockwise order ( that is , @xmath29 is a circular pair ) , consider any @xmath83 , @xmath84 , @xmath85 , @xmath86 with @xmath87 and @xmath88 . then : 1 . [ axside ] if @xmath89 , then either @xmath90 or @xmath91 . [ axcross ] if @xmath90 , then @xmath89 . 3 . [ axsep ] if @xmath91 , then @xmath89 . 1 . for @xmath92 and @xmath81 , with @xmath93 in clockwise order , consider any @xmath83 , @xmath84 , @xmath94 , @xmath86 with @xmath95 . then : 1 . [ axmid ] if @xmath96 , then either @xmath97 or @xmath98 . [ axleft ] if @xmath97 , then @xmath96 . [ axright ] if @xmath99 , then @xmath96 . finally : 1 . [ axsub ] ( * subset axiom * ) for @xmath100 and @xmath101 with @xmath29 a circular pair , if @xmath54 , then @xmath102 . [ axempty ] @xmath103 . [ circposthm ] a set @xmath6 of circular pairs is an electrical positroid if and only if there exists a response matrix whose positive circular minors are exactly those corresponding to @xmath6 . given a response matrix @xmath9 , it is straightforward to check that the set @xmath6 of circular pairs corresponding to the positive circular minors of @xmath9 satisfies the first six axioms , by proposition [ gp ] . @xmath6 also satisfies the subset axiom , by theorem [ circularminorspositive ] . finally , adopting the convention that the empty determinant is equal to 1 , we have the last axiom . to prove theorem [ circposthm ] , we thus need to show that any electrical positroid @xmath6 may be realized as the set of positive circular minors of a response matrix , or equivalently the set of connections in a circular planar graph . we now prove theorem [ circposthm ] . first , recall the important convention that @xmath104 . we leave it to the reader to check , whenever appropriate , that all of the definitions and statements we make in this section are compatible with this convention . fix a boundary circle with @xmath0 boundary vertices , which we label @xmath105 in clockwise order . in this section , all labels are considered modulo @xmath0 . we have shown , via the grassmann - plcker relations , that the set of circular pairs corresponding to the positive circular minors of a response matrix is an electrical positroid . we now prove that , for all electrical positroids @xmath6 , there exists a critical graph @xmath51 for which @xmath106 , which will establish theorem [ circposthm ] . the idea of the argument is as follows . assume , for sake of contradiction , that there exists some electrical positroid @xmath6 for which there does not exist such a critical graph @xmath51 with @xmath106 . then , let @xmath107 have maximal size among all such electrical positroids . note that @xmath107 does not contain all circular pairs @xmath29 , because otherwise @xmath108 , where @xmath109 denotes a critical representative of the top - rank element of @xmath10 ( see @xcite ) . we will then add circular pairs to @xmath107 according to the boundary edge and boundary spike properties ( cf . @xcite ) , discussed below , to form an electrical positroid @xmath110 . by the maximality of @xmath107 , @xmath111 for some critical graph @xmath112 . we will then delete a boundary edge or contract a boundary spike in @xmath112 to obtain a graph @xmath113 , and show that @xmath114 . we begin by defining two properties of circular pairs , the @xmath115-boundary edge property and the @xmath116-boundary spike property . let us first adopt a notational convention . given a circular pair @xmath29 , let @xmath117 denote the unique circular pair ( if it exists ) with @xmath118 and @xmath119 as sets . in the ordered sets @xmath120 , @xmath121 are inserted in the appropriate positions so that @xmath117 is indeed a circular pair . given arbitrary @xmath122 , @xmath117 may not be a circular pair . however , whenever we make reference to a pair of this form without commenting on its existence , we assert implicitly that it is , in fact , a circular pair . we say that a set @xmath6 of circular pairs has the * @xmath115-bep ( boundary edge property ) * if , for all circular pairs @xmath54 , if @xmath123 is a circular pair , then @xmath124 . according to remark [ flippq ] , if @xmath6 has the @xmath115-bep , then if @xmath54 and @xmath125 is a circular pair , then @xmath126 . [ graphbepbsp ] recall the definitions of boundary edges and boundary spikes from @xcite . let @xmath51 be a circular planar graph , and write @xmath127 . 1 . there exists @xmath128 with a boundary edge @xmath115 if and only if @xmath6 has the @xmath115-bep . there exists @xmath128 has a boundary spike at @xmath116 if and only if @xmath6 has the @xmath116-bsp . we prove ( a ) ; the proof of ( b ) is similar . without loss of generality , we may assume that @xmath129 . it easy to check that if @xmath51 has a boundary edge , then @xmath6 must have the corresponding bep . conversely , suppose that @xmath6 has the @xmath130-bep . then , let @xmath131 be the graph obtained by adding an edge @xmath130 in @xmath51 such that the added edge does not cut through any faces of @xmath51 . clearly , @xmath132 consists only of circular pairs @xmath29 such that @xmath133 or @xmath134 . however , @xmath6 contains all such circular pairs , so in fact @xmath135 . then , by theorem [ equivalentconnections ] , @xmath136 . it is left to check that @xmath137 , for some circular planar graph @xmath138 with the boundary edge @xmath130 . let @xmath60 be the two medial boundary vertices between @xmath139 and @xmath140 in @xmath141 . note that adding the edge @xmath130 to @xmath51 corresponds to introducing an additional crossing in @xmath141 between the ( distinct ) wires with endpoints @xmath62 and @xmath63 . introducing this new crossing yields an equivalent medial graph , so it must have created it a lens . from here , it is easily seen , after applying ( * ? ? ? * lemma 6.3 ) , that motions may be applied in @xmath142 so that this lens corresponds to parallel edges between the boundary vertices @xmath139 and @xmath140 in some @xmath143 . the desired conclusion follows . [ allbepbsp ] if @xmath6 has all @xmath0 beps and all @xmath0 bsps , then @xmath6 contains all circular pairs . we proceed by induction on the size of @xmath29 that @xmath54 for all circular pairs @xmath29 . first , suppose that @xmath144 . first , @xmath145 for all @xmath116 , because it has all beps and @xmath103 . then , because @xmath6 has the @xmath116-bsp , and @xmath146 we obtain @xmath147 . continuing in this way gives that @xmath6 contains all circular pairs @xmath29 with @xmath144 . now , suppose that @xmath6 contains all circular pairs of size @xmath148 . let @xmath149 be a circular pair of size @xmath31 . by assumption , @xmath150 . because @xmath6 has all beps , @xmath151 and @xmath152 , so by the @xmath153-bsp , @xmath154 . continuing in this way gives @xmath155 , so we have the desired claim . in particular , lemma [ allbepbsp ] tells us that there exists an @xmath116 such that @xmath107 , as defined in the beginning of this section , either does not have the @xmath115-bep for some @xmath116 , or does not have the @xmath116-bsp for some @xmath116 . we first assume that @xmath107 does not have all beps ; without loss of generality , suppose that @xmath107 does not have the @xmath156-bep . we will now add circular pairs to @xmath107 to obtain an electrical positroid @xmath110 that does have the @xmath156-bep . specifically , we add to @xmath107 every circular pair @xmath157 , where @xmath158 has @xmath159 ( here @xmath160 ) , to obtain @xmath110 . according to remark [ flippq ] , this construction also puts any @xmath161 , where @xmath158 and @xmath162 . @xmath163 is an electrical positroid , and has the @xmath156-bep . the proof is straightforward , so it is omitted . by assumption , @xmath107 is the maximal electrical positroid for which any circular planar graph @xmath51 has @xmath164 . thus , there exists a graph @xmath112 be a graph such that @xmath165 , and @xmath112 may be taken to have a boundary edge @xmath156 by lemma [ graphbepbsp ] . then , let @xmath113 be the result of deleting the boundary edge @xmath156 . to obtain a contradiction , it is enough to prove that @xmath166 . we now present a series of technical lemmas . [ completedef ] consider a circular pair @xmath158 for which @xmath167 . we will assume , for the rest of this section , that @xmath157 is a circular pair . @xmath29 is said to be is * incomplete * if @xmath168 , and * complete * if @xmath169 . [ crossing ] let @xmath170 be an incomplete circular pair , such that @xmath157 is a circular pair ( and is not in @xmath107 ) . furthermore , assume that @xmath29 is * minimal * , that is , @xmath171 is complete . then , for all @xmath172 , @xmath173 . immediate from axiom [ axside ] of definition [ posaxioms ] . [ trip ] let @xmath174 be a circular pair . then , if @xmath175 , then @xmath176 . immediate from axiom [ axcross ] . [ walk ] if @xmath177 , @xmath178 , and @xmath179 appear in clockwise order , then @xmath180 . if @xmath181 and @xmath182 , the claim follows from axiom [ axleft ] , axiom [ axright ] and induction on @xmath0 . otherwise , it follows from axiom [ axside ] and induction on @xmath0 . [ min ] let @xmath170 be a complete circular pair . then , @xmath183 is complete for all @xmath184 . applying axiom [ axright ] with @xmath185 to @xmath186 gives @xmath187 . then another application of axiom [ axright ] , to @xmath188 with @xmath189 gives the desired result . [ rtol ] let @xmath190 be a circular pair , where @xmath191 are sequences of boundary vertices . suppose that @xmath192 then , @xmath193 . first , write @xmath194 , where @xmath195 and @xmath196 are the last elements of the ordered sets @xmath197 , respectively . then , if @xmath198 , an application of axiom [ axleft ] on @xmath199 with @xmath200 yields @xmath201 . similarly , we find by induction that @xmath202 . then , we have @xmath203 by axiom [ axleft ] applied to @xmath204 with @xmath205 . similarly , write @xmath206 , where @xmath207 are the first elements of @xmath208 , respectively . by axiom [ axleft ] applied to @xmath209 with @xmath210 , we see that @xmath211 . the lemma then follows from axiom [ axright ] applied to @xmath212 with @xmath213 . let @xmath191 be sequences of indices , and let @xmath214 be a circular pair . suppose @xmath215 then @xmath216 . with the same notation as in the previous lemma , @xmath217 by axiom [ axright ] on @xmath218 with @xmath219 . then , an inductive argument shows that @xmath220 . a similar argument shows that @xmath221 . then , axiom [ axright ] applied to @xmath222 with @xmath223 implies that @xmath224 and applying axiom [ axright ] again to @xmath225 with @xmath226 yields the desired result . [ pa1qbn ] consider a circular pair @xmath227 , and let @xmath228 be an incomplete circular pair with @xmath229 in clockwise order . then , any electrical positroid @xmath230 satisfying @xmath231 contains @xmath232 . it is easy to see that any element of @xmath233 must be of the form @xmath234 , for some @xmath235 . by axiom [ axside ] , @xmath236 and @xmath237 implies that either @xmath238 , or @xmath239 and @xmath240 . we are done in the former case , so assume for sake of contradiction that we have the latter . @xmath241 are not of the form @xmath242 , so can not lie in @xmath243 ; thus , @xmath244 . finally , axiom [ axcross ] yields us @xmath245 , a contradiction , so we are done . two pairs of indices @xmath246 and @xmath247 are said to * cross * if @xmath248 and @xmath249 . [ betternotation ] for ease of notation , denote the sequence of indices @xmath250 by @xmath251 . we now algorithmically construct a set @xmath252 of circular pairs , which we will call * primary * circular pairs . we will use this notion to eventually prove lemma [ uniclos ] , a key ingredient in our proof of the main theorem . the construction is as follows : begin by placing @xmath253 . then , for each @xmath254 , if we also have @xmath158 , perform the following operation . * let @xmath62 be the first index appearing clockwise from @xmath255 such that there exists @xmath64 with @xmath256 crossing @xmath257 , and also @xmath258 . if @xmath62 does not exist , stop . otherwise , with @xmath62 fixed , take @xmath64 to be the first index appearing counterclockwise from @xmath259 satisfying these properties . * if @xmath62 exists , add @xmath260 to @xmath252 , and remove @xmath261 . * similarly , let @xmath63 to be the largest index counterclockwise from @xmath259 such that there exists @xmath65 with @xmath262 crossing @xmath257 and @xmath263 . if @xmath63 does not exist , stop . otherwise , with @xmath63 fixed , take @xmath65 to be the first index clockwise from @xmath255 with these properties . * if @xmath264 and @xmath265 ( note that if @xmath266 , then @xmath267 ) , then add @xmath268 to @xmath252 . note that @xmath269 or else , by [ axside ] , @xmath64 could originally have been set to @xmath65 . it is easily seen that at any time , the algorithm may be performed on the elements of @xmath252 in any order , and that it will eventually terminate , when the operation described above results in no change in @xmath252 for all @xmath270 . [ connset ] for a circular pair @xmath271 , define @xmath272 . we will take @xmath273 to be an ordered set and abusively refer to its elements as * connections*. [ pcpcont ] for any incomplete circular pair @xmath29 , there exists a circular pair @xmath274 such that any electrical positroid @xmath230 satisfying @xmath275 contains @xmath157 . by lemma [ pa1qbn ] , we may assume that @xmath29 is a minimal incomplete circular pair . let @xmath276 ( see definition [ betternotation ] ) . consider the primary circular pairs whose first @xmath116 connections are the same as those of @xmath29 . by the construction of @xmath252 , there are at most two such primary circular pairs , which we denote by @xmath277 by lemma [ crossing ] and the construction of @xmath252 , we have that @xmath278 and @xmath279 or @xmath280 and @xmath281 ( or else we would have been able to set @xmath282 or @xmath283 ) . furthermore , exactly one of these pairs of inequalities holds . let us assume that the former holds , as the latter case is identical , and in this case , call @xmath284 the primary circular pair associated to @xmath29 . if , on the other hand , @xmath284 ( as above ) is the only primary circular pair sharing its @xmath116 connections with @xmath29 , then @xmath278 and @xmath279 , and we still refer to @xmath284 as the primary circular pair associated to @xmath29 . we now prove the lemma by retrograde induction on @xmath116 , where here @xmath116 is such that the first @xmath116 connections of @xmath29 are shared with some primary circular pair . if @xmath285 , we are done by lemma [ pa1qbn ] , and if the first @xmath116 connections of @xmath29 are exactly the primary circular pair in question , we are done by the subset axiom . otherwise , we first need @xmath286 , which follows from lemmas [ trip ] and [ rtol ] , where the conditions of these lemmas are satisfied as a result of lemma [ crossing ] . it is easy to see that the primary circular pair associated to @xmath287 is the same as that for @xmath29 . it follows , then , by the inductive hypothesis , that @xmath288 . when @xmath289 and @xmath290 , lemma [ ltol ] yields the desired result , and if one of @xmath291 or @xmath292 , we are done by a similar argument . [ unpcp ] there is exactly one circular pair in @xmath252 that does not lie in @xmath107 , which we call the * @xmath107-primary circular pair*. by lemma [ pcpcont ] , @xmath293 has at least one element , because @xmath107 does not have the @xmath156-bep . assume , for sake of contradiction , that @xmath293 has two elements , of the form @xmath294 because @xmath295 and @xmath296 , we must have @xmath297 , by axiom [ axside ] . thus , @xmath298 , by the subset axiom . by the same argument applied to @xmath299 , we must have @xmath300 , so axiom [ axcross ] gives @xmath301 . however , because @xmath302 , we have a contradiction of the definition of @xmath65 . thus , @xmath303 . [ contpcp ] for any incomplete circular pair @xmath29 , any electrical positroid @xmath230 satisfying @xmath231 contains the @xmath107-primary circular pair . proceed by retrograde induction on @xmath116 , where @xmath116 is such that the first @xmath116 connections of @xmath29 are the same as those of some primary circular pair . by the subset axiom , we can assume that @xmath29 is minimal . the base case is immediate from the subset axiom , so suppose that @xmath304 . let @xmath305 then , we need to show that , if @xmath236 , then @xmath306 . first , suppose that both @xmath307 and @xmath308 . then , the desired claim is exactly lemma [ rtol ] , as long as @xmath309 . assume , then , that @xmath310 . first , an application of lemma [ rtol ] yields @xmath311 , which implies @xmath312 and @xmath313 . furthermore , if @xmath314 , then axiom [ axleft ] yields @xmath315 , a contradiction . similarly , we have @xmath316 . as a result , applying axiom [ axmid ] to @xmath317 with @xmath318 gives @xmath319 . one more application of axiom [ axmid ] to @xmath320 with @xmath321 yields the desired result . if one of the indices @xmath291 or @xmath292 , then we are also done by a similar argument . [ uniclos]for any two incomplete circular pairs @xmath29 and @xmath322 , any electrical positroid @xmath230 satisfying @xmath323 must also contain @xmath242 . by lemma [ contpcp ] , @xmath230 must contain the @xmath107-primary circular pair . the claim then follows by lemma [ pcpcont ] . by the above results , if we start with our set @xmath107 and some incomplete circular pair @xmath158 , `` completing '' @xmath29 by adding @xmath157 to @xmath107 will require that we have completed every incomplete pair . we now finish the proof of theorem [ circposthm ] , in the boundary edge case . let @xmath324 denote the subset of circular pairs in @xmath107 without the connection @xmath325 , and define @xmath326 similarly for @xmath327 , respectively . by construction , it is easily seen that @xmath328 . while @xmath329 may not necessarily be an electrical positroid , we have : [ findclos ] there exists an electrical positroid @xmath330 with @xmath331 . we give an algorithm to construct such an electrical positroid @xmath330 . we begin by setting @xmath332 ; note that @xmath330 satisfies the last two electrical positroid axioms , but may not satisfy the first six . each of the first six axioms are of the form @xmath333 , or otherwise @xmath333 or @xmath334 . at each step of the algorithm , if @xmath330 is an electrical positroid , we stop , and if not , we pick an electrical positroid axiom @xmath335 ( among the first six ) not satisfied by @xmath336 . we then show that we can add elements of @xmath337 to @xmath330 so that @xmath335 is satisfied by @xmath338 , and so that @xmath330 also still satisfies the subset axiom . it is clear that adding circular pairs to @xmath330 in this way is possible when @xmath335 is one of axioms [ axcross ] , [ axsep ] , [ axleft ] , and [ axright ] : we take the add circular pairs @xmath339 , as above , as well as all of the circular pairs formed by subsets of their respective connections . we will show that this operation is also possible when @xmath335 is one of axioms [ axside ] and [ axmid ] . from here , it will be clear that the algorithm must terminate , because we can only add finitely many elements to @xmath330 . therefore , we will eventually find @xmath330 with the desired properties . in each of the cases below , the circular pairs added to @xmath330 are always assumed to be added along with each of their subsets , that is , the circular pairs formed by subsets of their connections . in this way , the subset axiom is satisfied by @xmath330 at all steps in the algorithm . we first consider axiom [ axside ] , which we assume to fail in @xmath330 when applied to @xmath340 . if @xmath341 , either @xmath342 or @xmath343 . it is easy to see that @xmath344 and @xmath345 ( or vice versa , but we can swap @xmath34 and @xmath37 and reverse their orders ) , or else axiom [ axside ] already would have been satisfied by @xmath340 . we proceed by casework : * @xmath346 . then , because axiom [ axside ] fails , we have @xmath347 . thus , we may add @xmath348 to @xmath330 , and these lie in @xmath349 . * @xmath350 . we have @xmath351 . first , suppose that @xmath352 . because @xmath353 does not contain the connection @xmath325 , we have @xmath354 . then , @xmath355 , so we may add may @xmath29 to @xmath330 , so that axiom [ axside ] is satisfied with @xmath351 ( note that @xmath356 is already in @xmath330 ) . + now , suppose instead that @xmath357 . if @xmath358 , then @xmath359 by the subset axiom . then , @xmath355 , and so we may add @xmath29 to @xmath330 to satisfy axiom [ axside ] . now , assume that @xmath360 . for any electrical positroid @xmath361 , axiom [ axleft ] applied to @xmath362 and @xmath363 gives that @xmath364 and @xmath365 implies @xmath366 and @xmath367 . by the discussion above , we have @xmath368 , and so we may add @xmath369 to @xmath330 . the case in which @xmath370 is identical . * the case @xmath371 may be handled using similar logic ; the details are left to the reader . finally , consider axiom [ axmid ] , which we assume to fail for @xmath372 . as before , we may assume @xmath373 . * @xmath374 , or @xmath375 . similar to the first case above . * @xmath376 . then , we have @xmath377 . as in the second case for axiom [ axside ] , we may assume that we have @xmath378 and @xmath379 , or else both @xmath107 and @xmath380 would contain exactly one of @xmath381 and @xmath382 . moreover , we may assume that we have @xmath383 by similar logic . because @xmath384 and @xmath383 , we may apply axiom [ axsep ] to find that @xmath385 . thus , we can add @xmath386 to @xmath330 , so that we still have @xmath387 . the case @xmath388 is identical . * the cases @xmath389 and @xmath390 may be handled using similar logic ; we again omit the details . thus , in all cases , our algorithm is well - defined , and we are done . by lemma [ uniclos ] , we must in fact have @xmath391 , provided that neither @xmath107 nor @xmath380 is equal to @xmath110 , which is true by construction ( recall that @xmath392 is critical ) . the proof is complete , in the boundary edge case . it is left to consider the case in which @xmath107 has the @xmath115-bep , for each @xmath116 , but fails to have the @xmath116-bsp , for some @xmath116 . without loss of generality , suppose that @xmath107 does not have the @xmath139-bsp . we now form @xmath110 as the union of @xmath107 and the set of all circular pairs @xmath117 such that @xmath393 , where @xmath29 is a circular pair with @xmath394 . in appendix [ bspcase ] , we form a circular planar graph @xmath112 such that @xmath165 and @xmath112 has a boundary spike at 1 . then , contracting this boundary spike to obtain the graph @xmath113 , we find that @xmath114 . therefore , with the additional results of appendix [ bspcase ] , the theorem is proven . we now study the lp algebra @xmath11 . our starting point will be _ positivity tests _ ; a particular positivity test will form the initial seed in @xmath11 . we then proceed to investigae the algebraic and combinatorial properties of clusters in @xmath11 . let @xmath9 be a symmetric @xmath1 matrix with row and column sums equal to zero . in this section , we describe tests for deciding if @xmath9 is the response matrix for an electrical network @xmath14 in the top cell of @xmath10 , defined in @xcite . that is , we describe tests for deciding if all of the circular minors of @xmath9 are positive . these tests are similar to certain tests for total positivity described in @xcite . throughout the remainder of this section , all indices around the circle are considered modulo @xmath0 , and we will refer to circular pairs and their corresponding minors interchangeably . a set @xmath6 of circular pairs is a * positivity test * if , for all matrices @xmath9 whose minors corresponding to @xmath6 are positive , every circular minor of @xmath9 is positive ( equivalently , @xmath9 is the response matrix for a top - rank electrical network ) . we begin by describing a positivity test of size @xmath8 . fix @xmath0 vertices on a boundary circle , labeled @xmath105 in clockwise order . for two points @xmath395 $ ] , let @xmath396 denote the number of boundary vertices on the arc formed by starting at @xmath62 and moving clockwise to @xmath63 , inclusive . a circular pair @xmath397 is called * solid * if both sequences @xmath398 and @xmath399 appear consecutively in clockwise order around the circle . write @xmath400 , and @xmath401 . we will call a solid circular pair @xmath29 * picked * if one of the following conditions holds : * @xmath402 and @xmath403 , or * @xmath404 and @xmath405 let @xmath9 be a fixed symmetric @xmath1 matrix . define the set of * diametric pairs * @xmath12 to be the set of solid circular pairs @xmath29 such that either @xmath406 or @xmath407 and @xmath29 is picked . we will refer to the elements of @xmath12 as circular pairs and minors interchangeably . it is easily checked that @xmath408 . for a solid circular pair @xmath29 , we have that @xmath409 , so @xmath12 consists of the solid circular pairs with @xmath410 when @xmath0 is odd , and the solid circular pairs with either @xmath411 , or @xmath407 and @xmath29 is picked when @xmath0 is even . recall ( see remark [ flippq ] ) that the circular pairs @xmath29 and @xmath32 will be regarded as the same . note , for example , that @xmath412 if and only if @xmath413 , so the definition of @xmath12 is compatible with this convention . [ initialseed ] if @xmath9 is taken to be an @xmath1 symmetric matrix of indeterminates , any circular minor is a positive rational expression in the determinants of the elements of @xmath12 . we will make use of the grassmann - plcker relations , ( [ p1 ] ) and ( [ p2 ] ) , so we repeat them here : for @xmath73 a circular pair , @xmath414 and for @xmath415 in clockwise order , @xmath416 we will first show , by induction on @xmath417 , that any solid circular pair is a positive rational expression in the elements of @xmath12 . there is nothing to check when @xmath417 is equal to 0 ( when @xmath0 is even ) or 1 ( when @xmath0 is odd ) . if @xmath418 is a solid circular pair such that @xmath407 ( hence , @xmath0 is even ) and @xmath29 is not picked , then either @xmath419 or @xmath420 is a solid circular pair with @xmath407 , and it must be picked . assume , without loss of generality , that it is the former , and let @xmath421 , @xmath422 . letting @xmath423 , we have , by ( [ p1 ] ) , that : @xmath424 because @xmath425 is not picked , @xmath426 corresponds to a picked circular pair with @xmath407 , and @xmath427 and @xmath428 all have @xmath411 , so we have that @xmath29 is a rational expression of elements of @xmath12 . now , for @xmath429 , assume that all solid pairs with @xmath430 are positive rational expression in the elements of @xmath12 , and consider a solid pair @xmath29 with @xmath431 . then , either @xmath419 or @xmath420 is a solid circular pair with @xmath430 . assume , without loss of generality , that the former is the case . then , we may again set @xmath421 , @xmath422 , and @xmath432 , and ( [ apply1 ] ) still holds . each term on the right hand side corresponds to a solid pair with a smaller value of @xmath433 , so @xmath29 is a positive rational expression in the elements of @xmath12 , by the inductive hypothesis . we now show that any circular pair is a positive rational expression in the elements of @xmath12 . for a sequence @xmath434 of points ordered clockwise around the circle , let @xmath435 be the largest index such that @xmath436 are consecutive . if @xmath437 , then define @xmath438 and @xmath439 . if , on the other hand , @xmath440 , define @xmath441 . similarly , for a sequence @xmath442 of points ordered counterclockwise around the circle , let @xmath443 be the smallest index such that @xmath444 are consecutive . if @xmath445 , then define @xmath446 and @xmath447 . if , on the other hand , @xmath448 , define @xmath441 . for a circular pair @xmath29 , define @xmath449 , and note that @xmath450 , so @xmath451 is well - defined when we impose the convention @xmath104 . we finish the proof by showing that any circular pair is a positive rational expression in the elements of @xmath12 by induction on @xmath451 . if @xmath452 , then @xmath29 is solid and hence a rational expression in the elements of @xmath12 . now , assume that for any @xmath453 , every circular pair @xmath322 with @xmath454 is a rational expression in the elements of @xmath12 , and consider a circular pair @xmath29 with @xmath455 . assume , without loss of generality , that @xmath456 , and let @xmath457 . applying ( [ p1 ] ) to @xmath458 , we get : @xmath459 each term on the right hand side is easily seen to have a smaller value of @xmath451 than @xmath460 , so we are done by induction . [ postestcor ] @xmath12 is a positivity test . the positive rational expressions from the previous section are reminiscent of a cluster algebra structure ( see @xcite for definitions ) . in fact , ( [ p1 ] ) and ( [ p2 ] ) are exactly the exchange relations for the local moves in double wiring diagrams ( * ? ? ? * figure 9 ) . due to parity issues similar to those encountered when attempting to associate a cluster algebra to a non - orientable surface in @xcite , the structure of positivity tests is slightly different from a cluster algebra . we present the structure in two different ways : first , as a laurent phenomenon ( lp ) algebra @xmath11 ( see @xcite for definitions ) , and secondly as a cluster algebra @xmath13 similar to the double cover cluster algebra in @xcite . @xmath11 , we will find , is isomorphic to the polynomial ring on @xmath8 variables , but more importantly encodes the information of the positivity of the circular minors of a fixed @xmath1 matrix . we begin by describing an undirected graph @xmath461 that encodes the desired mutation relations among our initial seed . the vertex set of @xmath461 will be @xmath462 . a solid circular pair @xmath463 is called * maximal * if @xmath464 or @xmath465 and @xmath466 . a solid circular pair @xmath467 is called * limiting * if @xmath407 , @xmath29 is picked , and @xmath468 or @xmath469 . let us now describe the edges of @xmath470 : see figure [ q8 ] for an example . for each @xmath471 that is not maximal , limiting , or empty ( that is , equal to @xmath472 ) , there is a unique way to substitute values in [ p1 ] such that @xmath29 appears on the left hand side , and all four terms on the right hand side are in @xmath473 . we draw edges from @xmath29 to these four vertices in @xmath461 . finally , if @xmath474 are limiting , we draw an edge between them if their sizes differ by @xmath139 . the edges drawn in these two cases constitute all edges of @xmath475 . for any maximal circular pair @xmath29 , it can be proven that there exists a symmetric matrix @xmath3 such that @xmath3 is positive on any circular pair except @xmath476 . in fact , if @xmath29 is maximal and has @xmath406 , then the set of all circular pairs other than @xmath29 is an electrical positroid . hence , in our quivers , we will take the vertices corresponding to the maximal circular pairs and @xmath472 to be frozen . @xmath461 can then be embedded in the plane in a natural way with the circular pairs of size @xmath31 lying on the circle of radius @xmath31 centered at @xmath472 , and all edges except those between vertices corresponding to limiting circular pairs either along those circles or radially outward from @xmath472 . depicting the desired exhange relations among @xmath477 . vertices marked as squares correspond to frozen variables . @xmath478 , @xmath479 and @xmath480 are the limiting circular pairs . ] if we could orient the edges of @xmath461 such that they alternate between in- and out - edges at each non - frozen vertex , then the resulting quiver would give a cluster algebra such that mutations at vertices whose associated cluster variables are neither frozen not limiting correspond to the relation grassmann - plcker relation ( [ p1 ] ) . furthermore , these mutation relations among the vertices of @xmath473 constitute all of the grassmann - plcker relations for which five of the six terms on the right hand side are elements in @xmath473 , and the term which is not in @xmath475 is on the left hand side of the relation ( [ p1 ] ) or ( [ p2 ] ) . however , for @xmath481 , such an orientation of the edges of @xmath470 is impossible , because the dual graph of @xmath470 contains odd cycles . we thus define : [ deflmn ] let @xmath11 be the lp algebra constructed as follows : the initial seed @xmath482 has cluster variables equal to the minors in @xmath473 , with the maximal pairs and @xmath472 frozen , and , for any other @xmath471 , the exchange polynomial @xmath483 is the same as what is obtained from a quiver with underlying graph @xmath470 , such that the edges around the vertex associated to @xmath29 in @xmath461 alternate between in- and out - edges . for example , in @xmath484 , the exchange polynomial associated to the cluster variable @xmath485 is @xmath486 . we need the additional technical condition that @xmath483 is irreducible as a polynomial in the cluster variables @xmath475 , but the irreducibility is clear . we next define a cluster algebra @xmath13 which is a double cover of positivity tests , in the following sense : we begin consider an @xmath1 matrix @xmath57 , which we no longer assume to be symmetric . we write * non - symmetric circular pairs * in the row and column sets of @xmath57 as @xmath487 , so that @xmath487 and @xmath488 now represent different circular pairs . we will say that two expressions @xmath489 in the entries of @xmath57 * correspond * if swapping the rows and columns for each entry in @xmath3 gives @xmath490 , and we will write @xmath491 . for instance , @xmath492 . the set of cluster variables @xmath493 in our initial seed will consist of pairs @xmath487 such that @xmath494 . note that @xmath495 , as @xmath493 contains @xmath487 and @xmath488 for each @xmath496 , and finally @xmath472 . @xmath487 will be frozen in @xmath497 if @xmath29 was frozen in @xmath473 . we construct the undirected graph @xmath498 with vertex set @xmath497 by adding edges in the same way that @xmath461 was constructed . the only difference in our description is that if @xmath474 are limiting , then they will be adjacent only if their sizes differ by @xmath139 and @xmath499 . see figure [ qp ] for an example . . in the quiver @xmath500 , the edges alternate directions around a non - frozen vertex . ] unlike in @xmath461 , the edges of @xmath498 can be oriented such that they are alternating around each non - frozen vertex . let @xmath501 be the quiver from either orientation . then , let @xmath502 be the cluster algebra with initial quiver @xmath501 . breaking the symmetry of @xmath57 removed the parity problems from @xmath461 , so that we could define a cluster algebra , but we are still interested in using @xmath498 to study @xmath9 when @xmath9 is symmetric . toward this goal , we can restrict ourselves so that whenever we mutate at a cluster variable @xmath503 , we then mutate at @xmath504 immediately afterward . call this restriction the * symmetry restriction*. [ symm ] after the mutation sequence @xmath505 from the initial seed in @xmath502 , the number of edges from @xmath506 to @xmath507 in the quiver is equal to the number of edges from @xmath508 to @xmath509 for each @xmath121 in the final quiver . we proceed by induction on @xmath196 ; for @xmath510 , we have the claim by construction . now , suppose that we have performed the mutations @xmath511 and currently have the desired symmetry property . by the inductive hypothesis , @xmath512 and @xmath513 are not adjacent , or else we would have had edges between them in both directions , which would have been removed after mutations . thus , no edges incident to @xmath513 are created or removed upon mutating at @xmath512 . hence , mutating at @xmath513 afterward makes the symmetric changes to the graph , as desired . let @xmath514 $ ] and @xmath515 $ ] denote the polynomial rings in the off - diagonal entries of @xmath9 and @xmath57 respectively ; recall that @xmath9 is symmetric , so @xmath516 . then , we can define the * symmetrizing homomorphism * @xmath517 \to { \mathbb{c}}[m]$ ] by its action on the off - diagonal entries of @xmath57 : @xmath518 if @xmath6 is a set of polynomials in @xmath515 $ ] , then write @xmath519 . [ isom ] let @xmath520 be the cluster of @xmath502 that results from starting at the initial cluster and performing the sequence of mutations @xmath505 . let @xmath521 be the cluster of @xmath522 that results from starting at the initial cluster and performing the sequence of mutations @xmath523 . then , @xmath524 . using lemma [ symm ] , the proof is similar to ( * ? ? ? * proposition 4.4 ) . in light of lemma [ isom ] , we may understand the clusters in @xmath11 by forming `` double - cover '' clusters in @xmath13 . a sequence @xmath525 of mutations in @xmath11 corresponds to a sequence @xmath526 of twice as many mutations in @xmath13 , where we impose the symmetry restriction , and the cluster variables in @xmath11 after applying @xmath525 are the symmetrizations of those in @xmath13 after applying @xmath526 . [ lppos ] any cluster @xmath6 of @xmath11 consisting entirely of circular pairs is a positivity test . in @xmath502 , the exchange polynomial has only positive coefficients , so each variable in any cluster is a rational function with positive coefficients in the variables of any other cluster . in particular , each non - symmetric circular pair in @xmath527 is a rational function with positive coefficients in the variables of any cluster reachable under the symmetry restriction . hence , by lemma [ isom ] , each circular pair in @xmath528 can be written as a rational function with positive coefficients of the variables in @xmath6 . the desired result follows easily . as with double wiring diagrams for totally positive matrices @xcite , and plabic graphs for the totally nonnegative grassmannian @xcite , we now restricting ourselves to certain types of mutations in @xmath522 . a natural choice is mutations with exchange relations of the form [ p1 ] or [ p2 ] . these mutations keep us within clusters consisting entirely of circular minors , the `` plcker clusters . '' we begin by restricting ourselves only to mutations with exchange relations of the form [ p1 ] . because the initial seed @xmath529 consists only of solid circular pairs , we will only be able to mutate to other clusters consisting entirely of solid circular pairs . our goal is to chracterize these clusters . we will be able to write down such a characterization using corollary [ symallconsec ] and lemma [ isom ] , and give a more elegant description of the clusters in proposition [ wsconsec ] . let @xmath530 be a non - symmetric , non - empty circular pair . define the statistics @xmath531 , @xmath532 , and @xmath533 by : @xmath534 a non - symmetric solid circular pair @xmath487 is uniquely determined by the triple @xmath535 . a necessary condition for a triple @xmath536 to correspond to a non - symmetric solid circular pair is that @xmath537 . when the terms are non - symmetric solid circular pairs , ( [ p1 ] ) can be written using these triples as : @xmath538 after a mutation at @xmath539 , with non - symmetric solid circular pairs labeled by triples @xmath536 . in the quiver , the edges alternate directions around each non - frozen vertex . ] after a mutation at @xmath539 , with non - symmetric solid circular pairs labeled by triples @xmath536 . in the quiver , the edges alternate directions around each non - frozen vertex . ] we call two non - symmetric solid circular pairs corresponding to the triples @xmath540 and @xmath541 * adjacent * if @xmath542 and @xmath543 , or @xmath544 and @xmath545 . we call @xmath487 and @xmath546 * diagonally adjacent * if there are two non - symmetric solid circular pairs @xmath547 which are both adjacent to both @xmath487 and @xmath546 . we call @xmath547 the * connection * of @xmath548 . note that , in the initial quiver @xmath500 , adjacent and diagonally adjacent circular pairs correspond to vertices which are adjacent in particular ways . specifically , adjacent circular pairs correspond to vertices which are adjacent on the same concentric circle , or along the same radial spoke of @xmath549 . diagonally adjacent circular pairs correspond to those which are adjacent via all other edges , the `` diagonal '' edges . we can now classify clusters of @xmath13 which can be reached only using the mutations with exchange relation ( [ p1 ] ) . [ solidcluster ] we call a set @xmath6 of @xmath550 non - symmetric solid circular pairs a * solid cluster * if it has the following properties : * @xmath551 , * for each integer @xmath552 , and each @xmath553 , unless @xmath554 and @xmath330 is an integer , there is a @xmath15 such that the non - symmetric solid circular pair corresponding to @xmath536 is in @xmath6 , and * if @xmath555 and @xmath487 is adjacent to @xmath546 , then @xmath556 . [ embed ] there is a natural embedding of a solid cluster @xmath6 in the plane , similar to our embedding of @xmath498 . we place @xmath557 at any point , and then pairs of size @xmath31 on the circle of radius @xmath31 centered at that point . moreover , we place adjacent pairs of the same size consecutively around each circle , and adjacent pairs of different sizes collinear with @xmath557 . for a solid cluster @xmath6 of @xmath13 and associated quiver @xmath558 , we call @xmath559 a * solid seed * if it has the following properties : * vertices corresponding to maximal non - symmetric solid circular pairs are frozen , * there is an edge between any pair of adjacent vertices that are not both frozen , * there is an edge between diagonally adjacent vertices @xmath548 if their connection @xmath560 satisfies @xmath561 , * there is an edge from a size 1 vertex @xmath487 to @xmath557 if it would make the degree of @xmath487 even , * all edges of @xmath558 are in drawn in one of the four ways described above , and * all edges are oriented so that , in the embedding described in remark [ embed ] , edges alternate between in- and out - edges around any non - frozen vertex . if , furthermore , @xmath562 if and only if @xmath563 , or equivalently , the non - symmetric solid circular pair corresponding to @xmath536 is in @xmath6 if and only if that corresponding to @xmath564 is , then we call @xmath565 a * symmetric solid seed*. see , for example , figure [ bluesmutate ] . [ canmut ] in a solid seed @xmath566 , a variable @xmath567 has an exchange polynomial of the form ( [ p1 ] ) whenever its corresponding vertex in @xmath558 , and has edges to the vertices corresponding to its four adjacent variables in @xmath558 , and no other vertices . [ allconsec ] in @xmath502 , from the initial seed with cluster @xmath493 and quiver @xmath568 , mutations of the form ( [ p1 ] ) may be applied to obtain the seed @xmath569 if and only if @xmath569 is a solid seed . here , we do _ not _ impose the symmetry restriction . first , assume that @xmath569 via mutations of the form ( [ p1 ] ) . first , it is easy to check that @xmath570 is a solid seed . then , our mutations do indeed turn non - symmetric solid circular pairs into other non - symmetric solid circular pairs . furthermore , when we perform a mutation of the form ( [ p1 t ] ) at the vertex @xmath503 , the values of @xmath330 and @xmath31 do not change , and the value of @xmath15 changes from being either 2 more than the values of @xmath15 at the vertices adjacent to @xmath503 to being 2 less , or vice versa . hence , the resulting seed is also solid , so , by induction , @xmath569 is solid . conversely , assume @xmath569 is solid . we begin by noting that , by remark [ canmut ] , whenever the four terms on the right hand side of ( [ p1 t ] ) and one term on the left hand side are in our cluster , then we can perform the corresponding mutation . now , define @xmath571 to be the unique symmetric solid seed such that , for each @xmath572 , @xmath573 . for any solid seed @xmath569 , we give a mutation sequence @xmath574 using only mutations of the form ( [ p1 t ] ) that transforms @xmath569 into @xmath575 . hence , we will be able to get from the seed @xmath570 to @xmath569 by performing @xmath576 , followed by @xmath574 in reverse order . it is left to construct the desired mutation sequence . we define @xmath574 as follows : while the current seed is not @xmath571 , choose a vertex @xmath503 of the quiver , with associated cluster variable @xmath487 , for which the value of @xmath577 is maximized . we must have @xmath578 , and by maximality , for each vertex @xmath546 adjacent to @xmath487 , we must have @xmath579 . hence , we can mutate at @xmath29 to reduce @xmath577 by at least 2 . this process may be iterated to decrease the sum , over all cluster variables @xmath487 in our seed , of the @xmath577 , until we reach the seed @xmath571 . the proof is complete . [ symallconsec ] in @xmath502 , from our initial seed with cluster @xmath493 and quiver @xmath568 , we can apply symmetric pairs of mutations ( that is , mutations with the symmetry restriction ) of the form ( [ p1 ] ) to obtain the seed @xmath569 if and only if @xmath569 is a symmetric solid seed . the proof is almost identical to that of lemma [ allconsec ] . it suffices to note that in a symmetric solid seed , @xmath487 has a maximal value of @xmath580 if and only if @xmath581 does , so the mutation sequence @xmath574 can be selected to obey the symmetry restriction . using corollary [ symallconsec ] and lemma [ isom ] , we could also prove a similar result for the lp algebra @xmath522 . however , we will wait to do so until lemma [ wsconsec ] where we will describe it more elegantly using the notion of weak separation . we have now have the required machinery to prove our main theorem of this section . [ lpalg ] fix a symmetric @xmath1 matrix @xmath9 of distinct indeterminates . then , the @xmath582-laurent phenomenon algebra @xmath11 is isomorphic to the polynomial ring ( over @xmath582 ) on the @xmath8 non - diagonal entries of @xmath9 . we may directly apply the analogue of ( * ? ? ? * proposition 3.6 ) for lp algebras ( for which the proof is identical ) to @xmath522 as defined in definition [ deflmn ] . it is well - known that minors of a matrix of indeterminates are irreducible , so we immediately have that all of our seed variables are pairwise coprime . we also need to check that each initial seed variable is is coprime to the variable obtained by mutating its associated vertex in @xmath500 , and that this new variable is in the polynomial ring generated by the non - diagonal entries of @xmath9 . for non - limiting ( and non - frozen ) minors , this is clear , because each such mutation replaces a minor with another minor via ( [ p1 ] ) . for limiting minors , this is not the case , and we defer the proof to appendix [ nonplucker ] . it remains to check , then that each of the @xmath583 non - diagonal entries of @xmath57 appear as cluster variables in some cluster of @xmath13 . however , because @xmath74 minors are solid , the result is immediate from lemma [ symallconsec ] and lemma [ isom ] . because all @xmath74 minors are solid , mutations of the form ( [ p1 ] ) were sufficient to establish theorem [ lpalg ] . once we allow mutations of the form ( [ p2 ] ) , the clusters become more difficult to describe . however , let us propose the following conjecture , which has been established computationally for @xmath584 . [ getall ] every cluster of @xmath522 consisting entirely of circular pairs can be reached from the initial cluster using only mutations of the form ( [ p1 ] ) or ( [ p2 ] ) . we next introduce an analogue of weakly separated sets from @xcite for circular pairs . we recall the definition used in @xcite , @xcite , which is more natural in this case : two sets @xmath585 $ ] are * weakly separated * if there are no @xmath586 and @xmath587 such that @xmath588 or @xmath589 . our analogue is as follows : two circular pairs @xmath29 and @xmath590 are * weakly separated * if @xmath591 is weakly separated from @xmath592 , and @xmath593 is weakly separated from @xmath594 . note that @xmath29 is weakly separated from itself and from @xmath32 . furthermore , @xmath29 is weakly separated from @xmath590 if and only if @xmath32 is , so under the convention @xmath104 , weak separation is well - defined . [ wsconj ] let @xmath595 be a set of circular minors , for an @xmath1 generic response matrix . * @xmath595 is a minimal positivity test . * @xmath595 is a maximal set of pairwise weakly separated circular pairs . * @xmath595 is a cluster of @xmath11 . conjecture [ wsconj ] has been computationally verified for @xmath596 . we now prove various weak forms of this conjecture . first , for all clusters @xmath595 of @xmath11 that are reachable from the initial seed via grasmann - plcker relations ( cf . conjecture [ getall ] ) , the elements of @xmath595 are pairwise weakly separated : [ pluckerpreservesws ] if @xmath595 is a set of pairwise weakly separated circular pairs such that , for some substitution of values into ( [ p1 ] ) or ( [ p2 ] ) , all the terms on the right hand side , and one term @xmath29 on the left hand side , are in @xmath595 , then the remaining term @xmath590 on the left hand side is weakly separated from all of @xmath597 . let @xmath415 be as in ( [ p1 ] ) or ( [ p2 ] ) . it is clear that @xmath590 can only be non - weakly separated from an element of @xmath597 , if @xmath415 are boundary vertices forcing the non - weak separation .. however , this is easily seen to be impossible . when restricting ourselves to clusters of solid minors , the analogue of corollary [ symallconsec ] for @xmath522 matches exactly with a weak form of the equivalence @xmath598 in conjecture [ wsconj ] . [ wsconsec ] a set @xmath595 of solid circular pairs can be reached ( as a cluster ) from the initial cluster @xmath529 in @xmath522 using only mutations of the form ( [ p1 ] ) if and only if @xmath595 is a set of @xmath8 pairwise weakly separated solid circular pairs . the elements of the initial cluster @xmath529 in @xmath522 , which consists of the diametric pairs @xmath12 , are easily seen to be pairwise weakly separated . then , by proposition [ pluckerpreservesws ] , any cluster we can reach from @xmath529 using only mutations of the form ( [ p1 ] ) must also be pairwise weakly separated . conversely , consider any set @xmath595 of @xmath8 pairwise weakly separated solid circular pairs . let @xmath599 , and notice that @xmath600 . by corollary [ symallconsec ] and lemma [ isom ] , it is enough to prove that that @xmath601 is a solid cluster ( see definition [ solidcluster ] ) in @xmath13 . from here it will follow by definition that @xmath602 is a symmetric solid seed , meaning @xmath595 can be reached from @xmath529 in @xmath522 using only mutations of the form ( [ p1 ] ) , as desired . before proceeding , it is straightforward to check that circular pairs @xmath603 and @xmath604 are weakly separated if and only if the following four intersections are non - empty : @xmath605 we now prove that @xmath606 is a solid cluster . first , notice that if non - symmetric circular pairs @xmath607 and @xmath608 are such that @xmath609 and @xmath610 , but @xmath611 , then @xmath29 and @xmath612 are not weakly separated . hence , at most one of @xmath487 and @xmath546 is in @xmath602 . as there are exactly @xmath613 choices of @xmath330 and @xmath31 that give valid non - symmetric solid circular pairs , there must be one element of @xmath602 corresponding to each choice of @xmath614 . second , consider any adjacent @xmath487 and @xmath546 in @xmath602 . without loss of generality , one of * @xmath609 and @xmath615 , * @xmath610 and @xmath616 . holds . in either case , because @xmath487 and @xmath546 are weakly separated , we can see that @xmath556 . it follows that @xmath601 is a solid cluster , so we are done . we now relate @xmath617 and @xmath618 . recall that , by lemma [ lppos ] , any if @xmath595 satisfies @xmath617 , then @xmath595 is a positivity test . furthermore , @xmath619 . we can prove , similarly to ( * ? ? ? * theorem 1.2 ) , that : [ bound ] if @xmath595 satisfies @xmath620 , then @xmath621 . in fact , we can prove a slightly stronger result by interpreting a circular pair as a set of edges . for a circular pair @xmath271 , define @xmath272 ( cf . definition [ connset ] ) . similarly , for a set @xmath622 of edges such that no two edges in @xmath15 cross , let @xmath623 be the circular pair for which @xmath624 . [ strongbound ] if @xmath595 is a set of pairwise weakly separated circular pairs with @xmath625 then @xmath626 . procced by induction on @xmath627 . the case @xmath628 is trivial , so assume the result is true for @xmath629 . suppose that we have @xmath630 with @xmath631 , and assume for sake of contradiction that @xmath632 . choose some @xmath633 such that , for any other @xmath634 , @xmath64 and @xmath65 do not both lie on the arc drawn from @xmath62 to @xmath63 in the clockwise direction ( this arc is taken to include both @xmath62 and @xmath63 ) . now , letting @xmath635 , define the projection map @xmath636 by : @xmath637 we may define @xmath638 for circular pairs analogously : @xmath639 , and let @xmath640 . let us now prove two lemmas . [ wsinj ] the elements of @xmath602 are pairwise weakly separated . assume , for sake of contradiction , that we have @xmath641 , such that @xmath642 and @xmath643 are not weakly separated . if @xmath644 or @xmath645 , the claim is clear . thus , we may assume without loss of generality , that @xmath646 , @xmath647 , and @xmath648 . because @xmath642 and @xmath643 are not weakly separated , suppose , without loss of generality , that @xmath649 and @xmath650 such that @xmath651 are in clockwise order , and furthermore @xmath652 and @xmath653 . note that , if @xmath654 , then @xmath651 would also show that @xmath29 is not weakly separated from @xmath590 . assume that @xmath655 ; the other cases are similar . first , suppose that @xmath656 . then , we must have @xmath657 , and we obtain a similar contradiction to before . on the other hand , if @xmath658 , then @xmath659 and @xmath660 . but , since @xmath648 , @xmath62 and @xmath63 are in distinct non - intersecting edges in @xmath661 . because these edges are also in @xmath662 , we have a contradiction of the construction of @xmath76 . the lemma follows . [ atmost1j ] there is at most one @xmath663 with @xmath664 and @xmath665 . assume for sake of contradiction that we have distinct circular pairs @xmath641 , with @xmath666 , and @xmath667 . without loss of generality , assume that @xmath668 and @xmath669 . by the fact that @xmath667 and the construction of @xmath76 , there exist two points @xmath670 on the clockwise arc from @xmath63 to @xmath62 not containing its endpoints , which are both in exactly one of @xmath671 . first , if @xmath672 and @xmath673 , then the points @xmath674 force @xmath642 and @xmath590 not to be weakly separated . however , @xmath675 , so we have a contradiction . if @xmath672 and @xmath676 , then we get a similar contradiction if @xmath677 , so we have that @xmath674 are in clockwise order . because @xmath678 and @xmath679 , there must be an @xmath506 such that @xmath680 but @xmath681 . we have four cases for the position of @xmath506 , relative to @xmath674 . if @xmath682 are in clockwise order , then we get a contradiction of our construction of @xmath76 . if @xmath683 are in clockwise order , then @xmath684 either contradicts that @xmath29 is a circular pair or that @xmath29 is weakly separated from @xmath643 . the case in which @xmath685 are in clockwise order is similar . finally , if @xmath686 are in clockwise order , then either @xmath687 or @xmath688 contradicts that either @xmath29 is weakly separated from @xmath643 or that @xmath642 is weakly separated from @xmath590 . the other cases follow similarly . we can now finish the proof of proposition [ strongbound ] . by lemma [ wsinj ] , the elements of @xmath602 are pairwise weakly separated , and we also have @xmath689 . thus , by the inductive hypothesis , @xmath690 . however , it is easy to see from lemma [ atmost1j ] that @xmath691 , so the induction is complete . now , proposition [ bound ] follows easily by taking @xmath692 in proposition [ strongbound ] . proposition [ strongbound ] also has another natural corollary : for any set @xmath6 of pairwise weakly separated circular pairs , there is an injective map @xmath693 such that @xmath694 for each @xmath695 . proposition [ strongbound ] gives exactly the condition required to apply hall s marriage theorem . this work was undertaken at the reu ( research experiences for undergraduates ) program at the university of minnesota - twin cities , supported by nsf grants dms-1067183 and dms-1148634 . the authors thank joel lewis , gregg musiker , pavlo pylyavskyy , and dennis stanton for their leadership of the program , and are especially grateful to joel lewis and pavlo pylyavskyy for introducing them to this problem and for their invaluable insight and encouragement . the authors also thank to thomas mcconville for many helpful discussions . finally , the authors thank vic reiner , jonathan schneider , and dennis stanton for suggesting references , and damien jiang and ben zinberg for formatting suggestions . we now finish the proof of theorem [ circposthm ] . recall that @xmath107 is an electrical positroid for which no circular planar graph @xmath51 has @xmath696 , and that @xmath107 is chosen to be maximal among electrical positroids with this property . by assumption , @xmath107 has the @xmath115-bep for each @xmath116 , and does not have the @xmath139-bsp . furthermore , recall the construction of @xmath110 from the end of [ posproof ] . then , we have : @xmath110 is an electrical positroid , and has the @xmath139-bsp . straightforward . by assumption , @xmath111 , for some circular planar graph @xmath112 , which has a boundary spike at 1 . let @xmath113 be graph obtained after contracting the boundary spike in @xmath112 . we will prove that @xmath114 , which will yield the desired contradiction . the proof is similar to the case handled in [ posproof ] . recall the notation from definition [ betternotation ] , where we let @xmath251 denote the sequence @xmath250 . a circular pair @xmath697 is said to be * incomplete * if @xmath698 but @xmath699 and @xmath700 . if , on the other hand , @xmath54 in addition to @xmath701 and @xmath702 , @xmath29 is said to be * complete*. we also define the set @xmath252 of * primary * circular pairs as in [ posproof ] , where we take circular pairs of the form @xmath697 with the property that @xmath703 . it is easy to see that the analogue of lemma [ crossing ] holds when @xmath29 is incomplete . then , because @xmath107 has all beps , the primary circular pairs @xmath704 will all have @xmath705 . we now prove a series of lemmas , mirroring those in [ posproof ] . for an incomplete circular pair @xmath706 , any electrical positroid @xmath230 satisfying @xmath707 con @xmath228 with @xmath708 when @xmath228 is incomplete . first by axiom [ axside ] in @xmath230 , @xmath709 and @xmath710 implies that we either have our claim , or we have @xmath711 and @xmath712 . we first apply axiom [ axmid ] to @xmath713 on the circular pair @xmath714 . we have @xmath715 by definition , and we also have @xmath712 , so this implies that we either have our claim , or we have @xmath716 . the latter then implies that @xmath717 . a similar argument for axiom [ axmid ] on @xmath718 gives either our claim or that @xmath719 . then , axiom [ axmid ] implies that @xmath54 , a contradiction . thus , we have our claim . [ pcpsp ] for any incomplete circular pair @xmath29 , there exists a circular pair @xmath274 such that any electrical positroid @xmath230 satisfying @xmath707 contains @xmath29 . proceed by induction on @xmath116 , where @xmath116 is such that the first @xmath116 connections ( see definition [ connset ] ) of @xmath29 are the same as those of a primary circular pair . the base case may be handled similarly as in the proof of lemma [ pcpcont ] . now , let @xmath720 be the primary circular pair such that if @xmath721 , then @xmath722 for all @xmath116 ; @xmath723 exists by an identical argument as in the proof of lemma [ pcpcont ] . call @xmath723 the primary circular pair associated to @xmath29 . recalling that @xmath705 , we first need to show that @xmath724 . the same result replacing @xmath725 with @xmath325 would follow from an identical argument . by the definition of the @xmath130-bep , we need to show that @xmath726 . in the case that @xmath727 , we do so by applying lemmas [ crossing ] and lemma [ rtol ] . in the case that @xmath728 , we may apply the subset axiom . finally , in the case that @xmath129 , we may apply lemma [ walk ] . we now claim that , if @xmath729 , then @xmath730 . the lemma will then follow , because @xmath287 and @xmath29 are easily seen to have the same primary associated circular pair . if @xmath731 , the proof of the claim is identical to that of [ pcpcont ] , so assume that @xmath728 . in this case , we have @xmath732 . if @xmath733 , axiom [ axcross ] implies that we have @xmath734 . then , as @xmath735 , we are done by axiom [ axcross ] . [ primarybsp ] there is exactly one circular pair in @xmath252 that does not lie in @xmath107 , which we call the * @xmath107-primary circular pair . * the argument is the same to that of lemma [ unpcp ] . [ sppcp ] for any incomplete circular pair @xmath29 , any electrical positroid @xmath230 satisfying @xmath707 contains the @xmath107-primary circular pair . proceed by retrograde induction on @xmath116 , where @xmath116 is such that the first @xmath116 connections of @xmath29 are the same as those of the @xmath107-primary circular pair . if @xmath731 , we can argue exactly as in the proof of lemma [ contpcp ] . thus , assume that @xmath728 . let @xmath736 be the @xmath107-primary circular pair . by lemma [ primarybsp ] , @xmath737 has the property that @xmath722 for all @xmath116 . also , by how the construction of @xmath110 , for any circular pair @xmath738 , @xmath739 and @xmath740 . therefore , it follows that @xmath741 . then , by axiom [ axside ] , @xmath742 , and another application of axiom [ axside ] yields @xmath743 , completing the proof . [ uniclossp ] for any two incomplete circular pairs @xmath29 and @xmath322 any electrical positroid @xmath230 containing @xmath6 and contained in @xmath163 with @xmath29 must contain @xmath322 . by lemma [ pcpsp ] , @xmath230 contains the primary circular pair . the claim then follows by lemma [ sppcp ] . let @xmath744 . then , @xmath330 is an electrical positroid . the proof follows the same outline as that of lemma [ findclos ] ; here , we verify that @xmath330 satisfies each electrical positroid axiom . by construction , @xmath107 and @xmath380 only differ in the circular pairs @xmath29 for which @xmath745 . in particular , @xmath746 . @xmath330 is easily seen to satisfy the electrical positroid axioms other than [ axside ] and [ axmid ] . we first consider axiom [ axside ] : suppose that @xmath351 ; we show that either @xmath747 or @xmath748 . we have the following cases : * @xmath749 . suppose that @xmath750 . then , either @xmath751 or @xmath752 . axiom [ axcross ] applied to @xmath753 with @xmath754 gives @xmath751 . similarly , the roles of @xmath380 and @xmath107 may be swapped , and we may apply the same argument with @xmath369 . thus , either @xmath107 and @xmath380 both contain @xmath353 and @xmath369 or both do not , in which case they both contain @xmath29 and @xmath356 . * @xmath755 . suppose that @xmath750 and @xmath756 . then , @xmath757 . if @xmath751 , we are done . otherwise , if @xmath758 , as @xmath107 is an electrical positroid , we find @xmath158 and @xmath759 . then , we must have @xmath760 , in which case we are done , or either @xmath761 or @xmath762 . in the latter case , axiom [ axleft ] applied to @xmath763 with @xmath764 gives that @xmath761 . thus , @xmath107 and @xmath380 contain @xmath356 and @xmath29 , so we are done in this case as well . * the cases @xmath765 and @xmath766 are handled by similar logic ; the details are omitted . the case @xmath767 is symmetric with @xmath768 . we now consider axiom [ axmid ] : suppose that @xmath372 ; we show that either @xmath97 or @xmath769 . we have the following cases : * @xmath770 . if @xmath771 , then either @xmath772 or @xmath773 then , an application of axiom [ axcross ] to @xmath774 with @xmath754 yields that @xmath772 . the same argument holds if @xmath772 to show that @xmath771 does as well . now , suppose @xmath775 . then , either @xmath776 , which case we are done , or @xmath777 . then , the subset axiom , @xmath778 . applying axiom [ axsep ] to @xmath362 with @xmath779 then yields that @xmath776 , as desired . the same argument holds if we swap the roles of @xmath107 and @xmath380 . * @xmath780 . if @xmath771 , by the same argument as with the case @xmath781 , we have @xmath772 . if @xmath782 as well , then either @xmath776 , in which case we are done , or @xmath783 . in the latter case , because @xmath107 is an electrical positroid , @xmath784 . then , applying axiom [ axleft ] to @xmath785 with @xmath786 gives that @xmath776 , so we are done . * the cases @xmath768 and @xmath787 are handled with similar logic ; the details are omitted . we have exhausted all cases , so the proof of the lemma is complete . now , by lemma [ uniclossp ] , we have @xmath788 , so the proof of theorem [ circposthm ] is complete . recall that , to complete the proof of theorem [ lpalg ] , we need an additional technical result , which we state and prove here . [ mutlimit ] from the initial cluster of @xmath522 , mutating at a non - frozen limiting solid circular pair @xmath29 gives a new cluster variable which is a polynomial in the entries of @xmath9 , and relatively prime to @xmath29 . consider the limiting solid circular pair @xmath29 of size @xmath31 . fix the ground set @xmath789 so that @xmath20 denotes the determinant of the submatrix of @xmath9 with rows indexed by @xmath790 and columns indexed by @xmath638 . furthermore , let @xmath791 then , the cluster variable associated to the vertex @xmath29 is @xmath792 , and its corresponding exchange polynomial in the initial seed of @xmath11 is @xmath793 the new cluster variable from mutating at @xmath29 is @xmath794 where the last equality may be checked directly . we wish to show that @xmath795 is relatively prime to @xmath792 , which is irreducible , so it is enough to check that @xmath792 does not divide @xmath796 . let us outline the argument . if it is the case that @xmath792 divides @xmath797 , then each term in the expansion of @xmath797 must be divisible by a monomial in the expansion of @xmath792 . however , we claim that this can not be true . indeed , any monomial in the expansion of @xmath792 contains exactly one factor of a variable @xmath798 in the row of @xmath9 corresponding to @xmath63 , and the column of @xmath9 corresponding to @xmath799 . however , it is easily checked that there are terms of @xmath797 , after expanding and collecting like terms , containing variables @xmath800 with @xmath801 , so the claim is established .

the relative proximity of the galactic center ( henceforth gc ) provides a unique opportunity for `` close - up '' study of processes that are expected to be crucial in the formation of galaxies and black holes , yet remain essentially unobservable in more distant galaxies . recent advances in telescope technology have enabled us to resolve individual stars in the crowded gc environment , and the development of adaptive optics has allowed determination of both velocities and positions of such stars to be made with ever - increasing accuracy ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? these new data have presented several puzzles , notably the presence of a number of b - type stars very close ( @xmath4pc ) to the gc , and also the detection of one , and possibly two , coherent rings of massive o- & b - type stars at somewhat larger radii of @xmath5pc @xcite . these stars are known to be young , and are therefore presumed to be the result of recent star formation at or close to the gc . the environment at the gc is vastly different from the typical environment of ongoing star formation in the solar neighborhood , with much larger pressures , densities and temperatures , as well as strong tidal forces , so study of the formation of these stars raises a number of interesting issues ( see also the recent review by * ? ? ? a popular theory for the origin of these rings of stars is that they formed via fragmentation of accretion disks around the central black hole ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the stars are assumed to form on nearly circular orbits , as a result of gravitational instabilities in the disk . @xcite suggests that this will lead to a stellar mass function that is significantly more top - heavy ( i.e. with significantly more massive stars ) than that seen elsewhere in the galaxy , and recent observational studies suggest that this is indeed the case @xcite . other suggestions for the formation of these stellar rings exist , however , most notably the infalling star cluster scenario ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? this scenario can also result in a significantly top - heavy mass function , due to mass segregation and stellar collisions within the cluster ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in this paper we consider the dynamical evolution of a ring of stars such as those observed around the gc . previous models of the stellar dynamics of such a ring have considered a single stellar mass population @xcite , or the effect of processes such as resonant relaxation @xcite . here we investigate the effects of the stellar mass function on the evolution of the system , and use observational determinations of the stellar orbital parameters to constrain the mass function of the stellar rings . we find that the distribution of eccentricities can tell us about both the upper and lower ends of the stellar mass function , and discuss the consequences of this result for theories of star formation in the gc environment . the structure of the paper is as follows . in section [ sec : analytic ] we present a simple analytic model for the dynamical evolution of a mass - segregated ring of stars orbiting a massive central black hole . in section [ sec : numerical ] we describe numerical simulations of such a system , and compare the results to the predictions of our analytic model . we then apply our model to the gc system ( section [ sec : gc_app ] ) , and derive constraints on the initial conditions by comparing our model to recent observations . we discuss the consequences of our results , and the limitations of our analysis , in section [ sec : dis ] , and summarize our conclusions in section [ sec : conc ] . let us first consider the relaxation of a system of @xmath6 stars of mass @xmath7 , orbiting in a disk around a black hole of mass @xmath8 ( where @xmath9 ) . the disk ( or ring ) is centered at radius @xmath10 , with a radial width @xmath11 and a one - dimensional stellar velocity dispersion @xmath12 . we make the simplifying assumption that the velocity dispersion is isotropic . this is not strictly valid , but it has been shown that the ratio of the radial and vertical velocity dispersions can not become larger than 3 without the system becoming unstable @xcite . the relaxation time for such a system is given by @xmath13 ( e.g. * ? ? ? * ; * ? ? ? * ) where @xmath14 is the stellar density , @xmath15 is the coulomb logarithm and @xmath16 is an order - of - unity constant that depends on the geometry of the system . ( for a spherical system @xmath17 , @xcite . ) the stellar density is given by @xmath18 where @xmath19 is the scale - height of the disk . we can express @xmath20 in terms of the stellar velocity dispersion and the orbital angular velocity of the disk , @xmath21 , and so for a keplerian disk we have @xmath22 where @xmath23 is the keplerian orbital period at @xmath10 , and we have absorbed a factor of 2 into the constant such that @xmath24 . consequently , the relaxation of the system is governed by @xmath25 this form for @xmath26 allows the velocity dispersion to increase indefinitely , and in reality a cooling term should be included . however , this cooling arises when stars at the high - velocity tail of the distribution begin to escape from the system , and is only significant once the velocity dispersion becomes comparable to the orbital velocity . in our application of this model to the gc we are only interested in the behavior of the system at relatively early times , and in this case this simpler treatment remains accurate . the effects of this simplification are discussed in section [ sec : dis ] . we now extend this analysis to consider two different mass classes of stars , @xmath27 and @xmath28 , with @xmath29 and velocity dispersions @xmath30 and @xmath31 , stellar number @xmath32 and @xmath33 , and mean energies per particle @xmath34 and @xmath35 , respectively . each mass class is subject to the `` self - relaxation '' described above , but in addition energy can be exchanged between the classes . this type of analysis has previously been used to study the evolution of a distribution of planetesimals around the sun , and detailed calculations have been made ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite point out that the treatment of this problem depends on whether the velocity dispersion of the light stars , @xmath31 , is greater than or less than the hill velocity of the heavy stars , @xmath36 . the hill velocity is defined as @xmath37 where the hill radius @xmath38 . in the former case , referred to as the `` dispersion - dominated '' regime , the velocity dispersion is a good approximation for the speed of a single star and scattering encounters are well approximated by two - body dynamics . however , in the latter case , known as the `` shear dominated '' regime , the tidal gravity of the central black hole is important : in this regime the interactions are rather more subtle . in applying our model to the gc system ( see sections [ sec : numerical ] & [ sec : gc_app ] ) , we adopt @xmath39m@xmath2 and @xmath40pc . consequently , the keplerian orbital speed is @xmath41 km s@xmath42 , and the hill velocity scales as @xmath43 thus @xmath44 , so the hill velocity corresponds to orbits with very small eccentricities . in our models , @xmath45 at all but extremely early times , so we work in the `` dispersion - dominated '' regime throughout . by considering energy conservation in a two - body interaction , we can write the form for the `` exchange '' term as @xmath46 here @xmath47 is the mean of the two velocity dispersions ( and therefore the mean collision speed ) , @xmath48 is the appropriate coulomb logarithm for such a collision , @xmath49 is another order - of - unity constant , and the factor of 6 in the numerator arises because we are considering the energy rather than the velocity dispersion . we can then use @xmath50 to find expressions for the relaxation of the two velocity dispersions : @xmath51 @xmath52 here we have rewritten the constant terms for clarity , expressing them as @xmath53 for @xmath54,2 . the form of the exchange term can be considered as the product of a relaxation time ( containing three powers of @xmath55 , as in equation [ eq : t_rel ] ) , a volume scaling term which accounts for the different thicknesses of the two disks ( @xmath56 ) , and a normalized energy difference . as mentioned above , similar analyses are common in the study of planet formation . in this context other factors , such as physical collisions , are also significant , but if we neglect these terms we can compare the results to our model . the form of the solution is the same , and by comparison to the three - dimensional analysis of @xcite we see that that the constants @xmath57 and @xmath49 are related by the ratio @xmath58 . we adopt this ratio throughout . from these equations we can make some qualitative inferences about the evolution of this system . we see from equation [ eq : t_relax ] that the `` self - relaxation '' of the heavy stars will be more rapid than that of the light stars . however , this holds for any mass function @xmath59 where @xmath60 ( the salpeter slope is @xmath61 ) , and is therefore true for all cases considered in this paper . ] , so we expect the transfer term to boost the velocity dispersion of the light stars , @xmath31 , while at the same time damping the velocity dispersion of the heavy stars @xmath30 . however , the rate at which this energy transfer occurs depends on both the velocity dispersions and the distribution of mass in the system . in equation [ eq : sigma1 ] the damping term opposes the self - relaxation term when @xmath62 , and we see that the evolution of the heavy stars can be dominated by the damping term only if @xmath63 . since @xmath64 is at least of order @xmath30 , this means that the damping term can dominate if the total mass in light stars is greater than the total mass in heavy stars . in the case of the light stars , however , the exchange term acts in the same sense as the self - relaxation term , so we always expect the light stars to relax more rapidly than they would in isolation . this simple two - component model shows that , in the absence of other factors such as tidal and resonant effects , the stellar mass function is the critical factor in determining the relaxation of a stellar disk . in order to make detailed studies of the effect of a real mass function , it is necessary to extend this analysis to at least three mass classes . we add a third class with @xmath65 , and consider three exchange terms of the form seen in equation [ eq : ex_term ] ( 12 , 13 and 23 ) . in this case we see that the heaviest stars are damped by both lighter mass classes , while the velocity dispersion of the intermediate stars is boosted by the heaviest stars and damped by the lightest stars , and the lightest stars are boosted by both heavier mass classes . the distribution of stellar masses is critical to the evolution , but by considering three classes we can now emulate the effect of massive stars , which may only live for some fraction of the lifetime of the system . we apply this `` three - class '' model to the galactic center system in section [ sec : gc_app ] . our analytic model is rather simple , so in order to assess the validity of this approach we have conducted a number of numerical simulations . we use the @xmath66-body stellar dynamics code developed by @xcite , treating both the black hole and the stars as point masses . this code uses a hermite integration scheme , which enables us to retain high numerical accuracy . great care must be taken when using an @xmath66-body code to analyze such a problem , as the energy of the system is overwhelmingly dominated by the central object . if we assume that the stars have typical separation @xmath19 , then the ratio of the typical energy of a stellar encounter to that of the star black - hole system is approximately @xmath67 . therefore , for 10m@xmath2 stars orbiting a @xmath68m@xmath2 black hole in a disk with aspect ratio @xmath69 , the energy binding the stars to the black hole is some @xmath70 times larger than the typical energy of stellar encounters . consequently , we require very strict limits on the energy errors resulting from the numerical integration if we are to maintain accuracy in our simulations . this requirement , combined with restrictions on computational time , limit us to modeling systems with relatively few stars , typically 150 or fewer . m@xmath2 , @xmath71m@xmath2 and @xmath72 . the three grey curves show the results of the @xmath66-body simulations for three different random realizations of the initial conditions , and the black curve is the mean of these three realizations . the heavy curve shows the mean calculated in the same manner , but with the simulations run with more strict error tolerances ( see text ) . the dashed curve shows the best - fitting analytic model , with @xmath73 . ] we first consider a single mass of stars , in order to determine the value of the numerical constant @xmath57 in equation [ eq : t_relax ] ( and indeed to test whether a single constant is appropriate ) . we consider systems of 50 stars , orbiting around a black hole of mass @xmath74m@xmath2 . we set up the initial conditions as follows . the stars are distributed in a radial region between @xmath750.15pc ( i.e. a ring with radius @xmath76pc and width @xmath77pc ) , with uniform distributions in both radius and azimuth . the stars are given a gaussian distribution in @xmath78 , with scale - height @xmath79 . all stars are given zero - eccentricity keplerian orbital velocities in the @xmath80@xmath81 plane , with zero velocity in the @xmath78-direction , and this system is then integrated . we first consider a stellar mass of @xmath82m@xmath2 . we generated three sets of random initial conditions , and integrated each for 6000 orbital periods at @xmath10 ( @xmath83yr for the parameters specified ) . each model was computed using two different energy error tolerances , in order to check the numerical convergence . the cumulative fractional energy errors were typically @xmath84 in one case , and @xmath85 in the second case . translating the cumulative energy error into a measure of the reliability of the results is not straightforward , but in both cases we consider the estimated energy errors to be sufficiently small . we computed the orbital elements of each particle twice every orbital period in order to plot the results . figure [ fig : ecc_m25_n50 ] shows the evolution of the root - mean - square ( henceforth rms ) eccentricity in the simulations . the rms eccentricity is evaluated from the simulations by computing the instantaneous eccentricity of each star directly from the orbital elements . the discrepancy between the mean values obtained from two sets of simulations ( different error tolerances ) is around 15% , and is comparable to the differences between different random realizations of the same simulation . furthermore , this discrepancy is also comparable to the typical random fluctuations expected ( @xmath86% for @xmath87 ) , so we consider the simulations to be numerically converged . we see that the rms eccentricity of the stars rises from zero to around 0.15 after 10myr , and we are able to compare the results of the simulations to the analytic model presented in section [ sec : analytic ] . in a disk of small objects orbiting a massive central body , the rms eccentricity of the small bodies ( stars ) can be related to their velocity dispersion by @xmath88 ( e.g. * ? ? ? * ) where @xmath89 is the keplerian orbital speed . we performed a simple least - squares fit to determine the best - fitting value of the constant @xmath57 . we evaluate the model fit by integrating equation [ eq : sigma_star ] numerically . the coulomb logarithm is evaluated as the ratio of the maximum to minimum impact parameters . we assume that the maximum impact parameter is @xmath11 , and the minimum is @xmath90 ( e.g. * ? ? ? * ) , and therefore evaluate the coulomb logarithm as @xmath91 . however , we note that the solutions do not depend strongly on the form adopted for @xmath92 . for the `` low resolution '' simulations the best - fitting value is @xmath93 ; the `` high - resolution '' simulations give @xmath94 . however , we note that the velocity dispersion , and therefore the rms eccentricity , depends only on @xmath95 , so this apparently large discrepancy in the scaling constant corresponds only to around a 10% uncertainty in the velocity dispersion . by comparison , the more detailed analysis of @xcite suggests that @xmath57 lies in the range 1.92.8 , depending on the degree of anisotropy in the velocity dispersion , so we are satisfied that our simplified analytic form for the `` self - relaxation '' term is accurate to within 1015% . m@xmath2 , @xmath96m@xmath2 , @xmath97 , @xmath98m@xmath2 and @xmath99 . as in fig . [ fig : ecc_m25_n50 ] , the grey curves show the results of the @xmath66-body simulations for three different random realizations of the initial conditions , and the black curve is the mean of these three realizations . here the upper set of curves is for the 100 10m@xmath2 stars , and the lower curves are for the 10 50m@xmath2 stars . the dashed curves again show the best - fitting analytic model , with @xmath73 . ] we now consider a system with stars of two different stellar masses , in order to test the accuracy of the exchange term in our analytic model . we model the evolution of a system with @xmath100m@xmath101 stars and @xmath102m@xmath2 stars . thus the total mass in light , 10m@xmath2 stars is double that in heavy , 50m@xmath2 stars , so we expect the eccentricities of the heavy stars to be damped significantly . for comparison , we also consider a system with @xmath100m@xmath101 only . once again we evaluated three random realizations of the initial conditions for each ; the typical cumulative fractional energy errors were @xmath103 . however , with many more stars it was not practical to run a full convergence test in this case . runs with more stringent error constraints , which were run for much shorter times , suggest that these simulations have relaxed somewhat too rapidly , and that the accuracy of these simulations is around @xmath104% . however , we note that this comparison only considers the early evolution , where the velocity dispersion is rising very steeply . the results of the two - class simulations are shown in fig . [ fig : ecc_damp ] . the analytic model provides an excellent fit to the comparison ( @xmath100m@xmath101 only ) simulations and , as expected , the eccentricities of the heavy stars in the two - class model are significantly damped . the analytic curves are slightly below the simulated data , but we suggest that this is a result of slightly insufficient numerical accuracy in the simulations . even with this caveat , the agreement is good both in terms of the relative eccentricities of the two mass classes and the absolute numerical scaling . consequently , we are confident that our simple analytic model is accurate to within @xmath105% . we now apply our analysis to the gc system . one of the more surprising results of recent years was the detection of large numbers of young stars very close to the gc , and these stars have now been well - studied observationally . a population of massive o and b stars ( sometimes referred to in the literature as the gc `` hei stars '' ) is known to exist in one , and possibly two , coherent ring - like structures at a distance of around 0.1pc from the gc ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the young age ( @xmath106myr , * ? ? ? * ) of these stars suggests that they formed at or very close to their current location , but the environment at the gc poses a significant challenge to conventional theories of star formation . moreover , the detection of similar rings of stars in the center of m31 @xcite suggests that such systems may in fact be common , so their evolution warrants further study . one important goal of any theory of star formation is to predict the form of the initial stellar mass function [ henceforth ( i)mf ] . observational limitations ( primarily source confusion ) limit the study of the lower - end of the mass function at the gc through direct observations , but several indirect approaches have been taken to try to determine its form . @xcite infer the form of the mf from their observed @xmath107-band luminosity function , and conclude that the slope of the mf @xmath108 is in the range @xmath1091.35 ( i.e. @xmath59 ; in these units the salpeter slope is @xmath61 ) . @xcite argue that the integrated x - ray luminosity of the gc region observed by _ chandra _ sets a limit on the total mass in young , low - mass t tauri stars , of @xmath110m@xmath2 . we note , however , that young stars of earlier spectral type emit a much smaller fraction of their luminosity in x - rays than their t tauri counterparts ( e.g. * ? ? ? * ) , so formally this limit only applies to stars less massive than approximately 5m@xmath2 . @xcite consider the mutual interaction of the two stellar rings observed by @xcite and @xcite , and suggest that dynamical interactions between the two rings would destroy the observed structure unless the total mass of the rings was @xmath111m@xmath2 . finally , @xcite argue that @xmath66-body interactions within a stellar ring ( without considering mass segregation ) set a somewhat weaker upper limit on the total stellar mass , @xmath112m@xmath2 . these studies , combined with the knowledge that @xmath113m@xmath2 is present in more massive early - type stars @xcite , are strongly suggestive of a significantly top - heavy mf in the gc system , with many fewer low - mass ( @xmath114m@xmath2 ) stars than would be expected from a standard salpeter mf . @xcite suggested that such a top - heavy initial mf will arise naturally if stars are formed by the fragmentation of an accretion disk around the central bh , and models of the `` infalling cluster '' scenario predict a similarly top - heavy mf @xcite . here we investigate the formation of the gc system further by considering the effect of the mf on the dynamical evolution of a stellar ring . m@xmath2 and @xmath115 . the upper panel shows the evolution for a salpeter mass function with @xmath116m@xmath2 and @xmath117m@xmath2 : in this case we see that the light stars damp the eccentricities of the heavier stars significantly . the middle panel shows the same mass function slope @xmath61 , but this time with a low - mass cutoff at @xmath118m@xmath2 : we see that the damping effect is lessened in this case . the lower panel shows a top - heavy mass function , with @xmath119 : in this case the presence of many more massive stars results in much larger eccentricities . ] m@xmath2 and @xmath115 . the upper panel shows the evolution for a salpeter mass function with @xmath116m@xmath2 and @xmath117m@xmath2 : in this case we see that the light stars damp the eccentricities of the heavier stars significantly . the middle panel shows the same mass function slope @xmath61 , but this time with a low - mass cutoff at @xmath118m@xmath2 : we see that the damping effect is lessened in this case . the lower panel shows a top - heavy mass function , with @xmath119 : in this case the presence of many more massive stars results in much larger eccentricities . ] m@xmath2 and @xmath115 . the upper panel shows the evolution for a salpeter mass function with @xmath116m@xmath2 and @xmath117m@xmath2 : in this case we see that the light stars damp the eccentricities of the heavier stars significantly . the middle panel shows the same mass function slope @xmath61 , but this time with a low - mass cutoff at @xmath118m@xmath2 : we see that the damping effect is lessened in this case . the lower panel shows a top - heavy mass function , with @xmath119 : in this case the presence of many more massive stars results in much larger eccentricities . ] in order to study the effect of the stellar mf on the dynamical evolution of the gc system we use the `` three - class '' model presented in section [ sec : analytic ] . @xcite identify 53 ob and wolf - rayet ( wr ) stars in the clockwise stellar ring , and we use this observation to fix the central mass class in our model . the ob and wr stars have masses in the range @xmath12030m@xmath2 , so we adopt @xmath121m@xmath2 and @xmath122 . we model the effect of a varying mass function by allowing the masses @xmath27 and @xmath123 , and the slope of the mass function , @xmath108 , to be free parameters in our model . we choose @xmath32 and @xmath124 according to the number of stars prescribed by the mf at exactly @xmath27 and @xmath123 , and thus the choice of the mf slope fixes @xmath32 and @xmath124 . * see also @xcite ) find the rms eccentricity of the stars in the clockwise ring to be 0.20.3 . however , they find that the less well - defined counter - clockwise ring ( containing around 20 stars ) has significantly larger eccentricities , with an rms value of around 0.60.7 . however , we clearly see from fig . [ fig : ecc_m25_n50 ] that a ring of @xmath125m@xmath2 stars will reach a peak rms eccentricity of only 0.15 in 48myr . assuming that the stellar orbits were originally circular ( as expected from formation in an accretion disk : * ? ? ? * ) , we therefore require the presence of more massive stars in order to further excite the eccentricity of these stars to the level seen at the gc . we choose three initial parametrizations of our model . firstly , we choose a simple salpeter mf ( @xmath61 ) , with a maximum mass @xmath116m@xmath2 and a minimum mass @xmath117m@xmath2 . we then consider a salpeter slope with a `` low - mass cutoff '' , adopting @xmath118m@xmath2 . ( as the total mass of a salpeter mf diverges to low mass , this `` cutoff '' results in a lower total stellar mass . ) thirdly , we consider a significantly flatter mf , similar to that found by @xcite ( @xmath119 ) , while maintaining @xmath116m@xmath2 and @xmath118m@xmath2 m@xmath2 stars as a function of the mf index @xmath108 . @xmath116m@xmath2 and @xmath118m@xmath2 are fixed , and the three lines show the rms eccentricity of the 25m@xmath2 stars at 3.0 , 6.0 and 10.0myr ( bottom to top , respectively ) . we see that a significantly top - heavy mf is required in order to excite the eccentricity of the 25m@xmath2 stars to the level seen in the clockwise system @xcite , and also that the maximum rms eccentricity attainable is around 0.40.5 . ] the results of these models are shown in fig . [ fig : threeclass1 ] . we find that a standard salpeter mf results in significant damping of the eccentricities of the more massive stars . this is not altogether surprising , as the salpeter mass - function diverges to low mass , but we see clearly from fig . [ fig : threeclass1 ] that the rms eccentricity of the 25m@xmath2 stars is damped to @xmath126 at the age of the gc system . even allowing for the uncertainties both in our model and in the observed data , this is significantly lower than the observed eccentricities , and essentially rules out a standard imf for the gc rings if the stars were initially on circular orbits . when the low end of the salpeter mf is truncated at a higher mass the rms eccentricity of the 25m@xmath2 stars suffers noticeably less damping , but ( as noted above ) the rms eccentricity is still somewhat lower than observed , as there are not enough massive stars present to excite the eccentricities further . only when a significantly flatter mf is adopted , resulting in a much larger number of 125m@xmath2 stars , do the 25m@xmath2 stars reach eccentricities of @xmath127 in 5myr , as demanded by observations . we also note that these low to moderate eccentricities mean that few , if any , stars will be ejected from the ring over 10myr timescales , even at low stellar masses . [ fig : gamma_ecc ] shows the effect of varying the mass function slope on the rms eccentricity of the 25m@xmath2 stars , with @xmath116m@xmath2 and @xmath118m@xmath2 . we clearly see that a salpeter mf lacks sufficient massive stars to excite the rms eccentricity above 0.2 . however , even extremely top - heavy mass functions fail to excite the eccentricities to values much greater than @xmath128 at the age of the gc system , which suggests that the more eccentric counter - clockwise system @xcite may pose a problem for our model . we return to this issue in section [ sec : implications ] . ( `` class 2 '' ) stars . the upper panel shows the case where the heaviest ( 125m@xmath2 ) stars are simply removed at @xmath129myr . the solid line shows the case of a salpeter mf , truncated at @xmath118m@xmath2 ; the dashed line shows the case of a top - heavy mass function ( @xmath119 ) , also with the same low - mass cutoff . the lower panel is the same , but in this case the heaviest stars were subject to progressive mass - loss as specified in equation [ eq : massloss ] . in both cases we see that a top - heavy imf is required in order to produce rms eccentricities ( in the 25m@xmath2 stars ) larger than around 0.15 . ] ( `` class 2 '' ) stars . the upper panel shows the case where the heaviest ( 125m@xmath2 ) stars are simply removed at @xmath129myr . the solid line shows the case of a salpeter mf , truncated at @xmath118m@xmath2 ; the dashed line shows the case of a top - heavy mass function ( @xmath119 ) , also with the same low - mass cutoff . the lower panel is the same , but in this case the heaviest stars were subject to progressive mass - loss as specified in equation [ eq : massloss ] . in both cases we see that a top - heavy imf is required in order to produce rms eccentricities ( in the 25m@xmath2 stars ) larger than around 0.15 . ] this simple treatment of the gc system is unlikely to be valid , however , as stars born at the upper end of the imf are subject to significant mass - loss during their lifetimes . moreover , it is unlikely that stars of greater than @xmath130m@xmath2 will live for longer than 23myr ( e.g. * ? ? ? the effect of a supernova blast wave on the dynamics of the stellar rings is negligible ( e.g. * ? ? ? * ) , but the removal of the massive stars from the system can have a significant effect , as they are no longer able to excite the velocity dispersion of lower mass stars . we use two crude approximations to account for mass - loss and/or stellar death in our simple model . our first approach is simply to assume that the most massive stars are not subject to any mass - loss , but are removed from the system at the end of their lifetimes . we adopt a lifetime of 3myr for these stars , and model this by setting @xmath131 from @xmath129myr onwards . our second approach is to include mass - loss from the most massive stars . this is done simply by making the stellar mass @xmath27 a time - dependent function @xmath132 . the details of mass loss from very massive stars are not well understood , but the qualitative effect of mass - loss on the dynamical evolution of the system is not strongly dependent on the mass - loss rates adopted . for simplicity we adopt the mean mass - loss rate for wr stars derived by @xcite : @xmath133 we use this form for the mass - loss rate @xmath134 for @xmath135m@xmath2 , and assume that the stars mass remains constant once it drops to 25m@xmath2 . this results in a rather rapid decline in the mass of the heaviest stars , with the mass @xmath132 falling from @xmath136m@xmath2 to 25m@xmath2 in 1.25myr . we note , however , that this form was derived from observations of wr stars , the most massive of which were around 50m@xmath2 , and so is of uncertain validity at very large stellar mass . the results are shown in fig . [ fig : massloss ] . we see that the consequence of including mass - loss is , as expected , that the eccentricities of the observable stars are not excited as highly as in the case when no mass - loss was included . the exact eccentricities achieved are rather sensitive to some rather poorly - defined parameters in the model , namely the mass - loss rates and also the upper cutoff of the mf . however , in broad terms we see that the typical eccentricities of the observable stars are expected to be in the range @xmath00.3 , depending on the details of the upper end of the mf . taken together , these results have important consequences for the gc system . as noted above , @xcite find typical eccentricities in the clockwise ring of around 0.20.3 , and in the counter - clockwise ring of @xmath137 . our analysis has shown that , in order for the observed stars to reach even the moderate eccentricities seen in the clockwise ring , there must be a significant population of much more massive ( @xmath1m@xmath2 ) stars present in numbers greatly exceeding those expected from a salpeter mf . we also find that the total mass in low - mass ( @xmath114m@xmath2 ) stars must be less than the mass in ob stars in order to avoid significant damping of the eccentricities of the ob stars . thus we conclude that the clockwise system has a rather top - heavy mass function , in agreement with the previous observations of @xcite and @xcite , and find that the observed eccentricities are in agreement with those expected from star formation in a keplerian disk ( e.g. * ? ? ? * ; * ? ? ? as noted in section [ sec : basic ] , however , the much larger eccentricities in the counter - clockwise system @xcite pose a problem for our model . even in a `` best - case '' model , with no mass - loss from the heaviest stars , and a very top - heavy mf ( @xmath138 ) extending to very high stellar mass ( 175m@xmath2 ) , the observable 25m@xmath2 stars only reach an rms eccentricity of 0.5 in 10myr , and more realistic models suggest that rms eccentricities greater than 0.4 are unlikely ( see fig . [ fig : gamma_ecc ] ) . we therefore conclude that the counter - clockwise system was probably not formed from an initially circular disk , and suggest that some other mechanism must be responsible . as noted above , our model assumes initially circular stellar orbits , but it may be possible to produce larger eccentricities if the initial configuration of the system has significant eccentricity . such a configuration is also possible in the infalling cluster scenario @xcite . within the disk fragmentation scenario it may be possible to generate eccentric initial conditions if the disk is itself eccentric when it fragments . if the disk is formed by some individual `` accretion event '' , such as the capture of a molecular cloud ( as suggested by * ? ? ? * ) , then an eccentric stellar disk can form if the fragmentation timescale is shorter than the circularization timescale of the disk . models of eccentric accretion disks have shown that a disk can typically remain eccentric for many orbital times @xcite , so rapid fragmentation of such a disk may indeed provide a mechanism for generating the large observed stellar eccentricities . we note also that eccentric initial conditions may retain an observable signature , in the form of a significantly anisotropic velocity dispersion . observations of the stellar disk(s ) at the center of m31 suggest significant , coherent , eccentricities , and models of eccentric stellar disks have been shown to fit the observed data well ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? we note that the central black hole in m31 is approximately 100 times more massive than that in the galaxy , so an eccentric stellar disk will retain a preferred eccentricity vector for a much longer time in m31 than at the gc . however , detailed study of this problem , and associated issues to do with the evolution of initially coherent eccentric stellar disks , is beyond the scope of this investigation . our model predicts that few , if any , stars should be scattered inwards from the rings at 0.1pc , and we therefore make no attempt to explain the origin of the so - called `` s - stars '' ( e.g. * ? ? ? * ; * ? ? ? * ) , which orbit much closer to the gc ( at a radius of @xmath139pc ) . it has been suggested that these stars may have been scattered inwards from the rings by an intermediate - mass ( @xmath140@xmath70m@xmath2 ) black hole @xcite , but our model does not make any prediction in this regard . we note , however , that such an interaction would provide an additional means of exciting eccentricity in the stellar rings , and therefore may provide another means of generating the large eccentricities observed in the counter - clockwise ring . there are several obvious limitations to our analysis . our model assumes that the velocity dispersion is isotropic , and while this assumption is not unreasonable it is not strictly valid . similar analyses applied to planetesimals ( e.g. * ? ? ? * ) suggest that in equipartition the radial velocity dispersion is roughly double the vertical dispersion . our simulations are complicated by the fact that our initial conditions have zero velocity dispersion in the radial direction ( keplerian orbits ) , but non - zero dispersion in the vertical direction due to the finite thickness of the ring . the system does not reach equipartition over the timescales considered , but the general trend seems to be in broad agreement with previous analyses . this suggests that our relation between the velocity dispersion and the disk thickness is not exact , but we note that the manner in which we fit the scaling constant @xmath57 enables us to account for this . furthermore , we find good agreement between the model and the simulations , so we do not consider anisotropy in the system to be a significant problem . as mentioned in section [ sec : analytic ] , our analytic model does not allow for cooling , and consequently does not permit an equilibrium solution . as long as the velocity dispersion remains small relative to the orbital speed ( i.e. the rms eccentricity remains small ) this approximation is valid , and this is supported by the favorable comparison between the analytic model and the @xmath66-body simulations . this approach may result in an over - estimate of @xmath141 ( and therefore the eccentricity ) if @xmath141 becomes large . however , the largest eccentricities attained by any of our models are @xmath142 , so we do not consider this to be a significant problem . we also neglect cooling via physical stellar collisions , but note that at the stellar densities considered here this is unlikely to be significant . similarly , although we consider only the `` dispersion dominated '' regime for our two- and three - class models , we note that our initial conditions ( with circular orbits ) lie in the``shear - dominated '' regime . however , the relaxation of the system to the dispersion - dominated regime is very rapid , typically occurring within @xmath143 orbits , so we are satisfied that this simplification does not affect the results significantly . as noted in section [ sec : two_class ] , the convergence of our `` two - class '' numerical simulations is rather marginal , with tests indicating that the relaxation of the system is somewhat too fast . unfortunately it is not practical to run more stringent simulations with this type ( hermite ) of numerical algorithm , and is it not clear if more efficient algorithms ( such as tree - codes ) will provide sufficient accuracy to study this problem . our tests suggest that the accuracy of our simulations is likely no better than @xmath144% , but we are satisfied that this level of uncertainly does not affect the qualitative results of the model . we also note , as above , that our treatment of mass - loss from massive stars is somewhat arbitrary . however , the two models chosen span a significant fraction of the available parameter space , and more extreme parametrizations do not alter our conclusions significantly . as noted in section [ sec : analytic ] , our model neglects other dynamical effects such as tidal and resonant effects , and also treats the black hole as a point mass with a newtonian potential . at the large radius considered here ( 0.1pc ) the timescale for relativistic precession of orbits is much longer than the age of the system , so we can safely neglect relativistic effects . ( the expected precession rate is a few degrees in 10myr : @xcite . ) moreover , the non - keplerian component of the potential due to either remnant black holes @xcite or other stellar populations at the gc ( e.g. * ? ? ? * ; * ? ? ? * ) is not expected to be significant . the effect of resonances on the gc system has recently been studied by @xcite , who find that resonant relaxation can in fact dominate over the type of uncorrelated two - body interactions considered here . they note , however , that the stellar disks observed at the gc are sufficiently young as to be unaffected by resonant effects . the fact that our model , which neglects resonant effects , provides a good fit to our numerical simulations supports this conclusion . it may well be that resonant effects will dominate the future evolution of the stellar rings at the gc , but in the early stages of evolution considered here they do not . in this paper we have considered the dynamical evolution of rings of stars around a massive black hole . through analytic arguments and numerical simulations we have constructed a model for the evolution of a disk of stars of different masses , and shown that the stellar mass function is the dominant factor in determining the evolution of such a system . we have then applied our analysis to rings of stars observed to orbit the galactic center system . we find , in agreement with previous studies , that the total mass in low - mass ( @xmath114m@xmath2 ) stars must be significantly lower than expected from a salpeter mass function , and also find that a significant population of massive ( @xmath1m@xmath2 ) stars must have been present in order to produce eccentricities in the range 0.20.3 , as observed by @xcite . however , we find that dynamical relaxation alone is unlikely to produce rms eccentricities larger than @xmath145 in the gc system . consequently we conclude that rings with larger eccentricities , such as the counter - clockwise system observed by @xcite , are unlikely to have originated in a circular disk , and suggest that some other dynamical process must be responsible for such systems . alternatively , we suggest that star formation by fragmentation of an eccentric accretion disk could produce the observed eccentricities . we acknowledge useful discussions with elena rossi during the initial stages of this work , and we thank both the editor , frederic rasio , and an anonymous referee for useful comments . this work was supported by nasa under grants nag5 - 13207 , nng04gl01 g and nng05gi92 g from the origins of solar systems , astrophysics theory , and beyond einstein foundation science programs , and by the nsf under grants ast0307502 and ast0407040 . alexander , t. 2005 , phy . reports , 419 , 65 beloborodov , a.m. , levin , y. , eisenhauer , f. , genzel , r. , paumard , t. , gillessen , s. , & ott , t. 2006 , apj in press ( astro - ph/0601273 ) bender , r. , et al . 2005 , apj , 631 , 280 berukoff , s.j . , & hansen , b.m.s . 2006 , apj in press ( astro - ph/0607080 ) binney , j. , & tremaine , s. 1987 , _ galactic dynamics _ , princeton university press bonnell , i.a . , & davies , m.b . 1998 , mnras , 295 , 691 freitag , m. , grkan , m.a . , & rasio , f.a . 2006 , mnras , 368 , 141 genzel , r. , et al . 2003 , apj , 594 , 812 gerhard , o. 2001 , apj , 546 , l39 ghez , a.m. , klein , b.l . , morris , m. , & becklin , e.e . 1998 , apj , 509 , 678 ghez , a.m. , salim , s. , hornstein , d. , tanner , a. , lu , j.r . , morris , m. , becklin , e.e . , & duchne , g. 2005 , apj , 620 , 744 goldreich , p. , lithwick , y. , & sari , r. 2004 , ara&a , 42 , 549 goodman , j. 2003 , mnras , 339 , 937 goodman , j. , & tan , j.c . 2004 , apj , 608 , 108 grkan , m.a . , & rasio , f.a . 2005 , apj , 628 , 236 miralda - escud , j. , & gould , a. 2000 , apj , 545 , 847 hansen , b.m.s . , & milosavljevi , m. 2003 , apj , 593 , l77 hopman , c. , & alexander , t. 2006 , apj in press ( astro - ph/0601161 ) hut , p. , & makino , j. 2003 , in _ the art of computational science _ kulsrud , r.m . , mark , j.w.k . , & caruso , a. 1971 , ap . space sci . , 14 , 52 levin , y. , & beloborodov , a.m. 2003 , apj , 590 , l33 lissauer , j.j . 1993 , ara&a , 31 , 129 mcmillan , s.l.w . , & portegies zwart , s.f . 2003 , apj , 596 , 314 nayakshin , s. 2006 , mnras in press ( astro - ph/0512255 ) nayakshin , s. , & cuadra , j. 2005 , a&a , 437 , 437 nayakshin , s. , & sunyaev , r. 2005 , mnras , 364 , l23 nayakshin , s. , dehnen , w. , cuadra , j. , & genzel , r. 2006 , mnras , 366 , 1410 nugis , t. , & lamers , h.j.g.l.m . 2000 , a&a , 360 , 227 ogilvie , g.i . 2001 , mnras , 325 , 231 papaloizou , j.c.b . , & terquem , c. 2001 , mnras , 325 , 221 paumard , t. , et al . 2006 , apj , 643 , 1011 peiris , h.v . , & tremaine , s. 2003 , apj , 599 , 237 polyachenko , v.l . , & shukman , i.g . 1977 , soviet ast . letters , 3 , 134 schaller , g. , schaerer , d. , meynet , & g. , maeder , a. 1992 , a&as , 96 , 269 stewart , g.r . , & wetherill , g.w . 1988 , icarus , 74 , 542 stelzer , b. , micela , g. , hamaguchi , k. , & schmitt , j.h.h.m . 2006 , a&a in press ( astro - ph/0605590 ) syer , d. , & clarke , c.j . 1992 , mnras , 255 , 92 ( erratum : 1993 , mnras , 260 , 463 ) tremaine , s. 1995 , aj , 110 , 628 weinberg , n.n . , milosavljevi , m. , & ghez , a.m. 2005 , apj , 622 , 878 wheeler , j.c . , lecar , m. , & mckee , c.f . 1975 , apj , 200 , 145 ( erratum : 1976 , apj , 209 , 961 )

we consider the problem of reconstruction of refractive indices or dielectric constants of unknown targets placed in a homogeneous domain from blind backscattered experimental data . we work with time resolved backscattering experimental data of wave propagation for a 3-d hyperbolic coefficient inverse problem ( cip ) . our data are generated by a single location of the point source . the backscattering signal is measured on a part of a plane . we present a combination of the approximately globally convergent method of @xcite with a finite element method ( fem ) for the numerical solution of this cip . given a certain function computed by the technique of @xcite , the fem reconstructs the unknown coefficient in an explicit form . as a result , we can reconstruct refractive indices and locations of targets . in addition , we estimate their sizes . we believe that these results can be used as initial guesses for locally convergent methods in order to obtain better shapes , see , e.g. section 5.9 in @xcite , where the image obtained by the globally convergent method for transmitted experimental data was refined via a locally convergent adaptivity technique . experimental data were collected by the device which was recently built at university of north carolina at charlotte . in our experiments we image targets standing in the air . a potential application of our experiments is in imaging of explosives . note that explosives can be located in the air @xcite , e.g. improvised explosive devices ( ieds ) . the work on real data for the case when targets are hidden in a soil is ongoing . we have collected backscattering time resolved experimental data of electrical waves propagation in a non - attenuating medium . as it was pointed out in @xcite , the main difficulty of working with such data is caused by a huge mismatch between these data and ones produced by computational simulations . conventional data denoising techniques do not help in this case . therefore , it is unlikely that any numerical method would successfully invert the raw data . to get the data , which would look somewhat similar with ones obtained in computational simulations , a heuristic data pre - processing procedure should be applied . the pre - processed data are used as the input for the globally convergent method . previously our research group has applied the method of @xcite to the simpler case of transmitted experimental data which were produced by a similar device ( chapter 5 of @xcite ) . the backscattering real data are much harder to work with than transmitted ones since the backscattered signal is significantly weaker than the transmitted one , as well as because some unwanted signals are mixed up with the true one , see figure [ fig2]-a ) for the latter . we refer to our research in @xcite and section 6.9 of @xcite for the case of backscattering real data in 1-d . in the current paper we present results of reconstruction of the 3-d version of the method of @xcite . the approximately globally convergent method of @xcite relies on the structure of the underlying pde operator and does not use optimization techniques . each iterative step consists of solutions of two problems : the dirichlet boundary value problem for an elliptic pde and the cauchy problem for the underlying hyperbolic pde . approximate global convergence " ( global convergence in short ) means that we use a certain reasonable approximate mathematical model . approximation is used because of one inevitably faces with substantial challenges when trying to develop globally convergent numerical methods for multidimensional cips for hyperbolic pdes with single source . it is rigorously established in the framework of this model that the method of @xcite results in obtaining some points in a small neighborhood of the exact coefficient without a priori knowledge of any point in this neighborhood , see theorem 2.9.4 in @xcite and theorem 5.1 in @xcite . the distance between those points and the exact solution depends on the error in the data , the step size @xmath1 of a certain discretization of the pseudo - frequency interval and the computational domain @xmath2 where the inverse problem is solved ( see section [ sec:4.3 ] for definition of @xmath1 ) . a knowledge of the background medium in @xmath3 is also not required by this method . because of these theorems , convergence analysis is not presented here . a rigorous definition of the approximate global convergence property can be found in section 1.1.2 of @xcite and in @xcite . we use a mild approximation , since it amounts only to the truncation of a certain asymptotic series , and it is used only on the first iterative step ( section [ sec:4.2 ] ) . the validity of this approximate model was verified computationally on both synthetic and transmitted experimental data in @xcite as well as in the current work in the case of experimental backscattering data . different imaging methods are used to compute geometrical information of targets , such as their shapes , sizes and locations , see , e.g. @xcite . on the other hand , refractive indices , which is our main interest , characterize constituent materials of targets , and they are much more difficult to compute . as to the gradient - like methods , we refer to , e.g. @xcite and references therein . convergence of these methods is guaranteed only if the starting point of iterations is chosen to be sufficiently close to the correct solution . on the other hand , it was shown in section 5.8.4 of @xcite that the gradient method failed to work for transmitted experimental data of @xcite in the case when its starting point was the background medium . an outline of this paper is as follows . in section 2 we state forward and inverse problems . in section 3 we describe the experimental data and briefly outline the data pre - processing procedure . in section 4 we briefly outline the method of @xcite : for reader s convenience . in section 5 we describe a version of the fem which works for our case . in section 6 we describe our algorithm . in section 7 we outline some details of our numerical implementation . results are presented in section 8 and summary is in section 9 . let @xmath4 be a convex bounded domain with the boundary @xmath5 denote by @xmath6 we model the electromagnetic wave propagation in an isotropic and non - magnetic space @xmath7 with the dimensionless coefficient @xmath8 which describes the spatially distributed dielectric constant of the medium . we consider the following cauchy problem for the hyperbolic equation @xmath9@xmath10we assume that the coefficient @xmath11 of equation ( [ 2.1 ] ) is such that @xmath12,~~\varepsilon _ { r}(\mathbf{x})=1\text { for } \mathbf{x}\in \mathbb{r}^{3}\diagdown \omega , \label{2.20}\]]where @xmath13 we _ a priori _ assume knowledge of the constant @xmath14 which amounts to the knowledge of the set of admissible coefficients in ( 2.20 ) . however , we do not assume that the number @xmath15 is small , i.e. we do not impose smallness assumptions on the unknown coefficient @xmath16 . below @xmath17 are hlder spaces , where @xmath18 is an integer and @xmath19 let @xmath20 be a part of the boundary @xmath21 later we will designate @xmath22 as the backscattering side of @xmath2 and will explain how we deal with the absence of the data at @xmath23 * coefficient inverse problem ( cip ) . * _ suppose that the coefficient _ @xmath24 _ _ satisfies . determine the function _ _ @xmath24 _ _ for _ _ @xmath25 _ _ , assuming that the following function _ _ @xmath26 _ _ is known for a single source position _ _ @xmath27 @xmath28 the function @xmath26 in ( [ 2.5 ] ) models time dependent measurements of the wave field at the part @xmath22 of the boundary @xmath29 of the domain of interest @xmath2 . we assume below that the source position is fixed and @xmath30 . this assumption allows us to simplify the resulting integral - differential equation because @xmath31 in @xmath32 . the assumption @xmath33 for @xmath34 means that the coefficient @xmath11 has a known constant value outside of the domain of interest @xmath35 this is a cip with single measurement data . uniqueness theorem for such cips in the multidimensional case are currently known only if the function @xmath36 in ( [ 2.2 ] ) is replaced with a function @xmath37 such that @xmath38 @xmath39 a proper example of such function @xmath40 is a narrow gaussian centered around @xmath41 , which approximates the function @xmath36 in the distribution sense . from the physics standpoint this gaussian is equivalent to @xmath42 that uniqueness theorem can be proved by the method , which was originated in @xcite . this method is based on carleman estimates , also see , e.g. sections 1.10 , 1.11 of the book @xcite about this method . the authors believe that , because of applications , it still makes sense to develop numerical methods for this cip without completely addressing the uniqueness question . the function @xmath43 in ( [ 2.1 ] ) represents the voltage of one component of the electric field @xmath44 in our computer simulations the incident field has only one non - zero component @xmath45 . this component propagates along the @xmath46axis until it reaches the target , where it is scattered . so , we assume that in our experiment @xmath47 we now comment on five main discrepancies between our mathematical model ( [ 2.1])- ( [ 2.20 ] ) and the reality . the first discrepancy which causes the main difficulties , is the aforementioned huge mismatch between experimental data and computational simulations . the second one is that , although we realize that equation ( [ 2.1 ] ) can be derived from maxwell equations only in the 2-d case , we use it to model the full 3-d case . the reason is that our current receiver can measure only one of the polarization components of the scattered electric field @xmath48 . in addition , if using a more complicated mathematical model than the one of ( [ 2.1 ] ) , for example the one that includes vector scattering and thus depolarization effects on scattering , then one would need to develop a globally convergent inverse method for this case . the latter is a quite time consuming task with yet unknown outcome . * * * * equation ( [ 2.1 ] ) was used in chapter 5 of @xcite for the case of transmitted experimental data , and accurate solutions were obtained . a partial explanation of the latter can be found in @xcite , where the maxwell s system in a non - magnetic and non - conductive medium was solved numerically in time domain . it was shown numerically in section 7.2.2 of @xcite that the component of the vector @xmath49 which was initially incident upon the medium , dominates two other components . this is true for at least a rather simple medium such as ours . therefore , the function @xmath43 in ( [ 2.1 ] ) represents the voltage of the computed component @xmath50 of the electric field , which is emitted and measured by our antennas . the third discrepancy is that the condition @xmath51 is violated on the inclusion / background interface in our experiments . the fourth discrepancy is that formally equation ( [ 2.1 ] ) is invalid for the case when metallic targets are present . on the other hand , it was demonstrated computationally in kbksnf that one can treat metallic targets as dielectrics with large dielectric constants , which we call _ appearing dielectric constant _ , @xmath52modeling metallic targets as integral parts of the unknown coefficient @xmath24 is convenient for the above application to imaging of explosives . indeed , ieds usually consist of mixtures of some dielectrics with a number of metallic parts . such targets are heterogeneous ones , and we consider three heterogeneous cases in section [ sec:8.2 ] . on the other hand , modeling metallic parts of heterogeneous targets as a separate matter than the rest of an a priori unknown background medium would result in significant additional complications of the already difficult problem with yet unknown outcome . the fifth discrepancy is that we use the incident plane wave instead of the point source in our computations . we have discovered that the plane wave case works better in image reconstructions than the point source , while the point source case is more convenient for the convergence analysis in bk , bkjiip12 . in addition , since the distance between our measurement plane and targets is much larger than the wavelength of our signal , it is reasonable to approximate the incident wave as a plane wave . thus , our results of section [ sec:8.2 ] demonstrate the well known fact that computational results are often less pessimistic than the theory , since the theory can not grasp all nuances of the reality . in summary , we believe that accurate solutions of the above cip for experimental data justify our mathematical model . figure [ fig1]-a ) is a photograph explaining the data collection . the data collection is done in a regular room , which contains office furniture , computers , etc . keeping in mind our desired application ( see introduction ) , we intentionally did not arrange a special waveguide , which would protect our data from unwanted signals caused by reflections from various objects in the room . below @xmath53 and @xmath54 are horizontal and vertical axis respectively and the @xmath55 axis is perpendicular to the measurement plane , the positive direction of @xmath55 axis is in the direction from the target to the measurement plane . we dimensionalize our coordinates as @xmath56 where m " stands for meter . however , we do not change notations of coordinates for brevity . hence , below , e.g. 0.05 of length actually means 5 centimeters . `` the transmitter sends the pulse into the medium which contains targets of interest . the electric wave caused by the pulse is scattered by the targets , and the backscattered signal is detected by the detector . the detected signal is recorded by the real time oscilloscope . '' two main pieces of our device are picosecond pulse generator ( figure [ fig1]-b ) ) and textronix oscilloscope ( figure [ fig1]-c ) ) . the picosecond pulse generator generates electric pulses . the duration of each pulse is 300 picoseconds . this pulse goes to the transmitter , which is a horn antenna ( source ) . the transmitter sends the pulse into the medium which contains targets of interest . the electric wave caused by the pulse is scattered by the targets , and the backscattered signal is detected by the detector . the detected signal is recorded by the real time oscilloscope . the oscilloscope produces a digitized time resolved signal with the step size in time of 10 picoseconds . the total time of measurements for one pulse is 10 nanoseconds=10@xmath57 picoseconds=10@xmath58 second . to decrease the measurement noise , the pulse is generated 800 times for each position of the detector , the backscattering signal is also measured 800 times and resulting signals are averaged . the detector moves in both horizontal and vertical directions covering the square @xmath59 on the measurement plane . we have chosen the step size of this movement to be 0.02 . although we can choose any step size , we found that 0.02 provides a good compromise between the precision of measurements and the total time spent on data collection . the distance between our targets and the measurement plane is approximately 0.8 with about 0.05 deviations , and the wavelength of our signal is about 0.03 . therefore , the distance between the measurement plane and our targets is of about 26 wavelengths . this is in the far field zone . the main difficulty working with experimental data is that there is a huge mismatch between these data and computationally simulated ones . indeed , figure [ fig2]-a ) depicts a sample of experimentally measured data for a wooden block at one position of the detector , see figure [ fig1]-a ) for data collection scheme . on this figure , the direct signal is the signal going directly to the receiver . we use this direct signal as the time reference for data pre - processing . unwanted signals are due to reflections of the electric wave from several objects present in the room . figure [ fig2]-b ) presents the computationally simulated data for the same target , see section [ sec:7.1 ] for data simulations . these figures show a huge mismatch between real and computationally simulated data . therefore , data pre - processing is necessary . we refer to @xcite for details of our data pre - processing procedure . the main steps of this procedure include : 1 . _ time - zero correction_. the time - zero correction is to shift the measured data in time . so that its starting time is the same as when the incident pulse is emitted from the transmitter this is done using the direct signals from the transmitter to the detector as the time reference . 2 . _ extraction of scattered signals_. apart from the backscattered wave by the targets , our measured data also contain various types of signals , e.g. direct signals from the horn to the detector , scattered signals from structures inside the room , etc . what we need , however , is the scattered signals by the targets only . to obtain them , we single out the scattered signals caused by the targets only and remove all unwanted signals . data propagation_. after getting the scattered signals , the next step of data pre - processing is to propagate the data closer to the targets , i.e. to approximate the scattered wave on a plane which is much closer to the targets then the measurement plane . the distance between that propagated plane and the front surface of a target is usually between 0.02 and 0.06 ( compare with the 0.8 distance from the measurement plane ) . there are two reasons for doing this . the first one is that the method of @xcite works with the laplace transform of the function @xmath43 ( section [ sec:4 ] ) . that laplace transform decays exponentially in terms of the time delay , which is proportional to the distance from the target to the measurement plane . hence , the amplitude of the laplace transformed experimental data on the measurement plane is very small and can be dominated by computational round - off error . the second reason is that this propagation procedure helps to substantially reduce the computational cost since the computational domain for the inverse problem is reduced . 4 . _ data calibration . _ finally , since the amplitudes of the experimental incident and scattered waves are usually significantly different from simulations , we need to bring the former to the same level of the amplitude as the latter . this is done using a known target referred to as _ calibrating object_. in this paper , the result of data pre - processing is used as the measured data @xmath60 on the backscattering boundary @xmath22 of our computational domain @xmath2 for the inverse problem . in this section we briefly outline the globally convergent method for reader s convenience . we refer to sections 2.3 , 2.5 , 2.6.1 and 2.9.2 of bk as well as @xcite for details . the first step of our inverse algorithm is the laplace transform of the function @xmath61@xmath62where @xmath63 is a certain number . we assume that the number @xmath63 is sufficiently large , and we call the parameter @xmath64 _ pseudo frequency_. it follows from ( [ 2.1 ] ) , ( [ 2.2 ] ) and ( [ 2.6 ] ) that the function @xmath65 is the solution of the following problem @xmath66@xmath67the limit ( [ 2.8 ] ) is proved in theorem 2.7.1 of @xcite . in addition , it was proven in theorem 2.7.2 of @xcite that for the function @xmath24 satisfying ( [ 2.20 ] ) there exists unique solution @xmath68 of the problem ( 2.7 ) , ( [ 2.8 ] ) for every @xmath69 such that @xmath70where @xmath71 is the solution of the problem ( 2.7 ) , ( [ 2.8 ] ) for the case @xmath72@xmath73 it follows from theorem 2.7.2 of @xcite that @xmath74 hence , we can consider the functions @xmath75 , @xmath76 @xmath77substituting @xmath78 in ( [ 2.7 ] ) and keeping in mind that the source @xmath30 , we obtain @xmath79using ( [ transform ] ) we obtain @xmath80where the truncation pseudo frequency @xmath81 is a large number , which is chosen numerically , see section [ sec:8 ] for details . we call @xmath82 the _ tail function , _ and it is unknown . it follows from ( [ transform ] ) and ( 3.7 ) that @xmath83it follows from @xcite ( section 2.3 ) that , under some conditions , there exists a function @xmath84 such that the following asymptotic behavior with respect to @xmath85 holds for functions @xmath86 and @xmath87 @xmath88@xmath89 differentiating both sides of equation ( [ new3.2 ] ) with respect to @xmath64 then using ( [ transform ] ) and ( [ 3.7 ] ) , we obtain the following nonlinear integral differential equation @xmath90 . \end{split } \label{3.8}\]]in addition , ( [ 2.5 ] ) and ( [ transform ] ) lead to the following dirichlet boundary condition for the function @xmath87 @xmath91 , \label{3.9}\]]@xmath92here @xmath93 is the laplace transform ( 2.6 ) of the function @xmath60 in ( [ 2.5 ] ) . we now need to complement the boundary data ( [ 3.9 ] ) at the backscattering side @xmath22 with the boundary data at the rest of the boundary @xmath94 using computationally simulated data , it was shown numerically in section 6.8.5 of @xcite as well as in @xcite that it is reasonable to approximate the boundary data on @xmath95 by the solution of the forward problem for the homogeneous medium for the case @xmath96 : recall that this equality holds outside of the domain @xmath97 see ( [ 2.20 ] ) . thus , we use below the following dirichlet boundary condition for the function @xmath98@xmath99 , \label{3.91}\]]@xmath100 , \\ \psi ^{0}\left ( \mathbf{x},s\right ) , \forall \left ( \mathbf{x},s\right ) \in \left ( \partial \omega \setminus \gamma \right ) \times \left [ \underline{s},\overline{s}\right ] .\end{array}\right . \label{3.92}\]]where the function @xmath101 is the function @xmath102 in ( [ 3.90 ] ) computed for the case @xmath103 . even though equation ( [ 3.8 ] ) with the boundary condition ( [ 3.91 ] ) has two unknown functions @xmath87 and @xmath86 , we can approximate both of them because approximation procedures for them are different , see section 7.1 . suppose for a moment that functions @xmath87 and @xmath86 are approximated in @xmath2 together with their derivatives @xmath104 then the corresponding approximation for the coefficient @xmath11 can be found via backwards calculation using ( [ new3.2 ] ) . to start iterations , we need the first approximation @xmath105 for the tail function . in this section we show how to calculate @xmath106 this is the same choice as the one in section 2.9.2 of the book @xcite as well as in @xcite . let the function @xmath107 satisfying ( 2.20 ) be the exact solution of our cip for the exact data @xmath108 in ( [ 2.5 ] ) . let @xmath109 be the exact tail function defined as @xmath110let @xmath111 $ ] be the corresponding exact function @xmath98 satisfying equation ( [ 3.8 ] ) . let @xmath112 $ ] be the corresponding exact dirichlet boundary condition for @xmath113 as defined in ( 3.91 ) . following ( [ 3.92 ] ) , we assume that @xmath114 for @xmath115 .$]hence , ( [ 3.8 ] ) and ( [ 3.91 ] ) hold for functions @xmath116 setting in ( 3.8 ) @xmath117 we obtain @xmath118next , truncating the second term in each of the asymptotics ( [ 2.b ] ) and ( [ 2c ] ) , we obtain that there exists a function @xmath119 such that @xmath120substituting formulae ( [ asymptotic ] ) into ( [ 4.18 * * ] ) , we obtain the following approximate dirichlet boundary value problem for the function @xmath121 @xmath122@xmath123thus , using ( [ 4.17 * ] ) and ( [ asymptotic ] ) , we obtain the following approximate mathematical model . * approximate mathematical model . * _ we assume that there exists a function _ @xmath124 _ _ such that the exact tail function _ _ @xmath125 _ _ has the form _ _ @xmath126__and the function _ _ @xmath127 _ _ is__@xmath128 because of ( [ 2.917 ] ) , ( [ 2.918 ] ) and ( [ 2.914 ] ) , we set for the first tail @xmath129where the function @xmath130 is the solution of the following dirichlet boundary value problem @xmath131@xmath132we point out that we calculate @xmath105 without any advanced knowledge of a small neighborhood of the exact coefficient @xmath133 using ( [ asymptotic])-([2.925 ] ) and schauder theorem @xcite , we obtain @xmath134where the number @xmath135 depends only from the domain @xmath35 hence , the error in the calculation of @xmath105 depends only on the error in the boundary data @xmath136 on the other hand , since the boundary function @xmath137 is generated by the function @xmath26 in ( [ 2.5 ] ) , then the error in @xmath138 is generated by the error in measurements . the estimate ( [ 2.926 ] ) is one of elements of the proof of the approximate global convergence theorem for this numerical method , see theorem 2.9.4 in @xcite and theorem 5.1 in @xcite . although a good approximation for the exact solution @xmath107 can be derived from the function @xmath139 we have observed computationally that better approximations are delivered via iterations described below in sections [ sec:6.1 ] , [ sec:6.2 ] . to approximate both functions @xmath87 and @xmath86 using ( [ 3.8 ] ) and ( [ 3.91 ] ) , we consider the layer stripping procedure with respect to @xmath64 . we divide the interval @xmath140 $ ] into @xmath141 small subintervals with the uniform step size @xmath142 . here , @xmath143 we approximate the function @xmath144 as a piecewise constant function with respect to @xmath145 i.e. we assume that @xmath146 for @xmath147 hence , using ( [ 3.7 ] ) , we approximate the function @xmath148 as @xmath149to obtain a sequence of dirichlet boundary value problems for elliptic pdes for functions @xmath150 we introduce the @xmath151dependent carleman weight function ( cwf ) @xmath152 , $ ] where @xmath153 is a large parameter . in our numerical studies we take @xmath154 . this function mitigates the influence of the nonlinear term in the resulting integral - differential equations on every pseudo - frequency interval @xmath155 . multiply both sides of equation ( [ 3.8 ] ) by @xmath156 and integrate with respect to @xmath157 we obtain @xmath158here @xmath159 is such an approximation of the tail function @xmath160 which corresponds to the function @xmath161 ( section [ sec:6.1 ] ) . numbers @xmath162 are computed explicitly . furthermore , @xmath163 for this reason we ignore the nonlinear term in ( [ 4.21 ] ) , thus setting@xmath164note that ( [ 4.22 ] ) is not a linearization , since ( [ 4.21 ] ) contains products @xmath165 and also because the tail function @xmath166 depends nonlinearly on functions @xmath167 see ( [ 2.a ] ) and step 6 in section [ sec:6.1 ] . in this section we explain how we compute functions @xmath169 on every pseudo - frequency interval @xmath170 using the fem . once the functions @xmath171 along with the function @xmath172 in ( [ 4.21 ] ) are calculated , we compute the function @xmath173 using the direct analog of ( [ 4.20 ] ) , @xmath174using ( [ transform ] ) , we set @xmath175 . \label{3.104}\ ] ] to find the function @xmath176 we note that the function @xmath177 is the solution of the following analog of the problem ( [ 2.7 ] ) , ( [ 2.8])@xmath178@xmath179where @xmath180 \text { for } \mathbf{x}\in \partial \omega . \label{3.107}\]]to compute the function @xmath181 from ( [ 3.105 ] ) , ( [ 3.106 ] ) and ( [ 3.107 ] ) , we apply a version of the fem as described below in sections [ sec:5.1 ] , [ sec:5.2 ] . following @xcite we discretize in computations our bounded domain @xmath4 by an unstructured tetrahedral mesh @xmath182 using non - overlapping tetrahedral elements @xmath183 . the elements @xmath184 are such that @xmath185 , where @xmath186 is the total number of elements in @xmath2 , and @xmath187we associate with the mesh @xmath182 the mesh function @xmath188 as a piecewise - constant function such that @xmath189where @xmath190 is the diameter of @xmath184 which we define as the longest side of @xmath184 . we impose the following shape regularity assumption of the mesh @xmath182 for every element @xmath191 @xmath192where @xmath193 is the radius of the maximal sphere contained in the element @xmath184 . define the set of polynomials @xmath194 as @xmath195we introduce now the finite element space @xmath196 as @xmath197where @xmath198 denotes the set of linear functions on @xmath184 defined by ( polynom ) for @xmath199 . hence , the finite element space @xmath196 consists of continuous piecewise linear functions in @xmath2 . to approximate functions @xmath200 we introduce the space of piecewise constant functions @xmath201 , @xmath202where @xmath203 is the piecewise constant function on @xmath184 defined by ( polynom ) for @xmath204 . to compute the function @xmath205 from ( [ 3.105 ] ) , we formulate the finite element method for the problem ( [ 3.106])-([3.107 ] ) as : find the function @xmath206 for the known function @xmath207 such that @xmath208where @xmath209 is the scalar product in @xmath210 . we expand @xmath211 in terms of the standard continuous piecewise linear functions @xmath212 in the space @xmath196 as @xmath213where @xmath214 denote the nodal values of the function @xmath211 at the nodes @xmath215 of the elements @xmath184 in the mesh @xmath182 . we can determine @xmath214 by knowing already computed functions @xmath216 using the following analog of ( [ 3.104 ] ) @xmath217 , \forall \mathbf{x}\in \omega .\]]substitute ( [ 3.200 ] ) into ( [ 3.107_1 ] ) and choose @xmath218 then we obtain the following linear algebraic system of equations @xmath219 , \label{3.108}\]]where @xmath220 $ ] is the scalar product in @xmath221 the system ( [ 3.108 ] ) can be rewritten in the matrix form for the unknown vector @xmath222 and known vector @xmath223 as @xmath224here @xmath225 is the block mass matrix in space , @xmath226 is the stiffness matrix corresponding to the term containing @xmath227 in ( [ 3.108 ] ) and @xmath228 is the load vector . at the element @xmath184 the matrix entries in ( [ 3.108_1 ] ) are explicitly given by : @xmath229 to obtain an explicit scheme for the computation of coefficients @xmath205 , we approximate the matrix @xmath225 by the lumped mass matrix @xmath230 in space , i.e. , the diagonal approximation is obtained by taking the row sum of @xmath225 @xcite . we obtain @xmath231note that for the case of linear lagrange elements which are used in our computations in section [ sec:8 ] we have @xmath232 . thus , the lumping procedure does not include approximation errors in this case . we present now our algorithm for the numerical solution of equations ( 4.21 ) and computing the functions @xmath205 using the equation([3.109 ] ) . in this algorithm the index @xmath233 denotes the number of inner iterations inside every pseudo - frequency interval @xmath234 when we update tails . step 0 : : set @xmath235 . compute the initial tail function @xmath236 as in ( 2.923)-([2.925 ] ) . step 1 : : here we describe iterations which update tails inside every pseudo - frequency interval @xmath155 . let @xmath237 suppose that functions @xmath238 are computed . solve the dirichlet boundary value problem for the function @xmath239 @xmath240 step 2 : : compute functions @xmath241 and @xmath242 @xmath243@xmath244 .\ ] ] step 3 : : compute the function @xmath245 via backwards calculations , using the finite element formulation of equation ( [ 3.109 ] ) as @xmath246since by ( [ 2.20 ] ) we should have @xmath247 and also since we need to extend the function @xmath248 outside of the domain @xmath2 by unity , we set @xmath249 step 4 : : solve the forward problem ( [ 2.1])-([2.2 ] ) with @xmath250 and compute the laplace transform ( [ 2.6 ] ) for @xmath251 . we obtain the function @xmath252 . step 5 : : update the tail function as@xmath253continue inner iterations with respect to @xmath233 until the stopping criterion of step 1 of section [ sec:6.2 ] is met at @xmath254 . step 6 : : set for the pseudo - frequency interval @xmath255 @xmath256 step 7 : : if either the stopping criterion with respect to @xmath257 of step 4 of section [ sec:6.2 ] is met , or @xmath258 then set the resulting function @xmath259 as the solution of our cip . otherwise , set @xmath260 and go to step 1 . when testing the algorithm of section [ sec:6.1 ] on experimental data , we have developed a reliable stopping criterion for iterations @xmath261 in this algorithm . on every pseudo - frequency interval @xmath155 we define first norms @xmath262 as @xmath263 in ( [ first_norms ] ) the function @xmath264 is the computed tail functions at the inner iteration @xmath265 as in ( [ 7.104 ] ) . functions @xmath266 in ( [ first_norms ] ) are obtained from the known measured function @xmath26 in ( [ 2.5 ] ) as @xmath267where @xmath268 is the laplace transform of the function @xmath26 at @xmath269 . we have observed that computed first norms @xmath262 always achieve only one minimum at a certain @xmath270 , where the number @xmath271 depends on the specific set of experimental data . furthermore , in non - blind cases of non - metallic targets , the corresponding values of @xmath272 were in a good agreement with a priori known ones . however , in the cases of non - blind metallic targets we have observed that @xmath273 this contradicts with ( [ 2.51 ] ) . therefore , we have developed the following stopping criterion which consists of four steps . * the stopping criterion * the first step in our criterion is for stopping inner iterations with respect to @xmath233 in step 5 of section [ sec:6.1 ] . as to steps 2 - 4 , they are for stopping outer iterations with respect to @xmath257 ( step 7 in section [ sec:6.1 ] ) . first , we define numbers @xmath274 and @xmath275 as@xmath276@xmath277 in ( [ 7.6_1 ] ) functions @xmath278 are computed tail functions corresponding to @xmath279 ( step 6 in section [ sec:6.1 ] ) and functions @xmath280 are calculated using ( [ obstail ] ) . * * step 1 . * iterate with respect to @xmath233 and stop iterations at @xmath281 such that@xmath282or @xmath283 where @xmath284 is a chosen tolerance . * * for every @xmath257 * * * * compute final norms @xmath285 as @xmath286 in ( [ final_norms ] ) functions @xmath287 are computed as in ( [ 7.104 ] ) . * * step 3*. compute the number @xmath288 of the pseudo frequency interval such that the first norms @xmath262 in ( [ first_norms ] ) achieve its first minimum with respect to @xmath257 and get corresponding @xmath289 on this interval . compute the number @xmath290 of the pseudo frequency interval such that the final norms @xmath291 in ( [ final_norms ] ) achieve its first minimum or they are stabilized with respect to @xmath257 , and get corresponding @xmath292 on this interval . next , compute the number @xmath293 @xmath294 * * step 4*. if @xmath295 or @xmath296 then take the final reconstructed value of the refractive index @xmath297 as the computed function @xmath298 take @xmath299 and stop iterations . however , if @xmath300 then continue iterations and compute the number @xmath301 $ ] of the pseudo frequency interval such that the global minimum with respect to @xmath257 of final norms @xmath285 in ( [ final_norms ] ) is achieved . then , similarly with ( [ 7.7 ] ) , compute the number @xmath302 @xmath303and take @xmath304 as the final reconstructed value of the refractive index . also , take the function @xmath305 as the computed coefficient @xmath24 and stop iterations . we have observed in all our computations that conditions of our stopping criterion are always achieved . more precisely , one of conditions ( [ 7.5 ] ) , ( [ 7.5_1 ] ) is always achieved for iterations with respect to @xmath233 and the minimal values mentioned in steps 3 and 4 are always achieved . figure [ fig : finalnorms ] displays a typical @xmath306dependence of sequences @xmath307and @xmath308 in this section we present some additional details of our numerical implementation . because of ( [ 2.51 ] ) , we define in all our tests the upper value of the function @xmath309 as @xmath310 see ( [ 2.20 ] ) . thus , we set lower and upper bounds for the reconstructed function @xmath311 in @xmath3 as @xmath312 \}. \label{4.30}\]]as to the lower bound , we ensure it via ( [ 7.101 ] ) . we ensure the upper bound 15 similarly via truncating to 15 those values of @xmath313 which exceed this number . to solve dirichlet boundary value problems ( [ 7.102 ] ) , we use fem . we reconstruct refractive indices rather than dielectric constants of material since they can be directly measured . to compare our computational results with directly measured refractive indices @xmath314 of dielectric targets and with appearing dielectric constants of metallic targets ( see ( [ 2.51 ] ) ) , we consider maximal values of computed functions @xmath315 , @xmath316 see step 4 of section [ sec:6.2 ] for the definition of @xmath317 using experimental data for non - blind targets and comparing reconstruction results with cases of synthetic data , we have found that our algorithm provided accurate results with the following pseudo frequency interval , which we use in all our computations @xmath318,\underline{s}=8,\overline{s}=10,h=0.05.\ ] ] as it is clear from step 4 of section [ sec:6.1 ] , we need to solve the forward problem ( [ 2.1 ] ) , ( [ 2.2 ] ) on each iterative step of inner iterations to update the tail via ( [ 7.103 ] ) . since it is impossible to computationally solve equation ( [ 2.1 ] ) in the infinite space @xmath319 we work with a truncated domain . namely , we choose the domain @xmath226 as @xmath320 we use the hybrid fem / fdm method described in @xcite and the software package waves @xcite . we split @xmath226 into two subdomains @xmath321 and @xmath322 so that @xmath323 . we solve the forward problem in @xmath226 and the inverse problem via the algorithm of section [ sec:6.1 ] in @xmath324 the space mesh in @xmath325 and in @xmath322 consists of tetrahedral and cubes , respectively . below @xmath326 since by ( [ 2.20 ] ) @xmath33 in @xmath327 then it is computationally efficient to use fdm in @xmath322 and to use fem in @xmath328 as it is done in the hybrid method of @xcite . the front and back sides of the rectangular prism @xmath226 are @xmath329 and @xmath330 , respectively . the boundary of the domain @xmath226 is @xmath331 here , @xmath332 and @xmath333 are , respectively , front and back sides of the domain @xmath226 , and @xmath334 is the union of left , right , top and bottom sides of this domain . the front side @xmath22 of the rectangular prism @xmath2 where the propagated data @xmath60 in ( [ 2.5 ] ) are given , is @xmath335 now we describe the forward problem which is used in our computations . to compute tail functions @xmath336 via steps 4 , 5 of the algorithm of section [ sec:6.1 ] , we computationally solve the following forward problem in our tests : @xmath337 , \\ { n}u & = -\partial _ { t}u,~\mbox{on}~\partial _ { 1}g\times ( t_{1},t ) , \\ \partial _ { n}u & = -\partial _ { t}u,~\mbox{on}~\partial _ { 2}g\times ( 0,t ) , \\ \partial _ { n}u & = 0,~\mbox{on}~\partial _ { 3}g\times ( 0,t ) , \end{split } \label{8.2}\]]where @xmath338 is the amplitude of the initialized plane wave , @xmath339we use @xmath340 and @xmath341 we solve the problem ( [ 8.2 ] ) using the explicit scheme with the time step size @xmath342 which satisfies the cfl condition . our reconstruction procedure is done in two stages described in this section . in the first stage we follow the algorithm of section [ sec:6.1 ] . we have observed that this stage provides accurate locations of targets of interest . it also provides accurate values of refractive indices @xmath343 of dielectric targets and large values of appearing dielectric constants @xmath344 for metallic targets , see ( [ 7.7 ] ) and ( [ 7.8 ] ) . however , the algorithm of section [ sec:6.1 ] does not reconstruct well sizes / shapes of targets . thus , we need a postprocessing procedure , which is done in the second stage . let @xmath345 be the function in ( 7.101 ) . then we set @xmath346next , we determine minimal @xmath347 and maximal @xmath348 values in @xmath53 and @xmath54 directions , where the function @xmath349 next , we set@xmath350 , y\in \left [ y_{\min } , y_{\max } \right ] , \\ 1 & \text { otherwise}\end{cases}\]]and proceed with step 5 of the algorithm of section [ sec:6.1 ] . in this second stage we perform the same number of iterations with respect to both indices @xmath351 as ones of the first stage . we are concerned in the second stage only with sized and shapes of targets , and we are not concerned with values of @xmath352 rather , we take these values from the first stage . let @xmath353 be the function @xmath24 obtained at the last iteration of the second stage . then we form the image of the target based on the function @xmath354@xmath355 .__object names . _ _ [ cols="^,<,<",options="header " , ] we collected experimental backscattering time resolved data of electrical wave propagation and have applied the approximately globally convergent numerical method of @xcite to these data . results for four non - blind and seven blind cases show a good accuracy of reconstruction of refractive indices of dielectric targets and appearing dielectric constants of metallic targets . in the case of dielectrics , the average reconstruction error is at least three times less than the error of direct measurements . we confidently differentiate between metallic and dielectric targets . in particular , we have accurately computed maximal values of refractive indices / dielectric constants of three blind heterogeneous targets . these targets represent simplified models of improvised explosive devices ieds , which are heterogeneous ones . locations of targets and their sizes in @xmath356 directions are accurately reconstructed . the most difficult cases of sizes in the @xmath46direction ( depth ) are well reconstructed in some cases . in addition , shapes of some targets are well reconstructed in some cases . we believe that a follow up application of the locally convergent adaptivity technique might improve reconstructions of shapes of targets . the adaptivity takes the solution obtained by the approximately globally convergent method as the starting point for the minimization of the tikhonov functional on a sequence of adaptively refined meshes . a significant refinement via the adaptivity was demonstrated in section 5.9 of @xcite for the case of transmitted experimental data , see figures 5.13 and 5.16 in @xcite . this research was supported by us army research laboratory and us army research office grants w911nf-11 - 1 - 0325 and w911nf-11 - 1 - 0399 , the swedish research council , the swedish foundation for strategic research ( ssf ) through the gothenburg mathematical modelling centre ( gmmc ) and by the swedish institute , visby program . the authors are grateful to mr . steven kitchin for his excellent work on data collection . l. beilina , k. samuelsson and k. ahlander , efficiency of a hybrid method for the wave equation . in _ international conference on finite element methods _ , gakuto international series mathematical sciences and applications , gakkotosho co. , ltd , 2001 . l. beilina and m.v . klibanov , a new approximate mathematical model for global convergence for a coefficient inverse problem with backscattering data , _ j. inverse and ill - posed problems _ , 20 , 513 - 565 , 2012 . l. beilina , energy estimates and numerical verification of the stabilized domain decomposition finite element / finite difference approach for the maxwell s system in time domain , _ central european journal of mathematics _ , 11 , 702 - 733 , 2013 . kuzhuget , l. beilina , m.v . klibanov , a. sullivan , l. nguyen and m.a . fiddy , blind experimental data collected in the field and an approximately globally convergent inverse algorithm , _ inverse problems _ , 28 , 095007 , 2012 . n. t. thnh , l. beilina , m. v. klibanov , and m. a. fiddy . reconstruction of the refractive index from experimental backscattering data using a globally convergent inverse method , preprint , arxiv:1306.3150 [ math - ph ] , 2013 .

mass gap generation and ( chiral ) symmetry breaking in relativistic fermionic systems can arise from a variety of mechanisms which are often related to certain couplings or interaction channels becoming dominant . an apparent counter - example is catalyzed symmetry breaking , first studied in the context of magnetic catalysis @xcite , where mass gap generation is triggered by the presence of a magnetic field even for arbitrarily small values of the interaction strength . this phenomenon can be understood in various ways , the essence being that the long - range fluctuations driving the symmetry - breaking transitions are parametrically enhanced , see @xcite for a recent review . more concretely , a magnetic field induces a fermionic fluctuation spectrum with landau - levels containing a zero mode that leads to an enhancement of the density of states in the ir and to an effective dimensional reduction favoring symmetry breaking . magnetic catalysis has found a rich variety of applications both in particle physics ( chiral phases of qcd ) @xcite and condensed matter physics @xcite . another simple picture for magnetic catalysis has recently been developed within the framework of the functional renormalization group in the context of the @xmath0 gross - neveu model @xcite . in line with the fact that symmetry - breaking phase transitions are often related to fixed points of renormalization group transformations , also magnetic catalysis can be related to the behavior of rg fixed points as a function of the magnetic field . this rg picture has already successfully been applied in the context of qcd @xcite . in the present work , we verify the underlying rg picture of catalyzed symmetry breaking in the context of curved spacetimes . the fact that symmetry breaking and mass generation in fermionic systems can be influenced by negative curvature of the spacetime has been realized early @xcite , and is meanwhile reviewed in textbooks @xcite . the phenomenon is typically studied at mean - field level and occurs in many different fermionic models @xcite . in @xcite , the similarity to magnetic catalysis was realized in terms of an effective dimensional reduction mechanism of the spectral properties of the dirac operator . this justifies the use of the terminology `` gravitational catalysis '' @xcite . in fact , we find in the present work that the effective dimensional reduction and the corresponding enhancement of the density of states in the ir is directly related to the fixed point structure as identified below . from this rg viewpoint , symmetry breaking arises as a consequence of the fact that the coupling value required for criticality becomes arbitrarily small as a function of the curvature ( the catalyzer ) . hence , any finite value of the fermionic interactions ultimately becomes supercritical , typically driving the system towards the ordered phase . we investigate this rg mechanism within the simple @xmath0 gross - neveu model in the present work . we concentrate on two different curved backgrounds with constant negative curvature : a maximally symmetric spacetime ( anti de sitter ) and a purely spatially curved case ( lobachevsky plane ) . for both cases , mean - field studies are already available , see @xcite and @xcite , respectively . whereas the former allows for an analytic treatment in terms of simple functions , the latter is potentially relevant for curved layered condensed matter systems . for instance , the exitonic or anti - ferromagnetic instabilities in graphite and graphene have been associated with quantum phase transitions falling into the @xmath0 gross - neveu universality class @xcite . as catalyzed symmetry breaking is manifestly driven by the long - range modes , the rg analysis allows us to estimate the required curvature in relation to the length scale of the sample . this paper is organized as follows : in sect . [ sec : gnmodel ] , we briefly introduce the model and its formulation in curved space . section [ sec : frg ] is devoted to an evaluation of the rg flow in its simplest formulation . the manifestation of gravitational catalysis is discussed in sect . [ sec : gc ] . we estimate the influence of finite probe length on gravitational catalysis in sect . [ sec : pseudo ] by means of a pseudo - critical coupling . conclusions are drawn in sect . [ sec : conc ] . relevant technical details are deferred to the appendices . we aim at investigating the @xmath0 gross - neveu model @xcite in curved spacetime with metric @xmath1 ( greek indices running from @xmath2 to @xmath3 ) and signature @xmath4 using functional rg methods . the microscopic action functional @xmath5 at some ultraviolet ( uv ) scale @xmath6 depends on the bare coupling constant @xmath7 , the @xmath8 grassmann - valued fields @xmath9 and the @xmath8 conjugated fields @xmath10 , @xmath11 = { } & \!\ ! { \int \!\!\ ! { } _ { { } _ { { } _ { { } _ { { } _ \text{\small{\ensuremath{x } } } } } } } } \ ! \left [ \sum\limits_{i = 1}^{{n } } \bar{\psi}^{i } \slashed{\nabla } \psi^{i } + \frac{\bar{\lambda}_{\lambda}}{2 { n } } \left ( \sum\limits_{i = 1}^{{n } } \bar{\psi}^{i } \psi^{i } \right)^{\!\!\!2 } \right ] \notag \\ = { } & \!\ ! { \int \!\!\ ! { } _ { { } _ { { } _ { { } _ { { } _ \text{\small{\ensuremath{x } } } } } } } } \left [ \bar{\psi } \slashed{\nabla } \psi + \frac{\bar{\lambda}_{\lambda}}{2 { n } } ( \bar{\psi } \psi ) ^2 \right ] \ ! \text { , } \label{eq : gnaction}\end{aligned}\ ] ] where @xmath12 is a shorthand for the integral over the @xmath13-dimensional spacetime and @xmath14 is the determinant of the spacetime metric . the differential operator @xmath15 is composed from a set of @xmath16-dimensional gamma matrices satisfying the clifford algebra @xmath17 where @xmath18 is the identity in spinor space . in the present work , we use the irreducible representation , such that @xmath19 specifies the dimension of the @xmath20 matrices as well as the number of dirac components of the fermions . for our computations with fermions in curved space , we have actually used the weldon formalism @xcite which can be viewed as a generalization of the conventional vierbein formalism . however , a mapping to standard vierbein formulas is straightforward . the covariant derivative in the weldon formalism reads @xmath21 accounting for covariance with respect to the spinor and spacetime structure . here , @xmath22 is the affine spin connection , implicitly defined by @xmath23,\\ { \ensuremath{(\mathrm{ii})}}\quad { } & 0 = { \operatorname{tr}}\gamma_{\mu } \text { , } \end{aligned}\end{aligned}\ ] ] with @xmath24 the christoffel symbols . can additionally accommodate a @xmath25 gauge field @xmath26 @xcite . in this case , the right - hand side of @xmath27 in ( [ eq : defspincon ] ) would be given by @xmath28 , where @xmath29 is the charge under the @xmath25 gauge group . in the present work , we ignore such a @xmath25 gauge sector , setting @xmath30 . ] in the present work , we are interested in a discrete `` chiral '' @xmath31 symmetry , where the nontrivial transformation is defined by @xcite @xmath32 this symmetry acts simultaneously on all flavors . it can spontaneously be broken by a chiral condensate @xmath33 , which for finite interactions goes along with a mass gap generation . incidentally , the @xmath0 gross - neveu model actually has a much larger continuous u(@xmath8 ) flavor symmetry also allowing for more complicated breaking patterns @xcite . condensed matter systems where the gross - neveu model is considered as an effective theory , the low - energy degrees of freedom can be arranged into @xmath34 4-component dirac spinors , corresponding to a reducible representation of the dirac algebra . this reducible representation can be constructed from a suitable combination of 2-component spinors such that @xmath35 in terms of the counting of fermions of the present work , see , e.g. , @xcite for a review . note that the gross - neveu interaction term considered in this work @xmath36 corresponds to @xmath37 in the reducible 4-component notation of @xcite ( or to @xmath38 in the notation of @xcite ) for even @xmath8 . the critical properties of the discrete chiral transition are , however , identical to a 4-component gross - neveu model with a @xmath39 interaction as considered in @xcite . ] in the following , we use the functional renormalization group to compute the rg flow of the gross - neveu coupling as a function of the ( negative ) curvature . we employ the wetterich equation @xcite , describing the flow of a scale - dependent effective action functional @xmath40 as a function of an ir regulator scale @xmath41 , @xmath42 = \frac{{{\ensuremath{\mathrm{i}}}}}{2 } { \operatorname{str}}\left [ \left ( \gamma_{k}^{\ensuremath{\text{\tiny{$(2)$ } } } } + r_{k } \right)^{-1 } \partial_{k } r_{k } \right ] \ ! \text{. } \label{eq : wetteq}\end{aligned}\ ] ] the effective average action @xmath40 is related to the standard generating functional for 1pi correlation functions in the limit @xmath43 . towards the uv cutoff @xmath44 , @xmath40 approaches the microscopic action . the regularization is encoded in the regulator function @xmath45 , see below . for reviews of the functional rg adapted to the present context , see refs . @xcite . in the present work , we evaluate the flow within a rather simple approximation for the effective action . for this , we truncate the effective action to @xmath46 = { \int \!\!\ ! { } _ { { } _ { { } _ { { } _ { { } _ \text{\small{\ensuremath{x } } } } } } } } \left [ \bar{\psi } \slashed{\nabla } \psi + \frac{\bar{\lambda}_{k}}{2 { n } } ( \bar{\psi } \psi ) ^2 \right]\text{,}\end{aligned}\ ] ] where the only scale dependence lies in the four fermion coupling @xmath47 . furthermore , @xmath48 parametrically depends on the curvature of the background manifold . the ir regularization is ensured by a chirally symmetric regulator of the form @xmath49 where the superscript @xmath50 denotes transposition in dirac space , and @xmath51 represents a spin - valued delta distribution , keeping track of the spinor or conjugate - spinor transformation properties associated with the spacetime arguments . in app . [ app : beta_function ] , we briefly summarize our conventions . for practical computations , we use a callan - symanzik type regulator , that facilitates the use of proper time representations , @xmath52 within our investigations we restrict ourselves to negative curvature , giving rise to gravitational catalysis . it is intuitively clear , that positive curvature ( e.g. , a sphere ) generically suppresses ir modes and thus reduces the density of states of low lying modes @xcite . we also consider the system mainly in the large-@xmath8 limit . at finite @xmath8 , further pointlike fermionic self - interactions are generated which correspond to operators with an explicit curvature dependence . we expect no qualitative modifications from these operators and hence ignore them in the following . this approximation becomes exact in the limit @xmath53 . it is straightforward ( cf . [ app : beta_function ] ) to calculate the flow of the coupling as an implicit functional of the choice of the manifold , which enters via the spectrum of the dirac operator @xmath54 . \end{aligned}\ ] ] for this calculation , it suffices to project the flow onto constant fields @xmath55 , with @xmath56 . here , @xmath57 denotes the spacetime volume . the operator occurring in the trace is related to the square of the regularized fermionic green s function in curved spacetime . this has a direct correspondence to a feynman diagram representation , see fig . [ fig : feynman_graph ] , as the flow in the present simple truncation is driven by a single fermion bubble ( and rg - improved resummations thereof ) . in the following we distinguish between the cases of a maximally symmetric spacetime in sect . [ sec : maximally_symmetric ] which can be treated fully analytically , and a negatively curved space in sect . [ sec : flat_time ] , which is a more interesting case in view of two - dimensional condensed matter systems . the use of a callan - symanzik regulator shape function , cf . ( [ eq : csregulator ] ) , facilitates to rewrite the right - hand side of eq . ( [ eq : betafunction1 ] ) in terms of a simple proper time representation . other shape functions would still permit to use a proper time representation but would lead to more intricate @xmath41 dependencies . the laplace transform of the operator of ( [ eq : betafunction1 ] ) reads @xmath58 which upon insertion into ( [ eq : betafunction1 ] ) yields @xmath59 \text{. } \label{eq : lambdabar}\end{aligned}\ ] ] the expression inside the super trace is known as the heat kernel @xmath60 of the ( squared ) dirac operator . it satisfies @xmath61 and was calculated in @xcite for any maximally symmetric space , with euclidean signature . for our case the solution to this equation can be obtained in an easier way with a special ansatz ( cf . [ app : heat_kernel ] ) : @xmath62 where @xmath63 , and @xmath64 is the geodesic distance between the points @xmath65 and @xmath66 . the parallel transporter @xmath67 ( wegner - wilson line ) is defined by @xmath68 where @xmath69 denotes the path ordering prescription and @xmath70 is the geodesic between @xmath65 and @xmath66 with @xmath71 and @xmath72 . now we are able to calculate the supertrace as @xmath73 \ ! = - \frac{2 \omega { n}}{(4 \pi)^{\frac{3}{2 } } \sqrt{s } } \ ! \left ( \frac{1}{s } + { { \ensuremath{\mathrm{i}}}}\frac{{\ensuremath{\left\vertr\right\vert}}}{12 } \right ) \ ! { \mathrm{e}}^{- { { \ensuremath{\mathrm{i}}}}\frac{\pi}{4 } } \text{,}\end{aligned}\ ] ] and finally get @xmath74 in terms of the dimensionless coupling , @xmath75 we obtain the beta function @xmath76 @xmath77 this is an ordinary differential equation that parametrically depends on the curvature . it can be solved by straightforward integration : @xmath78 the initial value is given by the dimensionless coupling @xmath79 which in terms of the initial bare gross - neveu coupling reads @xmath80 . for the case of a manifold where the spatial part has a constant negative curvature , we choose a special set of coordinates such that the metric can be expressed as @xmath81 where @xmath82 ( latin indices running from @xmath83 to @xmath3 ) represents the metric of a two dimensional maximally symmetric space and therefore only depends on the spatial coordinates . hence the gamma matrices are time independent as well . the christoffel symbols vanish for every time component @xmath84 and also the curvature tensor vanishes if any index is zero @xmath85 . from eq . ( [ eq : defspincon ] ) @xmath86 , we infer that the spin connection @xmath87 has to be proportional to @xmath18 , @xmath88.\end{aligned}\ ] ] moreover , from eq . @xmath27 , we conclude that @xmath87 even has to vanish completely , @xmath89 this implies that the operator @xmath90 is separable into @xmath91 where the curvature is only induced by the spatial components . for a simpler calculation of the beta function , we perform a wick rotation @xmath92 and perform again a laplace transformation of the operator occurring in eq . ( [ eq : betafunction1 ] ) , @xmath93 we arrive at the euclidean analogue of eq . @xmath94 \text{.}\end{aligned}\ ] ] here , we calculate the super trace again with the aid of the heat kernel using that the differential operator @xmath90 is separable and the delta distribution factorizes in a time like and a spatial part @xmath95 where @xmath96 . the quantity @xmath97 denotes the spatial coordinates of @xmath65 and should be treated as a set of coordinates and not as a vector . both factors satisfy a heat - kernel equation and can be solved analytically , cf . @xcite , @xmath98 where @xmath99 , @xmath100 is the parallel transporter for the spatial part and @xmath101 is the nonnegative spatial geodesic distance between @xmath97 and @xmath102 with @xmath103 . plugging these relations into eq . ( [ eq : betafunction2 ] ) gives @xmath104 the @xmath105 integral is an integral representation of the modified bessel function of the second kind @xmath106 @xcite . therefore , we have @xmath107 with these results we are again able to derive the beta function for the dimensionless coupling @xmath108 , @xmath109 the integration of this ordinary differential equation depending parametrically on the curvature can be cast into an integral representation , @xmath110 this result is qualitatively similar to the maximally symmetric case of eq . . let us now analyze the consequences of the rg flows for the long - range properties of the gross - neveu model . for both background manifolds , @xmath76 considered as a function of @xmath111 is a parabola where the prefactor of the quadratic part is scale and curvature dependent , see fig . [ fig : betafunction ] . for vanishing curvature , the @xmath76 function vanishes at the two fixed points @xmath112 ( gauian ) and @xmath113 which corresponds to the well - known critical coupling of the gross - neveu model in flat space in this regularization scheme @xcite . this critical coupling separates the symmetric phase for @xmath114 where the long range behavior is controlled by the non - interacting gauian fixed point from the chiral symmetry broken phase for @xmath115 . in the latter case , @xmath111 runs to large values towards the infrared . in the present simple truncation , @xmath111 in fact diverges at a finite scale @xmath116 signaling the transition into the ordered regime . the scale @xmath116 is thus characteristic for the physical scales in the ordered phase . in @xcite it has been shown that @xmath116 actually agrees with the value of the dynamically generated fermion mass @xmath117 as obtained in mean - field approximation . since we are working in the large-@xmath8 limit anyway , we will use this mean - field identification in the following : @xmath118 . the existence of the non - gauian fixed point @xmath119 can be attributed to the competition between the power - counting scaling ( the linear coupling term in @xmath76 ) and the interaction terms @xmath120 . in our rg picture , the interaction terms are enhanced by negative curvature as soon as the wavelength of the fluctuations becomes of the order of the curvature scale . as a consequence , the interacting second zero of the @xmath121 function no longer is a true fixed point but becomes scale dependent . this `` pseudo - critical coupling '' @xmath122 moves towards the gauian fixed point for decreasing scale @xmath41 , see fig . [ fig : betafunction ] . any finite initial coupling strength @xmath79 will eventually become larger than @xmath123 for small rg scales @xmath124 . by this mechanism , the system is forced into the symmetry - broken phase even at the weakest initial coupling . we observe this mechanism in both cases of negative curvature , the maximally symmetric as well as the purely spatial curvature case . function of the coupling @xmath125 for different values of the scale dependent negative curvature @xmath126 ( from top to bottom : @xmath127 ) . the black lines depict the @xmath76 functions for the maximally symmetric spacetime ( ads ) , cf . eq . ( [ eq : betafunction_maximally_symmetric ] ) ; arrows indicate the flow towards the ir . the green dotted graphs show the flows for the case of purely spatial curvature ( lobachevsky plane ) , cf . ( [ eq : betafunction_flat_time ] ) . in addition to the gauian fixed point , there exists a non gauian fixed point ( full red circle at @xmath128 for the present regulator scheme ) , separating the symmetric phase for @xmath129 from the broken phase for @xmath130 for zero curvature . for finite curvature , this critical point becomes scale - dependent and moves towards the gauian fixed point for increasing scale dependent curvature , i.e. , with decreasing ir scale for fixed curvature . in the case of vanishing curvature , the symmetry is preserved and no mass is generated for initial values @xmath131 in the blue dashed region.,title="fig:",scaledwidth=48.0% ] as we see in fig . [ fig : betafunction ] the influence of the curvature is somewhat stronger in the maximally symmetric case . as discussed above , we calculate the symmetry breaking scale @xmath116 by searching for a zero of the inverse coupling . the fermion mass @xmath117 corresponding to this scale where the rg flow enters the symmetry - broken regime can thus be computed from the criterion @xmath132 upon partial bosonization ( hubbard - stratonovich transformation ) , @xmath133 is related to the mass parameter of a composite bosonic field . hence the divergence of the fermionic self - interaction simply corresponds to onset of the order parameter @xcite . let us now analyze the two different backgrounds under consideration in detail . in the maximally symmetric case , the fermion mass defined by the criterion eq . can be straightforwardly computed from the running coupling eq . , yielding @xmath134 plots of this gravitationally catalyzed fermion mass are shown as a function of the curvature as solid lines in fig . [ fig : comparison_generated_mass ] . let us discuss this result in various limits . in the zero - curvature limit @xmath135 , we find @xmath136 for @xmath137 . for super - critical couplings @xmath138 , we rediscover the standard mean - field result in @xmath0 , @xmath139 this is in perfect agreement with the known behavior in flat spacetime . provided the fermion system is initially weakly coupled , @xmath140 , a leading order expansion can be performed for any value of the curvature , resulting in @xmath141 } |r| \bar{\lambda}_\lambda } } , \quad \text{for}\,\ , \bar{\lambda}_\lambda \lambda \ll 1,\ ] ] where we have reinserted the dimensionful initial coupling @xmath142 . if we additionally consider the limit of small curvature , we find a linear dependence of the fermion mass on both the curvature as well as the coupling , @xmath143 by contrast , in the limit of large curvature , @xmath144 , we find that @xmath145 . in other words , large curvature induces immediate chiral symmetry breaking , such that the induced mass becomes of the order of the cutoff . incidentally , this result is similar for the large - coupling limit : for @xmath146 , we again find that @xmath147 to leading order independently of the curvature . the above results display explicit uv cutoff and regularization - scheme dependencies . since the @xmath0 gross - neveu model is asymptotically safe and thus non - perturbatively renormalizable @xcite , we can remove the uv cutoff by keeping an ir obvservable fixed while sending @xmath148 . this `` line of constant physics '' defines a renormalized trajectory . this can most conveniently be done in the super - critical regime where @xmath149 such that the fermion mass in flat - space @xmath150 of eq . defines a natural ir renormalization point . at a suitable renormalization point @xmath151 to a specific value . ] in this case , the generated fermion mass in the limit @xmath152 can be written as @xmath153 where the first line holds for arbitrary curvature , and the second line represents a weak - curvature expansion being in perfect agreement with @xcite . we emphasize that the fermions acquire a mass @xmath154 for any given @xmath79 as long as the curvature is nonvanishing . while we expect the tendency to drive the fermion system towards the broken phase through gravitational catalysis to remain also beyond our truncation , fluctuations of bosonic composites entering beyond the large-@xmath8 limit typically provide for an opposite tendency . hence , the status of gravitational catalysis beyond mean - field remains an interesting question . analogously , the effects of beyond - mean - field fluctuations on magnetic catalysis are under active current investigation @xcite . in the case of pure spatial curvature , the criterion eq . can not be resolved analytically , but we can give an implicit equation for the induced fermion mass @xmath117 @xmath155 which can be solved numerically . though the basic picture does not differ much from the maximally symmetric case , there are some interesting differences . as can already be inferred from the beta function in fig . [ fig : betafunction ] ( dotted lines ) , the curvature induced mass in the spatially curved case is smaller than in the maximally symmetric case , cf . fig . [ fig : comparison_generated_mass ] . from top to bottom ( @xmath128 ) . as long as the background manifold is negatively curved , @xmath156 , a finite fermion mass is generated.,title="fig:",scaledwidth=48.0% ] several limits can be discussed in an analytic fashion , using the series representation of @xmath157 developed in app . [ app : curvature_expansion ] . furthermore , we need the integral @xmath158 defined by @xmath159 where @xmath27 fixes the constant of integration . the explicit calculation is done in app . [ app : curvature_expansion ] . the defining equation for the fermion mass in terms of @xmath160 is @xmath161 this representation provides us with some interesting insight . first , we can show that there exists a unique solution to this equation with @xmath162 for any given negative curvature @xmath163 and @xmath164 . this can be seen in two steps . the uniqueness is because the function @xmath165 for @xmath166 is bijective . this property holds , because @xmath157 is positive and therefore @xmath160 is strictly monotonically increasing . owing to @xmath167 for small @xmath168 and @xmath169 for large @xmath170 , the function @xmath160 is not bounded . since by assumption @xmath171 , it follows from eq . that @xmath172 has to hold , which implies that @xmath173 because of the monotonic behavior this demonstrates that @xmath174 as claimed above . let us first check the flat spacetime limit @xmath175 . in complete agreement with eq . , we find again that @xmath176 = 1 - \frac{m_{\text{f},0}}{\lambda},\end{aligned}\ ] ] where we have used eq . . in the weak coupling regime , we expect the fermion mass to be small compared to the curvature scale , cf . . hence , we need @xmath158 for large argument as provided by eq . , @xmath177 using this asymptotic behavior for @xmath178 , we infer from eq . that @xmath179 , \,\ , \alpha_\lambda= \sqrt{\frac{|r|}{2\lambda^2}}. \end{aligned}\ ] ] if in addition , the curvature is small , i.e. , @xmath180 , we can use the corresponding expansion of eq . and replace @xmath181 . taking differences arising from the dirac representation into account , the exponential inverse - coupling dependence is in perfect agreement with the results of @xcite . despite the overall similarities to the maximally symmetric case , we observe that the weak - coupling and weak - curvature limit of the case with pure spatial curvature shows some distinct differences . in particular , there is an exponential non - analytic dependence of the fermion mass on the coupling as well as on the curvature . this difference is reminiscent to magnetic catalysis in @xmath182 and @xmath183 @xcite , where the fermion gap is analytic in @xmath182 , but shows an essential singularity in the coupling in @xmath183 . also in the present case , such a singularity shows up similar to bcs gap formation , as a consequence of the effective dimensional reduction of the fermionic fluctuation spectrum to @xmath184 @xcite . also in this respect , our functional rg picture is in perfect agreement with previous studies @xcite . at zero curvature , the @xmath0 gross - neveu model exhibits a critical coupling strength corresponding to a quantum critical point above which chiral symmetry is broken at large length scales . this critical coupling manifests itself as a non - gauian fixed point of the rg flow . in the present regularization scheme , we identified @xmath185 , or @xmath128 in dimensionless conventions . as illustrated in fig . [ fig : betafunction ] , the fixed point strictly speaking no longer exists at finite negative curvature . the nontrivial zero of the @xmath121 function becomes scale dependent and eventually merges with the gauian fixed point in the deep ir for @xmath186 , such that only the chirally broken branch of the @xmath121 function remains . let us call the nontrivial zero of @xmath121 a pseudo - critical coupling @xmath187 . we find @xmath188 for the maximally symmetric case , cf . . , and @xmath189^{a } } \right . \!\ ! \sqrt{\frac{|r|}{2k^2 } } \left . \vphantom{\left[\frac{a}{b}\right]^{a } } \right ) } \label{eq : lpsc}\ ] ] for the spatially curved case , cf . . since @xmath190 is a monotonically decreasing function of scale @xmath41 , the coupling @xmath125 can eventually exceed @xmath187 such that the system becomes critical and runs towards the ordered phase . in this sense , @xmath191 can be viewed as a criticality condition @xcite , defining a scale @xmath192 , where the system becomes critical . for lower scale , the system is driven towards the symmetry broken regime which is ultimately entered at @xmath193 defined above . the value of @xmath192 depends on the curvature as well as the initial coupling @xmath79 ( the latter is considered as initially subcritical here and in the following ) . the preceding discussion implicitly assumed that @xmath41 can run over all scales down to @xmath194 , such that the criticality condition can eventually always be satisfied . however , if , for instance , the system has a finite volume characterized by a finite length scale @xmath195 , also the fluctuation momenta are restricted , typically leading to an ir cutoff @xmath196 . one may think of a finite probe length , such as , e.g. , the size of a layer of graphene . this finite probe length @xmath195 can lead to a screening of the gravitationally catalyzed ordered regime if @xmath197 is larger than the would - be critical scale @xmath192 . hence , @xmath198 can be viewed as a lower bound for the coupling strength required for symmetry breaking in a real system of finite length . it thus generalizes the critical coupling at infinite volume and zero curvature to the situation of finite volume and finite curvature . it is instructive to study @xmath187 as a function of the length scale @xmath195 measured in units of a typical curvature length scale which we define by @xmath199 . for finite systems , this gives an estimate for how strongly a probe has to be curved in order to exhibit gravitational catalysis . in turn , for a given curvature of the probe , @xmath187 provides an estimate for the initial coupling strength required for symmetry breaking in the finite system . for instance , for the maximally symmetric case , we read off from eq . that @xmath200 for the spatially curved case , we incidentally find the same result in the limit of small curvature , i.e. , @xmath201 , using the expansion which is accurate even for values of @xmath202 . by contrast , the large curvature limit for the spatially curved case is different , cf . : @xmath203 from eq . , it is obvious that probe length to curvature ratios up to @xmath204 lead to pseudo - critical couplings @xmath187 which deviate from the zero - curvature critical coupling of the gross - neveu model only below the 1% level . significant deviations only occur for @xmath195 being an order of magnitude larger than the curvature scale @xmath205 . from the viewpoint of curvature - deformed condensed matter systems , a ratio of @xmath204 appears to be `` large '' in the sense that loosely speaking a spatially curved @xmath206 planar probe embedded in @xmath0 euclidean space would rather look like a @xmath0 object . another way to interpret these results is the following : consider a finite - probe system with a subcritical bare coupling , @xmath207 , thus being in the symmetric ( e.g. semimetal ) phase . in order to gravitationally catalyze a transition to a broken ( gapped or insulating ) phase , the criticality condition has to be met for a sufficiently large @xmath208 . in view of eq . this requires comparatively strong curvature , i.e. , a small curvature length scale @xmath205 compared with the probe length @xmath195 . we have investigated the phenomenon of gravitational catalysis in the @xmath0 gross - neveu model on specific manifolds with constant negative curvature . while the mechanism had already been studied frequently with mean - field methods as well as from the viewpoint of the fluctuation spectrum of the dirac operator , we have added a new renormalization group picture to the comprehensive understanding of this phenomenon . the essence of this picture is that the critical coupling of the fermionic system , corresponding to a quantum phase transition in flat spacetime , is transmuted into a scale - dependent pseudo - critical coupling that flows to zero as a consequence of long - wavelength fluctuations ( compared to the curvature scale ) . in this manner , the infinite - volume fermion system becomes critical for any arbitrarily weak coupling . we have identified the rg ( pseudo- ) fixed point mechanism for two example manifolds of constant negative curvature : a maximally symmetric spacetime ( anti de sitter ) and a purely spatially curved case ( lobachevsky plane ) . both manifolds support the mechanism of gravitational catalysis , but exhibit a rather different behavior as far as the dependence of chiral symmetry breaking on the coupling and the curvature are concerned . the maximally symmetric case shows a linear dependence ( to leading order ) on both quantities which makes clear that the phenomenon is essentially perturbative , being reminiscent of the `` quantum anomaly '' for fermions in a magnetic field @xcite . by contrast , order parameters indicating the symmetry broken state such as the induced fermion mass exhibit an essential singularity in both the coupling and the curvature for the case of the purely spatially curved case . this is in many respects similar to bcs - type gap formation . again , our renormalization group picture goes hand in hand here with properties of the fermionic fluctuation spectrum , as analyzed in @xcite . as a benefit , the functional renormalization group also gives a simple access to systems of finite extent by identifying the rg infrared cutoff with an inverse length scale . in this manner , we can estimate the fate of gravitational catalysis in finite systems . in fact , the phenomenon only occurs , as long as the curvature radius is sufficiently small compared to the systems length scale . we have been able to phrase this statement quantitatively by introducing a pseudo - critical coupling . thinking in terms of curved layered condensed matter systems , rather large curvatures are needed compared to a realizable probe length in order to drive a sub - critical system into a phase dominated by gravitational catalysis . we would like to emphasize that an immediate application of our results to condensed matter systems would only be possible for reparametrization invariant systems such as fluid membranes @xcite , curvature effects of which can be mapped onto the language of riemannian geometry . for tethered membranes or general lattice systems , further phenomena connected to extrinsic curvature or curvature related defects can be become relevant . in this context , it is interesting to note that an external strain exerted on a graphene sheet in flat space induces a pseudo - magnetic field @xcite , that may also support ( pseudo-)magnetic catalysis . for such systems , we hence expect an interesting interplay between these various effects if we expose them to negative curvature inducing strain . finally , gravitational catalysis may become relevant in the context of asymptotically safe quantum gravity @xcite . in conjunction with fermionic degrees of freedom @xcite , the uv fixed point determining the shape of the universe at highest energies might go along with a negative ( though scale - dependent ) curvature @xcite . whether or not gravitational catalysis in connection with gravitionally modified critical fermion interactions @xcite could become active and impose constraints on the matter content of the universe then is a highly involved question that deserves to be investigated in greater depth . the authors thank astrid eichhorn , lukas janssen , daniel scherer , ren sondenheimer , and andreas wipf for valuable discussions and acknowledge support by the dfg under grants gi 328/5 - 2 ( heisenberg program ) , grk1523 and for 723 . let us first introduce our conventions . in the weldon formalism @xcite , the dirac conjugate spinor @xmath209 is related to the hermitean conjugate of the spinor @xmath210 via the spin metric @xmath211 @xmath212 which is implicitly defined by @xmath213 this gives rise to the following properties of the building blocks of the gross - neveu action : @xmath214 choosing the spin metric @xmath211 to be anti - hermitean in eq . ( i ) is convenient for our metric and clifford algebra conventions , but differs from @xcite . also , the absence of any imaginary factor of `` @xmath215 '' in the fermion kinetic term is due to these conventions . the flow equation uses a rather condensed notation . more concretely , we work in field space parameterized by the collective fields @xmath216 representing grassmann - valued functions on the manifold , reminiscent to nambu - gorkov spinors . for instance , the classical action amended by the regulator term which is used for deriving the flow equation from the functional integral ( cf . @xcite ) , reads in these conventions @xmath217 = { } & s[\phi ] + \frac{1}{2 } { \int \!\!\ ! { } _ { { } _ { { } _ { { } _ { { } _ \text{\small{\ensuremath{x } } } } } } } } \!\!\ ! { \int \!\!\ ! { } _ { { } _ { { } _ { { } _ { { } _ \text{\small{\ensuremath{y } } } } } } } } \bar{\phi}(x ) r_{k}(x , y ) \phi(y).\end{aligned}\ ] ] these conventions differ slightly from those commonly used , see , e.g. , @xcite , but turn out to be advantageous for coordinate space computations on curved manifolds . for instance , the representation of the unit element in field space becomes rather intuitive , @xmath218 by @xmath51 , we denote the spin - valued delta distribution , which fulfills @xmath219 therefore , @xmath220 has to hold . with this notation , we indicate that @xmath51 transforms as a spinor in @xmath65 and a dirac - conjugated spinor in @xmath66 , i.e. @xmath51 can be interpreted as @xmath221 , with the standard scalar delta distribution @xmath222 . we remark that there is a difference between @xmath51 and @xmath223 in the spinor structure , since @xmath224 for the evaluation of the flow equation eq . , we proceed in a standard fashion . we decompose @xmath225 into a field - dependent part @xmath226 and a field - independent part @xmath227 in order to expand the wetterich equation in powers of the fields , @xmath228 \text{.}\end{aligned}\ ] ] since @xmath229 & \psi \psi^{{\ensuremath{\mathrm{t } } } } \\ \bar{\psi}^{{\ensuremath{\mathrm{t } } } } \bar{\psi } & { } \!\!\ ! [ ( \bar{\psi } \psi ) { \ensuremath{\mathrm{i}}}+ \psi \bar{\psi}]^{{\ensuremath{\mathrm{t } } } } \ ! { } \end{pmatrix } \ ! \mathbb{1},\\ { \ensuremath{\mathcal{p}}}_{k } = { } & { { \ensuremath{\mathrm{i}}}}k \ ! \begin{pmatrix } \sqrt{\tau}\big ( { \ensuremath{\mathrm{i}}}+ r(\tau ) \big ) & 0 \\ 0 & \sqrt{\tau^{{\ensuremath{\mathrm{t } } } } } \big ( { \ensuremath{\mathrm{i}}}+ r(\tau^{{\ensuremath{\mathrm{t } } } } ) \big ) \end{pmatrix } \ ! \mathbb{1},\end{aligned}\ ] ] we observe that only the term @xmath230 in eq . can contribute to the flow of @xmath47 . we need @xmath231 with @xmath232 , and therefore only the diagonal of @xmath233 is required . in the limit @xmath234 , only the following terms remain : @xmath235_{11}(x , y ) = - \frac{\frac{\bar{\lambda}_{k}^2}{{n}^{2 } k^{2 } } ( \bar{\psi } \psi)^2}{\tau \big ( { \ensuremath{\mathrm{i}}}+ r(\tau ) \big)^{2 } } \ , \frac{\delta(x , y)}{\sqrt{-g}},\\ & \left[({\ensuremath{\mathcal{p}}}_{k}^{-1 } { \ensuremath{\mathcal{f}}}_{k})^2\right]_{22}(x , y ) = \ ! \left[({\ensuremath{\mathcal{p}}}_{k}^{-1 } { \ensuremath{\mathcal{f}}}_{k})^2\right]^{{\ensuremath{\mathrm{t}}}}_{11}(y , x)\text{. } \ ] ] the lhs of eq . boils down to @xmath236 and the rhs yields @xmath237 \notag \\ & { } = { { \ensuremath{\mathrm{i}}}}\frac{2 \bar{\lambda}_{k}^2}{{n}^2 k^3 } ( \bar{\psi } \psi)^2 { \operatorname{str}}\ ! \left [ \frac { r'(\tau ) } { \big ( { \ensuremath{\mathrm{i}}}+ r(\tau ) \big)^{3 } } \frac{\delta(x , y)}{\sqrt{-g } } \right].\end{aligned}\ ] ] inserting the callan - symanzik regulator , we end up with eq . of the main text . following @xcite , we choose for the heat kernel satisfying eq . the ansatz @xmath238 where @xmath239 is a scalar function of @xmath240 the geodesic distance and the proper time @xmath105 . plugging this into equation ( [ eq : heatkernel ] ) using @xmath241 , @xmath242 and @xmath243 n^{\nu } u \text { , } \end{aligned}\end{aligned}\ ] ] cf . @xcite , we get @xmath244 with @xmath245 , @xmath246 . because of the boundary condition for @xmath247 and the regularity of @xmath248 we observe that @xmath249 has to hold , with @xmath222 representing the standard delta distribution . one suitable representation for the delta distribution in 3@xmath13 maximally symmetric space with negative curvature in the limit @xmath250 is @xmath251 next , we factorize @xmath239 into the delta part and an auxiliary function @xmath252 of @xmath240 and @xmath105 , @xmath253 expanding @xmath252 in powers of @xmath254 and using the boundary conditions , @xmath255 leads to @xmath256 with the boundary condition @xmath257 . from equation @xmath86 and the boundary condition , we see @xmath258 plugging this into @xmath27 gives @xmath259 where we have eliminated the constant of integration using equation @xmath260 . from this , we finally get @xmath261 for the detailed analysis of the spatially curved case , the integral representation of the running coupling eq . can be studied in various limits analytically . more specifically , @xmath157 as defined in eq . can be expanded for small and for large values of @xmath262 . let us first consider an expansion of this function about @xmath263 , starting with an expansion of the integrand , @xmath264 using the standard integral @xmath265 the small @xmath262 expansion of @xmath157 can be computed to any order . to order @xmath266 , we find @xmath267 due to the factorial growth of the coefficients , cf . eq . ( [ eq : bessel_integral ] ) , the expansion is an asymptotic series . by comparison with the numerical result , the accuracy of eq . ( [ eq : mfri_small_alpha ] ) turns out to be above 99% up to @xmath268 . a similar approximation can be done for large @xmath262 by expanding @xmath269 which holds for any @xmath270 . next , we employ @xmath271 which can be used for @xmath272 . for @xmath273 , an analytic continuation into the complex is implicitly understood , leading to a replacement of the term @xmath274 by @xmath275 \big ( 1 - ( n \alpha)^2 \big)^{-3/2}$ ] here and in the following . this leads to the convergent series @xmath276\end{aligned}\ ] ] for any @xmath277 . neglecting orders higher than @xmath278 and using @xmath279 for @xmath280 , we arrive at @xmath281 where @xmath282 is the euler - mascheroni constant and @xmath283 is the glaisher - kinkelin constant . the accuracy of this result is above 99% for @xmath284 . with the series ( [ eq : mfri_sum_rep ] ) it is possible to find a series representation for the required integral ( [ eq : def_mfrf ] ) @xmath285 \ ! \text{.}\end{aligned}\ ] ] this series is convergent for any @xmath286 . the expansions for small and large @xmath262 are @xmath287 where @xmath288 is defined as the limit @xmath289 v. p. gusynin , v. a. miransky and i. a. shovkovy , phys . d * 52 * , 4718 ( 1995 ) [ hep - th/9407168 ] . v. p. gusynin , v. a. miransky and i. a. shovkovy , phys . b * 349 * , 477 ( 1995 ) [ hep - ph/9412257 ] . v. p. gusynin , v. a. miransky and i. a. shovkovy , nucl . b * 462 * , 249 ( 1996 ) [ hep - ph/9509320 ] . i. a. shovkovy , arxiv:1207.5081 [ hep - ph ] . i. a. shushpanov and a. v. smilga , phys . b * 402 * , 351 ( 1997 ) [ hep - ph/9703201 ] . t. d. cohen , d. a. mcgady and e. s. werbos , phys . c * 76 * , 055201 ( 2007 ) [ arxiv:0706.3208 [ hep - ph ] ] . a. v. zayakin , jhep * 0807 * , 116 ( 2008 ) [ arxiv:0807.2917 [ hep - th ] ] . a. j. mizher , m. n. chernodub and e. s. fraga , phys . d * 82 * , 105016 ( 2010 ) [ arxiv:1004.2712 [ hep - ph ] ] . r. gatto and m. ruggieri , phys . d * 82 * , 054027 ( 2010 ) [ arxiv:1007.0790 [ hep - ph ] ] . j. k. boomsma and d. boer , phys . d * 81 * , 074005 ( 2010 ) [ arxiv:0911.2164 [ hep - ph ] ] . g. s. bali , f. bruckmann , g. endrodi , z. fodor , s. d. katz and a. schafer , phys . d * 86 * , 071502 ( 2012 ) [ arxiv:1206.4205 [ hep - lat ] ] . g. w. semenoff , i. a. shovkovy and l. c. r. wijewardhana , mod . lett . a * 13 * , 1143 ( 1998 ) [ hep - ph/9803371 ] . k. krishana , n. p. ong , y. zhang , _ et al . _ , phys . lett . * 82 * , 5108 ( 1999 ) [ cond - mat/9904138 ] . d. v. khveshchenko , phys . lett . * 87 * , 206401 ( 2001 ) [ cond - mat/0106261 ] . h. leal and d. v. khveshchenko , nucl . b * 687 * , 323 ( 2004 ) [ cond - mat/0302164 ] . i. f. herbut and b. roy , phys . b * 77 * , 245438 ( 2008 ) [ arxiv:0802.2546 [ cond-mat.mes-hall ] ] ; b. roy and i. f. herbut , phys . b * 83 * , 195422 ( 2011 ) [ arxiv:1102.3481 [ cond-mat.mes-hall ] ] . e. j. ferrer and v. de la incera , phys . * 102 * , 050402 ( 2009 ) [ arxiv:0807.4744 [ hep - ph ] ] ; nucl . b * 824 * , 217 ( 2010 ) [ arxiv:0905.1733 [ hep - ph ] ] ; e. j. ferrer , v. de la incera and a. sanchez , phys . lett . * 107 * , 041602 ( 2011 ) [ arxiv:1103.5152 [ hep - ph ] ] . d. d. scherer and h. gies , phys . b * 85 * , 195417 ( 2012 ) [ arxiv:1201.3746 [ cond-mat.str-el ] ] . k. fukushima and j. m. pawlowski , phys . d * 86 * , 076013 ( 2012 ) [ arxiv:1203.4330 [ hep - ph ] ] . i. l. buchbinder and e. n. kirillova , sov . j. * 32 * , 446 ( 1989 ) ; int . j. mod . phys . a * 4 * , 143 ( 1989 ) . i. l. buchbinder , s. d. odintsov and i. l. shapiro , `` effective action in quantum gravity , '' bristol , uk : iop ( 1992 ) . t. inagaki , t. muta and s. d. odintsov , mod . lett . a * 8 * , 2117 ( 1993 ) [ hep - th/9306023 ] . i. sachs and a. wipf , phys . b * 326 * , 105 ( 1994 ) [ hep - th/9310085 ] . e. elizalde , s. leseduarte , s. d. odintsov and y. i. shilnov , phys . d * 53 * , 1917 ( 1996 ) [ hep - th/9505065 ] . s. kanemura and h. -t . sato , mod . a * 11 * , 785 ( 1996 ) [ hep - th/9511059 ] . t. inagaki , int . j. mod . a * 11 * , 4561 ( 1996 ) [ hep - th/9512200 ] . t. inagaki and k. -i . ishikawa , phys . d * 56 * , 5097 ( 1997 ) . b. geyer and s. d. odintsov , phys . b * 376 * , 260 ( 1996 ) [ hep - th/9603172 ] ; phys . d * 53 * , 7321 ( 1996 ) [ hep - th/9602110 ] . g. miele and p. vitale , nucl . b * 494 * , 365 ( 1997 ) [ hep - th/9612168 ] ; p. vitale , nucl . b * 551 * , 490 ( 1999 ) [ hep - th/9812076 ] . t. inagaki , t. muta and s. d. odintsov , prog . suppl . * 127 * , 93 ( 1997 ) [ hep - th/9711084 ] . j. hashida , s. mukaigawa , t. muta , k. ohkura and k. yamamoto , phys . d * 61 * , 044015 ( 2000 ) [ gr - qc/9907014 ] . e. v. gorbar and v. p. gusynin , annals phys . * 323 * , 2132 ( 2008 ) [ arxiv:0710.2292 [ hep - ph ] ] . m. hayashi , t. inagaki and h. takata , arxiv:0812.0900 [ hep - ph ] ; t. inagaki and m. hayashi , arxiv:1003.1173 [ hep - ph ] . s. sasagawa and h. tanaka , prog . . phys . * 128 * , 925 ( 2012 ) [ arxiv:1209.2782 [ hep - ph ] ] . e. v. gorbar , phys . d * 61 * , 024013 ( 2000 ) [ hep - th/9904180 ] ; ukr . j. phys . * 54 * , 541 ( 2009 ) [ arxiv:0809.2558 [ hep - th ] ] . d. ebert , a. v. tyukov and v. c. zhukovsky , phys . d * 80 * , 085019 ( 2009 ) [ arxiv:0808.2961 [ hep - th ] ] . d. v. khveshchenko , phys . lett . * 87 * , 246802 ( 2001 ) [ cond - mat/0101306 ] . i. f. herbut , phys . lett . * 97 * , 146401 ( 2006 ) [ cond - mat/0606195 ] . d. j. gross and a. neveu , phys . d * 10 * , 3235 ( 1974 ) . h. a. weldon , phys . d * 63 * , 104010 ( 2001 ) [ gr - qc/0009086 ] . f. hofling , c. nowak and c. wetterich , phys . b * 66 * , 205111 ( 2002 ) [ cond - mat/0203588 ] . l. janssen and h. gies , phys . d * 86 * , 105007 ( 2012 ) [ arxiv:1208.3327 [ hep - th ] ] ; h. gies and l. janssen , phys . d * 82 * , 085018 ( 2010 ) [ arxiv:1006.3747 [ hep - th ] ] . l. janssen , `` critical phenomena in ( 2 + 1)-dimensional relativistic fermion systems , '' phd thesis , fsu jena , http://www.db-thueringen.de/servlets/document + servlet?id=20856 ( 2012 ) . j. braun , h. gies and d. d. scherer , phys . d * 83 * , 085012 ( 2011 ) [ arxiv:1011.1456 [ hep - th ] ] . c. wetterich , phys . b * 301 * , 90 ( 1993 ) . j. berges , n. tetradis and c. wetterich , phys . rept . * 363 * , 223 ( 2002 ) [ hep - ph/0005122 ] . k. aoki , int . j. mod . b * 14 * , 1249 ( 2000 ) . j. m. pawlowski , ann . * 322 * , 2831 ( 2007 ) [ hep - th/0512261 ] . h. gies , lect . notes phys . * 852 * , 287 ( 2012 ) [ hep - ph/0611146 ] . b. delamotte , lect . notes phys . * 852 * , 49 ( 2012 ) [ cond - mat/0702365 [ cond - mat ] ] . p. kopietz , l. bartosch and f. schutz , lect . notes phys . * 798 * , 1 ( 2010 ) . w. metzner , m. salmhofer , c. honerkamp , v. meden and k. schonhammer , rev . phys . * 84 * , 299 ( 2012 ) [ arxiv:1105.5289 [ cond-mat.str-el ] ] . j. braun , j. phys . g * 39 * , 033001 ( 2012 ) [ arxiv:1108.4449 ] . r. camporesi , commun . phys . * 148 * , 283 ( 1992 ) . i. s. gradshteyn and i. m. ryzhik , `` table of integrals , series , and products , '' new york , us : academic press ( 2000 ) . j. jaeckel and c. wetterich , phys . d * 68 * , 025020 ( 2003 ) [ hep - ph/0207094 ] . u. ellwanger and c. wetterich , nucl . b * 423 * , 137 ( 1994 ) [ hep - ph/9402221 ] . h. gies and c. wetterich , phys . d * 65 * , 065001 ( 2002 ) [ hep - th/0107221 ] . v. skokov , phys . d * 85 * , 034026 ( 2012 ) [ arxiv:1112.5137 [ hep - ph ] ] . j. braun , c. s. fischer and h. gies , phys . rev . d * 84 * , 034045 ( 2011 ) [ arxiv:1012.4279 [ hep - ph ] ] . v. p. gusynin and s. g. sharapov , phys . lett . * 95 * , 146801 ( 2005 ) [ cond - mat/0506575 ] . d. nelson , t. piran and s. weinberg , `` statistical mechanics of membranes and surfaces , '' singapore , sg : world scientific ( 1989 ) . f. guinea , m. i. katsnelson and a. k. geim , nature phys . * 6 * , 30 ( 2010 ) [ arxiv:0909.1787 [ cond-mat.mes-hall ] ] ; f. guinea , a. k. geim , m. i. katsnelson , k. s. novoselov , phys . b * 81 * , 035408 ( 2010 ) ; n. levy _ et al . _ , science * 329 * , 544 ( 2010 ) . s. weinberg , _ in * hawking , s.w . , israel , w. : general relativity * , 790 - 831 _ ; m. reuter , phys . d * 57 * , 971 ( 1998 ) [ arxiv : hep - th/9605030 ] ; m. niedermaier and m. reuter , living rev . * 9 * , 5 ( 2006 ) ; r. percacci , in * oriti , d. ( ed . ) : approaches to quantum gravity * 111 - 128 [ arxiv:0709.3851 [ hep - th ] ] ; m. reuter and f. saueressig , new j. phys . * 14 * , 055022 ( 2012 ) [ arxiv:1202.2274 [ hep - th ] ] . dona and r. percacci , arxiv:1209.3649 [ hep - th ] ; u. harst and m. reuter , jhep * 1205 * , 005 ( 2012 ) [ arxiv:1203.2158 [ hep - th ] ] ; r. percacci and d. perini , phys . d * 67 * , 081503 ( 2003 ) [ hep - th/0207033 ] . o. lauscher and m. reuter , jhep * 0510 * , 050 ( 2005 ) [ hep - th/0508202 ] . a. eichhorn and h. gies , new j. phys . * 13 * , 125012 ( 2011 ) [ arxiv:1104.5366 [ hep - th ] ] .

recently there was much interest in rectification in nanoscale systems . @xcite nonlinear mesoscopic transport exhibits much interesting physics such as a peculiar magnetic field dependence of the current @xcite and negative differential resistance for the rectification current at low voltages.@xcite another motivation for the investigation of mesoscopic diode or ratchet @xcite effect comes from possible practical applications in nanoelectronics and energy conversion . following the pioneering paper by christen and bttiker @xcite most attention has focused on a simpler case of a fermi - liquid conductor . at the same time , the diode effect requires a combination of spacial asymmetry and strong electron interactions in the conductor . hence , one may expect a stronger ratchet current in strongly interacting luttinger liquid systems . this expectation has been confirmed by a recent study of transport asymmetries in one - channel quantum wires.@xcite refs . have focused on a one - channel luttinger liquid in a linear conductor in the presence of a single asymmetric scatterer . this is the conceptually simplest situation giving rise to rectification . at the same time , changing geometry may increase asymmetry and hence the rectification current . in this paper we consider an asymmetric setup based on a y - junction of three quantum wires with three impurities . we show that a stronger diode effect can be achieved in such system than in a linear luttinger liquid and rectification is possible even in the case of symmetric point scatterers . we focus on the simplest one - channel y - junctions . more complicated luttinger liquid junctions , such as y - junctions of single - wall carbon nanotubes , are also of interest . in particular , it might be easier to make such junctions in a reproducible way . y - junctions are among basic elements of electric circuits , however , a theoretical investigation of luttinger liquid y - junctions@xcite has begun only during the last decade . by now , there is a good understanding of linear conductance near various fixed points as well as tunneling density of states.@xcite in this paper we extend the previous research to the problem of transport asymmetries . specifically , we consider a setup of the type shown in fig . we assume that one of the three terminals is kept at zero voltage . ac voltages with amplitudes @xmath2 and @xmath3 , @xmath4 , are applied to the remaining two terminals . in a general case , a dc - current is generated in each of the three wires in the junction , fig . 1 . the three currents are different but can be computed in a similar way . we calculate such rectification dc currents for various types of y - junctions . we focus on the limit of a low - frequency ac voltage bias . in order to determine the amplitude of the rectification effect in that limit , it is sufficient to find the difference of the dc currents at the opposite dc bias voltages , i.e. , compare the current when the potentials at the terminals are time - independent and equal 0 , @xmath2 and @xmath3 with the current when the potentials are 0 , @xmath5 and @xmath6 ( cf . refs . ) . this corresponds to a dc current generated by low - frequency square voltage waves . we find that in some classes of y - junctions the rectification current is higher than in a linear wire with the same strength of the repulsive electron interaction at the same voltage bias . in a certain interval of low voltages the rectification current exhibits a power - law dependence on the bias : @xmath7 . the exponent @xmath8 can be negative . clearly , such dependence with a negative exponent can not extend all the way to zero voltage as @xmath9 at @xmath10 . hence , the rectification current reaches a maximum at a certain voltage . we demonstrate that in junctions without time - reversal symmetry , the maximal rectification current can be comparable with the total current @xmath0 even for low voltages and modest interaction strength in the wires . in particular , such rectification current can be achieved in the `` island setup '' , illustrated in fig . 1 , which can be experimentally realized in quantum hall systems @xcite . for comparison , in a linear wire , the rectification current @xcite is always much lower than @xmath11 . the diode effect in y - junctions is not as strong in the presence of time - reversal symmetry as in its absence . still , the maximal rectification current is higher than in a linear conductor . the paper is organized as follows : first , after a brief qualitative discussion , we describe our setup and formulate a model . we then review the properties of y - junctions and derive a general expression for the rectification current . next , we apply the general formalism to the model of three weakly connected wires . then we determine the leading contributions to the rectification current at different interaction strengths and matching conditions at the junction . we discover several regimes with different voltage dependences of the current . finally , we discuss how to build a junction with the maximal rectification current . is applied to the first wire . wires 2 and 3 are connected to the ground . we calculate the dc current in wire 1 . one can view ab , cd and ef as edges of an integer quantum hall liquid . this corresponds to an `` island junction''.@xcite , width=192 ] we consider a junction of @xmath12 one - channel wires . in the subsequent sections we will specialize to the case of @xmath13 and assume that the system is spin - polarized or , equivalently , that charge carriers are spinless fermions . in all cases we assume that long range coulomb interaction in the conductor is screened by the gates . in such situation the applied voltage bias affects not only the current but also the charge density in the system . rectification is possible , if a change of the voltage bias sign results in a change of not only the sign but also absolute value of the current . the latter obviously requires left - right asymmetry in the system . asymmetry may have two origins : asymmetry caused by the way how the bias is applied and geometric asymmetry due to , e.g. , an asymmetric scatterer.@xcite in particular , one can consider a situation in which several terminals are kept at zero electric potential and the potential of the other terminals changes between @xmath14 and @xmath5 . the charge density is different for the opposite voltage signs . this in turn affects the current and leads to rectification . such density - driven rectification is possible in a broad range of situations including systems without electron interaction in the presence of time - reversal symmetry . it has been investigated in the context of a carbon nanotube junction in the noninteracting - electrons approximation in ref . . at a high voltage bias a strong rectification effect was found . in this paper , on the other hand , we focus on the low - voltage regime and , in particular , the universal behavior at low bias . in that limit the density - driven rectification effect is of little importance as it results in a small rectification current @xmath15 in a noninteracting system . besides , the density - driven rectification requires a specific way to apply bias . if we are interested , for example , in the transformation of incoming electromagnetic radiation into a dc current then clearly the potential oscillates in all terminals . thus , in this paper we focus on the rectification mechanism due to geometric asymmetry . y - junctions have a `` built - in '' geometric asymmetry . indeed , let us focus on the current @xmath16 in one of the three wires connected by the junction . the current @xmath17 equals the sum of the currents in the other two wires . thus , the same current enters the junction through two wires and leaves through one only . this means that the two sides of the junction are not equivalent since they correspond to 1 and 2 wires . thus , we may expect that geometric mechanism of rectification applies to any y - junction . however , as we demonstrate below , the geometric mechanism only works , if electron interaction is present or time - reversal symmetry is broken . indeed , let us consider a time - reversal invariant system of non - interacting spinless fermions in an @xmath12-terminal mesoscopic junction . we calculate the current between terminals @xmath18 and @xmath19 . we compare the current in the situation when terminals @xmath18 are kept at the voltage @xmath2 and terminals @xmath20 are kept at zero bias ( case 1 ) with the current in the situation when terminals @xmath21 are kept at zero bias and terminals @xmath19 are kept at the voltage @xmath2 ( case 2 ) . for non - interacting fermions the current reduces to the sum of the single particle contributions corresponding to each energy in the window @xmath22 . we thus compare such contributions for two opposite voltage biases . the wave function of an electron incoming from terminal @xmath23 with the momentum @xmath24 is @xmath25 , where @xmath26 is the coordinate in wire number @xmath27 and @xmath28 is a unitary scattering matrix . the above notations for the wave function imply that the probability to find an electron with the wave function @xmath29 in wire number @xmath30 is determined by the outgoing wave and is proportional to @xmath31 . for @xmath32 , the probability is determined by both the incoming and outgoing waves . the time - reversal symmetry implies that @xmath33 is also a solution of the schrdinger equation . the solution is made of @xmath12 incoming waves and one outgoing wave . it can be represented as a linear combination of waves @xmath34 with different @xmath27 . hence , the outgoing wave satisfies the equation @xmath35 . the absence of outgoing waves in the channels with numbers @xmath36 implies that @xmath37 . hence , @xmath38 , i.e. , @xmath39 is a symmetric matrix . now we can compare the contributions to the current from particles with the energy @xmath40 for two opposite signs of the bias . electrons , incoming from different terminals , are not coherent . hence , in case 1 , @xmath41 . in case 2 , @xmath42 . the currents are equal from the symmetry of the scattering matrix and hence the diode effect is absent . in the presence of electron interaction , the rectification effect becomes possible . this can be understood already from a simple model with two point scatterers of unequal strength in a linear wire ( fig . similar rectification mechanisms operate in more complex junctions of luttinger liquids . we will assume that long - range coulomb interactions are screened by the gates and thus the charge density depends on the voltage . in fig . 2a we consider the situation with the incoming current from the left . backscattering off two impurities results in a `` staircase '' charge density profile ( we have averaged over friedel oscillations ) . the `` staircase '' goes up as one moves from the right to the left . for the opposite voltage sign , the incoming current arrives from the right , fig . we again have a `` staircase '' charge density profile but now the `` staircase '' goes down as one moves from the right to the left . in the absence of electron interactions , the transmission coefficients must be the same in both cases and rectification is absent . let us now consider the effect of electron interactions in a simple mean - field hartree picture . since long - range coulomb interactions are screened , the relation between the charge density @xmath43 and the electric potential @xmath44 is local . assuming small charge density variations , we thus find @xmath45 . incoming electrons are scattered by a combined potential of the two impurities and the hartree potential @xmath44 . the latter depends on the incoming charge density and hence the applied voltage . hence , for opposite voltage signs , electrons feel different backscattering potentials . this results in different transmission coefficients and thus rectification . another mechanism of transport asymmetries comes from time - reversal symmetry breaking . the latter is possible in the presence of a magnetic field . we illustrate asymmetric transport in the absence of the time - reversal symmetry with a model system depicted in fig . while it is not a realistic model of any mesoscopic conductor , it is very simple and exhibits strong transport asymmetries . we assume no electron interactions in the model . we consider a box with four holes in its sides . each hole corresponds to a terminal in a more realistic description of a junction . we assume that all charge carriers , entering the box through the holes , have exactly the same speed , perpendicular to the wall in which the hole is made . the speed @xmath46 , where @xmath2 plays a role , similar to the electrostatic potential of a terminal in a realistic junction . at @xmath10 the speed @xmath47 and hence no current is injected . in a y - junction , charge density , injected from each terminal , depends on its electrostatic potential . a comparison between opposite voltage signs in a realistic junction corresponds in our model to the comparison between the situations with the charge injected through the right and left holes ( figs . 3a , c ) and through the upper and lower holes ( figs . 3b , d ) . in the absence of a magnetic field , the current through each hole is zero in both situations : currents of the particles , injected through the opposite holes , cancel as shown in figs . let us now turn on a magnetic field in shaded areas ( figs . if the charge carriers are injected from the upper and lower holes then the current remains zero even in the presence of a magnetic field ( fig . 3d ) . indeed , the carriers never enter the region with the field . however , if the carriers arrive from the left and right holes then they are deflected by the field ( fig . 3c ) and the electric currents are nonzero in all four holes . this illustrates how magnetic field can result in transport asymmetry . the above arguments make it plausible that the rectification effect is the strongest if both strong electron interaction is present and time - reversal symmetry is broken . this is confirmed by our calculations below for y - junctions of quantum wires with and without time - reversal symmetry . in this section we formulate our model and review basic properties of y - junctions . we consider a y - junction with the action @xmath48,\ ] ] where @xmath49 are the actions of three uniform wires and @xmath50 describe three impurities in the wires close to the junction point @xmath51 . the above action must be supplied by matching conditions for three wires . they will be discussed below . the action of a uniform wire is given by the equation @xmath52,\end{aligned}\ ] ] where @xmath53 are the operators of incoming and outgoing chiral electron fields , @xmath54 is the effective fermi velocity , @xmath55 defines the interaction strength . we set @xmath56 and the electron charge @xmath57 in most of the following text . we assume that the long range part of the coulomb force is screened by the gates . hence , the electron density depends on the voltage bias . the impurity hamiltonians @xmath58,\ ] ] where @xmath59 is the potential of the impurity in wire number @xmath24 , the impurity being located close to the junction at @xmath51 . we only take the backscattering part of the impurity hamiltonian into account as the forward scattering terms do not affect our results . we assume the same interaction strength and fermi velocity in each wire . in order to treat the case of strong interaction it is convenient to bosonize @xcite the action in terms of the chiral fields @xmath60 such that @xmath61 , where the opposite signs should be chosen for the in- and out - fields , the commutator @xmath62=\mp2\pi i\delta(y - x)\delta_{kl}\ ] ] and @xmath63 plays the role of the effective fermi momentum and determines the average charge density in the wires . @xmath64 are klein factors , necessary to ensure the proper fermi commutation relations . the local densities of in- and out - moving particles in point @xmath65 are @xmath66 . the action now assumes the form @xmath67;\end{aligned}\ ] ] @xmath68)+h.c.,\ ] ] where @xmath69 , with real @xmath70 and @xmath71 , are of the order of the fourier components of the asymmetric potential , @xmath72 . note that @xmath71 can be nonzero even for a symmetric potential @xmath73 in contrast to the situation considered in refs . . for example , for @xmath74 , @xmath75 . the above expression for the backscattering operators @xmath50 includes multi - particle backscattering processes.@xcite such multi - particle contributions to the action are inevitably generated under the action of the renormalization group by the interplay of a short - range coulomb interaction and impurity potential ( see ref . for a discussion ) . all prefactors @xmath70 have the same order of magnitude and are proportional to @xmath73 . we assume that backscattering amplitudes @xmath70 have the same order of magnitude in all three wires . the use of the fields @xmath76 at @xmath51 in the backscattering operators is justified , if the distance from the impurities to the junction is lower than the scale @xmath77 set by the voltage bias . we assume that @xmath70 are sufficiently small so that a perturbative expansion in powers of @xmath70 can be developed for the calculation of the current . the conditions on @xmath70 will be formulated below . following the standard notation conventions , the hamiltonian , corresponding to eq . ( [ bos ] ) , can be written as @xmath78,\ ] ] where the dimensionless interaction strength @xmath79 and @xmath80 . the above action alone is not enough to describe the system and matching conditions at the junction are necessary . as we will see , matching conditions are subject to several restrictions . the most general matching condition has the form @xmath81 , where @xmath82 is an arbitrary function . below we will focus on the situation in which the system is close to a fixed point . then only most relevant operators should be kept in the matching conditions and hence all derivatives of the bose - fields @xmath83 can be neglected . next , we note that the shift of any field @xmath84 by a constant @xmath85 is a gauge transformation that does not change the physics of the system . hence , the matching conditions must be invariant with respect to such gauge transformations . this singles out the boundary conditions of the form @xcite @xmath86 where @xmath87 is a matrix with real matrix elements . with this matching condition , it is easy to solve the equations of motion for the chiral fields @xmath88 in the absence of the interaction ( @xmath89 ) and impurities , for arbitrary initial conditions for fields @xmath84 . substituting the solution into the commutation relations ( [ comm ] ) for in- and outgoing fields , one finds that they are compatible , provided that the matrix @xmath87 is orthogonal . the same condition for the matching matrix must hold for a general case with an arbitrary interaction strength @xcite since the problem can be reduced to a model of non - interacting bosons by diagonalizing eq . ( [ ham ] ) in each wire ( see the next section ) . thus , we established that @xmath87 is a real orthogonal matrix . we next demonstrate that the sum of the elements of each of its rows and columns equals one.@xcite it is again convenient to consider the situation without electron interaction and tunneling . in that case the currents of in- and outgoing electrons in each wire are @xmath90 . charge conservation implies that the sum of the currents in all wires is zero at @xmath51 . taking into account eq . ( [ cond ] ) one finds that @xmath91 . multiplying eq . ( [ cond ] ) by @xmath92 and repeating the same argument one also finds that @xmath93 . again , electron interactions do not affect the above result @xcite . the origin of the matching conditions can be understood if one notes that they can be imposed by adding to the hamiltonian a term of the form @xmath94),\ ] ] where @xmath95 is a large constant . each cosine tends to keep its argument at zero . this can be achieved simultaneously for each cosine only if their arguments commute . this happens for a unitary @xmath87 . the cosines describe tunneling between different wires . thus , charge conservation implies @xmath93 . this provides an alternative derivation of the matching conditions . since we plan to separately investigate time - reversal invariant and non - invariant systems , we next need to determine what matching conditions satisfy the time - reversal symmetry this is easy as the time - reversal transformation corresponds to the change of variables @xmath96 . from eq . ( [ cond ] ) we then see that in time - reversal - invariant systems @xmath97 . taking into account that @xmath87 is orthogonal we conclude that in time - reversal invariant systems it is also symmetric.@xcite the action ( [ l ] ) is always time - reversal invariant . thus , the behavior of the whole junction with respect to time reversal is determined solely by the symmetry of the matching matrix @xmath87 . at this point we are in the position to give a full classification of fixed - point matching matrices.@xcite for time - reversal invariant systems , it is convenient to use a parametrization from ref . . we discover three possibilities with time reversal symmetry : @xmath98 @xmath99 @xmath100 where @xmath101 , @xmath102 and @xmath103 . eq . ( [ identity ] ) corresponds to three disconnected wires . the cases of @xmath104 correspond to a junction of two wires and a detached third wire . the physics is the same as for one linear wire and will not be discussed below . in the absence of time - reversal symmetry we use the parametrization from ref . : @xmath105 where @xmath106 , @xmath107 and @xmath108 . the case of @xmath109 , @xmath110 corresponds to the island setup , depicted in fig . 1 . note that the matching matrix ( [ 122 ] ) has the form ( [ no - time ] ) with @xmath111 so we will consider it with the case of no time - reversal symmetry . in general , matching matrices contain negative matrix elements . this means that the incoming current in one wire may suppress outgoing current in that or other wires . such situation is possible in the presence of andreev scattering , if , for example , a part of the junction is superconducting . negative matrix elements were also predicted in y - shaped beam splitters for cold bosonic atoms @xcite . note that for noninteracting electrons no negative matrix elements are allowed @xcite since in the absence of interaction the elements of the matrix @xmath87 reduce to the squares of the absolute values of the elements of the scattering matrix ( cf . in order to calculate the current we need to include a voltage bias . we model fermi - liquid leads by assuming that electron interaction in the wires is zero at large distances @xmath65 from the junction.@xcite we will assume that leads 2 and 3 are kept at a zero voltage . the bias voltage @xmath112 is applied to lead 1 . the results do not change in a more general model with the potential @xmath112 applied to lead 1 , the potential @xmath113 , @xmath4 , applied to lead 2 , and a zero potential applied to lead 3 . we will use the language of zero @xmath114 since it is simpler . our approach can be easily extended to a general @xmath114 . the calculations will be based on a renormalization group approach with the voltage @xmath2 playing the role of the infrared cutoff . thus , we will assume that the temperature @xmath115 . we would like to emphasize that the role of the voltage does not reduce solely to that of a cut - off ; otherwise the density - driven rectification would be lost ( see a discussion after eq . ( [ int - repr ] ) ) . it is well known that in luttinger liquids a voltage bias can play a more prominent role than just a cut - off ( see , e.g. , refs . ) . at the same time , the leading contribution to the total current at zero temperature can be estimated by simply setting the renormalization group cutoff to the value of the voltage @xcite . below we find that the rectification current is comparable with the total current in certain regimes in the absence of the time - reversal symmetry . this certainly means that in those regimes the rectification current can be computed by assuming that the voltage plays the role of a cutoff only ( we however do not make such an assumption ) . for a general @xmath114 , we do not expect the current to depend significantly on the temperature at @xmath115 . such dependence can emerge at particular values of @xmath114 . in particular , in the island setup ( fig . 1 ) , this happens at @xmath116 . this can be seen from the keldysh perturbation theory @xcite . the infrared cutoff in some integrals in the perturbation expansion is set by @xmath117 and in some others by @xmath118v)$ ] . hence , at @xmath116 our approach applies only for @xmath119 in the island setup . we want to find the current in the first wire . at low frequencies the current conserves and hence it is sufficient to find the current in lead 1 . it is given by the sum of the chiral incoming and outgoing currents @xmath120 . the incoming current can be found from the landauer formula and is linear in voltage . it does not contribute to the rectification current . we thus focus on the outgoing current in lead 1 . it can be found with a generalization of the approach of ref . . our approach is also related to that of ref . . in what follows we set the temperature @xmath121 to simplify notations . our method can be easily generalized to finite temperatures . let us introduce an auxiliary field @xmath122 : @xmath123 the field @xmath124 satisfies a simple matching condition @xmath125 . thus , the auxiliary field is a chiral field propagating through the junction . in the stationary regime the average time - derivative of any operator is zero . let us now consider the operator @xmath126 , where the integration extends between points taken in the noninteracting leads . its meaning is the charge carried by the mode @xmath124 . the equation @xmath127\rangle$ ] , where the hamiltonian @xmath128 describes the whole system including the leads , wires and scatterers , reduces to the following relation : @xmath129\rangle.\ ] ] the left hand side is the outgoing current we want to find . the first term in the right hand side is linear in the voltage bias and can not contribute to the rectification current . thus , we have to calculate only the second contribution to the right hand side . in other words , our problem reduces to the calculation of the average backscattering current whose operator equals @xmath130 , where @xmath131 \nonumber\\ = & i[u_l(f^{k\dagger}_of^k_i)^n\exp(i\alpha_l+in\{\sum_p ( m^{-1})_{kp}\phi^p_o(0 ) -\phi^k_o(0)\ } ) \nonumber \\ & + h.c.,\hat o ] \nonumber \\ = & i n u_l ( f^{k\dagger}_of^k_i)^n \exp(i\alpha_l+in[\sum_p ( m^{-1})_{kp}\phi^p_o(0)-\phi^k_o(0 ) ] ) \nonumber\\ & \times [ ( m^{-1})_{k1}-\delta_{k1}]+ h.c . \ , .\end{aligned}\ ] ] note that for any @xmath23 the current operator expresses via _ all _ backscattering operators in all three wires . in order to find the average current @xmath132 , we will assume that the backscattering operators @xmath50 are absent at @xmath133 and are then gradually turned on . without the operators @xmath50 the system can be viewed as an equilibrium one in the ground state of an appropriate effective hamiltonian . in order to find it we introduce another auxiliary field with the structure , similar to @xmath124 : @xmath134 in the absence of the backscattering operators @xmath50 , the operator @xmath135 commutes with the hamiltonian . it can be understood as the charge of a chiral mode propagating through the junction . in other words , it is an additive integral of motion . hence , the system can be described by a gibbs distribution with an appropriate thermodynamic potential conjugated to @xmath136 . the physical meaning of that thermodynamic potential is the applied voltage bias @xmath2 . at zero temperature , one finds that the system is in the ground state of the effective hamiltonian @xmath137 , where @xmath138 is the actual hamiltonian of the y - junction . the current , i.e. , the average of the sum of the operators @xmath139 $ ] , can be now calculated with the keldysh technique @xcite . it is convenient to apply the interaction representation @xmath140 . the interaction representation introduces time - dependence into the operators @xmath50 and @xmath141 according to the rule : @xmath142)\rightarrow \nonumber\\ & \exp(iv\hat at)\exp(in[\phi^k_i(0)-\sum_p m_{kp}\phi^p_i(0)])\exp(-iv\hat at ) \nonumber\\ & = \exp(in[\sum_p ( m^{-1})_{kp}\phi^p_o(0)-\phi^k_o(0)+vt(m_{k1}-\delta_{k1})]).\end{aligned}\ ] ] after such time - dependence is added into all backscattering operators , the contribution of the form @xmath143 can be removed from the action by a linear shift of all fields @xmath144 . this does not mean , however , that the voltage would only enter the action through the time - dependence of @xmath50 . indeed , since the charge density depends on the voltage , the amplitudes @xmath145 of the operators @xmath50 may get corrections proportional to the small voltage bias . this effect is discussed in ref . . the current can now be found with a perturbative expansion in powers of @xmath145 from the standard expression @xmath146 where @xmath147 is the evolution operator from time @xmath148 to time @xmath149 and @xmath150 is the ground state of the effective hamiltonian @xmath151 . to complete the calculation , we need the green functions determined by the quadratic part of the action ( [ lk ] ) and the matching conditions ( [ cond ] ) . they have been found in ref . . a bogoliubov transformation of the form @xmath152 $ ] in each wire allows one to obtain free chiral fields @xmath153 with correct luttinger liquid commutation relations.@xcite the correlation function of the incoming fields at @xmath51 is given by the free particle relation @xmath154/\tau_c+\delta),\ ] ] where @xmath155 is the ultraviolet cutoff time of the order of the inverse bandwidth , @xmath156 is infinitesimal . the matching conditions for the new fields have the form @xcite @xmath157 @xmath158^{-1}[(1+g)m-(1-g)].\ ] ] in the time - reversal invariant case @xmath159 . in the absence of the time - reversal symmetry , @xmath160 has the same general structure ( [ no - time ] ) as @xmath87 and satisfies the same set of constraints but the matrix elements are different : @xmath161 @xmath162 at this point one can write an expression for the current as an expansion in powers of backscattering amplitudes @xmath163 . evaluation of the terms of that expansion is technically difficult and not very informative . indeed , neither the amplitudes nor the ultra - violet cutoff are known exactly . as a result , it is only possible to estimate the order of magnitude of each contribution . at the same time , such estimation can be performed without explicitly calculating integrals over the keldysh contour and will be sufficient to find the leading power dependence of the rectification current on the bias voltage at small @xmath163 and low voltages near each of the fixed points . such estimation will be the focus of the remaining sections . it is convenient to use a renormalization group point of view for the calculation of the current.@xcite in the renormalization group procedure , the coefficient @xmath163 in the hamiltonian and the operators @xmath141 scale as @xmath164 , where @xmath165 is the energy scale and @xmath166 the scaling dimension . at the energy scale @xmath167 the renormalization group procedure stops . different perturbative contributions to the current can be estimated from the scattering theory.@xcite the current can be expressed as an infinite sum of the contributions , proportional to the products of different combinations of @xmath163 : @xmath168 strictly speaking , it is not enough to include only backscattering operators from the bare tunneling hamiltonian ( [ tun ] ) in the expansion ( [ current - scale ] ) . all operators generated by the renormalization group must also be included . their general structure is @xmath169 ) \nonumber\\ = & w_l\exp(i\alpha_l+i\sqrt{g}\sum_k\sum_p n_k[\tilde m_{kp}-\delta_{kp}]\tilde\phi^p_i).\end{aligned}\ ] ] for simplicity we use the same notation @xmath170 for the phases of @xmath171 and @xmath145 . note that it is always sufficient to keep only two different values of @xmath24 in the sum in eq . ( [ gen ] ) , since @xmath172=0 $ ] . this means that each possible operator @xmath173 can be generated from a product of two operators @xmath50 ( [ tun ] ) and hence @xmath174 . since klein factors do not change the scaling dimensions of @xmath171 , we omit them in eq . ( [ gen ] ) . to complete the calculation of the current we need to find @xmath166 . a straightforward calculation based on eq . ( [ correl ] ) yields for the operator ( [ gen ] ) with an arbitrary choice of @xmath175 : @xmath176 in the absence of the time - reversal symmetry the above expression greatly simplifies : @xmath177.\ ] ] in the low - voltage regime , the main contribution comes from the operators with the smallest @xmath166 . one easily sees that it equals @xmath178 and is achieved , if two of the coefficients @xmath179 with the remaining one being @xmath180 . in the time - reversal invariant system , the expression for the scaling dimensions is more complicated . it can be simplified by setting @xmath181 and @xmath182 . using the matching matrix ( [ time - reverse ] ) , one finds : @xmath183 ^ 2}{b+c}=\frac{g}{1+\alpha+\alpha^2}(k\alpha - m)^2.\ ] ] we will assume for simplicity that @xmath184 is irrational . we will thus avoid situations with @xmath185 at some @xmath24 and @xmath27 . certainly , such situations correspond to the operators @xmath171 with no @xmath76-dependence . such operators can not affect transport . the renormalization group procedure only applies if all @xmath163 are small at the initial energy scale @xmath186 and remain small up to the scale @xmath2 . in the case without time - reversal symmetry , this condition can be easily expressed in terms of @xmath187 : @xmath188 the above equation assumes that @xmath189 , i.e. , @xmath163 is relevant . if @xmath190 then the only restriction on @xmath163 is @xmath191 . what happens in the presence of the time - reversal symmetry will be addressed in subsequent sections . the above discussion applies to all contributions to the current , both even and odd in the voltage bias . we are interested in the even contribution , i.e. , the rectification current @xmath192/2 $ ] . what terms of the perturbation expansion contribute to the rectification current ? the answer to this question can be obtained from symmetry considerations . in what follows we will need the commutation relations for the klein factors in ( [ tun ] ) . using the commutation relations @xmath193 and @xmath194=-i\pi { \rm sign}(x - y ) $ ] , an expression for @xmath195 in terms of @xmath196 and the baker - hausdorff formula , one finds @xmath197)f_i^k f_o^{k\dagger}.\ ] ] next , some terms of the perturbative expansion ( [ current - scale ] ) are zero identically as only certain combinations of the vertex operators in @xmath50 and @xmath141 produce non - zero results after averaging with respect to the quadratic part of the action ( [ bos ] ) . the condition is well known : @xmath198 let us now apply the above results to possible contributions of different orders of the perturbation theory to the rectification current . strong limitations emerge for second order contributions . indeed , eq . ( [ nonzero ] ) implies that any second order contribution , proportional to an operator @xmath163 , must also contain its hermitian conjugated operator @xmath199 . hence , the phase factors @xmath200 cancel each other and drop out from the expression for the current . thus , we can just set @xmath201 . let us now compare the currents at the bias voltages @xmath2 and @xmath5 . in the case with the bias @xmath5 , we make the change of variables @xmath202 . this transformation does not affect the form of the quadratic part of the action ( [ bos ] ) and the linear matching conditions . let us also change the order of the klein factors in the tunneling and current operators using the commutation relation ( [ klein - comm ] ) and redefine @xmath203 . since @xmath163 pairs up with @xmath204 in the nonzero second - order contribution to the current , one can easily see that the phase factors @xmath205)$ ] in the commutation relation ( [ klein - comm ] ) drop out . at the same time , if we omit the phase factors @xmath200 and @xmath205)$ ] from the action and current operators , we discover that the action assumes precisely the same form as for the voltage bias @xmath2 . on the other hand , the current operator @xmath141 ( [ backscattering ] ) changes its sign . this means that the average of @xmath141 also changes its sign and hence the second order term of the order @xmath206 does not contribute to the rectification current . the above argument assumes that the backscattering amplitudes @xmath163 are independent of the voltage bias . since the injected charge density depends on the voltage in our setup , @xmath163 can exhibit a weak linear dependence on the bias voltage . this means that second order terms can in fact lead to a `` density - driven '' rectification effect but its amplitude is suppressed by an additional factor @xmath207 , where the ultra - violet cutoff @xmath155 is of the order of the inverse band width . thus , the second order contribution to the ratchet current scales as @xmath208 beyond the second order , no additional general restrictions on possible contributions to the rectification current can be derived . particular contributions to the ratchet current can disappear at particular values of the phases @xmath170 and interaction strength @xmath209 . is applied to wire 1 , and bias @xmath210 to wire 2 . wire 3 is connected to the ground . we calculate the dc current in wire 1.,width=192 ] so far we discussed general features of y - junctions . we now determine the rectification current in different setups . as a warm - up exercise we investigate the simplest situation of three almost disconnected wires , fig . this situation corresponds to the matching matrix ( [ identity ] ) . the rectification current is much weaker in this limit than for other matching conditions . at the same time , the qualitative picture is quite similar . there is no current at all in our original model with the action ( [ l ] ) and backscattering operators ( [ tk ] ) as it describes three disconnected conductors . we thus modify the model in this section . instead of the backscattering operators ( [ tk ] ) we consider weak tunneling between the wires . the tunneling operators have the form @xmath211 where @xmath212 , @xmath213 are klein factors and we use the convention @xmath214 . the commutation relations for the klein factors are @xmath215 . we assume that wires 3 is kept at zero voltage . the chemical potential of the first wire is @xmath112 . the second wire experiences the bias @xmath113 . note a different meaning of the amplitudes @xmath145 and phases @xmath216 from the model with three weak scatterers discussed above . in eq . ( [ 3-tun ] ) , the phases @xmath216 describe the aharonov - bohm effect due to the magnetic field through the junction . on the other hand , in the model with weak scatterers , all information about the magnetic flux is contained in the matching conditions . the voltage bias can be included in the action in the form of the time - dependence of the tunneling operators , similar to the discussion above . we will not need explicit expressions below . the operator of the current , tunneling between wire 1 and wires 2 and 3 , is @xmath217 the current can be estimated from a renormalization group procedure that stops at the scale @xmath167 . at that scale @xmath218 , where @xmath219 characterizes the interaction strength . the second order contribution to the rectification current can now be easily found and scales as @xmath220 where @xmath221 . the @xmath222-th order contribution to the current can not exceed @xmath223 , @xmath224 . thus , the leading higher - order contribution corresponds to @xmath225 , @xmath226 interestingly , at low voltages and @xmath227 , the third order contribution exceeds the second order contribution . the above discussion ignored the issue of the time - reversal symmetry . depending on the presence or absence of the magnetic field through the junction , the system can have or have no time - reversal symmetry.@xcite it is instructive to investigate the effect of the symmetry breaking . in the presence of the time - reversal symmetry the magnetic flux is zero and all @xmath228 . indeed , the time - reversal transformation can be represented as @xmath229 . since @xmath230 , we conclude that @xmath231 at zero magnetic field . on the other hand , in the presence of the magnetic field @xmath232 . the second order contribution to the rectification current does not depend on the phases @xmath216 . let us compare the third order contributions for different values of @xmath216 . nonzero third order contributions to the current originate only from the product of all three operators @xmath50 or all three operators @xmath233 in the perturbative expansion of the keldysh expression for the current ( [ keldysh ] ) . the first contribution , @xmath234 , is proportional to @xmath235 , @xmath236 , and the second contribution is proportional to @xmath237 , @xmath238 . let us change the voltage sign . simultaneously , we change @xmath239 , change the order of klein factors in all operators and redefine @xmath240 . this transformation is equivalent to changing the sign of @xmath216 in all operators and simultaneously changing the overall sign of the tunneling hamiltonian ( [ 3-tun ] ) but not the sign of the current operator . hence , the third order current @xmath241 . we see that the current is an even function of the voltage in the presence of the time - reversal symmetry , i.e. , at @xmath242 . the third order current is an odd function of the voltage and does not contribute to the rectification effect , if @xmath243 . thus , for three weakly connected wires , magnetic field suppresses rectification . we will see below that typically it has an opposite effect and rectification is stronger in the absence of the time - reversal symmetry . note that in the above example , the ratchet current returns to its maximal value at @xmath244 . let us finally compare our results with the case of a linear wire with a high asymmetric potential barrier or , equivalently , two weakly connected wires . in that case the rectification current is given @xcite by eq . ( [ 3-i-2 ] ) . thus , at @xmath227 the rectification effect is stronger in a y - junction than in a linear wire with the same interaction strength . here we address matching matrices of the form ( [ no - time ] ) . in particular , the results of this section are relevant for the island setup , fig . our analysis also applies to the time - reversal - invariant situation with the matching matrix ( [ 122 ] ) . below we find that at sufficiently low voltages the rectification current scales as eq . ( [ 3-no - time ] ) with @xmath245 being the impurity potential strength , @xmath2 the voltage , @xmath219 the luttinger liquid parameter , characterizing electrostatic interactions , @xmath155 the ultraviolet cutoff of the order of the inverse band width , and @xmath246 is determined by the matching conditions according to eq . ( [ tildea ] ) . similar to the previous section we need to compare the leading second and third order contributions to the rectification current . they are determined by the operators of the form @xmath247)$ ] with the minimal scaling dimension ( [ z - min - no - time ] ) . the leading second order contribution scales as @xmath248 , where @xmath246 is given by eq . ( [ tildea ] ) and satisfies @xmath249 . the leading third order contribution comes from the terms in the perturbation expansion proportional to the product of all three @xmath50 ( or all three @xmath233 ) . it scales as @xmath250 at small @xmath245 the second order contribution always exceeds higher - order contributions . interestingly , however , at greater @xmath245 which are still within the region of validity of the perturbation expansion , the third order contribution is leading . let us show that the third order contribution dominates . we will focus on the largest @xmath163 accessible with the perturbation expansion : @xmath251 at @xmath252 when @xmath163 is irrelevant ; and @xmath253 at @xmath254 when @xmath163 is relevant . in all cases the electron interaction is repulsive , i.e. , @xmath219 . let us first consider the case of @xmath252 . according to eq . ( [ tildea ] ) , @xmath255 . . the ratio of the second order contribution to the third order contribution scales as @xmath257 . thus , the third order contribution dominates indeed . the case of @xmath254 is also easy . in the limit of @xmath253 we find @xmath258 and @xmath259 . interestingly , the third order current may become comparable to the total ac current @xmath0 through the junction at @xmath254 . note also that the exponent @xmath260 in the voltage dependence of the dc current ( [ 3-no - time ] ) is negative at @xmath261 . this corresponds to the dc current _ increase _ as the ac voltage decreases . the above calculation applies for the voltage interval @xmath262 at @xmath252 and @xmath263}/\tau_c$ ] at @xmath254 . the left inequality is dictated by the applicability of the luttinger liquid model . the right inequality is determined by the validity of the perturbation theory . one can not calculate the rectification current at lower voltages with the above perturbative approach . however , it is obvious that @xmath9 at @xmath10 . from this we conclude that there is a bump on the voltage dependence of the rectification current at @xmath261 : it grows as a function of the voltage at low voltages and achieves its maximal value of the order of @xmath11 at @xmath264}$ ] . in the last equation we omitted a power of @xmath155 as this does not lead to a confusion . finally , let us compare the result with the case of a linear wire with the same interaction strength @xmath209 in the presence of an asymmetric impurity . in that case the maximal rectification current , calculated in the perturbative regime in ref . , scales as @xmath265 at @xmath266 ( we omit powers of @xmath155 ) . the rectification current does not exceed @xmath267 at @xmath268 . in a y - junction , at @xmath269 the rectification current @xmath270 at @xmath271 is always greater than in a linear wire . on the other hand , a negative @xmath272 implies @xmath273 . the maximal rectification current corresponds then to @xmath274 and scales as @xmath275 . thus , the maximal ratio of the rectification and total currents is always higher in a y - junction than in a linear wire in the luttinger liquid regime @xmath276 . this case is more complicated than the situation without the symmetry . the summary of the results is given in section ix.c . we will need to take into account many different backscattering operators . thus , it is important to classify them . the classification relies on the matching conditions at the junction . the matching matrix is given by equation ( [ time - reverse ] ) where the expressions for @xmath277 in the parametrization @xcite are written under the equation . it is always possible to redefine @xmath184 in such a way that @xmath278 , where @xmath279 is an arbitrary transposition of @xmath280 . indeed , the change of the variable @xmath281 exchanges @xmath282 and @xmath283 ; @xmath284 exchanges @xmath282 and @xmath285 ; @xmath286 exchanges @xmath283 and @xmath285 . any other transposition is a superposition of the above three . since @xmath87 is an orthogonal matrix , at least one of the elements must be negative or zero . otherwise , different rows can not be orthogonal . without loss of generality we may assume that @xmath287 and hence @xmath288 . this can always be achieved by renumbering the wires and redefining @xmath184 . note that @xmath289 . similarly , without loss of generality we may assume that @xmath290 . hence , @xmath291 . note that @xmath292 . thus , the ranges of the parameters that we consider are : @xmath293 , @xmath294 , @xmath295 and @xmath296 . as has already been discussed , we need to deal with two types of backscattering operators in the action and the current operator : @xmath297 @xmath298)$ ] ; @xmath299 @xmath300+in_3[\phi^3_o(x=0)-\phi^3_i(x=0)])$ ] . note that we do not need to include a contribution of the form @xmath301 $ ] in the exponent in @xmath171 . we will first concentrate on the operators of the first type and determine the leading contribution to the ratchet current in the absence of the operators of the form @xmath171 . next , we will check what changes after @xmath171 are taken into account . we summarize our findings in sec . ix.c . the scaling dimensions of the operators @xmath163 are simply @xmath302 the lowest scaling dimension is @xmath303 with @xmath304 . it sets the scale of the maximal @xmath245 in the perturbation theory , @xmath305 . strictly speaking we need to know the scaling behavior of the operators @xmath171 to determine the maximal allowed value of @xmath245 in the perturbative regime . we will see in the next subsection that the operators @xmath171 do not change the above expression for the maximal @xmath245 . we thus immediately find the leading second - order contribution to the rectification current @xmath306 what about the third and higher orders ? it turns out that the leading contribution is third order and comes from the operators @xmath307)$ ] , @xmath308)$ ] and @xmath308)$ ] . it scales as @xmath309 - 2}\ ] ] to see that ( [ third - order - main ] ) is the leading contribution , we first note that @xmath310 , eq . ( [ third - order - main ] ) , exceeds @xmath311 at the maximal allowed @xmath305 . indeed , at such @xmath245 , @xmath312 while @xmath313}>(v\tau_c)^{1 + 3g\times 1/3}>(v\tau_c)^2 $ ] . next , let us compare the contribution ( [ third - order - main ] ) with other higher order contributions to the current , @xmath314^{z_l - g[1-a]}\right).\ ] ] if an operator with the scaling dimension @xmath315 $ ] enters the above expression then ( [ third ] ) is smaller than ( [ third - order - main ] ) . indeed , any contribution to the rectification current that contains such operators satisfies the inequality @xmath316}=(v\tau_c)^{1+g[1 + 2\alpha]/[1+\alpha+\alpha^2]}<(v\tau_c)^{1 + 3g[1-a]}$ ] . thus , only contributions with the operators @xmath317)$ ] and @xmath318)$ ] have to be considered . if one of the contributing operators has @xmath319 or @xmath320 then it is obvious that the contribution can not exceed ( [ third - order - main ] ) . we thus consider the case with all @xmath321 . a nonzero contribution to the current requires @xmath322+(\sum_l n_{3l})[\phi^3_o(x=0)-\phi^3_i(x=0)]=0 $ ] . for a general transcendental @xmath184 this equality is satisfied only if @xmath323 . thus , we consider contributions in which the operator @xmath324)$ ] enters the same number of times as the operator @xmath325 and the operator @xmath326)$ ] enters the same number of times as the operator @xmath327 . hence , the phases @xmath328 drop out from the final answer . let us now change the voltage sign , perform the transformation @xmath329 , change the order of the klein factors in each term and redefine @xmath330 . changing the order of the klein factors introduces complex conjugate phase factors into @xmath145 and @xmath331 . hence , we can ignore both those phase factors and @xmath216 since neither affects the final result . on the other hand , if we ignore the phase factors , we discover that the backscattering part of the hamiltonian does not change under our transformation while the current operators @xmath141 change their signs . this means , in turn , that the contribution to the total current we are calculating changes its sign when the bias voltage changes its sign . hence , it does not contribute to the _ rectification _ thus , eq . ( [ third - order - main ] ) describes the main contribution to the rectification current . we now find the leading contribution to the rectification current that contains operators @xmath171 . first of all , let us check that renormalized operators @xmath332 remain small at @xmath333}/\tau_c$ ] . indeed , @xmath334 and @xmath335 . hence , @xmath336}<(v\tau_c)^{1 - 2g/3}<(v\tau_c)^{1/3}\ll 1 $ ] . thus , all renormalized operators remain small and the maximal @xmath245 for which the perturbation theory can be used was found correctly in the previous subsection . in what follows we focus on the case with @xmath245 of the order of its maximal allowed value . the opposite limit @xmath337 is trivial since in that limit the second order contribution in @xmath245 always dominates the ratchet current . next , let us check how the operators @xmath171 affect second order contributions to the rectification current . the corresponding second order contribution @xmath338 ^ 2v^2/\tau_c^3\le ( u\tau_c)^4/\tau_c\sim ( v\tau_c)^{4 - 4g[1-a]}/\tau_c$ ] . we need to compare it with the current ( [ third - order - main ] ) , @xmath339}/\tau_c$ ] . one easily sees that @xmath340>1 + 3g[1-a]$ ] since @xmath341<1/3 $ ] . thus , the second order contribution to the rectification current can be neglected in comparison with the third order contribution even after @xmath171 are taken into account . at the same time , in a certain region of parameters the dominant higher - order contribution to the rectification current contains an operator @xmath171 . as we will see , such contribution contains exactly one operator @xmath171 . we thus begin our analysis by excluding contributions which contain three or more @xmath171 s . indeed , any such contribution @xmath342}/\tau_c$ ] . we need to compare this estimate with @xmath339}/\tau_c$ ] . one easily sees that @xmath343>[1 + 3g(1-a)]$ ] and hence we do not need to take into account contributions with three or more operators @xmath171 . thus , it remains to consider higher - order contributions with one or two operators @xmath171 . we first consider contributions with exactly one operator @xmath344)$ ] , where @xmath345 can be set to zero as discussed above . the contribution also contains at least two operators @xmath163 . it is sufficient to consider the case with all @xmath346 in the definition of the operators @xmath163 @xmath297 . indeed , a contribution with an operator with @xmath347 can be increased by substituting it with @xmath348 operators @xmath349)$ ] . moreover , we can assume that for each @xmath24 only operators @xmath350 or only operators @xmath351 enter . let us denote the number of the operators of the form @xmath352)$ ] in the expression for the contribution under consideration as @xmath353 . then we can estimate the contribution to the rectification current as @xmath354^{l_3}[uv^{g[1-b]-1}]^{l_1}[uv^{g(1-c)-1}]^{l_2},\ ] ] where @xmath355 ^ 2/(1+\alpha+\alpha^2)$ ] . the same argument as in the previous subsection shows that the maximal contribution corresponds to @xmath356 . then for a general @xmath184 , the product of the vertex operators @xmath357 gives a nonzero result after averaging with respect to the quadratic part of the action only if @xmath358 and @xmath359 . note that we can assume that both @xmath360 and @xmath361 are nonzero and @xmath362 . otherwise the operator @xmath171 would have the same from as one of the operators @xmath163 and the analysis from the previous subsection would apply . we want to compare the contributions ( [ third - order - main ] ) and ( [ est - one - w ] ) . at @xmath363}/\tau_c$ ] we find @xmath364 = \gamma(s_1,s_3)\log ( v\tau_c ) ; \nonumber \\ & \gamma = 1+\frac{g}{1+\alpha+\alpha^2}[-5\alpha^2+|s_1|(1-\alpha^2)+(s_1+s_3\alpha)^2].\end{aligned}\ ] ] if @xmath114 is positive for every choice of @xmath365 then @xmath310 is the main contribution to the rectification current . if @xmath366 for a certain choice of @xmath367 then the leading contribution comes from @xmath368 . we thus want to investigate at what conditions @xmath366 and find what choice of @xmath360 and @xmath345 minimizes @xmath114 . that choice determines the power dependence of the rectification current ( [ est - one - w ] ) on the voltage . first , let us prove that @xmath114 is minimal , if @xmath369 . indeed , let us compare @xmath370 with @xmath371 , where @xmath372 and @xmath373 is arbitrary . one finds @xmath374 \nonumber\\ & \ge\frac{g}{1+\alpha+\alpha^2}(1-\alpha)[(|p_1|-1)(1+\alpha)-(1-\alpha)]>0.\end{aligned}\ ] ] thus , we can focus on @xmath375 ( the case of @xmath376 is completely analogous ) . next , we prove that a negative @xmath377 is minimal at @xmath378 . indeed , a negative @xmath377 implies that @xmath379/[1+\alpha+\alpha^2]<0 $ ] . hence , @xmath380 . taking into account that @xmath381 , we find that @xmath382 is minimal at @xmath378 . this allows one to establish that @xmath114 is minimal at @xmath378 . the remaining task is simple . we just determine at what conditions @xmath383 is negative , i.e. , we need to investigate the inequality @xmath384}{1+\alpha+\alpha^2}<0.\ ] ] one easily sees that @xmath114 can only be negative , if @xmath385 , i.e. , for relatively weak repulsive electron interaction . @xmath114 is negative in the largest interval of @xmath184 at @xmath386 . in that case , @xmath366 for @xmath387 . the contributions with @xmath171 matter only if @xmath388 . they give rise to the rectification current of the form ( [ answer2 ] ) at negative @xmath114 . the last question we must address in this subsection concerns the role of the contributions to the rectification current with two operators of the form @xmath299 . an estimation of such contributions is similar to ( [ est - one - w ] ) : @xmath389^{l_3}[uv^{g[1-b]-1}]^{l_1},\ ] ] where @xmath390 ^ 2/(1+\alpha+\alpha^2)$ ] are the scaling dimensions of the operators @xmath391+is^{x , y}_3[\phi^3_o(x=0)-\phi^3_i(x=0)])$ ] . as above , we first divide @xmath392 by @xmath310 and ask when the ratio is greater than 1 : @xmath393=\mu(s^x_1,s^x_3,s_1^y , s_3^y)\log ( v\tau_c ) ; \nonumber \\ & \mu=2+\frac{g}{1+\alpha+\alpha^2 } ( -7\alpha^2+l_1(1-\alpha^2)+\sum_{r = x , y}(s_1^r+s_3^r\alpha)^2).\end{aligned}\ ] ] if @xmath394 then @xmath395 . obviously , a negative @xmath396 implies that @xmath397 and hence @xmath398 . let us now compare @xmath396 and @xmath383 at @xmath399 . one finds @xmath400 one easily checks that the above difference is positive at @xmath399 . hence , contributions with one or no operators @xmath171 are always more important than contributions with two such operators . we found that the leading contribution to the rectification current depends on the sign of @xmath383 , eq . ( [ inv - ineq ] ) . for strong repulsive interaction ( @xmath401 ) , @xmath383 is always positive . if @xmath402 , at sufficiently low voltages @xmath403}/\tau_c$ ] the current scales as @xmath404 at a negative @xmath114 and low voltages @xmath271 @xmath405 the exponents in both voltage dependences are negative . this is related to the fact that our calculations are only valid in an interval of low voltages , @xmath406}/\tau_c$ ] . similar to section viii the @xmath1-@xmath2 curve for the rectification current exhibits a bump at @xmath271 . the maximal rectification current at @xmath402 @xmath407 at @xmath366 @xmath408 one can easily verify that the maximal current ( [ answer3],[answer4 ] ) exceeds the maximal possible rectification current at the same voltage @xmath2 in a linear wire @xcite @xmath409^{3g},[v\tau_c])$ ] . we have found the rectification current in the absence ( [ 3-no - time ] ) and in the presence ( [ answer1],[answer2 ] ) of the time - reversal symmetry in y - junctions . in all cases the maximal rectification current is greater than in a linear wire with the same interaction strength and bias voltage . in the absence of the time - reversal symmetry the rectification current can be comparable with the total ac current @xmath0 for sufficiently strong interaction strength , i.e. , it achieves its maximal possible order of magnitude . this reflects the fact that both electron interaction and time - reversal symmetry breaking facilitate rectification . for most values of parameters the rectification current is a nonmonotonous function of the bias voltage . our calculations are valid in the vicinity of various fixed points in the low voltage regime . in a general case , a junction is controlled by a stable fixed point at low voltages . for repulsive interaction of spin - polarized particles there is only one stable fixed point : three disconnected wires.@xcite we found a stronger rectification effect near that fixed point than for two weakly connected wires . however , the current is low for weakly connected wires . the diode effect is much stronger in the vicinity of unstable fixed points . thus , it is important to understand how to tune the system close to those fixed points . some of them may be tricky to realize experimentally . indeed , as discussed above , negative elements in the matching matrix @xmath87 imply andreev reflection and could be obtained in a hybrid normal - superconductor structure or in cold atom systems @xcite . at the same time , it is straightforward to make an `` island junction '' @xcite , fig . 1 , with positive matching matrix elements @xmath410 for example , one can use three line junctions between three quantum hall systems . the role of impurities is played by three constrictions in the junctions . alternatively , one can use a single quantum hall island confined between edges ab , cd and ef ( fig . 1 ) . the rectification current scales as @xmath411 at @xmath412 . time - reversal symmetry is broken in such setup . thus , the rectification current can be made comparable to the total ac current even in the low voltage regime . this is the main result of the paper . most of the time we ignored phase factors @xmath413 in tunneling operators . as the example of three weakly connected wires shows , for special values of the phases the rectification effect is suppressed . it has the same order of magnitude for other values of the phases . we considered the simplest example of a junction : 3 spin - polarized wires . it would be interesting to generalize our results to a system with spin . this may result in a more complicated behavior as the minimal model with spin includes 6 channels : two for each wire . still , we expect a similar physics . in particular , a time - reversal - invariant system with spin can be obtained from two copies of the `` island junctions '' with opposite spins and chiralities . in the absence of the interaction between the copies , the problem reduces to a spin - polarized island junction . an actual realization , based , e.g. , on a topological insulator , must involve interaction between opposite spins . we expect that such interaction does not change the qualitative picture .

variational principles have a long and distinguished history in physics . apart from global formulations of physical principles , equivalent to local differential equations , they also are useful to approximate problems too difficult for analytic solutions . in the past century quantum variational calculations have been ubiquitous as an approximate method for the ground state of many difficult systems @xcite . in recent years there has been some progress in using classical variational principles ( action principles ) to approximate the motion of classical systems such as classical molecules . we review recent developments in this area of classical mechanics , although we also overlap with quantum mechanics . in particular we shall also discuss the use of the quantum variational principle for excited states , and the connection to classical action principles . in classical mechanics , variational principles are often called least action principles , because the quantity subject to variations is traditionally the action . to confuse matters there are two classical actions , corresponding to two main action principles , which are called respectively , hamilton s action @xmath0 and maupertuis action @xmath1 ; these differ from each other ( they are related by a legendre transformation ) , and the notation @xmath2 is not universal ( we follow current usage but in fact some authors use the same symbols reversed @xcite ) . both actions , @xmath3 and @xmath4 , have the same dimensions ( i.e. energy@xmath5time , or angular momentum ) . maupertuis least action principle is the older of the two ( 1744 ) by about a century , and as we shall discuss , it is inconveniently formulated in all textbooks @xcite . for clear statements of these old action principles , see the textbooks by arnold @xcite , goldstein et al @xcite , and sommerfeld @xcite . we shall review these action principles in the next section . some mathematics and notation required to deal with variational problems are summarized in appendix i. the distinguishing feature of variational problems is their character @xcite . one is searching for the which gives a minimum ( or stationary value ) to an integral , as opposed to extrema , where one is searching for the value of the variable which minimizes ( or makes stationary ) a function . of course this dichotomy is not so sharp in practice , as the global problem is equivalent to a differential equation ( the euler - lagrange equation ) which is of course local . the relation has even found its way into modern parlance @xcite : `` think globally , act locally ! '' . in quantum mechanics , where ( usually ) the aim is for approximations to the energy by choosing a wave - function , one speaks of wave - functions , which are optimized according to a variational principle . in classical action principles , the equivalent notion is that of a or trajectory , which should be optimized in accordance with the action principle . for uniformity we shall talk of in the classical domain , to emphasize the similarity of these ideas . as noted above , maupertuis action principle was formulated in precise form first by euler and lagrange some two and a half centuries ago ; for the history , see the books by yourgrau and mandelstam and by terral @xcite . however , even the formulation of these eminent mathematicians remained problematic , as emphasized by jacobi , who stated in his lectures on dynamics : `` in almost all text - books , even the best , this principle is presented so that it is impossible to understand '' @xcite . to understand at least in part the problem ( see @xcite for a detailed discussion ) , it is sufficient to note that maupertuis principle assumes conservation of energy , whereas a well formulated principle , like hamilton s principle , implies energy conservation . as we shall see , it is possible to reformulate maupertuis principle in a more general form to remedy this problem and also make it more useful in applications @xcite . it should be noted that hamilton s principle does not suffer from any such drawbacks and is useful both conceptually ( to derive equations of motion ) and as a tool for approximations . the reformulated maupertuis principle has the advantage of being closely related to the classical limit of schrdinger s variational principle of wave mechanics , and thereby lends itself easily to semiclassical applications . even though we emphasize two action principles for classical mechanics , it is worth noting that there are many other formulations and reformulations of these principles . for a list of references on this point see ref.@xcite . we note in particular the elegant work of percival in the 1970 s on variational principles for invariant tori ( see ref.@xcite for a review ) . the emphasis on maupertuis and hamilton reflects the two main approaches of mechanics , one based on the lagrangian @xmath6 and the other based on the hamiltonian @xmath7 . in the above and in section 2 we use the traditional terminology for the variational principles , i.e. `` least '' action principles , but , since jacobi s work it has been recognized @xcite that the action is in fact in general for the true trajectories . this means the first - order variation vanishes , and the action may be a minimum , a maximum or a saddle , depending on the second - order variation @xcite . hence `` stationary '' action principles would be more accurate terminology . similar remarks apply to the general maupertuis and reciprocal variational principles discussed in section 2 . in sections 2 - 4 we restrict ourselves to conservative holonomic systems . various problems are solved using classical and semiclassical variational methods and some comparisons are made with the results from the quantum variational method for excited states @xcite . in section 5 we discuss nonconservative ( but nondissipative ) systems , and in section 6 we mention briefly particular nonholonomic systems ( which have velocity constraints ) , in connection with relativistic systems . the relations between classical and quantum variational principles ( vp s ) are discussed in section 7 . hamilton s least action principle ( hp ) states that , for a true trajectory of a system , hamilton s action @xmath3 is stationary for trajectories which run from the fixed initial space - time point @xmath8 to the fixed final space - time point @xmath9 , @xmath10 where @xmath11 is the duration . here the ( hamilton ) action @xmath3 is the time integral of the lagrangian @xmath12 from the initial point to the final point on the trial trajectory @xmath13 , @xmath14 where @xmath13 is the generalized coordinate , @xmath15 the generalized velocity and @xmath16 the time . in practice we choose @xmath17 and @xmath18 for convenience . in general @xmath19 stands for the complete set of independent generalized coordinates @xmath20 , where @xmath21 is the number of degrees of freedom @xcite . @xmath22 the latter convention for the end - positions is also followed below in ( 2.2 ) , ( 2.3 ) , etc . from the action principle ( 2.1 ) one can derive lagrange s equation(s ) for the trajectory , which we shall not do in this review ; see for example @xcite . the reformulated ( or general ) maupertuis least action principle ( gmp ) states that , for a true trajectory maupertuis action @xmath4 is stationary on trajectories with fixed end - positions @xmath23 and @xmath24 and fixed mean energy @xmath25 : @xmath26 here maupertuis action @xmath4 is given by @xmath27 where @xmath28 is the canonical momentum , and in general @xmath29 stands for @xmath30 . the second form for @xmath4 is valid for normal systems , for which @xmath31 and @xmath32 is quadratic in the @xmath33 s , where @xmath32 is the kinetic energy and @xmath34 the potential energy . the mean energy @xmath25 is the time average of the hamiltonian @xmath35 over the trial trajectory of duration @xmath36 , @xmath37 from the initial position @xmath23 to the final position @xmath24 which are common to all trial trajectories . @xmath38 we use the notation @xmath25 rather than @xmath39 because in the early sections we restrict @xcite ourselves to dynamical variables where the hamiltonian is equal to the energy @xmath40 ; the general case is discussed in note @xcite and later sections . the reciprocal maupertuis principle ( rmp ) , or principle of least ( stationary ) mean energy , is @xmath41 for stationary ( i.e. steady - state or bound ) motions ( periodic , quasiperiodic , chaotic ) ( 2.3 ) is the classical limit @xcite of the schrdinger variational principle of quantum mechanics : @xmath42 , ( see also sec.7 ) . the connection is clear intuitively : for bound motions , in the large quantum number limit the expectation value of the hamiltonian becomes the classical mean energy and the quantum number @xmath43 becomes proportional to the classical action @xmath4 over the motion . this connection enables the rmp to lend itself naturally to semiclassical applications ( sec . 2.3 ) . for periodic and quasiperiodic motions the reciprocal maupertuis principle ( 2.3 ) is equivalent @xcite to percival s principle for invariant tori @xcite . the rmp is , however , more general and is valid for chaotic motions , scattering orbits , arbitrary segments of trajectories , etc . percival s principle has been derived by klein and co - workers from matrix mechanics @xcite in the classical limit . hamilton s principle also has a reciprocal version @xcite - see next section . for a summary of reciprocity in variational calculus , see appendix i. the textbook euler - lagrange version of maupertuis principle differs from ( 2.2 ) ; the constraint of fixed mean energy @xmath25 is replaced by one of fixed energy @xmath44 , @xmath45 this is not erroneous , since a true trajectory does indeed conserve energy , but it is the source of the inconveniences referred to by jacobi @xcite . because , as we show in the next section , the gmp is equivalent to the hp , and therefore to the equations of motion from which energy conservation follows , it will become clear that energy conservation is a consequence of the gmp ( 2.2 ) , rather than an assumption as in the original mp ( 2.4 ) . it is well known that conservation of energy is a consequence of symmetry under time translation ( noether s theorem ) , either via the lagrangian and equations of motion @xcite , or from the action and hamilton s principle @xcite . similarly energy conservation can be derived directly from the gmp @xcite . in the early sections we assume @xmath31 and that @xmath34 is time - independent ( conservative ) . later these restrictions are relaxed ( see sections 5 and 6 ) . within mechanics one can show that different action principles such as ( 2.1 ) , ( 2.2 ) and ( 2.3 ) are equivalent to each other . this we sketch below ( see @xcite for more details ) . alternatively , one can show @xcite from hamilton s principle ( 2.1 ) that the lagrange equations of motion , in some coordinates are essentially newton s equations in the same set of coordinates . reference @xcite shows that ( 2.4 ) is equivalent to newton s equations , quoting an argument due to lagrange , and in @xcite we derive ( 2.3 ) from similar arguments . on a more lofty level one can derive the action principles from quantum mechanics in the classical limit . dirac and feynman @xcite have derived hamilton s principle from a path integral formulation of quantum mechanics , and similarly the rmp ( 2.3 ) can be derived from wave mechanics ( see @xcite ) . further connections between quantum and classical variational principles are given in sec . the equivalence of hamilton s principle ( 2.1 ) with the general maupertuis principle ( 2.2 ) is worth spelling out in a little more detail . ( the equivalence of the mp with the gmp is discussed in appendix ii ) . the starting point of the derivation is the legendre transformation between the lagrangian of the system and the corresponding hamiltonian , @xmath46 here @xmath47 runs over the various degrees of freedom , and @xmath19 stands for @xmath48 and @xmath49 for @xmath50 . upon integration over time along a trial trajectory starting at @xmath51 and ending at @xmath52 ( where we could have @xmath53 for a closed trajectory ) , and with the definitions given above we obtain @xmath54 where @xmath36 is the duration along the trial trajectory . thus @xmath4 and @xmath3 are also related by a legendre transformation . if we now vary the trial trajectory and the duration @xmath36 , keeping @xmath23 and @xmath24 fixed , and compute the variation of the different terms in ( 2.6 ) , we obtain @xmath55 which is the relation between the different variations . for conservative systems , near a true trajectory hamilton s principle is equivalent to the vanishing of the left hand side . in fact a vanishing left hand side is the unconstrained form @xcite of hamilton s principle ( the uhp - see @xcite for proof ) : @xmath56 where we have used the fact @xmath57 a true trajectory for conservative systems . eq ( 2.8 ) has the unconstrained form @xmath58 , with @xmath59 ( the energy of the true trajectory ) a constant lagrange multiplier . for fixed @xmath36 ( i.e. @xmath60 ) , we recover the hp @xmath61 , and for fixed @xmath3 ( i.e. @xmath62 ) we obtain the reciprocal hamilton principle ( rhp ) @xmath63 the rhp ( principle of least time ) is discussed in detail in ref.@xcite . the right hand side of ( 2.7 ) ( which therefore must also vanish near a true trajectory ) gives the unconstrained form of maupertuis principle ( ump ) for conservative systems : @xmath64 this has the unconstrained form @xmath65 , with @xmath66 ( the duration of the true trajectory ) a constant lagrange multiplier . for fixed @xmath67 ( 2.10 ) gives the gmp of equation ( 2.2 ) while for fixed @xmath4 it gives the reciprocal , the rmp of equation ( 2.3 ) . an alternative derivation of the ump ( 2.10 ) is given in ref.@xcite with traditional variational procedures , employing the general first variation theorem of variational calculus , which includes end - point variations @xcite . equation ( 2.7 ) connecting the different variations is reminiscent of equations of thermodynamics , where two conditions are derived from a single equation . there are many other analogies with thermodynamics , e.g. two actions in mechanics analogous to two free energies in thermodynamics , with a legendre transform relation in each case @xcite , adiabatic processes and reversible processes in both mechanics and thermodynamics , and reciprocal variational principles in both mechanics and thermodynamics ( see appendix i ) . beginning with helmholtz @xcite , many of these analogies have been explored by various authors @xcite . as we have stressed , we hold the end - positions @xmath23 and @xmath24 fixed in all the variational principles discussed above . it is possible to relax these constraints by generalizing further the uhp and ump ( see notes @xcite , @xcite , and @xcite ) , but we shall not need these generalizations for the applications we discuss . the derivation we have sketched shows that the four principles ( 2.1 ) , ( 2.2 ) , ( 2.3 ) , ( 2.9 ) are equivalent to each other . mathematically , the hamilton principles can be regarded as legendre transformations of the maupertuis principles , since the legendre transformation relation @xmath68 allows us to change independent variables from @xmath69 to @xmath36 . thus in classical mechanics we have a set of four variational principles that are symmetric under legendre and reciprocal transformations : @xmath70 + after a short interlude on action principles in semiclassical mechanics we proceed to discuss some simple examples . the rmp ( 2.3 ) is the basis of a very simple semiclassical quantization method . as we shall see , the solution of ( 2.3 ) yields , for the true trajectory or approximations to it , @xmath25 as a function of the action @xmath4 , @xmath71 . for bound motions we then assume the standard einstein - brillouin - keller ( ebk ) @xcite quantization rule for the action over a cycle @xmath72 where @xmath73 , @xmath74 is essentially the morse - maslov index ( e.g. @xmath75 for a harmonic oscillator ) , and @xmath76 is planck s constant . ( ebk ( or torus ) quantization is the generalization of bohr - sommerfeld - wilson quantization from separable to arbitrary integrable systems ) . thus we have expressed @xmath25 as a function of the quantum number @xmath43 , @xmath77 . for multidimensional systems , ( 2.11 ) is applied to each action @xmath78 in @xmath79 , so that we get @xmath80 as a function of the quantum numbers @xmath81 . this method has been found to be reasonably accurate even for nonintegrable systems , where strictly speaking the good actions @xmath78 do not exist . the ump ( 2.10 ) can also be used to obtain semiclassical expressions for @xmath80 . examples are given later in sections 4.1 - 4.3 . here we discuss how to use variational principles to solve practical problems . we use the so - called direct method @xcite of the calculus of variations ( e.g. rayleigh - ritz ) , which operates directly with the variational principle and makes no use of the associated euler - lagrange differential equation . historically @xcite , the method originates with euler , hamilton , rayleigh , ritz and others ( see @xcite for some early references).we assume a trial solution , which contains one or more adjustable parameters @xmath82 . we then use the variational principle to optimize the choice of the @xmath82 . generally speaking , the more parameters @xmath82 , the better the solution . the direct method is well known in quantum mechanics , but is also very useful in classical mechanics , as we shall see . we give a few simple examples in this section , mainly for bound motions ; further simple examples , including for scattering problems , are given in refs . @xcite and in the references therein . the simplest example is the free motion of a particle of unit mass on the line segment @xmath83 to @xmath84 . with the traditional maupertuis principle ( 2.4 ) once we fix the energy @xmath85 there is essentially nothing left to vary . with the gmp ( 2.2 ) or its reciprocal ( 2.3 ) we can choose a variety of trial trajectories . for example we can take velocity @xmath86 on the sub - segment @xmath87 and velocity @xmath88 on the sub - segment @xmath89 . the action @xmath4 is @xmath90 and the mean energy @xmath25 is @xmath91 , where we can vary @xmath86 to change @xmath4 while keeping the mean energy fixed . we then find that @xmath92 , a simple example of energy conservation for this particular trial trajectory . by iteration we can now see that energy is conserved for the whole trajectory on the segment @xmath93 , as a consequence of the general maupertuis principle ( 2.2 ) . an instructive example is the one - dimensional ( 1d ) quartic oscillator , with hamiltonian @xmath94 using a harmonic oscillator trial trajectory with @xmath95 we find for a complete cycle with period @xmath96 @xmath97 @xmath98 from ( 3.3 ) , with the use of the rmp ( 2.3 ) we can obtain the best " frequency @xmath99 solving @xmath100 , which can be substituted into ( 3.3 ) to obtain @xmath25 @xmath101^{1/3 } , \;\;\ ; \bar{e } = \frac{1}{2}(c/{m^2})^{1/3 } ( { 3 w}/{4 \pi})^{4/3 } \eqno(3.4)\ ] ] and the period @xmath36 as a function of mean energy : @xmath102^{1/4}. \eqno(3.5)\ ] ] this variational estimate for the period is in error by @xmath103 when compared with the exact result , as noted in reference @xcite . systematic improvements can be obtained by including terms @xmath104 , @xmath105 , etc . , in the trial trajectory @xmath106 ; see @xcite for a related example . we have used the rmp @xmath107 here since ( 3.3 ) made it very convenient to do so . the gmp @xmath108 and the hp are also viable for this problem , but the original mp @xmath109 is not , since the constraint of fixed @xmath44 leaves essentially no freedom for variation for 1d problems @xcite . one can get around this by relaxing the constraint via a lagrange multiplier , but this is then equivalent to the gmp ( see app . ii ) or the ump . in 2d etc . , the constraint of fixed @xmath44 does allow some freedom for variation , but the constraint is very cumbersome and the gmp is always much more convenient . the use of the action principle ( 2.3 ) to obtain the mean energy as a function of the action ( 3.4 ) or the period as a function of mean energy ( 3.5 ) , with harmonic oscillator trial trajectories , is very similar to the use of the variational principle in quantum mechanics to estimate the energies of states of the quartic oscillator with harmonic oscillator trial wave - functions . we show this explicitly for the eigenstates of the hamiltonian ( 3.1 ) . in both cases an optimum frequency for the trial harmonic case is sought and the final errors are also similar . we start by computing the quantum expectation value of ( 3.1 ) using a harmonic oscillator state with trial frequency @xmath110 , @xmath111 and find the optimum frequency @xmath99 by demanding the vanishing of the derivative of @xmath112 with respect to the frequency . we substitute this frequency in ( 3.6 ) to estimate the energy of excited states @xmath113 taking the asymptotic limit of ( 3.7 ) for @xmath114 and replacing @xmath115 by @xmath4 we obtain the second equation ( 3.4 ) . this is not as trivial as it seems , since ( 3.7 ) was obtained from the quantum variational principle ( for an arbitrary state @xcite ) , while ( 3.4 ) was obtained from a classical action principle . this example illustrates neatly that the two variational principles are related : the reciprocal maupertuis principle is the classical limit of the schrdinger variational principle ( see sec.7 for further discussion ) . the example illustrates also that the classical action principle can be used as a method of approximation , and semiclassically to estimate @xmath116 as described in sec.2.3 . according to ( 2.11 ) , we quantize @xmath117 in ( 3.4 ) semiclassically by the replacement @xmath118 ; as just discussed , this leads to agreement asymptotically ( @xmath119 ) with the quantum result ( 3.7 ) . reference @xcite discusses the similar relation between quantum and classical variations for a different one - dimensional system , with a linear potential . the application of classical action principles to other simple systems is discussed in reference @xcite . the application of quantum variations for excited states has a long history : the earliest reference is mcweeny and coulson , who applied it to the quartic oscillator described above @xcite more than fifty years ago ( for later references see @xcite ) . for a pendulum with two degrees of freedom ( @xmath120 ) , there is an additional term in the kinetic energy compared to that of a plane pendulum , i.e. @xmath121 where @xmath122 is the polar angle ( measured from the downward vertical axis ) and @xmath123 is the azimuthal angle , @xmath124 is the bob mass , @xmath6 the length , and @xmath125 the gravitational acceleration . using the axial symmetry around the @xmath126 ( vertical ) axis , we introduce the cylindrical coordinate @xmath127 , and the angular momentum component @xmath128 , which is a constant of the motion . we also expand @xmath129 , keeping up to quartic terms , which are @xmath130 . this gives @xmath131 where @xmath132 . with two degrees of freedom and two constants of the motion ( @xmath133 and the energy ) , the system is integrable . if the quartic terms are neglected in ( 3.9 ) by assuming @xmath134 , the @xmath135 or @xmath136 motion is that of a 2d isotropic harmonic oscillator with frequency @xmath99 . in particular , an elliptical orbit , with semiaxes @xmath137 and @xmath138 , is fixed in space , and the motion is periodic . if the quartic terms are not negligible , the oscillator frequency shifts from @xmath99 to a lower value @xmath110 , as in the plane pendulum case , where @xmath139 and the orbit precesses in the prograde sense with frequency @xmath140 , where @xmath141 the motion is now in general quasiperiodic . the precession is one of the major sources of error to be eliminated in the design of foucault pendula @xcite . ( the foucault pendulum is discussed in sec . because the precession rate is @xmath130 , this precession is particularly relevant for recently developed short foucault pendula , which are less than 1 m in length . equations ( 3.10 ) and ( 3.11 ) have been derived by perturbation theory @xcite . to derive ( 3.10 ) and ( 3.11 ) variationally @xcite , we first choose a trial trajectory ; we take an ellipse which is closed in the coordinate system @xmath142 which rotates with respect to the fixed @xmath143 system with the ( unknown ) precession frequency @xmath140 . in the @xmath142 system we therefore have @xmath144 and @xmath145 , where @xmath110 may differ from @xmath99 . because the @xmath142 and @xmath146 systems differ by a rotation , we have @xmath147 , and therefore our trial trajectory in the fixed coordinate system is @xmath148 to apply the maupertuis principles , we calculate the action @xmath4 and mean energy @xmath25 over one period @xmath149 of the rotating ellipse , @xmath150 where @xmath151 denotes a time average over one period , and @xmath32 and @xmath34 are given by ( 3.9 ) . the averages of the terms in ( 3.9 ) are easily calculated using ( 3.12 ) , which gives @xmath152 @xmath153 @xmath154 the precession frequency @xmath140 can be related to the angular momentum using @xmath155 , and @xmath156 from ( 3.14 ) , giving @xmath157 as initial and final positions 1 and 2 ( see fig.1 ) , we take two successive aphelia at @xmath158 , i.e. @xmath159 and @xmath160 , where @xmath161.the positions 1 and 2 and hence @xmath137 and @xmath162 are thus fixed , which leaves @xmath138 and @xmath110 available as variational parameters . note that ( 3.15 ) then implies that @xmath133 is varied with the trial trajectory ( although it is a constant of the motion on the individual trial trajectories ) . we find it easiest to use the unconstrained version of the maupertuis principles [ i.e. the ump ( 2.10 ) ] , @xmath163 , where @xmath164 is the time to reach position 2 from position 1 . thus we use @xmath165 where , choosing units such that @xmath166 , @xmath167 , @xmath168 , we have from ( 3.13 ) to ( 3.15 ) @xmath169 @xmath170 and @xmath171 the first of the variational equations ( 3.16 ) is satisfied identically by ( 3.17 ) and ( 3.18 ) , and gives no useful information . from the second , setting @xmath172 , where @xmath173 is the frequency shift , and retaining terms only to second order in @xmath137 and @xmath138 , i.e. , @xmath130 in general units , we get the condition @xmath174 when @xmath175 , ( 3.19 ) gives @xmath176 as it should [ recall the plane pendulum result @xcite ] . by symmetry , @xmath173 must reduce to @xmath177 when @xmath178 . this requires that @xmath179 in ( 3.19 ) , and from this result we then get @xmath180 from ( 3.19 ) . when we restore the dimensional quantities @xmath99 and @xmath6 , these results agree with ( 3.10 ) and ( 3.11 ) , so that with a harmonic oscillator trial trajectory , the variational method gives results correct to @xmath130 . hamilton s principle can also be applied to this problem . here , we need the hamilton action @xmath181 , where @xmath36 ( and hence @xmath110 ) is now fixed , in addition to the positions 1 and 2 ( and hence @xmath137 and @xmath162 ) . this leaves @xmath138 available as a variational parameter , and setting @xmath182 leads to the same results as above . more complicated examples are easy to find , and are especially interesting in multidimensional systems which exhibit chaotic motion . such systems are nonintegrable , and have fewer constants of the motion than the number of degrees of freedom . we illustrate the use of classical and semiclassical variational principles together with the application of quantum variational principles to excited states . the 2d @xmath184 oscillator is a simple example of this class , which has been studied in the literature for some time ( see , e.g. @xcite ) . there is only one constant of the motion , the energy . classically , most trajectories in this system are chaotic @xcite . fig.2 shows the contours of the @xmath184 potential , and fig.3 shows a typical ( chaotic ) classical trajectory . . the darkened contour corresponds to @xmath185 . ( from ref.@xcite ) ] -axis . the trajectory eventually reverses its @xmath186-direction and returns to the vicinity of the origin before entering another arm of the potential . ( from ref.@xcite ) ] refs.@xcite give classical and semiclassical discussions of the motion using the rmp . here we discuss the quantum states using variational methods for the excited states , as done above for a simpler example . comparison with the semiclassical results is given later . the hamiltonian is @xmath187 we choose units where @xmath166 and @xmath188 , where @xmath189 is the potential . below we also choose @xmath190 . we take as an approximation a separable trial wave - function @xmath191 @xmath192 where the wave - function @xmath193 is an eigenstate of a harmonic oscillator with frequency @xmath194 : @xmath195 the frequencies @xmath194 and @xmath196 are variational parameters , which will differ from state to state , and as a result the set ( 4.2 ) is nonorthogonal . the computation of matrix elements and expectation values with the wave functions ( 4.3 ) is rather easy , @xmath197 and therefore the expectation value of the hamiltonian ( 4.1 ) with the wave - functions ( 4.2 ) is @xmath198 . \eqno(4.5)\ ] ] we then find the best frequencies @xmath194 and @xmath196 by setting to zero the partial derivatives of @xmath112 with respect to the two @xmath110 s to obtain the frequency conditions @xmath199 from which we obtain @xmath200 eqs . ( 4.7 ) and ( 4.3 ) show that at the variational minimum the ratio of the two frequencies is rational and the mean energy of the @xmath186-oscillator is the same as the mean energy of the @xmath201-oscillator . it follows immediately that the mean kinetic energies of the two oscillators are also equal to each other at the variational minimum : @xmath202 . from ( 4.6 ) we also find @xmath203^{1/3 } , \eqno(4.8)\ ] ] which gives the separate frequencies @xmath204 and the variational estimate for the energy : @xmath205^{2/3}. \eqno(4.10)\ ] ] therefore the variational states are labelled by the two integers @xmath206 and @xmath207 and the energy at high quantum numbers increases as the two thirds power of their product . the estimate ( 4.10 ) can also be obtained semiclassically by using the rmp to find @xmath208 , and then quantizing the actions @xmath78 with the ebk relation ( 2.11 ) ; see ref.@xcite for a detailed discussion . it is to be noted that this method yields the quite reasonable first approximation ( 4.10 ) for @xmath209 despite the fact that strictly speaking the good actions @xmath210 and @xmath211 do not exist for the @xmath184 oscillator . in essence , one is approximating a chaotic trajectory by a quasiperiodic one , and the rmp picks out the best such trial trajectory @xcite . the estimate ( 4.10 ) differs only in the leading coefficient , 3/2 vs. 1.405 , from a semiclassical adiabatic formula ( sca ) obtained by martens et al @xcite . the difference between the two estimates is about @xmath212 . as noted by martens et al @xcite formula ( 4.10 ) gives rise to degeneracies . for example @xmath213 = ( 7,0 ) , ( 0,7 ) , ( 2,1 ) , ( 1,2 ) give the same energies - a four - fold degeneracy . however , numerical estimates of the first 50 energy levels , obtained in @xcite , do not show such high degeneracies . also , because the symmetry group of the hamiltonian @xmath214 ( a fourfold axis of rotation , and four reflection planes ) has only one- and two - dimensional irreducible representations @xcite , we expect at most two - fold degeneracy . we shall see that the basis ( 4.2 ) permits an evaluation of the splitting of these spurious degeneracies . the pattern of degeneracies is easy to understand if the bracket is rewritten @xmath215 = n/4,\ ] ] where the number @xmath216 ( theprincipal " quantum number ) , is an odd integer . the degeneracy pattern is determined by the decomposition of @xmath216 into prime factors . for example @xmath213=(2,1 ) corresponds to @xmath216 = 15 which can be factored as @xmath217 , @xmath218 , @xmath219 , @xmath220 and these products correspond to ( 7,0 ) etc . if @xmath216 is a prime number there are only two states , corresponding to @xmath221 and @xmath222 , while every nontrivial factorization of a nonprime @xmath216 allows further states . the integer @xmath216 belongs to one of two possible classes : @xmath223 or @xmath224 . for the case @xmath225 ( e.g. @xmath216 = 15 ) one of @xmath226 is even and the other is odd . this class of states eo and oe ( of @xcite ) corresponds to the two dimensional irreducible representation of @xmath214 . for @xmath227 ( e.g. @xmath216=21 ) the two quantum numbers @xmath228 are both even or both odd ( e.g. @xmath228 = ( 10,0 ) or ( 0,10 ) or ( 1,3 ) or ( 3,1 ) for @xmath216=21 , which is therefore four - fold degenerate ) this degeneracy is split into states of symmetry eee , eeo , ooe and ooo , in the notation of martens et al @xcite for the one dimensional irreducible representations . in this notation the first two letters give respectively the parity ( even or odd ) of @xmath206 and @xmath207 ( or equivalently the behaviour of the wave - function under the reflections ( @xmath229 , @xmath230 ) , while the third letter stands for the behavior under the interchange of @xmath186 with @xmath201 . we discuss later the splitting of degeneracies resulting from the expression ( 4.10 ) . at this point we wish only to note that the degeneracy grows very slowly with @xmath216 , like lnln@xmath216 , and even for @xmath216 quite large where quasiclassical behaviour is expected ( say @xmath231 ) the typical number of degenerate states is quite modest , 4 - 6 . the main points of discussion arising from the result ( 4.10 ) are the splitting of the degenerate levels , perturbative corrections to the variational energies , and the density of states implied by this formula . for the first fifty low lying states one has the accurate numerical estimates of martens et al @xcite with which to compare . we can improve the variational result ( 4.10 ) if we rewrite the hamiltonian @xmath7 ( 4.1 ) in the form @xmath232 here , the perturbation @xmath34 is the last term above , and its effect is already included in first order , in the variational estimate ( 4.10 ) . therefore the first correction comes in second order in @xmath34 . we must also remember that that the frequencies @xmath233 are appropriate to a fixed pair of quantum numbers @xmath213 , and change if we go to a different pair . the simplest way to do perturbation theory in these circumstances is to estimate corrections to each state @xmath213 by considering a set of orthogonal intermediate states @xmath234 all with the same frequency @xmath233 appropriate to the state @xmath213 which we consider . therefore we take the second order correction @xmath235}. \eqno(4.12)\ ] ] due to the form of @xmath34 ( quadratic in the coordinates @xmath143 ) , we must have @xmath236 or @xmath237 or @xmath238 for nonzero contributions and similarly for @xmath239 . it is an interesting property of the variational solution that only intermediate states with both @xmath240 different from @xmath206 and @xmath239 different from @xmath207 give nonvanishing contributions , because of the frequency conditions . this can be seen from an evaluation of matrix elements with only one quantum number different , for example @xmath241 , but @xmath242 . then one finds @xmath243 , \eqno(4.12^{'})\ ] ] and the bracket on the right hand side vanishes because of ( 4.6 ) . only the perturbation @xmath244 contributes to nonvanishing matrix elements . there are only four intermediate states which contribute : @xmath245 , where @xmath246 , etc . therefore ( 4.12 ) becomes after a little computation @xmath247 and @xmath248^{2/3 } , \eqno(4.14)\ ] ] where @xmath249 denotes a second - order shift and @xmath250 denotes the energy up to the second - order . therefore , in second - order perturbation theory , the coefficient @xmath251 is diminished by about @xmath252 to 1.437 , closer to the adiabatic result 1.405 of martens et al @xcite . the remaining difference is about @xmath253 . the degeneracies of ( 4.14 ) are still the same as those of ( 4.10 ) . if we apply perturbation theory to the ground state @xmath254 we find @xmath255 the sequence ( 4.15 ) indicates that the perturbation expansion does not converge , and it is a good idea to stop at second order . the accurate numerical result for @xmath256 ( from @xcite ) is 0.5541 , which shows that stopping at second order gives an error of about @xmath257 for the ground state . some examples of the estimates @xmath258 , and @xmath259 are given in table 1 , and compared with numerical estimates from @xcite , for a set of states which are also relevant to the next section . * table 1 * [ cols="^,^,^,^,^,^,^",options="header " , ] comparison of selected energy levels calculated via quantum and semiclassical methods demonstrating the failure of the simple variational calculation to produce level splitting observed in the quantum results.(from ref.@xcite ) in this section we consider nonconservative mechanical systems . for example consider a system of particles moving in the time - dependent external field , or a system whose parameters depend explicitly on time . other examples include time - dependent constraints , e.g. a bead on a rotating hoop whose angular velocity is a prescribed function of time , and variable mass systems ( for which @xmath260 in general @xcite ) . such systems are not closed and the energy is not conserved . the evolution of a nonconservative system is completely determined by a time - dependent lagrangian @xmath261 or by a time - dependent hamiltonian @xmath262 . ( we exclude dissipative systems from our discussion . we do note that lagrangians and hamiltonians have been found for such systems in some cases @xcite , so that the action principles discussed here would apply in these cases . alternatively @xcite , dissipative terms can be added to the lagrange equations of motion or to the hamilton principle . ) @xmath7 and @xmath6 are related to each other by the legendre transformation ( see eq . ( 2.5 ) ) @xmath263 locally these lead to the usual euler - lagrange or hamilton equations of motion that govern the dynamics of the system . we consider the global description of nonconservative systems . it is well known that the hamilton principle @xmath264 is applicable for nonconservative systems as well as for conservative ones . here subscript @xmath36 reminds us that variations near the actual path are constrained by fixing the travel time @xmath36 from the initial point @xmath265 to the final point @xmath52 . again we have chosen @xmath266 and @xmath267 , and , as before , it is understood here and below that the end - positions @xmath23 and @xmath24 are also fixed . as before , we can relax the constraint of fixed @xmath36 and rewrite ( 5.2)@xcite as an unconstrained hamilton principle ( uhp ) @xmath268 where the lagrange multiplier @xmath269 is the true value of the energy at the time of arrival at the final point @xmath270 . ( elsewhere in this paper @xmath269 is denoted variously as @xmath271 , @xmath272 or @xmath273 ; see notes @xcite and @xcite ) . we assume that all trial trajectories start at the same time @xmath274 and also have the same end positions @xmath23 and @xmath24 . using eq.(2.7 ) , which is derived from ( 5.1 ) as before , we transform the uhp ( 5.3 ) to the form @xmath275\delta{t } = 0 . \eqno(5.4)\ ] ] this is the unconstrained maupertuis principle ( ump ) for nonconservative systems , which seems not to have been given in the literature explicitly ( it is implicit in an appendix in @xcite - see also @xcite ) . it relates three variations - the variation of mean energy @xmath276 , the variation of the action @xmath4 and the variation of the travel time @xmath36 . the lagrange multipliers are the true travel time @xmath36 and the difference of energy and mean energy of the true trajectory at the final point , @xmath277 $ ] . the latter can be rewritten as @xmath278= < t\frac{\partial h}{\partial t } > , \eqno(5.5)\ ] ] where the bar and brackets @xmath279 both denote a time average over the interval ( 0,t ) . the generalized ump ( 5.4 ) can also be derived by traditional variational procedures ( @xcite , eq.(a.6 ) ) . one can derive constrained principles from ( 5.4 ) as before , e.g. @xmath280 , but these would be complicated to use in practice because of the severe constraints on the trial trajectories . the version @xmath281 is closest to the original maupertuis principle . the constrained principle @xmath282\right)_{e(0),e(t)}=0\ ] ] is mentioned by whittaker @xcite , and is easily derived from ( 5.4 ) and ( 5.5 ) . but since ( 5.4 ) is also valid without fixing @xmath283 and @xmath269 , ( 5.4 ) is more convenient . formally the unconstrained maupertuis principle ( 5.4 ) is equivalent to the hamilton principle , but it also appears too complicated to be useful in practice . this is not quite true and we demonstrate how the ump can be used for adiabatic perturbations . for simplicity we start with a one - dimensional periodic system that depends on a parameter @xmath284 . let @xmath284 vary with time adiabatically , i.e. the change @xmath285 during a period of oscillation @xmath36 . it is well known that in this case the energy @xmath44 as well as the period of oscillation @xmath36 depend slightly on time , but the action @xmath286 over a cycle , where @xmath287 remains constant to an excellent approximation ( adiabatic invariant ) when the parameter @xmath284 varies . consider what kind of information can be extracted from the ump ( 5.4 ) . both the energy @xmath44 and the period @xmath36 of an actual motion are functions of the action @xmath286 ( recall the action - angle variables ) : @xmath288 , @xmath289 . consider two actual trajectories with actions @xmath286 and @xmath290 as two trial trajectories for the ump ( 5.4 ) . then the variational equation can be rewritten as a differential equation @xmath291 it is more convenient to rewrite this in terms of the frequency @xmath292 , @xmath293 this equation generalizes the standard relation @xcite for integrable conservative systems @xmath294 to integrable nonconservative systems . up to this point ( 5.7 ) and ( 5.8 ) are general and valid for any nonconservative system for which a good action can be defined . consider now the adiabatic regime where the variation of the parameter @xmath295 over the short period @xmath36 , @xmath296 , is much less than the parameter itself : @xmath297 . in this case @xmath298 where the averaging @xmath299 is to be performed at a fixed value of @xmath300 since @xmath301 varies little over a period @xmath36 . the action @xmath286 is conserved for any given parameter @xmath301 , and hence the derivative @xmath302 is also a constant of the motion . as a result in this approximation we get @xmath303 in the same approximation we have for @xmath304 in ( 5.8 ) @xmath305 thus to the first order approximation in @xmath306 we can rewrite ( 5.8 ) as @xmath307[\frac { d}{di } ( \frac{1}{\omega^{(0)}}\frac{\partial e}{\partial \lambda})- \frac{\partial e}{\partial \lambda}\frac{d}{di } ( \frac{1}{\omega^{(0)}})],\ ] ] or finally as @xmath308 here @xmath309 is the zero - order term , and @xmath310 is the variation of the parameter @xmath301 during the given period . we get corrections ( of the first - order in the nonadiabaticity parameter @xmath311 ) to the frequency of oscillation , for adiabatic systems . the result seems very simple . it appears that @xmath312 is a continuous function of time @xmath16 and we calculate some average frequency over the period . but actually @xmath312 is not always a continuous function of time ; it is a parameter of a ( slightly ) aperiodic system . thus ( 5.10 ) is nontrivial in general . to illustrate these results it is instructive to consider a simple exactly solvable example . we consider the 1d motion of a perfectly elastic ball of mass @xmath124 bouncing between two planes . we suppose that one plane is fixed and the other is slowly moving with constant velocity @xmath313 . we suppose also that the planes are heavy enough so that they are unaffected by collisions with the ball . the distance between the two planes @xmath314 is the adiabatic parameter . let us define a as the motion of the ball from the fixed plane to the moving plane , collision , and the motion of the ball in the backward direction up to the collision with the fixed plane . the n-@xmath315 cycle can be described in the following way . at time @xmath316 ( beginning of the cycle ) the ball has velocity @xmath317 , and the distance between planes is @xmath318 . at time @xmath319 the ball collides with the moving plane and after the collision it moves to the fixed plane with velocity @xmath320 ( the collision with the fixed plane is elastic and the backward velocity is equal to the forward velocity after collision , i.e. @xmath321 ) . the backward motion takes time @xmath322 . the period of the whole cycle is @xmath323 . at time @xmath324 ( the initial time for ( n+1)-th cycle ) the distance between the two planes is equal to @xmath325 . this means that @xmath326 is an exact ( adiabatic ) invariant . it easy to see that the action @xmath286 is proportional to this invariant , @xmath327 i.e. the action @xmath286 is an exact ( adiabatic ) invariant for our special initial condition @xcite . in terms of the action @xmath286 and the adiabatic parameter @xmath6 the complete set of relevant quantities is as follows : @xmath328 @xmath329 ^ 2,\ ] ] @xmath330 i.e. @xmath331 is linear function of the nonadiabaticity parameter @xmath332 and the approximate relation ( 5.10 ) must be exact here . indeed , we have @xmath333 , @xmath334 and @xmath335 thus the approximate relation ( 5.10 ) is valid here . it is also not difficult to check after a little algebra that in the exact relation ( 5.8 ) all higher powers of @xmath313 cancel each other and the linear approximation is exact . if @xmath336 is not extremely small , @xmath286 ceases to be an adiabatic invariant and the system becomes nonintegrable . if @xmath337 is varied nonadiabatically and periodically around @xmath338 , the system ( termed the fermi model @xcite ) can show steady energy growth and chaotic motion . the generalization of ( 5.10 ) to multidimensional integrable systems with actions @xmath339, ... ,@xmath340 is evident , as is the generalization to multiparameter systems , with parameters @xmath341 . the @xmath342 corrections in @xmath312 have been derived by other methods @xcite , but in different forms from ( 5.10 ) . if we assume @xmath295 is periodic , with a long period @xmath343 , the @xmath342 terms in @xmath312 , when integrated over @xmath344 , define the @xmath345 , the classical analogue of the quantum berry phase @xcite . the hannay angle is the extra ( geometric ) phase undergone by the system in one cycle of @xmath295 , in addition to the more obvious dynamical phase @xmath346 obtained by integrating the instantaneous frequency @xmath347 . in general one requires more than one @xmath301-type parameter to obtain a nonvanishing result for @xmath345 , but hannay and berry and coworkers give a few examples @xcite where one parameter suffices . we discuss two examples with nonvanishing @xmath345 . consider a bead of mass @xmath124 sliding without friction on a closed planar wire loop which is slowly rotating with angular velocity @xmath140 in the horizontal plane . for a loop of arbitrary shape , hannay and berry @xcite show that the hannay angle @xmath345 defined above is given by @xmath348 , where @xmath349 is the area and @xmath6 the perimeter of the loop . for a circular loop rotating about an axis perpendicular to the plane of the loop and through the center , @xmath345 is simply @xmath350 , as is clear intuitively : the `` starting line '' for the particle moves ahead by @xmath351 while the particle rapidly slides around the loop . plane with angular velocity @xmath352 about axis through @xmath353 perpendicular to the @xmath143 plane . at @xmath316 , the particle starts with @xmath354 and the rotating @xmath355 axis ( a diameter ) coincides with the fixed axis @xmath186.,width=226 ] we consider a circular loop of radius @xmath356 ( fig.8 ) , but with the vertical axis of rotation through a point @xmath353 on the wire instead of the center @xmath357 of the loop , to make the answer less obvious . the adiabatic parameter here is the prescribed orientation of the loop @xmath295 , where @xmath358 is small . we do not at this stage assume @xmath359 is constant . using the horizontal @xmath360 lab - frame ( fig.8 ) , we can easily write the lagrangian , which is just the lab kinetic energy of the particle . using @xmath122 in fig.8 as the generalized coordinate we find @xmath361 \;\ ; . \eqno(5.11)\ ] ] from ( 5.11 ) we obtain the equation of motion @xmath362 if the loop rotates uniformly ( @xmath363 ) , ( 5.12 ) is just the pendulum equation , as one expects intuitively . because we are using a rotating coordinate system ( @xmath122 is measured from @xmath364 ) @xmath6 is not equal to @xmath365 in this system , so that the hamiltonian @xmath7 is not equal to the energy ( @xmath366 ) in this system either . to find @xmath7 , we first calculate the canonical momentum @xmath367 , where @xmath368 and then @xmath369 is @xmath370 in terms of the canonical variables @xmath122 , @xmath371 the hamiltonian @xmath7 is @xmath372 where we drop @xmath373 terms since we assume @xmath374 . we can not apply ( 5.10 ) as it stands , since @xmath375 was used in the derivation . the generalization to include cases where @xmath376 is @xmath377 where @xmath39 is the average of @xmath7 over the fast motion ( here @xmath122 ) . alternatively , we can use ( 5.10 ) , with now @xmath378 . from ( 5.15 ) we have @xmath379 since @xmath122 varies rapidly and @xmath371 is nearly constant . we now assume @xmath140 is constant . for @xmath380 , @xmath39 is a function of the good action @xmath381 only ( no @xmath122 dependence ) , and when @xmath382 we see @xmath371 remains a good action . note that @xmath39 is independent of angle @xmath301 , so that the second term in ( 5.16 ) will not contribute to @xmath383 , which is @xmath384 the total phase undergone after a long return time @xmath385 is thus @xmath386 the first term is the dynamical phase , the second is the hannay ( geometric ) phase . for a foucault pendulum the hannay angle is the angle by which the plane of the pendulum vibration shifts in one day , i.e. @xmath387 , where @xmath122 is the co - latitude of the point on earth where the pendulum is located . this is a clockwise rotation in the northern hemisphere . this result for @xmath388 has been derived by berry @xcite and others @xcite by various arguments . there are also some pre - hannay purely geometric arguments for the foucault pendulum precession @xcite . here we show the result follows simply from ( 5.16 ) . we consider a pendulum of length @xmath6 and bob mass @xmath124 oscillating at a point on the earth s surface with co - latitude @xmath122 . we use axes x , y , z attached to the earth , with x , y in the horizontal plane and z vertical . when at rest in the x , y , z system , the pendulum hangs along the z - axis , and when in motion the bob oscillates very nearly in the x , y plane . neglecting the z motion and assuming @xmath389 is small , where @xmath390 is the earth s angular velocity , we easily find @xcite the approximate horizontal equations of motion @xmath391 where @xmath392 . we have kept the @xmath393 coriolis terms , but dropped the @xmath373 centrifugal terms , since @xmath140 is small compared to the pendulum frequency @xmath99 . here @xmath140 is the local vertical component of the earth s angular velocity . we are interested in particular in the orbit precession rate predicted by ( 5.20 ) . it is easily checked that ( 5.20 ) follow from the lagrangian @xmath394 and of course ( 5.21 ) can be derived directly @xcite . changing to polar coordinates @xmath395 , @xmath123 via @xmath396 , @xmath397 changes @xmath6 to the form @xmath398 two constants of the motion are obvious from ( 5.22 ) : since @xmath6 does not depend on @xmath123 , @xmath399 is a constant , where @xmath400 and because @xmath6 does not depend on @xmath16 explicitly , the hamiltonian @xmath401 is a constant , where @xmath402 and @xmath403 two constants of the motion with two degrees of freedom imply that the system is integrable and that two good actions exist . to express @xmath7 in terms of the canonical variables @xmath395 , @xmath123 , @xmath404 , @xmath399 we use ( 5.23 ) and @xmath405 to get @xmath406 where again we drop an @xmath373 term . we wish to use ( 5.16 ) ( not ( 5.10 ) here as @xmath407 ) to find the frequencies @xmath408 and @xmath409 , where @xmath301 is the orientational angle of the earth multiplied by @xmath129 . note that for @xmath410 , ( 5.26 ) is simply the hamiltonian for an isotropic 2d harmonic oscillator . calculation of the actions for the 2d oscillator , @xmath411 and @xmath412 , is a standard exercise @xcite , and the hamiltonian @xmath413 in terms of these actions is @xmath414 the term @xmath415 in ( 5.26 ) will not disturb the good actions , so that ( 5.26 ) becomes @xmath416 application of ( 5.16 ) with @xmath417 and @xmath418 then gives @xmath419 where a positive ( + ) frequency in @xmath420 corresponds to a counterclockwise rotation . note that the foucault precession frequency @xmath421 is true for orbits of any shape , e.g. near linear ( the usual case ) , elliptical , or circular . the factor of 2 in @xmath422 in ( 5.29 ) is clear intuitively : for @xmath423 , in one circuit of an elliptical orbit , @xmath395 goes through two cycles while @xmath123 goes through one . computing the phase of the pendulum @xmath424 after one complete rotation of the earth ( period @xmath425 ) then gives @xmath426 the term @xmath427 is the dynamical phase @xmath428 and @xmath429 is the geometric hannay angle @xmath388 . the examples in this section and the previous one led to a nonvanishing hannay angle @xmath345 . it is easy to show that the earlier example ( particle in a 1d box with a slowly moving wall ) gives a vanishing result . the same is true for a particle in a slowly rotating 1d box ( fixed walls ) . golin @xcite has proved that , for 1d systems with one adiabatic parameter @xmath301 , one always has @xmath430 unless @xmath301 undergoes a rotational excursion through @xmath351 . however the proof does not cover our rotating 1d box example ( which has a hard wall potential ) , since golin @xcite assumes a smooth potential . the case of a slowly rotating 2d box of elliptical shape has been worked out by koiller @xcite , and gives a nonvanishing @xmath345 depending in a complicated way on the area of the box . nonintegrable systems have fewer than @xmath21 independent constants of the motion or good actions , where @xmath21 is the number of degrees of freedom . for such systems , at least some of the possible motions are chaotic , and for these at least some of the good action variables do not exist . hence we have not been able to proceed rigorously beyond the general relation ( 5.4 ) for chaotic motions . ( despite the nonexistence of good actions @xmath286 for nonintegrable systems , formal use of them and their adiabatic invariance have been used successfully in semiclassical quantization schemes ; see sections 2.3 , 3.1 , 4.1 , 4.3 and ref . @xcite for a method based on the maupertuis principles for conservative systems , and see refs . @xcite for a method based on adiabatic switching . ) as stated , we do not have rigorous arguments to draw consequences from the ump ( 5.4 ) for nonintegrable systems , but some heuristic arguments can be given . for example , by choosing @xmath431 to vary slowly and linearly with @xmath16 , e.g. @xmath432 , where @xmath306 is small , we can the adiabatic invariance of the action @xmath286 for a 1d periodic system ( earlier we that @xmath286 is an adiabatic invariant for such systems ) . further , by considering long trajectories of multidimensional systems , which start and end in some small region in phase space , and which are not quite periodic in general , we can formally show that , in a poincar cycle time or in the limit @xmath433 the long - path action @xmath4 is an adiabatic invariant for quasiperiodic and chaotic trajectories of multidimensional systems . ehrenfest @xcite has previously shown that @xmath4 is an adiabatic invariant for multidimensional periodic systems . for systems , the phase volume @xmath434 is an adiabatic invariant ( see next paragraph ) , and since the adiabatic invariant is apparently unique for such systems ( kasuga @xcite ) , it should be possible to derive the adiabatic invariance of @xmath4 from that of @xmath140 for these systems , but we have been so far unable to do this @xcite . we note that counterparts of the hannay angle have been established for adiabatically evolving nonintegrable @xcite and ergodic @xcite systems . for one limiting case , complete chaos ( ergodic motion ) where the energy is the only constant of the motion , and long trajectories typically cover essentially the whole energy surface , a rigorous adiabatic invariant has been found @xcite , i.e. the phase volume @xmath435 , enclosed by the energy surface @xmath436 : @xmath437 , \eqno ( 5.31)\ ] ] where @xmath438 is the step function , which is unity for positive argument . ( for multidimensional systems , @xmath439 stands for @xmath440 . ) if @xmath441 depends on a slowly varying parameter @xmath295 so that @xmath442 , then @xmath443 . the adiabatic invariance of @xmath434 is intuitively plausible @xcite . first consider the initial ( @xmath316 ) energy surface @xmath444 , and its image under an time development ( i.e.not necessarily slow ) . the final ( image ) surface at the time @xmath36 is not in general a constant energy surface , and hence not in general a dynamical surface . however , by liouville s theorem the volume enclosed by the final surface is equal to that enclosed by the initial surface . the time - development from @xmath316 to @xmath445 sufficiently slow then at each stage the surface which has evolved from the @xmath316 surface a dynamical surface , by definition of adiabatic evolution , where allowed motions transform into allowed motions @xcite . in this case the final @xmath445 surface is a true constant energy dynamical surface , and hence @xmath446 from liouville s theorem . for ergodic systems , there are no good quantum numbers except for the energy - ordering quantum number @xmath447 , where @xmath448 . a rough but useful semiclassical quantization scheme for ergodic systems based on @xmath434 is the weyl rule @xcite @xmath449 which implicitly defines @xmath450 . this is just a restatement of the early rule due to planck @xcite that each quantum state occupies a volume of size @xmath451 in classical phase space . in practice , various improvements such as replacing @xmath216 by @xmath452 as in ( 2.4 ) are used @xcite . more subtle improvements can be obtained from gutzwiller s trace formula @xcite . these relate fluctuations from the smooth dependence of @xmath44 on @xmath216 predicted by ( 5.32 ) to the classical periodic orbits . in cases such as the @xmath184 potential , where the classical @xmath434 diverges @xcite due to the infinite channels of the potential ( see fig.1 ) , one can estimate the smooth part of @xmath453 ( the cumulative number of states up to @xmath44 ) quantum mechanically @xcite at low energy by an asymptotic expansion @xcite , with leading terms @xmath454 and @xmath455 , where @xmath456 is dimensionless . application of the generalization of ( 5.32 ) , i.e. @xmath457 , then gives reasonable estimates of the low - lying state energies @xcite . another consequence of the ump ( 5.4 ) is the classical version of the hellmann - feynman theorem @xcite @xmath458 here @xmath459 depends on a parameter @xmath301 , so that the energy @xmath460 depends on @xmath301 , as well as on the ( constant ) action @xmath286 , and @xmath461 denotes a time - average over a period @xmath462 at fixed action @xmath286 . we assume a one - dimensional periodic system for simplicity . the quantum version of the theorem @xcite is nowadays much better known , i.e. @xmath463 where @xmath464 , and where the hamiltonian operator @xmath465 depends on some parameter @xmath301 , so that the energy eigenvalue @xmath466 and eigenfunction @xmath467 do also , because of @xmath468 proof of ( 5.34 ) is simple ; we apply @xmath469 to both sides of the relation @xmath470 , and note that the two terms involving @xmath471 add to zero because of ( 5.35 ) , the hermiticity of @xmath465 , and the fact that we assume @xmath472 for all @xmath301 . it is clear that ( 5.33 ) is the classical limit of ( 5.34 ) , since as @xmath473 ( or @xmath474 ) @xmath475 approaches a time average over a period , and fixing @xmath43 corresponds to fixing @xmath286 , since @xmath476 . to derive ( 5.33 ) from ( 5.4 ) we consider @xmath431 to evolve adiabatically , so that @xmath301 and @xmath36 in ( 5.4 ) can be related . the derivation is intricate , however , and it is simpler to proceed instead from the rmp ( 2.3 ) for conservative systems . we apply ( 2.3 ) to two periodic trajectories of @xmath477 , the true one with period @xmath478 , and a virtual one taken as the true trajectory of @xmath479 with period @xmath480 . since first - order variations in @xmath481 vanish ( at fixed @xmath482 ) , we have @xmath483 where @xmath484 is an average over the trajectory with period @xmath478 , and it is understood that @xmath286 is the same for both @xmath484 and @xmath485 . from ( 5.36 ) we get for @xmath486 to @xmath487 accuracy @xmath488 again neglecting @xmath489 terms we have @xmath490 eq(5.37 ) yields @xmath491 , and since @xmath492 for the true trajectory we get @xmath493 , the hellmann - feynman relation ( 5.33 ) apart from notation . as a simple example of ( 5.33 ) consider a 1d harmonic oscillator , with hamiltonian @xmath494 and @xmath495 , where @xmath496 is the angular frequency . choose @xmath497 , the force constant . the left side of ( 5.33 ) is @xmath498 and the right side is @xmath499 . thus we have @xmath500 , as is well known from the virial theorem @xcite . if we choose @xmath501 , we find similarly @xmath502 . as a second example , consider the bound motion of a 1d system with hamiltonian @xmath503 . we change variables @xmath504 via the transformation @xmath505 , @xmath506 , where @xmath301 is an arbitrary scale factor . this transformation is canonical and preserves the hamiltonian @xcite , i.e. @xmath507 , where @xmath508 the energy of a given state of motion is unchanged by the change from @xmath7 to @xmath509 for any value of @xmath301 . we now use ( 5.38 ) in the hellmann - feynman relation ( 5.33 ) . the left hand side vanishes . the right hand side derivative is @xmath510 where @xmath511 , and we have reverted to the original variables in the second form . hence we get @xmath512 i.e. the virial theorem for an arbitrary potential @xmath513 . hamilton s principle is widely used in classical relativistic mechanics and in the classical theory of electromagnetic and gravitational fields to derive covariant equations of motion ( see , e.g.@xcite ) . as for maupertuis principle , it is widely believed that it is not well suited for that purpose because when operating with the energy one loses explicit covariance . in this section we demonstrate , first , how to use maupertuis principle for the derivation of the covariant equations of motion and , the second , how to use it for the solution of specific problems . consider a relativistic particle with mass @xmath124 and charge @xmath514 moving in an arbitrary external electromagnetic field , described by a four - potential with contravariant components @xmath515 , and covariant components @xmath516 , where @xmath123 and @xmath517 ( for @xmath518 ) are the usual scalar and vector potentials . hamilton s action for this system can be written in the lorentz invariant form @xcite @xmath519 where we use the sign of lanczos @xcite for @xmath3 , opposite to that of landau and lifshitz @xcite . the sign can be chosen arbitrarily ; our choice allows us to use below the orthodox definitions @xmath520 and @xmath521 . the four - dimensional path runs from the initial point @xmath522 to the final point @xmath523 in four - dimensional space - time with corresponding proper times @xmath524 and @xmath525 . here @xmath526 is the infinitesimal interval of the path ( or of the proper time ) @xmath527 , the metric has signature @xmath528 and we use the summation convention and units with @xmath529 , where @xmath530 is the velocity of light . @xmath3 is itself not gauge invariant , but a gauge transformation @xmath531 ( for arbitrary @xmath21 ) adds only constant boundary points terms to @xmath3 , so that @xmath532 is unchanged . the hp is thus gauge invariant . if we introduce a parameter @xmath533 along the four - dimensional path ( proper time @xmath534 along the true or any particular virtual path are valid choices ) the action @xmath3 can be rewritten in the form @xmath535^{1/2 } { d\tau } + e \int\ a_{\alpha}v^{\alpha}{d\tau } , \eqno(6.2)\ ] ] where @xmath536 . in general , @xmath533 is both frame - independent and path - independent . a path - independent parameter is here invaluable for variational purposes , since we then do not have to vary the parameter when we vary the path . after the variations have been performed , we can then choose @xmath537 , etc . the limits of the integrals are the invariants @xmath538 and @xmath539 . with respect to parameter @xmath533 we get the covariant lagrangian @xmath540^{1/2 } + e v_{\alpha}a^{\alpha } , \eqno(6.3)\ ] ] and corresponding conjugate momenta @xmath541^{1/2 } + ea^{\alpha } , \eqno(6.4)\ ] ] and hamiltonian @xmath542 thus in this particular covariant treatment the hamiltonian is trivial . the dynamics is hidden in a constraint . one can see that the four conjugate momenta are not independent variables but satisfy the constraint @xmath543 which is obvious from ( 6.4 ) . this constraint can be taken into account with the help of the lagrange multiplier method , i.e. with the help of an effective hamiltonian @xcite @xmath544 . \eqno(6.7)\ ] ] we easily find that the hamilton equations of motion with @xmath545 are equivalent to @xmath546 ( for @xmath547 , the true trajectory proper time ) and to the covariant lorentz equations of motion for a charged particle in an external field : @xmath548 @xmath549 i.e. the same equations as are found from the lagrangian ( 6.3 ) ( see , e.g. @xcite ) . on the other hand the hamilton equations with @xmath545 are equivalent to the mp or to the rmp with the same @xmath545 . thus one can use any of the four variational principles ( hp , rhp , gmp , rmp ) either with lagrangian ( 6.3 ) or with effective hamiltonian ( 6.7 ) for the derivation of the covariant equations of motion . it is to be noted that in the covariant formulation of relativistic mechanics , the hamiltonian ( a lorentz scalar ) is not equal to the energy ( the time component of a four - vector ) . thus @xmath25 in the maupertuis principle must be interpreted as @xmath550 where @xmath551 . by contrast , in the noncovariant formulation discussed below , the hamiltonian is equal to the energy @xmath552 . there are other , related , formal difficulties arising from ( 6.5 ) . this condition ( @xmath553 ) is due to the fact that the lagrangian @xmath6 in ( 6.3 ) is homogeneous and of first degree in @xmath554 ( i.e. essentially linear in velocities @xmath554 ) , so that @xmath555 and hence the difference @xmath556 ( the hamiltonian ) vanishes @xcite . since @xmath553 , the legendre transform relation ( 6.5 ) between @xmath6 and @xmath7 simplifies to @xmath557 and hence the legendre transform relation between @xmath4 and @xmath3 , i.e. @xmath558 ( where @xmath559 ) , simplifies to @xmath560 because @xmath561 , direct application of the gmp ( @xmath562 would be cumbersome , and direct application of the rmp ( @xmath563 would be impossible . all these difficulties can be traced to the well studied difficulty @xcite of applying an action principle with kinematic @xcite nonholonomic constraints , such as ( 6.6 ) for the momenta , or for the velocities @xmath564 for proper time @xmath534 , or @xmath565 for arbitrary parameter @xmath533 . one way out of the difficulties is to introduce the constraint with a lagrange multiplier for either the lagrangian or hamiltonian ( as is done above for @xmath545 ) . another way out is to exploit the nonuniqueness of the lagrangian . for example @xcite , the covariant lagrangian @xmath6 ( 6.3 ) is equivalent to the covariant lagrangian @xmath566 , where ( with @xmath533 as parameter ) @xmath567 where @xmath568 . it is easily verified that ( 6.12 ) generates the same euler - lagrange equations of motion as ( 6.3 ) . for @xmath537 ( the proper time of the path in question ) , ( 6.12 ) simplifies to @xmath569 which is often employed @xcite . the advantage of ( 6.12 ) and ( 6.13 ) is that they are nonhomogeneous in @xmath554 , and hence will generate a nontrivial corresponding hamiltonian @xmath570 . for ( 6.13 ) , for example , we find @xmath571 note that @xmath570 ( 6.14 ) and @xmath545 ( 6.7 ) ( for @xmath572 ) differ only by a constant . use of the covariant @xmath573 and @xmath570 in place of @xmath6 and @xmath7 removes the difficulties of using all four variational principles ( hp , rhp , gmp , rmp ) . for @xmath6 and @xmath7 , only the first two can be used . note again that @xmath25 is computed from @xmath570 as in ( 6.9 ) , with @xmath570 in place of @xmath545 . now we come to the question of the solution of specific physical problems . in this case explicit covariance usually is not of great importance and it is more useful to find an appropriate convenient frame . for any given frame we can choose @xmath574 ( the coordinate time ) and obtain the complete set of nontrivial quantities @xmath575 @xmath576 @xmath577 , \eqno(6.15)\ ] ] @xmath578 the expression for @xmath6 in ( 6.15 ) is derived from ( 6.3 ) or ( 6.12 ) by choosing @xmath574 , except that we have used the freedom to change the sign of @xmath6 to agree with the standard choice in the noncovariant case @xcite . again we have chosen @xmath17 and @xmath18 . note that the nonrelativistic form @xmath579 for the maupertius action is not valid here , and also note that the hamiltonian @xmath7 is equal to the energy @xmath580 , despite the fact that @xmath581 . as a result the complete set of four variational principles is available in any particular frame . it is matter of taste or practical convenience to use one principle or another . from a practical point of view there is no difference in the usage of variational principles in nonrelativistic mechanics and in relativistic mechanics in a given frame . as an example of the usage of the gmp for relativistic mechanics consider the motion of an electron in a uniform magnetic field @xmath582 . a corresponding vector potential in the coulomb gauge ( @xmath583 ) is @xmath584 , or @xmath585 @xcite . the solution of ( 6.8 ) in this case is well known - the electron moves with constant velocity along the direction of the field ( along the z - axis in our case ) and along a circular orbit in the @xmath586 plane with the cyclotron frequency @xmath587 where @xmath44 is the total energy of the particle including the kinetic energy of the motion in the z -direction . to calculate this frequency from the gmp we take as a trial trajectory @xmath588 where @xmath110 and @xmath349 are free parameters . after some simple algebra we get @xmath589 @xmath590 . \eqno(6.17)\ ] ] in this case it seems that an economical method is to use the gmp ( 2.2 ) . for this purpose one has to rewrite @xmath4 as a function of @xmath25 @xmath591 [ \omega \bar e - \frac{|e|}{2}b ] , \eqno(6.18)\ ] ] and to calculate the variation @xmath592 at fixed @xmath25 . the gmp ( 2.2 ) translates here to @xmath593 . one then easily finds the analytical solution @xmath594 the exact solution corresponds to the case when @xmath492 . for our trial trajectory , we happen to have @xmath492 and the approximate solution coincides with the exact one ( similar to the case of a harmonic oscillator ) . this simple example demonstrates how to use variational principles for relativistic systems . in section 2.1 we mentioned that the classical limit of the schrdinger quantum variational principle is the reciprocal maupertuis principle for periodic or other steady - state motions , i.e. @xmath595 where @xmath596 is the hamiltonian operator corresponding to the classical hamiltonian @xmath35 . here @xmath36 is the period for periodic motions , or @xmath597 for other steady - state motions ( quasiperiodic or chaotic ) . solution of the left - hand side yields @xcite the stationary state @xmath598 with the energy eigenvalue @xmath116 , i.e. @xmath599 the time - independent schrdinger equation . details of the proof of ( 7.1 ) ( which is straightforward only for integrable systems ) are given in @xcite . in ref.@xcite it is explained how schrdinger , in his first paper on wave mechanics in 1926 , in essence worked backwards , i.e. starting from the right hand side of ( 7.1 ) , he derived " the left hand side , and thereby discovered wave mechanics . in his second paper on wave mechanics in 1926 he retracted this argument because his version of the right hand side of ( 7.1 ) was not quite correct . in this second paper he presented an alternative argument as a basis for wave mechanics ( now in the standard textbooks ) based on the analogy between geometric and wave optics on the one hand , and particle and wave mechanics on the other . we consider now the generalization of ( 7.1 ) to nonstationary states @xmath600 , where the true dynamical states satisfy the time - dependent schrdinger equation @xmath601 where @xmath602 , and the initial state @xmath603 is given . for notational simplicity we consider a single particle in three dimensions , so that @xmath604 , and @xmath605 where @xmath606 and @xmath607 = @xmath608 . the arguments which follow are also valid for nonconservative systems , where @xmath609 . we shall consider here the quantum time - dependent variational principle due to frenkel and dirac @xcite for the wave function @xmath600 , but we note the existence of other quantum time - dependent variational principles , e.g. for the time - evolution operator and transition amplitudes @xcite , for the heisenberg operator version @xcite and for the matrix mechanics version @xcite of the hamilton principle , for the density matrix @xcite , for the mean value of an arbitrary observable @xcite , and for time - dependent density functional theory @xcite . ( for completeness , we note that there are also other @xcite quantum time - independent variational principles , in addition to that of schrdinger . ) the original frenkel - dirac variational principle is @xcite @xmath610 for arbitrary variations @xmath611 around the true dynamical state @xmath612 . since ( 7.5 ) is true for all possible variations @xmath613 , then @xmath614 for the true states , as in ( 7.3 ) . by adding ( 7.5 ) and its complex conjugate and using the hermiticity of @xmath596 we can transform ( 7.5 ) to @xcite @xmath615 integrating over an arbitrary but fixed time interval @xmath616 , and assuming @xmath617 is fixed , we find @xcite from ( 7.6 ) @xmath618 note that ( 7.7 ) is not in the traditional form for a variational principle @xmath619 , because of the right hand side . it is more like the unconstrained principles discussed earlier . some authors drop the right hand side without comment @xcite , but it is required if arbitrary variations @xmath613 in @xmath620 are allowed ; otherwise ( 7.7 ) will not generate the schrdinger equation ( 7.3 ) and its complex conjugate @xcite . it is possible to obtain a vanishing right - hand side by restricting the variations @xmath613 in various ways @xcite , e.g. requiring the trial wave functions @xmath600 to be normalized at all times . another way is to write the right hand side as @xmath621 , formally regard @xmath620 and its complex conjugate @xmath622 as independent ( see below ) , and assume @xmath623 ( as well as @xmath603 ) is fixed . however , we find it simplest to retain the general form ( 7.7 ) , with no constraints on @xmath600 except for the fixing of the initial condition @xmath603 . ( when dealing explicitly with bound states later , we assume boundary conditions on @xmath624 . ) by writing out the scalar products @xmath625 in ( 7.7 ) and using the hermiticity of @xmath626 , we can write ( 7.7 ) in field theory form @xmath627 where the lagrangian density @xmath628 is @xmath629 since @xmath620 is a complex function , whose real and imaginary parts can be varied independently , it is well known @xcite that @xmath620 and @xmath630 can be regarded as independent for variational purposes . these two variations in ( 7.8 ) with @xmath628 given by ( 7.9 ) will generate the schrdinger equation and its complex conjugate . following feynman @xcite , we change variables @xmath631 ( or @xmath632 ) in the lagrangian density @xmath633 , where @xmath634 is the amplitude and @xmath635 the phase of the wave function , via the madelung transformation @xcite : @xmath636 where @xmath637 is the probability density . in terms of @xmath395 and @xmath3 we have @xmath638 in terms of @xmath632 the boundary term in ( 7.8 ) is @xmath639 we now take the classical ( @xmath473 ) limit of ( 7.8 ) . using ( 7.11 ) and ( 7.12 ) we get @xmath640 = \int d{\bf r } \rho(t)\delta s(t ) \;\ ; . \eqno ( 7.13)\ ] ] we have assumed that in the limit @xmath473 , @xmath395 and @xmath3 approach well defined limits ( which we continue to denote by @xmath395 and @xmath3 ) . in particular we assume that @xmath641 does not contain an @xmath642 term @xcite . we vary @xmath395 and @xmath3 independently in ( 7.13 ) . varying @xmath395 , with @xmath3 fixed , gives @xmath643 = 0.\ ] ] since @xmath644 is arbitrary , the bracket [ ... ] must vanish , so that @xmath645 this is the classical hamilton - jacobi equation @xcite for the classical action @xmath646 along true paths . next we vary @xmath3 in ( 7.13 ) with @xmath395 fixed . we integrate the @xmath647 term by parts using @xmath648 where we drop a surface term by assuming @xmath649 for @xmath650 . this assumes the corresponding quantum state has finite norm . we integrate the @xmath651 term by parts using @xmath652 where we use @xmath653 which follows from the fixing of @xmath617 . using ( 7.15 ) and ( 7.16 ) in ( 7.13 ) thus gives @xmath654\delta s = 0.\ ] ] since this is true for arbitrary @xmath655 , we get @xmath656 this is the continuity equation , reflecting conservation of total probability , since @xcite @xmath657 is the momentum of the particle . note that if the right hand side of ( 7.13 ) were missing , we would not get the correct continuity equation . in contrast to what we have done here , feynman @xcite keeps the @xmath658 terms in ( 7.11 ) , and the variational principle then leads to quantum generalizations @xcite of the hamilton - jacobi ( 7.14 ) and continuity ( 7.17 ) equations . thus we see that the classical limit of the quantum time - dependent variational principle is the classical variational principle for the hamilton - jacobi and continuity equations . interestingly the classical limit of the quantum time - dependent variational principle leads to a classical variational principle for the hamilton action @xmath3 along the true trajectories , whereas the classical limit of the quantum time - independent variational principle in the form of the left side of ( 7.1 ) leads to a classical variational principle for the trajectories themselves ( rmp -the right side of ( 7.1 ) ) . an alternative form @xcite of the quantum time - independent variational principle , @xmath659 will generate in the classical limit a variational principle for the maupertuis ( or time - independent ) action @xmath660 along the true paths . in ( 7.18 ) @xmath44 is a constant lagrange multiplier introduced to remove the normalization constraint @xmath661 . the extremizing @xmath662 s which satisfy ( 7.18 ) obey @xmath663 so that @xmath44 is the energy eigenvalue corresponding to the eigenfunction @xmath620 . to derive the classical limit of ( 7.18 ) we proceed much as before . ( we assume @xmath596 is time - independent in this section . ) we first apply the madelung transformation to @xmath662 , i.e. @xmath664 where @xmath665 is the amplitude and @xmath666 the phase . following similar steps as above , we find ( 7.18 ) becomes in the classical ( @xmath473 ) limit @xmath667 = 0 \;\ ; , \eqno ( 7.21)\ ] ] where @xmath668 . varying @xmath669 in ( 7.21 ) with @xmath666 fixed gives @xmath670 the time - independent hamilton - jacobi equation @xcite for the maupertuis action @xmath671 along true paths . if we vary @xmath666 in ( 7.21 ) with fixed @xmath669 we get the continuity equation @xmath672 since @xcite @xmath673 is the momentum of the particle . in ref.@xcite , we show how the classical limit of ( 7.18 ) can be taken in a different way ( using standard 1d wkb wave functions ) and for a 1d periodic orbit leads to @xmath674 , i.e. the uhp for a periodic orbit of period @xmath36 . we note that , just as for the right hand side of ( 7.1 ) , the derivation of the classical limits ( 7.14 ) and ( 7.22 ) is somewhat formal for the chaotic motions of nonintegrable systems , since classical values of @xmath3 and @xmath4 may not exist as well defined functions for such systems . unlike the nonrelativistic limit ( @xmath675 ) , which is regular , the limit @xmath676 or @xmath677 is singular , which makes the classical limit subtle and interesting @xcite . we summarize the two groups of classical variational principles reviewed in this paper . first we have the maupertuis principles : @xmath678 the first of these is the unconstrained form ( ump ) , the second the general maupertuis principle ( gmp ) , and the third is the reciprocal maupertuis principle ( rmp ) . by legendre transformation of the principles in ( 8.1 ) we obtain the hamilton principles : @xmath679 the first is the unconstrained form ( uhp ) , the second the hamilton principle ( hp ) , and the third the reciprocal hamilton principle ( rhp ) . in all these principles , the end - positions of the trajectory @xmath23 and @xmath24 are held fixed , and the notation indicates what else is fixed and what is varied . we have given a number of examples showing how to use these principles to solve practical problems , focussing particularly on the three maupertuis principles . both classical and semiclassical applications have been given , and some comparisons are made with the results from quantum variational principle calculations . the classical limits of two quantum variational principles ( time - dependent and time - independent ) are given . in particular , the rmp is the classical limit of the schrdinger time - independent quantum variational principle . the rmp is thereby naturally adapted to semiclassical applications in the applications , the variational principles ( both classical and quantum ) are solved by the direct variational method , where one guesses a trial solution containing adjustable parameters , and adjusts the parameters to satisfy the variational principle . the euler - lagrange differential equation is not required in this method . when we choose a simple trial solution , the results can be found simply and analytically . using the direct variational method the rmp and ump have been used to estimate semiclassically the energy levels in several examples . the levels @xmath680 can be expressed in analytical form , and the accuracy is quite acceptable in most cases , even for nonintegrable systems which are usually treated by much more complicated semiclassical methods , e.g. the gutzwiller trace formula @xcite , or quantizing the birkhoff - gustavson normal form @xcite . it would be useful to have comparisons of the results from the maupertuis principle - based method with those from the latter two and other methods for a number of specific problems . the methods should be compared for accuracy of results and computational complexity . some limited comparisons of the maupertuis principle and normal form methods have been done for the @xmath681 oscillator of sec . 4.1 in refs . @xcite , and limited comparisons for this model of results from the maupertuis principle and the semiclassical adiabatic methods @xcite are presented in table 1 of sec . 4.1.1 , but much more extensive comparisons would be of interest . variational principles appeal to various individuals for differing reasons , both aesthetic @xcite and practical . we have stressed the utility for classical and semiclassical particle dynamics calculations in this review . engineers too make heavy use of variational principles in solving practical problems of classical continuum mechanics @xcite . on rare occasions variational principles have even led to new laws of physics @xcite . we gratefully acknowledge the natural sciences and engineering research council of canada for financial support of this work . we thank donald sprung for a careful reading of the manuscript , david garfinkle for helpful remarks on relativistic mechanics and stepan bulanov for help with tex . ... since @xmath682 is the variation of @xmath201 with @xmath186 kept constant , we could conveniently write it as @xmath683 , in accordance with a convention used in thermodynamics ... it is curious that in spite of the obvious need in partial differentiation for precise statement of what is being kept constant , such statement is not embodied in customary notation of pure mathematics , though it is provided in thermodynamics , ... " _ h.jeffreys and b.s.jeffreys + methods of mathematical physics + cambridge u.p.,1946 . _ variational problems ask for the function which makes stationary an integral involving the function and its derivatives . usually there is a constraint which is fixed . for example , we are to find a function @xmath684 , describing a closed planar curve , of fixed length @xmath6 , which maximizes the area @xmath349 enclosed by the curve @xmath684 . @xmath349 and @xmath6 are functionals of @xmath684 : @xmath685 , l = l[y(x)]$ ] . for our purpose here , we do not need the precise form of @xmath349 or @xmath6 . @xmath6 is the constraint and @xmath349 is the functional to be varied . we use the following notation to describe this variation @xmath686 where the constraint @xmath6 is indicated as a subscript on the bracket enclosing the variation @xmath687 of the functional @xmath349 . the variation @xmath687 is first - order in @xmath682 , and is called the first variation when higher - order variations are considered @xcite . the constraint @xmath6 is denoted explicitly in ( a1.1 ) because it is the most important , or the one we wish to focus on . there may be other constraints which are left implicit , e.g. that the planar curve pass through a given point ( @xmath688 ) . in this paper we always leave fixed end - position constraints on curves ( closed or open ) implicit . the solution of the problem ( a1.1 ) is the function @xmath684 , which satisfies a differential equation , the euler - lagrange equation . the reciprocal variational problem has the constraint @xmath6 and the varied functional @xmath349 interchanged . in other words , it is described by the equation @xmath689 the two equations , ( a1.1 ) and ( a1.2 ) , @xmath684 . the reciprocal variation is searching for the function @xmath684 which makes stationary ( here minimizes ) the length @xmath6 of a closed curve with fixed enclosed area @xmath349 . the common solution for both problems is a circle , as is clear intuitively . one way to prove reciprocity is to use a trick , due to lagrange , called the method of lagrange multipliers . the problem ( a1.1 ) is first restated in unconstrained form as @xmath690 where the variations of both @xmath349 and @xmath6 are unconstrained and @xmath301 is an undetermined multiplier . to obtain ( a1.2 ) from ( a1.3 ) , we simply specify that we now fix @xmath349 ( i.e. set @xmath691 ) and vary @xmath6 , i.e. @xmath692 . this proves the reciprocity theorem . reciprocity is clear intuitively in the isoperimetric problem ( a1.1 ) . an example of reciprocity from physics is the gibbs conditions for equilibrium in thermodynamics @xcite . the reciprocal conditions @xmath693 and @xmath694 are both conditions for thermal equilibrium , where @xmath3 is the entropy and @xmath695 the internal energy . these express the facts that the state of thermal equilibrium is that state which maximizes the entropy for fixed internal energy , and also is that state which minimizes the internal energy for fixed entropy . ( the unconstrained form @xcite is @xmath696 , where the lagrange multiplier is the temperature @xmath36 . here the implicit constraints in all cases are fixed system volume @xmath34 and mole numbers @xmath697 ) reciprocal vp s in mechanics have only recently been studied @xcite - see sec.2 . this is all we shall need from the calculus of variations ; for more details one should consult a text - book @xcite . in sec . 2.2 , for conservative systems , we discussed the equivalence of the hamilton principle ( hp ) @xmath698 with the general maupertuis principle ( gmp ) @xmath699 here we show directly the equivalence of the gmp ( a2.2 ) with the original maupertuis principle ( mp ) @xmath700 in all these principles the end - positions @xmath23 and @xmath24 are held fixed . since ( a2.1 ) , ( a2.2 ) and ( a2.3 ) all generate the same ( euler - lagrange ) equations of motion , it is clear they must be equivalent , and it is interesting to show directly the equivalence of ( a2.2 ) and ( a2.3 ) . we must show that relaxing the constraint of fixed @xmath44 in ( a2.3 ) to fixed @xmath25 in ( a2.2 ) does not generate spurious or unphysical solutions . the constraint of fixed @xmath701 in ( a2.3 ) applies along and between the trial trajectories . to relax this constraint , we use the method of lagrange multipliers @xcite . since @xmath702 , we have @xmath703 , and hence @xmath704 where @xmath301 is an arbitrary constant . the trajectories all start at @xmath316 at fixed point @xmath349 and end at @xmath445 ( which may vary with the trial trajectory ) at fixed point @xmath705 . we release the constraint of fixed @xmath44 in ( a2.3 ) and choose @xmath301 such that where @xmath710 , and where we recall that the end - time @xmath36 is not fixed in ( a2.3 ) . recalling the definition @xmath711 enables us to write ( a2.5 ) as @xmath712 where we have used the fact that @xmath713 the true trajectory . expanding ( a2.6 ) gives to see that @xmath717 in ( a2.8 ) , we apply ( a2.8 ) to the special case of two true trajectories , with actions @xmath4 and @xmath718 and energies @xmath44 and @xmath719 . for this case ( a2.8 ) reads @xmath720 . comparison of this with the standard relation @xcite @xmath721 gives @xmath722 . hence we have thus ( a2.2 ) follows from ( a2.3 ) . the converse , the derivation of ( a2.3 ) from ( a2.2 ) is easy ; we simply restrict the trial trajectories in ( a2.2 ) to those of fixed energy @xmath44 , since the equations of motion following from ( a2.2 ) imply energy conservation . hence ( a2.2 ) and ( a2.3 ) are equivalent . nesbet r k 2003 _ variational principles and methods in theoretical physics and chemistry _ ( new york : cambridge u.p . ) ; l.r . ram - mohan l r 2002 _ finite element and boundary element applications in quantum mechanics _ ( oxford : oxford u.p . ) ; drake g w f 1998 _ encyclopedia of applied physics _ ( new york : wiley - vch ) * 23 * 121 ; gerjuoy e , rau a r p and spruch l 1983 _ rev.mod.phys . _ * 55 * 725 ; epstein s t 1974 _ the variation method in quantum chemistry _ ( new york : academic ) . quantum vp s for scattering states are reviewed in adhikari s k 1998 _ variational principles for the numerical solution of scattering problems _ ( new york : wiley ) . classical vp s for scattering are discussed in ref . @xcite . for example , in yourgrau and mandelstam @xcite the notation is reversed . hamilton himself used the current notation , but , for the actions along true paths , termed @xmath3 and @xmath4 the principal and characteristic functions , respectively . there is a recent exception : papastavridis j g 2002 _ analytical mechanics _ ( new york : oxford u.p . ) , p. 1044 , where the general maupertuis principle is discussed . this book contains a masterly and historical account of the various variational principles . in this review we focus on global ( integral ) vp s . we do not discuss the various local ( differential ) vp s due to dalembert , gauss , hertz , gibbs , jourdain and others @xcite . jourdain s principle , and its connections to others , is reviewed by wang l - s and pao y - h 2003 _ am . j. phys _ * 71 * 72 yourgrau w and mandelstam s 1968 _ variational principles in dynamics and quantum theory _ 3rd edition ( london : pitman , philadelphia : w.b.saunders ) , reprinted , dover , new york , 1979 ; terral m 2002 _ the man who flattened the earth _ ( chicago : university of chicago press ) c.g.j.jacobi 1884 _ vorlesungen ber dynamik _ , zweite ausgabe , herausgegeben von a.clebsch ( braunschweig/ wiesbaden : f.vieweg ) , reprinted by chelsea 1969 , p 44 . the quotation in this text is more explicit than @xcite : it names the textbooks of poisson , lagrange and laplace as some of the best ! percival i c 1977 _ adv . phys . _ * 36 * 1 , and references therein van brunt b 2004 _ the calculus of variations _ ( new york : springer ) ch 10 karl g and novikov v a 1995 _ phys . * 51 * 5069 ; 1995 _ j. exp . phys . _ * 80 * 783 holonomic ( coordinate ) constraints are assumed to be taken into account by the choice of the @xmath20 . nonholonomic ( velocity ) constraints ( see note @xcite ) are assumed absent in the early sections but are considered briefly in sec.6 and app.ii . we also exclude quasi - coordinates ( or nonholonomic coordinates ) , e.g. for a rigid body , the three rotation angles around space - fixed @xmath724 axes . vp s using such variables are discussed by papastavridis @xcite , and by marsden j e and ratiu t s 1994 _ introduction to mechanics and symmetry _ ( new york : springer ) p 390 . a seminal paper is poincar h 1901 _ c.r.acad.sci . _ * 132 * 369 the condition @xmath725 is valid for normal systems , for which @xmath31 , the holonomic constraints ( if any ) are time - independent and the reference frame is inertial . if these conditions are satisfied @xmath32 is quadratic in the @xmath726 s . note that the single condition @xmath32 quadratic in the @xmath726 s is not sufficient to have @xmath725 since for some systems with time - dependent constraints @xmath32 is quadratic in the @xmath726 s : see , e.g. de castro a s 2000 _ eur . j. phys . _ * 21 * 23 ; ferrario c and passerini a 2002 _ eur . j. phys . _ * 22 * 11 . percival i c 1974 _ j. phys . _ a * 7 * 794 ; ibid 1979 * 12 * 157 ; see also klein a and li c - t 1979 _ j. math . phys . _ * 20 * 572 . percival later extended his vp for invariant tori to cantori : 1980 _ amer . _ * 55 * 302 . a cantorus is , roughly speaking , a torus - shaped surface with holes such that what remains forms a cantor set of points . a trajectory confined to a torus surface is quasiperiodic . a trajectory starting on a cantorus will slowly leak away " because of the holes , hence is not confined to the surface so that the quasiperiodicity is destroyed in the long run , and is thus weakly chaotic . see arnold et al @xcite , p.194 , for a precise definition . also see percival i c 1991 in _ chaos and quantum physics giannoni m - j , voros a and zinn - justin j 1991 ( amsterdam : north - holland ) p 11 goldstein et al , ref . @xcite , p 597 ; lanczos , ref . @xcite , p 402 ; oliver d 1994 _ the shaggy steed of physics _ ( new york : springer ) p 39 ; schwinger j , de raad l l jr . , milton k a and tsai w - y 1998 _ classical electrodynamics _ ( reading ma : perseus books ) p 90 the unconstrained hamilton principle ( 2.8 ) is closely related to what hamilton termed the law of varying action ( see landau and lifshitz @xcite , p.134 ) , and also to what some authors refer to as the weiss action principle : see sudarshan e c g and mukunda n 1974 _ classical dynamics _ ( new york : wiley ) , reprinted by krieger 1983 , p 16 . if end - position variations are allowed , as well as end - time variations , the uhp ( 2.8 ) generalizes to @xmath730 , where @xmath731 is the true value of @xmath49 at end - point @xmath705 , etc , and for conservative systems @xmath732 . ( this generalization can be derived from the general first variational theorem @xcite of variational calculus , which includes end - point variations - see appendix a of @xcite for a related discussion of the ump . ) what hamilton termed the law of varying action is the special case for the variation between two true trajectories , so that @xmath733 and @xmath734 are exact differentials . this particular analogy can be pushed further . just as there are numerous other thermodynamic potentials besides the helmholtz free energy @xmath735 and the gibbs free energy @xmath736 , ( where @xmath49 is the pressure and @xmath34 the volume ) , e.g. the internal energy @xmath737 , ( where @xmath36 is the temperature and @xmath3 the entropy ) , we can introduce in mechanics further actions via further legendre transformations . for example , define @xmath738 , where @xmath24 is the final end - point position and @xmath731 the true momentum at @xmath705 . if we relax the constraint of fixed @xmath24 , the uhp generalizes further to @xmath739 . in terms of @xmath740 this transforms to @xmath741 . in terms of @xmath740 , the usual fully constrained hp is @xmath742 , replacing @xmath743 . here , as always , the constraint on @xmath23 is still left implicit in both . further legendre ( using @xmath744 and/or @xmath4 ) and reciprocal transformations are obvious . in this review we focus on the four basic variational principles summarized at the end of sec.2.2 . szamosi g 1973 _ found . phys . _ * 3 * 241 ; peterson m a 1979 _ am . _ * 47 * 488 ; toffoli t in 1999 _ feynman and computation _ ed . hey a j g ( reading ma : perseus books ) p 349 ; bailyn m 1990 _ a survey of thermodynamics _ ( new york : aip press ) p 416 . in particular , authors interested in linking the principle of least action to the second law of thermodynamics include ( see bailyn ) boltzmann , clausius , helmholtz , ehrenfest @xcite and planck : see also , e.g. , planck m 1915 _ eight lectures on theoretical physics _ ( new york : columbia univ . press ) , reprinted by dover , 1998 , p 97 ; thomson j j 1888 _ applications of dynamics to physics and chemistry _ ( london : macmillan ) reprinted by dawson , 1968 ; fnyes i 1952 _ zeit . f. phys . _ * 132 * 140 , and de broglie l 1995 _ diverses questions de mcanique et de thermodynamique classiques et relativistes _ ( berlin : springer ) gutzwiller m c 1990 _ chaos in classical and quantum mechanics _ ( new york : springer ) ; child m s 1991 _ semiclassical mechanics with molecular applications_(oxford : oxford u.p . ) ; brack m and bhaduri r k 1997 _ semiclassical physics _ ( reading ma : addison - wesley ) euler first described the method of discretizing time in the action integral , treating the @xmath745 as discrete variables , and obtaining discretized ( finite difference ) versions of the euler - lagrange ( and newton ) equations , to be solved with the two end - positions and the duration fixed . this method has also been much used in recent years : see , e.g. , percival 1991 @xcite ; beck t l , doll j d and freeman d l 1989 _ j. chem . phys . _ * 90 * 3181 ; basille a g and gray c g 1992 j.comp.phys . * 101 * 80 ; marsden j e and west m 2001 _ acta numerica _ * 10 * 357 ; monforte j c 2002 _ geometric , control and numerical aspects of nonholonomic systems _ ( new york : springer ) p 143 ; elber r , crdenas a , ghosh a and stern h a 2003 _ adv . phys . _ * 126 * 93 . courant r ( bull . * 49 * , 1 ( 1943 ) ) was one of the first to suggest spacial discretization of variational functionals involving fields @xmath746 , a precursor to the modern finite element method ( ram - mohan , ref.@xcite ) which uses a basis of localized trial functions . unfortunately , in some good textbooks the authors misapply the original mp @xmath747 for 1d problems by employing trial trajectories which do not conserve the energy @xmath44 : see , e.g. , tabarrok and rimrott @xcite , p. 180 , or park d 1990 _ classical dynamics and its quantum analogues _ 2nd ed.(new york : springer ) p 80 . olsson m g 1978 _ am . _ * 46 * 1118 ; ibid 1981 * 49 * 531 ; synge j l and griffiths b a 1959 _ principles of mechanics _ ( new york : mcgraw hill ) p 335 ; routh e j 1898 _ a treatise on the dynamics of a particle _ ( cambridge : cambridge u.p . ) , reprinted by dover 1960 p 349 ; airy g b 1851 _ phil.mag . _ series 4 * 2 * 141 . in the second edition of _ mcanique analytique _ ( 1815 ) , lagrange gave an incorrect result for eq ( 3.11 ) . for an english translation of the second edition see lagrange j l 1997 _ analytical mechanics _ , eds . boissonnade a and vagliente v n ( dordrecht : kluwer ) p 462 . the error was corrected by bertrand , the editor of the third edition ( 1855 ) , following a communiqu by bravais . the rmp method yields @xmath748 and hence after semiclassical quantization @xmath749 , the same as ( 4.7 ) found by the fully quantum vp . since @xmath750 is therefore rational , the optimum quantized trajectories are periodic , at least at this level of approximation . periodic trajectories also play a key role in gutzwiller s trace formula method of semiclassical quantization @xcite . rajan m and junkins j l 1983 _ int . j. nonlinear mech . _ * 18 * 335 bukta d , karl g and nickel b g 2000 _ can . j. phys . _ * 78 * 449 szeredi t and goodings d a 1992 _ phys . rev . lett . _ * 69 * 1640 ; ibid . 1993 rev._e * 48 * 3529 burnell f , mann r b and ohta t 2003 _ phys . lett . _ * 90 * 134101 pauli w 1973 _ pauli lectures in physics _ vol.5 ( cambridge : mit press ) shinohara h 2000 _ rep . phys . _ * 63 * 843 ; eletskii a v 2000 _ uspekhi _ * 43 * , 111 dresselhaus m s , dresselhaus g and eklund p s 1996 _ science of fullerenes and carbon nanotubes _ ( san diego : academic press ) cioslowski j 1995 _ electronic structure calculations on fullerenes and their derivatives _ ( new york : oxford u.p . ) joslin c g , gray c g , poll j d , goldman s and buckingham a d 1995 _ collision and interaction - induced spectroscopy _ eds tabisz g c and neuman m n nato asi ser . c * 452 * ( dordrecht : kluwer ) p 261 joslin c g , yang j , gray c g , poll j d and goldman s 1995 _ lectures on thermodynamics and statistical mechanics _ eds castas m , rodriguez r and benavides a l ( singapore : world scientific ) p 146 complexes xc@xmath751 , where x is on the of the cage , have also been studied , e.g. h@xmath752c@xmath753 @xcite and h@xmath752c@xmath754 ( nanotubes ) @xcite . complexes such as k@xmath755 c@xmath753 are superconductors in the solid state @xcite . joslin c g , gray c g , goddard j d , goldman s , yang j and poll j d 1993 _ chem . lett . _ * 213 * 377 ; joslin c g , yang j , gray c g , goldman s , and poll j d 1993 _ chem . lett . _ * 208 * 86 ; joslin c g , yang j , gray c g , goldman s , and poll j d 1993 _ chem . lett . _ * 211 * 587 ; joslin c g , gray c g , goldman s , yang j and poll j d 1993 _ chem . lett . _ * 215 * 44 joslin c g , gray c g and goldman s 1995 _ chem . phys.lett._ * 244 * 93 ; ibid 1996 * 251 * 414 ( e ) lee - dadswell g r and gray c g 1995 _ can . j. phys . _ * 78 * 599 lee - dadswell g r and gray c g 2003 _ can . _ submitted mandziuk m and bacic z 1994 _ j. chem . _ * 101 * 2126 joslin c g , gray c g and goldman s 1994 _ chem . phys.lett._ * 227 * 405 for li@xmath756@c@xmath753 and lih@c@xmath753 , anisotropic perturbations of the isotropic potential , and the consequent chaotic motions , have been studied by hernndez - rojas et al . yakobson b i and smalley r e 1997 _ american scientist _ * 85 * 324 bug a l r , wilson a and voth g a 1992 _ j. phys . chem . _ * 96 * 7864 ; cardini g , procacci p , salvi p r and schettino v 1992 _ chem . lett . _ * 200 * 39 ; li y s and tomnek d 1994 _ chem . lett . _ * 221 * , 453 the potential @xmath757 can be fit @xcite to simple analytic forms(polynomial , morse , etc ) . in some cases ( e.g. li@xmath756@c@xmath753 ) the potential has its minimum off - centre . for k@xmath756@c@xmath753 the potential is essentially @xmath758 , a rare example of a pure quartic oscillator in nature . lee - dadswell g r 1999 _ m.sc.thesis _ university of guelph , p 58 leubner c and p. krumm p 1990_eur . j. phys . _ * 11 * 31 ; anderson n and arthurs a m 1997 _ eur . j. phys . _ * 18 * 404 ; kobe et al . @xcite ; del valle g , campos i and jimnez j l 1996 _ eur . _ * 17 * 253 bateman h 1931 _ phys . rev . _ * 38 * 815 ; havas p 1957 _ nuovo cimento suppl . _ * 5 * 363 ; leitmann g 1963 _ j.appl.mech . _ * 30 * 623 ; denman h h 1966 _ am . j. phys . _ * 34 * 1147 ; ray j r 1974 _ am . _ * 47 * 626 ; greenberger d m 1979 it j. math * 20 * 762 ; kobussen j 1979 _ acta phys . aust . _ * 51 * 293 ; bahar l y and kwatny h g 1981 _ am . j. phys . _ * 49 * 1062 ; dekker h 1981 _ phys.rep . _ * 80 * , 1 ; lemos n a 1981 _ am . j. phys . _ * 49 * 1181 ; kobe d h , reali g and sieniutycz s 1986 _ am . j. phys . _ * 54 * 997 ; huang y - s and lin c - l 2002 _ am . _ * 70 * 741 ; vujanovic b d and jones s e 1989 _ variational methods in nonconservative systems _ ( boston : academic press ) ; edelen d g b 1977 _ lagrangian mechanics of nonconservative nonholonomic systems _ ( leyden : noordhoff ) ; soper d e 1976 _ classical field theory _ ( new york : wiley ) p 126 ; santilli r m 1978 _ foundations of theoretical mechanics _ vol . 1 ( new york : springer ) pp . 183 , 199 , 206 ; logan j d 1977 _ invariant variational principles _ ( new york : academic press ) p 56 . these workers go beyond the rayleigh dissipation function method @xcite , and incorporate the friction into the lagrangian ( or hamiltonian ) . for example , for a harmonic oscillator with linear damping ( equation of motion @xmath759 ) , one easily verifies that a suitable lagrangian is @xmath760 $ ] , and the corresponding hamiltonian is @xmath761 . buch l h and denman h h 1976 _ phys . lett . a _ * 55 * 325 . see also yourgrau and mandelstam @xcite , p.47 , who note essentially ( 5.3 ) without remarking explicitly that it is valid for nonconservative systems . similarly see landau and lifshitz @xcite , p.139 , routh @xcite , p.301 , and boltzmann @xcite , p. 121 , where the nonconservative aspect is more explicit . the form @xmath762 for the ump , which is easily shown to be equivalent to ( 5.4 ) , essentially appears in passing in the equations of yourgrau and mandelstam @xcite , p.46 , but they do not note explicitly its validity for nonconservative systems . for the special case of periodic systems evolving adiabatically with parameter @xmath295 , ehrenfest @xcite gives a relation related to ( 5.4 ) ; he has @xmath763 rather than @xmath764 ( as in the original mp ) , and our @xmath765 is replaced by a @xmath766 term . ehrenfest attributes his result to boltzmann and clausius . as before , in the ump ( 5.4 ) we have fixed the initial time @xmath17 and allow the final time @xmath767 to vary . if both @xmath768 and @xmath769 are allowed to vary , ( 5.4 ) generalizes to @xmath770 \delta t_b + [ e(t_a)- \bar e ] \delta t_a = 0 $ ] . this reduces to ( 5.4 ) for fixed @xmath768 , and to ( 2.10 ) for conservative systems ( @xmath771 ) . further generalizations , allowing @xmath24 and/or @xmath23 to vary , are discussed in notes @xcite if we start the cycle at the moving wall , for example , instead of the fixed one , the action @xmath772 is not an exact invariant . see crawford f s 1990 _ am.j.phys _ * 58 * 337 ; gignout c and brut f 1989 _ am.j.phys . _ * 57 * , 422 ; percival i and richards d 1982 _ introduction to dynamics _ ( cambridge : cambridge u.p . ) p 142 ; lochak and meunier @xcite , p.18 ; brack and bhaduri @xcite , p. 10 . lichtenberg a j and lieberman m a 1992 _ regular and chaotic dynamics _ 2 ed ( new york : springer ) p 57 and 216 . krainov v p 2002 _ selected mathematical methods in theoretical physics _ ( london : taylor and francis ) p 143 . hannay j h 1985 _ j.phys.a _ * 18 * 221 ; berry m v 1985 _ j.phys.a _ * 18 * 15 . these papers and others on the berry phase are reprinted in shapere a and wilczek f eds 1989 _ geometric phases in physics _ ( singapore : world scientific ) shapere a and wilczek f , ref . [ 30 ] , p 3 ; marsden j e and ratiu t s , ref . @xcite , p 44 ; khein a and nelson d 1992 _ am . j. phys . _ * 61 * 170 ; kugler m 1989 _ am . j. phys . _ * 57 * 247 ; montgomery r 1988 _ comun . phys . _ * 120 * 269 ; greenberger d m 1979 _ j. math . phys . _ * 20 * 762 ; koiller j 1989 _ contemp . _ * 97 * 159 grimsehl e 1932 _ a textbook of physics _ vol . 1 ( london : blackie and son ) p 165 ; sommerville w b 1972 _ quart ; hart j b , miller r e and mills r l 1987 _ am . j. phys . _ * 55 * 67 the converse can be made plausible : from an assumed adiabatic invariance of the long path @xmath4 , the adiabatic invariance of @xmath140 follows . we use @xcite , for @xmath433 , @xmath773 , where @xmath774 is the projected area of curve @xmath357 on the @xmath775 plane , and the fact that the dynamical invariance ( poincar invariant ) of @xmath776 can be derived @xcite from the dynamical invariance of @xmath777 the adiabatic invariance of long - path @xmath4 s for ergodic systems follows heuristically from the ump ( 5.4 ) as described in the text . alternatively , we can heuristically argue that for @xmath597 , the long path is essentially closed in phase space , so that the poincar dynamical invariant @xmath778 remains essentially adiabatically constant as the energy surface and the path on it slowly evolve . obviously the limiting procedures are delicate and problematic . even if our conjecture is true , the adiabatic invariance may appear to be of the seemingly uninformative type @xmath779 , since @xmath780 for @xmath433 . to extract finite results , @xmath36 will need to be taken as , say , the poincar cycle time . we note that , in another situation where @xmath780 as @xmath433 , i.e. scattering orbits , useful results can still be extracted @xcite . the arguments in the above paragraph , as well as the one in the text , apply also more generally to the chaotic motions of nonergodic nonintegrable systems . since adiabatic invariants have not been found previously for such systems , it would be of interest to prove , or disprove , these suggestions . if proven , it would then be of interest to establish a general semiclassical quantization scheme for nonintegrable systems based on @xmath4 ( generalizing ( 5.32 ) which holds for ergodic systems ) . hertz p 1910 _ ann.der phys . _ * 33 * 225 , 537 . hertz s proof of the adiabatic invariance of the phase volume @xmath434 made explicit what was implicit in the statistical mechanical works of boltzmann , gibbs , and others . ( recall the boltzmann relation for the entropy @xmath781log@xmath434 ; reversible adiabatic processes ( @xmath782 ) correspond to constant @xmath434 . ) hertz s argument is available in a few textbooks : kahan t 1960 _ physique thoretique _ tome premier , vol . ii , ( paris : presses universitaires ) p 482 ; becker r 1969 _ theory of heat _ ( new york : springer ) p 131 ; mnster a 1969 _ statistical thermodynamics _ , vol.i ( berlin : springer ) p 52 ; toda m , kubo r and saito n 1983 _ statistical physics i _ ( berlin : springer ) p 48 . for a rigorous proof , see kasuga t 1961_proc . japan acad . _ * 37 * 366 , 372 , 377 : see also lochak p and c.meunier c 1988 _ multiphase averaging for classical systems _ ( new york : springer ) p 239 . reinhardt w p 1994 _ prog . _ * 116 * 179 , and references therein skodje r t and cary j r 1988 _ computer phys . _ * 8 * 221 berry m v , in 1985 _ theoretical and applied mechanics _ eds . f.i.niordson and n.olhoff ( amsterdam : elsevier ) p 83 , and references therein brown r , ott e and grebogi c 1987 _ phys . lett . _ * 59 * 1173 ehrenfest p 1917 _ phil.mag . _ * 33 * 500 . reprinted in _ sources of quantum mechanics _ ed . van der waerden b l 1967 ( amsterdam : north holland ) , reprinted by dover 1968 , p 79 . an extended version of this paper is available in _ paul ehrenfest collected scientific papers _ , klein m j 1959 ( amsterdam : north holland ) p 378 . later ehrenfest s student j.m.burgers showed that the one - cycle ( or torus ) actions @xmath783 are adiabatic invariants for multidimensional quasiperiodic systems ( burgers j m 1918 _ proc.amsterdam acad . _ * 20 * 163 ; for a textbook treatment , see landau and lifshitz @xcite , p 161 . ) the subtle conditions ( concerning the nondegeneracy of the system frequencies @xmath784 during the adiabatic evolution ) required for the @xmath783 to be adiabatic invariants are discussed by dirac p a m 1925 _ proc . soc . _ * 107 * 725 and von laue m 1925 _ ann . der phys . _ * 76 * 619 the cumulative number of states up to @xmath44 , @xmath785 , is defined quantum mechanically by @xmath786 , where @xmath596 is the hamiltonian operator and tr denotes a trace . tomsovic s 1991 _ j.phys . _ a * 24 * 733 mcrae s m and vrscay e r 1992 _ j. math . phys . _ * 33 * 3004 ; mckinley w a 1971 _ am . j. phys . _ * 39 * 405 ; weisskopf v 1934 _ zeit.f.phys . _ * 89 * 27 . for periodic systems ehrenfest @xcite showed that the classical hellmann - feynman theorem is equivalent to the adiabatic invariance of the action @xmath4 . ehrenfest ( who was boltzmann s student ) attributes this result to boltzmann and clausius . the classical theorem can also be derived ( epstein @xcite ) from classical perturbation theory , using the result ( ter haar @xcite , p.156 ) that the first - order energy shift @xmath787 due to a small change in the hamiltonian @xmath477 by the amount @xmath788 is @xmath789 . our derivation , and that of epstein @xcite , allow @xmath301 to occur anywhere in the hamiltonian . other explicit published derivations assume @xmath301 occurs in the potential . implicit derivations are contained in a number of places : van vleck j h 1926 _ quantum principles and line spectra , bull . nat.res . council _ , 10 part 4 , p 205 ; becker @xcite , p 129 ; mnster @xcite , p51 ; landau and lifshitz @xcite , p 155 . the classical and quantum hellmann - feynman relations can be generalized to relations ; for systems in thermal equilibrium we find @xmath790 , where @xmath735 is the helmholtz free energy and @xmath791 denotes a ( classical or quantum ) canonical ensemble average . see gubbins k e and gray c g 1984 _ theory of molecular fluids _ ( oxford : oxford u.p . ) p 236 , and references therein . thus in statistical mechanics we do have @xmath792 in the ensemble ; the correct relation is @xmath793 $ ] , where @xmath794 , with @xmath795 boltzmann s constant and @xmath36 the temperature . ( in the ensemble , where @xmath796 because of the absence of energy fluctuations , we do find @xmath797 , both classically and quantum mechanically . ) hellmann h 1937 _ einfhrung in die quantenchemie _ ( vienna : denticke ) ; feynman r p 1939 _ phys . rev . _ * 56 * 340 ; see also van vleck j h 1928 _ phys . rev . _ * 31 * 587 , and pauli w 1933 _ handbuch der physik _ * 24 * , part 1 , eds . geiger h and scheel k ( berlin : springer ) , translated edition 1980 _ general principles of quantum mechanics _ ( berlin : springer ) p 86 . for a review of other derivations , applications and extensions , see epstein s t 1981 , in _ the force concept in chemistry _ ( new york : van nostrand reinhold ) ed . deb b m , p 1 . the condition @xmath553 also occurs nonrelativistically if a parameter @xmath533 is used in place of @xmath16 ; see , e.g. , goldstein et al @xcite , p.320 , or lanczos @xcite , p. 186 . doughty n a 1990 _ lagrangian interaction _ ( reading ma : addison - wesley ) p 416 ; leubner c 1986 _ eur . j. phys . _ * 7 * 17 ; johns o d 1985 _ am . _ * 53 * 982 dirac p a m 1964 _ lectures on quantum mechanics _ ( new york : yeshiva university and academic press ) p 9 ; arnold v i , kozlov v v and neishtadt a i 1988 in _ dynamical systems iii _ ed . arnold v i ( berlin : springer ) p 39 nonholonomic constraints @xmath798 contain the velocities @xmath799 , such that the constraints can not be integrated up to global coordinate constraints @xmath800 . nonholonomic constraints arise from introducing redundant variables as in sec.6 on relativistic mechanics , or by introducing constants of the motion as constraints @xcite as in app.ii for the mp , or by imposing the kinematic constraint between @xmath33 and the velocity or the momentum after first having released the constraint @xcite . nonholonomic constraints arise from physical forces of constraint , such as arise in rolling motion @xcite . kinematic nonholonomic constraints can be handled by an effective lagrangian @xmath801 , where @xmath802 are lagrange multipliers @xcite . dynamic nonholonomic constraints can not be handled by such an effective lagrangian ( contrary to what is stated in some excellent textbooks ) and construction of an hp or mp is highly problematic @xcite . papastavridis @xcite gives a clear discusion and references to the historical literature ; tabarrok b and rimrott f p j 1994 _ variational methods and complementary formulation in dynamics _ ( dordrecht : kluwer ) p 142 ; rund h 1966 _ the hamilton - jacobi theory in calculus of variations _ ( new york : van nostrund reinhold ) , reprinted by krieger 1973 , p 352 ; pars l a 1965 _ a treatise on analytic dynamics _ ( cambridge : cambridge u.p . ) , reprinted by ox bow press , 1979 , p 528 ; rumiantsev v v 1978 _ j. appl . math . and mech . _ * 42 * , 407 ; ibid . 1979 * 43 * 625 ; whittaker @xcite , p 249 ; saletan e j and cromer a h 1970 _ am . j. phys . _ * 38 * 892 another common choice is @xmath803 . this choice is known as the landau gauge ( in this gauge * a * is invariant under translations along the y axis ) . the first choice is known as the symmetric gauge ( because of the rotational invariance of * a * about the z axis ) . gibbs j w 1906 _ the scientific papers of j. willard gibbs , vol.1 , thermodynamics _ ( london : longmans , green ) , reprinted by dover , 1961 , p 56 . for modern treatments , see callen h b 1985 _ thermodynamics _ 2nd edition ( new york : wiley ) p 133 , and bailyn @xcite , p 224 . there are various ways to constrain the variation so that @xmath598 is selected . in principle , one can minimize the energy @xmath804 with the trial @xmath620 s restricted to be orthogonal to the exact @xmath805 . in practice , since @xmath805 are rarely known exactly , one can find an approximate @xmath598 by choosing trial @xmath620 s with @xmath43 nodes ( for 1d problems ) . alternatively , one can choose a trial @xmath620 which is known to be the @xmath43th eigenstate of a similar hamiltonian ( see sec . 3.1 for an example ) . frenkel j 1934 _ wave mechanics , advanced general theory _ ( oxford : oxford u.p . ) frenkel states that he took his argument from the appendix to the russian edition of dirac p a m 1930 _ the principles of quantum mechanics _ ( oxford : oxford u.p . ) , but we are unable to locate the argument in this reference . see also heisenberg w 1930 _ the physical principles of the quantum theory _ ( chicago : university of chicago press ) , reprinted by dover , 1960 , p 163 a minimum residual principle @xmath806 is similar to that of frenkel and dirac , and was proposed by mclachlan a d and bell m a 1964 _ rev . phys . _ * 36 * 844 . for recent applications of the frenkel - dirac variational principle to molecular and condensed matter problems , see e.g. refs @xcite and also beck m h , jackle a , worth g a and meyer h - d 2000 _ phys . rep . _ * 324 * 1 ; wang h 2000 _ j. chem . phys . _ * 113 * 9948 . schwinger j 1951 _ phys . rev . _ * 82 * 914 . the generalized operator uhp is @xmath807 ( allowing both final end parameters @xmath808 to vary , and fixing the initial ones @xmath809 ) , where @xmath810 is a heisenberg time - dependent operator , @xmath811 the action operator , and @xmath812 the lagrangian operator . if @xmath813 we get the operator uhp , @xmath814 , and if the end parameter variations both vanish , we get the operator hp , @xmath815 . from the generalized operator uhp schwinger derives what is nowadays known as the schwinger action principle for the variation of the propagator @xmath816 , where @xmath817 is the evolution operator , i.e. @xmath818 . here an arbitrary number of intermediate state parameters @xmath819 are allowed to vary on the left hand side , e.g. @xmath820 in @xmath821 , as well as the final end - state parameters @xmath822 . ( schwinger also discusses other possible variations ) . if there are no end - state parameter variations , then @xmath823 and hence @xmath824 . the analogy of the last relation with the classical hp is clearer when we write the classical hp as @xmath825 . schwinger s original paper discusses the case of relativistic quantum field theory . for a nonrelativistic quantum mechanics discussion , with examples , see schwinger s posthumously published lecture notes , schwinger j 2001 _ quantum mechanics _ ed englert b - g ( berlin : springer ) p 207 . for the connection to feynman s path integral expression for the propagator , see yourgrau and mandelstam , ref . @xcite , p 139 . for vp s for diagonal and off - diagonal matrix elements of arbitrary operators , and for vp s for wave - functions , see gerjuoy et al @xcite and references therein . for a ground state energy vp based on path integrals , see feynman r p and hibbs a r 1965 _ quantum mechanics and path integrals _ ( new york : mcgraw - hill ) p 303 . for a vp for the ground state energy as a functional of the particle density ( density functional theory ) , see hohenberg p and kohn w 1964 _ phys . rev . _ * 136*b 864 . lwdin p o and mukherjee p k 1972 _ chem . * 14 * 1 ; kramer p and saraceno m 1981 _ geometry of the time - dependent variational principle in quantum mechanics _ ( new york : springer ) ; deumens e , diz a , longo r and hrn y 1994 _ rev . phys . _ * 66 * 917 these are often referred to as the de broglie - bohm or quantum hydrodynamic equations : see , e.g. , holland @xcite , p 15 and p 69 ; wyatt r e 2002 _ j. chem . phys _ * 117 * 956 ; kendrick b k 2003 _ j. chem . phys _ * 119 * 5805 , and references therein . the quantum continuity equation retains the same form ( 7.17 ) , and the quantum hamilton - jacobi equation takes the form ( 7.14 ) with an additional so - called quantum potential term @xmath826 added to @xmath34 . these equations can also be ( and are usually ) derived by substituting the madelung form ( 7.10 ) into the schrdinger equation ( 7.3 ) and using ( 7.4 ) . in practice ( 7.18 ) is less useful for bound states than ( 7.1 ) since @xmath44 and @xmath620 are both unknown for bound states . for scattering states @xmath44 is known , and @xmath827 causes no problem in ( 7.18 ) , as it does in ( 7.1 ) . in the present context , the normal form is a perturbation series , often generated by poisson bracket and lie transform methods , which expresses the system energy @xmath44 as a power series in the good actions @xmath783 , @xmath828 semiclassical quantization is achieved by the ebk substitution @xmath829 . in molecular physics the quantized version of the normal form is usually referred to as the dunham expansion : see , e.g. , toda m 2002 _ adv . phys . _ * 123 * 153 ; herman m , lievin j , auwera j v and campargue a 1999 , _ adv . phys . _ * 108 * , 1 reddy j n 2002 _ energy principles and variational methods in applied mechanics _ ( new york : wiley ) the derivation of the einstein gravitational field equations from the einstein - hilbert action is the classic example : see , e.g. , landau and lifshitz @xcite , p 307 routh uses this method to , in effect , transform the original mp ( a2.3 ) into the unconstrained hamilton principle ( 2.8 ) , and from this he obtains the euler - lagrange equations of motions . pars uses a similar method to derive euler - lagrange equations from the original mp ; he however , allows @xmath301 to be a variable @xmath295 , and later finds it to be a constant after all . see routh e j 1905 _ advanced dynamics of a system of rigid bodies _ 6 ed ( london : macmillan ) , reprinted by dover , 1955 , p.306 , and pars @xcite , p 249 .

the question of the causal behavior of superluminal propagations has been often debated , in rather different contexts : tachyonic particles ( e.g. @xcite ) , field theories @xcite and quantum induced superluminal propagations @xcite . the issue is however controversial , even on specific contexts , and we intend to clarify it . we will focus on classical field theories _ whose full action is lorentz invariant _ ( or diffeomorphism invariant whenever einstein s gravity is taken into account ) . in that case , superluminal behavior can only arises from spontaneous breaking of local lorentz invariance by non trivial backgrounds ( or vacua ) . superluminal behavior then occurs when some sector of the theory is left unbroken , so that some degrees of freedom can propagate faster than other ones @xcite . superluminal propagations induced by vacuum polarization provide well - known examples of this phenomenon . note that we define superluminal behavior as going faster than gravitons [ i.e. , gravitational waves ] . we denote @xmath0 the speed of gravitons in vacuum . this maybe unconventional definition will not affect our arguments . in the standard theory of gravity photons propagate along the gravitational metric so that this definition is the usual one in that case . our general analysis will enable us to investigate the causal behavior of any field coupled to standard gravity . we will illustrate our arguments in three important cases : k - essence scalar fields , bimetric theories of gravity and quantum induced superluminal propagations . we will make constant use of many important ideas and tools that were developed in order to investigate the causal behavior of gr , and we refer notably to @xcite . in sec . [ causalite ] , we analyze in detail the meaning of causality in classical field theories . we stress the important role played by the notions of chronology , time arrow and cauchy problem in its definition . as an illustration , we show how causality is usually formulated in the theory of general relativity ( gr ) . we stress that superluminal propagations are automatically discarded if one assumes that causality should be defined with respect to the chronology induced by the gravitational metric . this _ postulate _ actually `` sets the [ gravitational ] metric apart from the other fields on @xmath1 and gives it its distinctive geometrical character '' @xcite . this prior assumption can not be supported by any mathematical reason but only by experiment . if superluminal propagations were found in the laboratory , it would thus _ not _ mean that causality , locality , or the entire framework of special relativity would be lost , but simply that the gravitational field does not have any fundamental geochronological character . in particular , causality should not be expressed in terms of the chronology induced by the gravitational metric field . as a consequence , it is necessary to drop such prior assumptions to address the issue of the causal behavior of superluminal fields . in sec . [ causalityandsuperluminalbehav ] we thus look for a minimal expression of causality in which no prior chronology is assumed . in that framework , superluminal propagations are generically allowed . before moving to superluminal and causal theories , we display two well known examples of noncausal theories and notably the tachyonic one . it enables us to stress how noncausal behaviors are related to constraints on initial data . we also comment on the causal paradoxes that arise in a non causal theory . we finally briefly investigate what kind of field theory can lead to superluminal propagation . we consider in sec . [ sectionkessence ] a k - essence scalar field @xmath2 , which can propagate superluminally along an effective metric @xmath3 $ ] . we find that the scalar field is causal provided that the spacetime embedded with the effective metric @xmath4 is globally hyperbolic . here @xmath4 is the effective metric evaluated on a solution @xmath5 to the field equation ( the background ) . this condition puts some generic constraints on the free function defining the theory , and is generally satisfied on reasonable backgrounds . we emphasize that the claim that such scalar fields are not causal even on globally hyperbolic backgrounds @xcite can only hold if a prior role is attributed to the gravitational ( or flat ) metric , an assumption that , we believe , was implicit in these works . the only threat for causality actually lies in global properties of backgrounds @xcite , and this can be related to the so - called chronology protection conjecture . in sec . [ bimetrictheories ] , we investigate the question of causality in a bimetric theory of gravity , in which the matter sector is universally coupled to a metric @xmath6 that can differ from the einstein - hilbert metric @xmath7 . these two metrics generally define two distinct `` causal '' cones . photons can travel faster than gravitons , and conversely . if causality is defined with respect to the chronology induced by the gravitational metric , then causality forces the matter metric @xmath6 to define a cone which coincide or lie within the gravitational cone , and conversely . these two choices of chronology have been considered in the literature , thus leading to opposite requirements on the theory . this will illustrate the fact that there is no clear reason why a metric , or a chronology , should be preferred to the other . we argue that the theory is actually causal if no prior chronology is assumed , but the solution to the cauchy problem depends , of course , on the precise dynamics of the matter sector . in sec . [ sectionqed ] , we briefly show how our analysis apply to the case of quantum induced superluminal propagations . this enables us to conclude that superluminal velocities do not threaten causality , a point that were still controversial . let us emphasize that the question of causality in k - essences or bimetric theories is not just of academical interest , since these theories have drawn much attention recently . quite generally , it is claimed that causality forces one of the cone to be wider than the others ones ( be it the gravitational or the matter one ) @xcite and this puts some constraints on the theory . our motivated definition of causality will however not support these claims . such a k - essence scalar field was notably used as a ( dynamical ) dark energy fluid responsible of the late time acceleration of the universe @xcite , as well as a fluid that drives inflation @xcite . it is also used as a new gravitational field in addition to the metric one in some relativistic theories of modified newtonian dynamics ( mond ) , that are intended to account for the mass discrepancy at astrophysical scales without invoking dark matter @xcite . some of these models were notably discarded because of the presence of superluminal propagations . bimetric theories are also an essential piece of some recent mond - like theories @xcite ( see also @xcite ) . moreover , it represents the best motivated framework that reproduces a varying speed of light scenario @xcite that may address the problems of standard cosmology in a rather different way than inflation does . indeed some vsl theories , in which the speed of light @xmath0 is replaced by a changing velocity @xmath8 _ inside _ the equation of motions @xcite , however interesting phenomenologically , are not satisfying theoretically @xcite ( the resulting theory can not be derived from an action ) . in the last section [ applications ] , we consider some applications of our work to these field theories . we investigate the links between causality and stability in k - essences theories . we show that ( k - essence ) ghosts stabilization might suffer from a serious problem . we also show that , if one reproduces the mond phenomenology with the help of a k - essence scalar field , then a slight modification of milgrom s law is necessary for the theory to be causally well - behaved . it leads to a non trivial modification of the phenomenology in the very low acceleration regime . we also comment on a simpler theory that reproduces the mond phenomenology , using only one scalar field . the initial value formulation of this theory has still to be checked and we leave it for further work . we briefly show how this framework can account for the pioneer anomaly , and we finally make some comments about vsl theories . we use , throughout the paper , the sign conventions of @xcite and notably the mostly @xmath9 signature . the flat metric is denoted @xmath10 or @xmath11 in a coordinate system . we denote @xmath7 the gravitational metric field ( which obeys the einstein equation ) and @xmath6 the matter metric to which matter field couples . unless specified , indices are moved with the help of the gravitational metric @xmath7 . in the present paper , we are interested in the superluminal behavior of some matter fields in a gr - like context . the metric field @xmath7 follows the dynamics induced by the einstein - hilbert action . by spacetime we will always mean a couple ( @xmath12 ) , where @xmath1 is a four - dimensional differentiable manifold and @xmath13 some non degenerate lorentzian metric on it . a superluminal signal , _ by definition _ , propagates along spacelike curves of the gravitational metric @xmath7 . by causality we usually mean the ability to find a cause to an effect . since cause and effects are both described in terms of some physical variables , the usual principle of causality states that physical variables should be unambiguously determined at a given time from their values at a time before . we also demand that this retrodiction can be converted forward in time , so that we should also be able to predict a future situation from a present one . the principle of causality thus requires that determinism holds `` in both directions of time '' . causality demands the existence and uniqueness of solutions to the equations of motion given some initial data . in mathematical words , equations of motion must have a well - posed cauchy problem , or initial value formulation . note that this definition shows that a time - ordering , or chronology , must exist between two spacetime points that are causally connected . unlike newton s theory where a global chronology preexists to dynamics because of the prior topology assigned to spacetime , namely @xmath14 , field theories ( including einstein s gravity ) do _ not _ involve a preexistent notion of time and chronology . on the contrary , any relativistic field @xmath15 defines its own chronology on @xmath1 by means of the metric @xmath16 along which it propagates . any lorentzian metric indeed induces a local chronology in the tangent space through the usual special relativistic notions of absolute ( i.e. , lorentz invariant ) future and past . let us consider one particular field @xmath17 that propagates along a metric @xmath13 . it is important to note that this metric may not induce a global chronology on the whole spacetime . this is the case if local causal cones of @xmath13 are distributed on the manifold in such a way that there exists a curve over @xmath1 which is everywhere future - directed timelike but closed . in that case , the field @xmath17 propagates along a closed curve and an event could be _ both _ the cause and the effect of another event . causality requires that it does not happen . the strongest way to prevent it is to require the global hyperbolicity exists @xcite . it can then be proved that the spacetime has the topology of @xmath18 , so that a global time function , i.e. , a global chronology exists . ] of the spacetime ( @xmath12 ) . let us emphasize that causality also requires a notion of time flow . it is indeed worth recalling that even if the above metric @xmath13 defines a global notion of future and past , we have also to require that the field @xmath17 can only propagate in the future . if it could , on the contrary , propagate both in future and in past , one could always form a closed timelike curve with it . on the other hand , global hyperbolicity is enough to guarantee that no closed _ future - directed _ timelike curves ( cftc ) exists . when there is a finite number of metrics @xmath16 on @xmath1 , the discussion of global properties is slightly more involved , see sec . [ causalityandsuperluminalbehav ] . let us illustrate the previous discussion . in gr , causality is generally expressed by the following properties @xcite : 1 . the gravitational metric @xmath7 must define a global chronology on spacetime . 2 . `` the null cones of the matter equations coincide or lie within the null cone of the spacetime metric @xmath7 '' . @xcite 3 . the whole set of equations of motion must admit a well - posed cauchy problem . the point ( a ) guarantees the existence of a global time ordering on @xmath1 and the point ( c ) is the formal expression of determinism . note that it is a non trivial mathematical property of systems of differential equations . its satisfaction in gr therefore critically depends on the precise dynamics of the matter sector . the point ( b ) excludes faster - than - graviton propagation . equivalently then , it means that any initial data set on spacelike hypersurfaces with respect to the gravitational metric @xmath7 are allowed . since the metric @xmath7 defines a global chronology on spacetime , so does any other metric @xmath16 associated to the propagation of some matter fields ) is globally hyperbolic and if @xmath19 is a metric whose null cone coincide or lie within the null cone of the metric @xmath13 _ everywhere on _ @xmath1 , then ( @xmath20 ) is globally hyperbolic . ] . the conditions ( a ) and ( b ) therefore ensures that no cftc exist for any metric ( be it @xmath7 or any metric @xmath16 ) . let us emphasize that , in the above formulation ( a)(c ) , the gravitational metric field clearly plays a preferred role . both conditions ( a ) and ( b ) are such that causality is actually defined with respect to the chronology induced by the metric @xmath7 . this can be understood as a `` _ _ postulate _ _ which sets the metric * g * apart from the other fields on @xmath1 and gives it its distinctive geometrical character '' ( @xcite @xmath21 , emphasis added ) . the chronology defined by @xmath7 is thus assumed to be a preferred one on @xmath1 . this postulate is however arguable because , as we pointed out , any relativistic field induces its own chronology on spacetime . it notably means that if a signal made of waves of some field @xmath15 propagates between two spacetime points , then these points can be time - ordered with the help of the metric @xmath16 , and causality may be preserved even if the field propagates superluminally . the gravitational metric field @xmath7 is just one particular field on @xmath1 and there is no clear reasons why it should be favored . let us finally stress that the three conditions ( a)(c ) are actually not sufficient to guarantee ( usual ) causality . indeed , if information were allowed to travel as well forward as backward in time , as it is the case in any time reversal invariant theories , then it would be always possible to have information running along a closed curve in spacetime . thus the absence of cftc does not suffice to guarantee the usual notion of causality , unless one specifies that a _ flow _ of time exists and that information can only go forward in time . the three points ( a)(c ) are clearly time - reversal invariant statements and therefore can not address this issue . note that both classical and quantum field theories are time - reversal invariant so that they may not be considered as `` causal '' in the usual sense . the emergence of the flow of time from thermodynamics in quantum field theory is a quite involved question that we will not consider here . in the present paper , we will therefore restrict ourselves to such a time - reversal invariant notion of causality , but one should keep in mind that thermodynamics may be the key to understand fully the ( coarse - grained ) notion of causality . we stressed that superluminal behaviors are automatically discarded by the above postulate . one may try to justify it by arguing that , since @xmath7 reduces locally to @xmath10 and therefore gives to the tangent space its special relativistic ( minkowskian ) structure , going faster than gravitons would `` undermine the entire framework of relativity theory '' @xcite . let us however stress that any other metric @xmath16 can also be reduced locally to its fundamental form in adapted `` inertial '' coordinates . these coordinates transform under the action of the lorentz group @xmath22 with an invariant speed @xmath23 _ that can differ from _ @xmath0 , if the cones of @xmath16 and @xmath7 do not coincide [ due for instance to some spontaneous breaking of local lorentz invariance , see sec . [ causatheoriessuperluminiques ] ] . these coordinates are actually relevant if one uses rods and clocks made of the field @xmath17 that propagates along @xmath13 . the fact that @xmath7 reduces locally to the flat spacetime metric can therefore not be used to claim that @xmath7 should be a preferred field on @xmath1 . we will come back to this important point later , see sec . [ bimetricsuperlum ] . we may therefore drop the above postulate and look for a minimal expression of causality in which no prior chronology is assumed . let us consider a collection of fields @xmath15 that propagate along some metrics @xmath16 ( gravity included ) . we are led to the picture of a finite set of `` causal '' cones at each points of spacetime . we do not want to prefer one metric with respect to the others , because there is no reason why , locally , some sets of coordinates , or some rods and clocks should be preferred : coordinates are meaningless in gr . the cones defined by the metrics @xmath16 may therefore be in any relative position with respect to one another . moreover , these cones may even tip over each others depending on the location on spacetime . we have however to guarantee that no signal can propagate along a closed curve . for this it is not sufficient to require the global hyperbolicity of each spacetime ( @xmath24 ) . indeed , fields may interact together , and we may form a physical signal which propagates on spacetime along different metrics . we shall rather define a extended notion of future - directed timelike curves , by requiring that at each point of the curve , the tangent vector is future - directed timelike with respect to _ at least _ one of the metrics @xmath16 . such a construction have been already advocated in @xcite . all the ( extended ) notions of future , past , domains of dependence , achronal sets , cauchy surfaces , global hyperbolicity then follow . this corresponds to a `` mixed '' notion of chronology : a point @xmath25 is in the ( extended ) future of @xmath26 if it is in the future of @xmath26 for at least one of the metrics @xmath16 , see fig . ( [ figurefuture ] ) . a spacetime interval is thus spacelike in the extended sense if it is spacelike with respect to _ all _ metrics @xmath16 . the hatched part shows the extended future defined by two metrics ( solid and dashed lines ) in the case where one metric defines a wider cone than the other one ( left ) , and in the opposite case ( right).,width=6 ] causality then requires that a global ( mixed ) chronology on spacetime exists . in other words , spacetime must be globally hyperbolic in that extended sense . this straightforward generalization enables us to express the miniminal formulation of causality by the following properties : 1 . the mixed chronology defined by the set of metrics @xmath16 must be a global chronology on spacetime . the whole set of equations of motion must admit a well - posed cauchy problem . the formulation of causality ( a)(c ) is immediately obtained from these more general requirements whenever the matter light cones coincide or lie within the gravitational one . we thus see that fields can propagate superluminally without threatening causality provided that we do not refer to any preferred chronology on spacetime . the cauchy problem of some field @xmath17 will be well posed depending on the precise equation of motion , and if initial data are set on surfaces that are spacelike with respect to the metric @xmath13 along which it propagates . moreover , if we are interested in the whole theory ( gravity included ) , the cauchy problem will be well posed if initial data are set on surfaces that are spacelike in the extended sense , that is _ spacelike with respect to all metrics @xmath16 _ ) . it must be stressed that , in gr , the fact that @xmath7 defines a global chronology can not be proved , for the reason that local physics does not determine the topology of spacetime , which could however prevent the existence of a global chronology . this the case for instance , of non time - orientable spacetimes . moreover , there exists exact solutions to the einstein equations that do not describe globally hyperbolic spacetimes , as explicitly shown by the kerr solution , which possess closed timelike curves , or gdel s universe . einstein equations thus admit solutions that violate causality . this difficulty is entirely subsumed into the so - called chronology protection conjecture @xcite that asserts that the local laws of physics are such that they prevent the formation of cftcs in spacetime . this has not been proved yet , so that we usually assume from the beginning that spacetime ( @xmath27 ) is globally hyperbolic . as long as the conjecture is not proved , such a restriction is not dynamical , but of epistemological nature@xcite . let us stress that in our extended framework , and notably in presence of superluminal matter fields , we will not be able to prove either that the condition ( a ) holds . we will have to _ assume _ that spacetime is globally hyperbolic in the extended sense . again , if we were able to prove the chronology protection conjecture , we would not have to make such a non trivial assumption . we will discuss further this point in sec . [ kessencesglobal ] . in other words , the condition ( a ) will have to be imposed by hand , just as in gr . as we already explained , we will not consider the issue of the existence of the flow of time , and thus , in the following , we will mostly be concerned by the point ( b ) . we provide in the next section two examples of theories that do not admit a well - posed cauchy problem . it enables us to show the deep relationship between non causal behaviors , closed curves in spacetime and constraints on initial data . an obviously non causal theory is an elliptic klein - gordon scalar field with an equation of motion @xmath28 , and where the metric @xmath13 has the signature @xmath29 . this elliptic equation does not have a well posed cauchy problem . it can be simply interpreted as the fact that this euclidean metric does not select the time coordinate as special compared to the spatial one . there is thus no available notion of propagation and initial data on three - surfaces can not be propagated in four dimensions . on the contrary , a lorentzian signature for @xmath13 guarantees that the equation is hyperbolic and that , by virtue of well - known theorems @xcite , the cauchy problem is well posed . we consider now the more interesting case of tachyonic particles . by tachyons we mean particles or signals that can be sent at a superluminal speed relative to the emitter ) ] . as correctly recognized in many papers in the literature , tachyons are not causal ( see , e.g. @xcite ) . indeed , they can always be used construct closed curves in spacetime along which a ( tachyonic ) signal propagate . consider for instance an observer @xmath30 who sends at time @xmath31 a signal to observer @xmath32 ( event @xmath33 ) , who in turn sends a signal back to @xmath30 , at time @xmath34 ( event @xmath35 ) . this last signal can be received by @xmath30 at a time @xmath36 ( event @xmath37 ) , if signals can be sent from @xmath32 at a superluminal speed and if @xmath30 and @xmath32 are in some relative motion at a speed @xmath38 , see fig . ( [ figuretachyons ] ) . information can therefore propagate along the closed curve in spacetime ( @xmath39 ) . the closed curve followed by the tachyonic signal viewed in the rest frame of @xmath30 . the tachyon is sent by @xmath30 at time @xmath31 ( event @xmath33 ) and received by @xmath32 at time @xmath34 ( event @xmath35 ) . since @xmath32 moves with respect to @xmath30 , the tachyonic signal he sends back to @xmath30 is actually received before it was sent , at time @xmath40 ( event @xmath37 ) . the thin line represents the minkowski cone , and horizontal and vertical lines are the space and time axis in the frame of @xmath30.,width=6 ] the theory is not causal because an event along this curve will influence itself , after the signal had gone one time round the curve . such an event is thus not freely specifiable , and we see that propagation of information along a closed curve in spacetime _ always _ lead to constraints on initial data . on the contrary , if the cauchy problem were well posed for such a theory , it would notably mean that there exists surfaces on which initial data can be freely specified ( up to possible constraints arising from gauge invariance , if any ) and unambiguously evolved with time . it shows that the present theory can not possess a well posed cauchy problem . before going further , let us comment the previous experimental device . suppose ( situation @xmath41 ) that , if @xmath32 receives a signal from @xmath30 , he sends a signal backwards in time to @xmath30 . then @xmath30 receives a signal at time @xmath42 and is thereby _ forced _ to send a signal to @xmath32 an instant later . it explicitly shows the constraints that appear in the case of propagations along closed curves . a temporal ( or causal ) paradox can even arise if we consider the following experimental device ( situation @xmath43 ) : when @xmath32 receives a signal from @xmath30 , he sends him back a signal , and @xmath30 sends a signal to @xmath32 _ only if _ he did not receive anything from @xmath32 . one may wonder what happens in that case . if @xmath30 receives a signal from @xmath32 , he will not send a signal to @xmath32 who in turn would not send a signal back to @xmath30 , and this contradicts the hypothesis . the other possibility , when @xmath30 does not receive anything from @xmath32 , is also inconsistent . this is in fact a version of the famous `` grandfather paradox '' . remarkably , there is no answer to the above question `` what happens ? '' . it is however immediate to see that this question implicitly assumes that the knowledge of the initial data at time @xmath42 for instance ( @xmath30 receives a signal from @xmath32 or not ) is enough to determine what will happen `` later '' along the curve . this question thus assumes that causality ( or determinism ) holds , whereas it can not even be defined , since there is no available notion of time - ordering along a closed curve . on the contrary , we have seen that such a closed curve leads to constraints on initial data . the event at time @xmath42 is strongly correlated to itself so that the only relevant question reads `` are any of the two initial conditions : @xmath30 receives -or does not receive- a signal from @xmath32 allowed ? '' . clearly the answer to this last question is negative . let us emphasize that situation @xmath43 ( the grandfather s paradox ) is often taken as an example of bad causal behavior induced by tachyonic particles ( or accordingly , induced by the use of time machines ) . this point is however not justifiable because the paradox only arises if we ask the wrong question . this experimental device simply _ can not _ exist since there is no initial data consistent with it . there is nothing to do here with causality , but only with logics : temporal paradoxes do not exist . on the other hand , the noncausal situation @xmath41 can exist ( constrained initial data exists that correspond to this experiment ) . tachyonic particles propagate superluminally and violate causality . this does not mean however that any superluminal propagations does . indeed , the closed curve in fig . ( [ figuretachyons ] ) can only be constructed if the speed of the tachyonic signal depends on the speed of the emitter in the rest frame of @xmath32 , and then deduces , using a lorentz boost , that it travels backward in time in the frame of @xmath30 , see @xcite . ] . this is not the case in metric description of superluminal propagations that we will mostly be concerned with . here we wish to investigate briefly what may look like a field theory that involves superluminal propagation . in a theory whose whole action is lorentz - invariant , the only possibility to open up the causal cones is to modify the propagational part of the dispersion relation . it can be achieved by adding higher - order derivatives in the action , but unless these terms arise from the low energy limit of some uv - complete theory , the theory is generically unstable @xcite . superluminal behavior for instance arises from quantum corrections to electrodynamics in non trivial vacua , see sec . [ sectionqed ] . on the other hand , we could still have second order , but nonlinear , equations of motions . we will consider two examples of such theories in the following sections . this nonlinearity , which is essential to obtain unusual dispersion relations , leads to a background - dependent cauchy problem . accordingly , lorentz invariance of the action is spontaneously broken through non trivial backgrounds and it allows superluminal propagations . note also that the speed of small perturbations around the background only depends on the background and _ not _ of the motion of the emitter . superluminal signals in field theories are not tachyonic in the sense indicated in the previous section . note that we could also have considered non - lorentz invariant lagrangians like @xmath44 , where @xmath45 . we will not consider such lagrangian in the present paper . in practice , the theories we will consider will lead to similar field equations , but only through a spontaneous breaking of local lorentz invariance by non trivial backgrounds . let us consider a so - called k - essence scalar field theory , whose action reads @xmath46.\ ] ] where @xmath47 and @xmath48 and @xmath49 are some functions , @xmath50 some mass scale and @xmath51 . hereafter we also take @xmath52 . note that the more general form @xmath53 is often considered , but the following arguments also apply to this case . let us first neglect the coupling to gravity which will be considered later , and consider that @xmath7 is flat . the field equation then reads @xmath54 where the effective metric @xmath55 is given by @xmath56 and a prime denotes derivation with respect to @xmath57 . scalar waves propagate inside the `` scalar cone '' defined by this effective metric . this cone generally differs from the gravitational one unless @xmath58 , see sec . [ superluminalkessence ] . this equation is a special case of quasilinear second order differential equation . a theorem due to leray then proves @xcite that this equation has a well posed cauchy problem if spacetime ( @xmath59 ) is globally hyperbolic , where @xmath4 is the effective metric evaluated on a solution @xmath5 to the field equation eq . ( [ eqchampkessencemat ] ) . hereafter , we will refer to @xmath5 as the background . the initial data have to be specified on three - surfaces that are spacelike with respect to the background metric @xmath4 , in deep connection with the discussion of sec . [ causalityandsuperluminalbehav ] . a necessary but not sufficient condition for global hyperbolicity is the lorentzian signature of the background metric @xmath4 , which notably ensures that the field equation is hyperbolic . by diagonalization of the matrix @xmath4 , we find that its signature is @xmath60 everywhere over @xmath1 if and only if [ condhyperkessencemat ] @xmath61 @xmath62 where @xmath63 . _ in particular _ , if the function @xmath48 is such that the above inequalities are satisfied for all @xmath57 , the metric @xmath4 will have a signature @xmath60 on any background . notice that the sign of the above inequalities eq . ( [ condhyperkessencemat ] ) can be reversed . this would however correspond to a metric @xmath4 with signature @xmath64 and the scalar waves would carry negative energy . the theory would therefore be unstable when coupled to other fields ( notably gravity ) . we exclude here this possibility , although it should be stressed that it comes from a stability argument , and not from some causal requirement . in presence of gravity the cauchy problem has to be solved simultaneously for the gravitational variables and for the matter ones . it has been proved ( see for instance @xcite ) that the cauchy problem is well posed if matter fields satisfy `` reasonable '' equations of motion and if the stress - energy tensor of the matter fields only involve matter variables , the gravitational metric , and their first derivatives . this is notably the case of k - essence theories , as can be easily checked . locally therefore , the whole theory has a well - posed cauchy problem . the analysis of the characteristic of the field equation eq . ( [ eqchampkessencemat ] ) shows that the scalar field propagates superluminally if @xmath65 , where we used that @xmath66 by virtue of eq . ( [ condhyperpositive ] ) . as it will be useful later , let us remind that any field which propagates along an effective metric of the form ( up to some positive conformal factor ) : @xmath67 where @xmath32 and @xmath68 are respectively some scalar and vector field , is superluminal if @xmath69 . notice that the opposite sign for @xmath32 is found in @xcite because of the opposite choice of signature . the above theorem and our analysis of the meaning of causality in sec . [ causalite ] enable us to conclude that k - essences theories are causal in flat spacetime whenever the background ( @xmath70 ) is globally hyperbolic , even in presence of superluminal scalar waves . again , this conclusion holds only if we do not refer to any preferred chronology . it was however claimed @xcite that k - essences theories are not causal even if the background is globally hyperbolic . here we emphasize that these claims can only be supported by the postulate that the gravitational ( here , the flat ) metric induces a preferred chronology on @xmath1 , a point that , we believe , is implicitly assumed in these references . the main argument which rules out superluminal propagations is indeed that the cauchy problem for the scalar field is not well posed for initial data that are set on surfaces which are spacelike _ with respect to the flat metric _ but timelike or null with respect to the background metric . in that case , initial data can not be evolved because of caustics @xcite , and the hamiltonian formalism is singular @xcite . let us however stress that it is not surprising in view of leray s theorem , since it proves that initial data surfaces for the scalar field must be spacelike _ with respect to the background metric_. if initial data are set on these surfaces , the theory is free of such singular behaviors . this claim therefore only arises from an unadapted choice of initial data surfaces , i.e. , from the postulate that the gravitational metric defines a preferred chronology . this postulate actually prevents _ any _ superluminal propagation , as we explained in sec . [ causalitegr ] . in such a case , superluminal behavior is ruled out _ from the beginning _ , but not by some intrinsic ( mathematical ) arguments . the only threat for causality lies in fact in global properties of the background . it has been correctly recognized in @xcite that the spacetime ( @xmath71 ) may not be always globally hyperbolic . in particular closed timelike curves ( with respect to the effective metric ) could exist . let us first stress that in general however , the spacetime ( @xmath59 ) is globally hyperbolic . it is notably the case in trivial backgrounds @xmath72 , and in non trivial but homogeneous ( in a certain lorentz frame ) backgrounds @xmath73 , which may be relevant in a cosmological context . note that these backgrounds have to be solutions to the equation of motion eq . ( [ eqchampkessencemat ] ) . in the first case , @xmath5 has thus to be an extremum of the potential , whereas in the second one , @xmath5 is not constant in spacetime and the potential must be flat over some range , or more simply vanish . in these two important cases , the spacetime ( @xmath74 ) is globally hyperbolic and , by virtue of the previous theorem , the cauchy problem is well posed and the theory is causal . since the global hyperbolicity of spacetime ( @xmath75 ) must be assumed in gr , as we explained in sec . [ causaliteglobal ] , we may as well assume that spacetime is globally hyperbolic in the extended sense , see sec . [ causalityandsuperluminalbehav ] . such an assumption obviously requires that eq . ( [ condhyperkessencemat ] ) hold , and automatically ensure that the whole theory of the k - essence scalar field and gravity is causal . the relevant cauchy surfaces in that case are hypersurfaces that are spacelike with respect to all metrics ( gravitational , scalar and others matter metrics ) . note that this is a non trivial ( and maybe arguable ) assumption . let us however stress that it must also be assumed in the context of gr and standard matter alone so that the fact that non trivial global properties may break causality does not appear to be rooted in superluminal propagations . this assumption may actually be linked to the chronology protection conjecture , as shown by the following example . let us consider the highly non trivial and non globally hyperbolic background invoked in @xcite . it consists of two `` bubbles '' of non trivial backgrounds @xmath76 that move rapidly in opposite directions with a finite impact parameter . the space is otherwise empty ( trivial background @xmath77 ) . small scalar perturbations thus travel superluminally inside the two bubbles , and along null rays of the flat metric outside them . the fact that the two bubbles are in relative motion implies the existence of cftc ( timelike with respect to the background metric @xmath4 ) . it should however be stressed that such a background is _ not _ relevant since it is not a solution to the equation of motion eq . ( [ eqchampkessencemat ] ) . indeed , the derivative of the field @xmath5 is not continuous at the transition between the bubbles and the empty space . some dirac - like source terms should be added to the equation of motion to accommodate such a background . on the contrary , a physical solution involving such two bubbles should exhibit a continuous transition from their interior to the empty space . of course it may turns out that even in this more realistic situation the background is still not globally hyperbolic , but the contrary could also happen , and it would be an illustration of the chronology protection conjecture . we did not try to perform this analysis , but it may be interesting . a very similar case was discussed in @xcite in the context of casimir experiment , see sec . [ sectionqed ] . note that an analogous case has also been found in the context of gr by gott @xcite , who showed that two straight infinite cosmic strings moving in opposite directions with a finite impact parameter lead to the formation of cftc in spacetime . again , it shows that these difficulties with causality at a global level already exists in the context of gr and subluminal matter , and should not be related to some intrinsic problems of superluminal propagations . by multi - metric theories of gravity we mean theories of gravity in which some degrees of freedom of the matter sector are coupled to some matter metrics @xmath78 distinct from the gravitational one @xmath7 . the fact that different matter fields are coupled to different metrics breaks the weak equivalence principle ( wep ) , which has been tested with great accuracy . as a special case , a bimetric theory of gravity is a theory where all the matter fields are coupled to the same metric @xmath6 . this ensures that the wep is satisfied . gr is just the special case @xmath79 . a typical example is scalar - tensor theories of gravity in which @xmath80 , at each point of @xmath1 , and where @xmath81 is a smooth real - valued function over @xmath1 . dynamics of the scalar field @xmath81 arises from a standard kinetic term ( but could also be of k - essence type ) . because of the conformal relationship between @xmath7 and @xmath6 , the two cones defined by these two metrics coincide , and there is no superluminal propagations . in general however , we could have non conformal relation between these two metrics . consider for instance , the so - called disformal relation @xcite @xmath82 where @xmath30 and @xmath32 are some functions of the scalar quantity @xmath83 , and @xmath84 is a vector field . when this matter metric is of lorentzian signature , the matter light - cone ( defined by @xmath6 ) can be wider than the gravitational cone ( defined by @xmath7 ) , depending on the sign of @xmath32 . this disformal relation is an essential piece of some recent relativistic field theories of the mond paradigm @xcite . throughout this paper , we defined superluminality as the propagation along spacelike curves of the gravitational metric . but we could also have defined it as going faster than light ( photons ) . these definitions coincide in gr but not in bimetric theories since @xmath85 . in that framework , there are two opposite definitions of superluminal behaviors . photons can travel faster than gravitational waves and conversely . it is thus not clear which of these two superluminal behavior we have to worry about . this is closely related , again , to the choice of a preferred chronology . if one assumes that the gravitational metric field induces a preferred chronology with respect to which causality should be defined , then , as in sec . [ causalitegr ] , the matter light - cone must coincide or lie within the gravitational one everywhere on spacetime . it was , on the contrary , claimed in @xcite that , since rods and clocks are made of matter , the matter metric @xmath6 should be somewhat favored , in the sense that in order to preserve causality , no propagations should escape the matter light - cone . in particular , this cone should be wider than the gravitational one . very interesting is the fact that this postulate is the exact opposite of the previous one . it is perhaps the best way to show that there is no clear reason why one metric should be preferred to the other . this confusion can be easily understood . let us go further in the discussion of sec . [ causalityandsuperluminalbehav ] . whenever @xmath6 is lorentzian and disformally related to @xmath7 , we have at hand two metrics which reduce locally to constant metrics @xmath86 and @xmath87 . as a consequence , there exist two classes of inertial coordinates for which one of these two metrics _ but not both _ , reduce to its fundamental form @xmath10 . inertial coordinates of each class transform under the action of the lorentz group @xmath22 with a _ _ different invariant speed _ _ , see @xcite . ] . let us emphasize that in gr , superluminality is defined as the propagation on spacelike curves with respect to the gravitational metric mainly because the latter is thought to be `` the '' spacetime metric , since it reduces locally to @xmath10 . we already stressed in sec . [ causalityandsuperluminalbehav ] that any lorentzian can actually take this form locally , so that it could also be viewed as the spacetime metric . the framework of bimetric theories considerably enlighten this point . in a sense , indeed , we have two `` natural '' metrics and there is simply no way to decide which of these two metrics should be `` the '' spacetime metric . accordingly , there are two natural chronologies and two locally invariant speeds . the above two classes of inertial coordinates corresponds to rods and clocks made of matter or gravitons . bimetric theories thus clearly show the irrelevance of postulating a preferred chronology . let us also stress that without such a postulate , the notion of superluminal behavior becomes itself irrelevant . the only point , as far as bimetric theories are concerned , is actually that the ratio of the speed of gravitational waves to the speed of photons for instance , vary in space and time . this is the reason why such a framework should be considered as the best motivated one that reproduces varying speed of light theories , see sec . [ bimetricvsl ] . following the discussion of sec . [ causalityandsuperluminalbehav ] , let us define causality by ( a)(c ) . it is clear that the matter metric has to be lorentzian for the cauchy problem to be well posed . this condition reads @xmath88 and @xmath30 must be non zero . we wrote the conformal factor as @xmath89 in order to ensure that matter fields carry positive energy . the matter stress - energy tensor only depends on the matter fields , the vector field , the gravitational metric , and their first derivatives . depending on the precise form of the action of the vector and matter fields , the whole set of equations of motion ( including gravity ) may be a diagonal , second order , and quasilinear hyperbolic system . note that when @xmath84 is given by the gradient of a scalar field @xmath90 , the initial value formulation of the scalar equation is a quite involved question , see sec . [ bimetricscalaire ] . on the other hand , when @xmath84 is a `` true '' vector field , the cauchy problem is generically well posed and is thereby not diagonal . it does not mean however that the cauchy problem is not well posed , but rather than a careful analysis is in order . generic vector field actions have been considered in @xcite ( and references therein ) . ] . we have also to consider the global structure of spacetime . as in the case of gr , or gr and k - essences , we have to assume that spacetime is globally hyperbolic in the extended sense of sec . [ causalityandsuperluminalbehav ] . this assumption , together with the above conditions on equations of motion of the matter fields and the vector field ensures that the whole theory is causal . in quantum electrodynamics , effects of temperature , electromagnetic or gravitational backgrounds , or boundaries ( e.g. casimir plates ) break local lorentz invariance at the loop level through vacuum polarization , thus leading , in some cases , to faster - than-@xmath0 light propagation @xcite ; see also @xcite and references therein . note that these results are derived within some range of approximation using the effective action formalism , up to the one or two loop level . generically the results only hold at frequencies less than the electron mass @xmath91 . one therefore only derives the phase velocity of soft photons , whereas the actually relevant speed is the wavefront velocity , which is the phase velocity at infinite frequency and which corresponds to the analysis of the characteristics @xcite . the computation of wavefront velocities is a non - perturbative task that we will not be concerned with . an argument based on the standard kramers - kronig relation @xcite actually gives some hints on the value of the wavefront velocity . in the case of casimir vacua , it was shown that the wavefront velocity might be greater than @xmath0 in the direction orthogonal to the plates ( breaking of lorentz invariance by the boundaries ) . the wavefront velocity has to be equal to @xmath0 in the parallel direction because lorentz invariance is left unbroken in that direction , at least if the casimir plates are infinite ( or if boundary effects are negligible ) . let us assume the validity of this result . of course , `` light does not travel faster than light '' , but the point is that gravity does not see the plates in first approximation , so that lorentz invariance is not broken in the gravitational sector and gravitons still propagate at the velocity @xmath0 . the ratio of the speed of photons to the one of gravitons is thus greater than one . our discussion of sec . [ causalite ] enables to give an immediate answer to the question of causality in that case . it can be shown that photons propagate inside the plates along an effective metric of the form of eq . ( [ metriqueeffective ] ) , where @xmath32 is some positive constant in that case , and @xmath68 is the unit spacelike vector orthogonal to the plates ( see @xcite and references therein ) . outside the plates , photons propagate along the flat metric @xmath10 ( here we neglect curvature ) . it was correctly recognized in @xcite that such a metric is stably causal , so that photons _ can not _ propagate along closed curves , contrary to the claim made in @xcite ( see also the criticism of @xcite ) . actually , this spacetime is even globally hyperbolic and therefore _ perfectly causal _ , even if photons propagate faster - than-@xmath0 inside the plates . causality may however be lost if two casimir vacua are moving rapidly towards each other @xcite . in that case , spacetime may possess cftc ( with respect to the effective metric ) . this case is strictly analogous to the case of two bubbles made of non trivial background in k - essence theory , see sec . [ kessencesglobal ] . authors of @xcite then invoked the chronology protection conjecture , and notably noted that the two casimir vacua should be confined within plates that can not be infinite , so that non trivial boundaries effects may prevent the formation of cftc . we essentially reached the same conclusion in sec . [ kessencesglobal ] . similar arguments may be applied to others vacua . in particular , note that vacuum polarization induces some effective metric along which photons propagate . this effective metric only differs from the flat one by terms of order @xmath92 where @xmath93 is the fine structure constant . these corrections are therefore small ( at least in `` reasonable '' vacua ) , so that the effective metric is still lorentzian and the spacetime is still globally hyperbolic . we briefly comment some applications of our results to recent interesting developments on possible modification of gravity . k - essence scalar field theories have been suggested as promising candidates of the dark energy fluid @xcite ( see also the review @xcite and relevant references therein ) . such an effective action can also be motivated by the low energy regime of some string theories @xcite . the hyperbolicity conditions eq . ( [ condhyperkessencemat ] ) on the function @xmath48 have been correctly derived in the literature , but in a different way ( except in @xcite ) . authors usually require the stability of scalar perturbations around some backgrounds . it depends on the reality of the `` sound speed '' @xmath94 and is thus equivalent to the lorentzian character of the effective metric @xmath55 . this ( un)stability is therefore directly related to the hyperbolic ( resp . elliptic ) character of the scalar field equation . note that the signature can still be @xmath60 or @xmath64 . authors then demand that these small perturbations carry positive energy , and it implies @xmath95 . here we wish to stress that these results also arise from the analysis of causality , and , moreover , that they are actually non - perturbative results . it can be shown , indeed , that the positivity of the whole hamiltonian and not only the one of perturbations , is guaranteed whenever the two conditions eq . ( [ condhyperkessencemat ] ) hold @xcite . unusual kinetic terms have also been considerably debated because phantom scalar fields can reproduce an equation of state @xmath96 @xcite , which indeed reads @xmath97 whenever the density @xmath98 is positive @xmath99 is equivalent to @xmath100 . thus only phantom ( ghost ) matter can lead to super - acceleration of the universe . since ghosts are usually associated with a fatal instability at the quantum level and notably in the uv regime , authors have suggested a stabilization mechanism of the ghost field at the uv scale . the function @xmath48 could be such that @xmath101 ( we reestablish the mass scale @xmath50 ) . in that case the ghost only appear at low energy @xmath102 , but could be stabilized at higher energies by higher - order terms , and the timescale of the instability can be made arbitrarily high @xcite . such a mechanism may however suffers serious diseases . notice indeed that the sign of @xmath103 must change for some values of @xmath104 and the squared sound speed may become negative . accordingly , the effective metric may become euclidean and the cauchy problem will not well posed anymore . the hyperbolicity seems however still guaranteed if @xmath105 also changes of sign at the same value @xmath104 . notice that the function @xmath48 must then be quite fine - tuned : @xmath104 may be equal to @xmath106 or @xmath104 must be a turning point as well as an inflexion point of @xmath48 . moreover , in that already fine tuned case , the effective metric becomes totally degenerate ( vanishes ) at the point @xmath104 . there is a caustic , and the theory is not well defined . it has been shown in @xcite that this point @xmath104 is however not reached through the cosmic evolution ( or at a time @xmath107 ) , and the theory may thus be free of singular behaviors . however the cosmological background is not the only relevant one . the theory must also apply at local ( e.g. astrophysical ) scales , and we expect the scalar field to have some inhomogeneities . local physics then drives @xmath57 to positive values , and the above singular point @xmath104 will generically be crossed . moreover , at a quantum level , it is not clear if we can still make sense of summing momentum from zero to some cutoff , whereas the propagator is not defined for some value of the momentum . in conclusion , k - essences field theories are quite relevant causal theories that can account for the dynamics of the dark energy fluid with an equation of state @xmath108 . on the other hand , k - essences phantom theories , even if stabilized , suffer serious diseases and it is very unlikely that such a fluid could drive the super - acceleration of the universe @xmath99 , if any . note that it was recently found that inhomogeneities of the matter distribution in the universe may be described by an effective scalar field which could be a phantom in certain cases @xcite . because only effective , this field does not suffer from quantum instabilities . k - essences theories are also used to account for the mass discrepancy in galaxies and clusters , without the need for dark matter . milgrom @xcite first pointed out that rotation curves of spiral galaxies exhibit a discrepancy at an universal acceleration scale @xmath109 , and therefore that a modification of newton dynamics ( mond ) of the form @xmath110 where @xmath111 is the newtonian gravitational field and @xmath112 the acceleration , could account for the observed discrepancy without dark matter . the @xmath113 function must behave asymptotically as @xmath114 if @xmath115 and @xmath116 if @xmath117 . the @xmath113 function is otherwise free . the choice @xmath118 is a quite standard one which fits well the data . such a behavior in the low acceleration regime automatically leads to flat rotation curves far from the source , and also reproduces the well established tully - fisher law @xmath119 , where @xmath120 is the plateau velocity and @xmath121 is the luminosity of the galaxy . this successful phenomenology @xcite can be reproduced with the help of relativistic aquadratic lagrangians ( raqual ) @xcite , i.e. , a k - essence scalar field . consider indeed a scalar field @xmath2 coupled to matter via a conformal metric @xmath122 . the scalar field equation then reads @xmath123 where @xmath124 is the covariant derivative corresponding to the metric @xmath7 , @xmath125 is newton s constant appearing in the eintein - hilbert action , and @xmath126 is the trace of the stress - energy tensor of the matter fields ( defined by variation and contraction with respect to @xmath7 ) . if we consider a static distribution of matter @xmath127 , this equation reduces to a static modified poisson equation for the scalar gravitational potential @xmath2 : @xmath128 it is then clear , after restoring the appropriate dimensionfull constants , that the function @xmath129 plays the role of the milgrom s function @xmath130 . note that @xmath131 in the static case . if we require that @xmath132 behaves as @xmath57 if @xmath133 , and as @xmath134 if @xmath135 , we thus recover the mond phenomenology . note that , since the @xmath113 function is generally taken as a monotonic ( increasing ) function of its argument , the second derivative of @xmath48 is positive and the scalar waves propagate superluminally . the theory were thus thought to be acausal @xcite . we however argued in sec . [ causalite ] and sec . [ sectionkessence ] that this conclusion is correct only if the gravitational metric defines a favored chronology , an assumption that may be dropped . a critical point is however that the free function @xmath48 ( related to @xmath113 ) must satisfy the conditions eq . ( [ condhyperkessencemat ] ) . it is however immediate to note that the asymptotic conditions on @xmath113 implies that when @xmath57 goes to @xmath106 , @xmath103 and @xmath136 also go to zero and the effective metric is completely degenerate . what it means is that , in such a theory , in an astrophysical context , there must exist around each galaxy or cluster a singular surface on which the scalar degree of freedom does not propagate . the reason is that , near the source , @xmath57 must be positive , but negative far from it due to the cosmological background . this theory can therefore not lead to a consistent picture of local physics imbedded into a cosmological background . this major objection also applies to the recent relativistic model of mond of @xcite . a trivial modification of the asymptotic form of the @xmath113 function however cures the problem . let us consider that @xmath137 if @xmath117 , or equivalently that @xmath138 if @xmath135 . it ensures that the theory is well behaved at the transition between local and cosmological physics . this slight modification induces a significant change in the phenomenology because there is a return to a newtonian behavior very far from the source , with a renormalized value of the gravitational constant . such a theory thus predicts that rotation curves are only approximatively flat on a finite range of @xmath139 , and current data requires @xmath140 to be at most of order @xmath141 . more details on this can be found in @xcite . interestingly , in that kind of theory , mond only appears as an intermediate regime between two newtonian ones only differing by the value of the gravitational constant , the transition , driven by the scalar field , occurring at milgrom s acceleration scale . mond - like theories of gravity must also predict enhanced light deflection in order to be consistent with the data . in the previous theory however , the conformal coupling of the scalar field to the matter metric implies that light is not coupled to the scalar field because of the conformal invariance of electromagnetism in four dimensions . this led authors to consider more general bimetric theory of gravity in which matter is coupled to a disformal metric of the type eq . ( [ disformalmetric ] ) @xcite . in these first models , the vector field @xmath84 in eq . ( [ disformalmetric ] ) was assumed to be the gradient of the k - essence scalar field , and @xmath30 and @xmath32 were some functions of @xmath2 and @xmath57 . it was however proved that when the matter light - cone is wider than the gravitational one ( see sec . [ bimetrictheories ] ) then there is actually less light deflection than in gr . we have seen however that causality does not require the matter light - cone to be wider than the gravitational one . on the contrary , it could be inside the gravitational one and the light deflection would be enhanced ( compared to gr ) . such a framework is therefore relevant for relativistic theories of the mond paradigm , as we stress in @xcite . the cauchy problem is very likely to be well posed , since the ( einstein ) equation of the metric field is still diagonalized , hyperbolic and of the second order . the scalar field equation in vacuum has a well - posed cauchy problem . the scalar field equation inside the matter is however quite complicated and is not diagonalized . the cauchy problem inside the matter is thus a quite involved question , but we expect that some generic requirements on the free functions @xmath30 and @xmath32 and on energy conditions of the matter sector will guarantee it , and by the way , that the whole theory has a well posed cauchy problem . this is left for further investigation ( see also @xcite ) . let us briefly consider another application of bimetric theories of gravity . conventional physics did not succeeded so far in explaining the anomalous motion of the two pioneer spacecrafts @xcite that experience a small anomalous acceleration ( roughly directed towards the sun ) . unconventional theories of gravity could however explain it . the two pioneer spacecrafts are moving along geodesics of the matter metric @xmath6 , and we shall look for a slight modification of the schwarzschild solution to explain the pioneer anomaly . it is worth noting that this anomaly was detected soon after jupiter s flyby @xcite so that we have to check that the suggested modification is compatible with the motion of outer planets . outer planets are moving along quasi - circular orbits , and are thereby mostly sensitive to the time - time component of the matter metric @xmath142 . very sensitive tests of kepler s third law @xcite then put stringent bounds on the deviation from the leading order of the time - time component of the schwarzschild metric . these bounds are actually too small to account for the pioneer anomaly . this fact leads to the wrong statement in @xcite that the pioneer anomaly can not be of gravitational origin . this is indeed not justifiable since the two spacecrafts evolve on hyperbolic trajectories and are thereby also sensitive to the radial - radial component of the matter metric , which is not well constrained in the outer solar system . let us stress that , if spherical symmetry is assumed , a slight modification of @xmath143 compared to the schwarzschild solution is the only way to give the pioneer anomaly a gravitational origin in a metric theory of gravity . this was notably realized in @xcite . here we provide , with the help of bimetric theories of gravity , a framework that realizes this modification of the radial - radial component . the action of the theory is given by the eintein - hilbert action , a canonical action for the scalar field , and the matter fields are coupled to the matter metric @xmath144 where @xmath57 is still defined by eq . ( [ defdex ] ) . as we already stressed , non trivial properties of @xmath30 , @xmath32 and energy conditions of the matter sector may be required to ensure the hyperbolicity of the scalar equation inside matter . in addition , the matter metric has to be lorentzian and this reads @xmath145 . in a static and spherical symmetric situation , we have @xmath146 and @xmath147 . the bare metric @xmath7 is given by the solution of einstein equations and it can be shown that it coincides with the schwarzschild solution to the leading order . this theory is thus a realization of the above phenomenology . such models predict `` pioneer - like '' anomalies in the precession of perihelion that could notably be found in precise measures of the orbit of mars . more details on the pioneer anomaly and disformal theories can be found in @xcite . the first varying speed of light theories were constructed by replacing @xmath0 by @xmath8 in the equations of motion of gr , where @xmath148 could be the cosmic time @xcite . this however leads to equations of motion that do not conserve stress - energy anymore . in other words , this theory can not be obtained from a variational principle @xcite . some authors thus made use of the disformal matter metric eq . ( [ metriquedisfscalaire ] ) to reproduce vsl within a consistent framework @xcite . note that in all of these models , the free functions were taken as constants @xmath149 and @xmath150 . let us stress that if one insists on the weak equivalence principle there can not be any coupling between @xmath2 and the standard model matter other than through the matter metric . then , by varying the action with respect to @xmath2 , one finds that the scalar field is not created by the matter sources ( @xmath151 is always a solution ) . the scalar field can only be generated by some ( non constant ) function @xmath152 , like in scalar - tensor theories . note also that the choice of @xmath153 in these works does not guarantee the lorentzian character of the matter metric , which can actually be euclidean in a cosmological background such that @xmath154 . one could however circumvent this problem by arguing that the theory is only an effective model valid for @xmath155 , and where @xmath50 could be of the order of the gut s scale . as we already stressed , the initial value formulation of the scalar field equation in such models is a quite involved question . this point has not been pointed out although the existence of an initial value formulation is a basic requirement that must be checked in order to give any sense to such vsl models . let us finally remark that , since the fine structure constant @xmath93 is proportional to the inverse of the speed of light , it has been argued that vsl theories could also account for the variation of @xmath93 with cosmic time , if any @xcite . however , in this bimetric framework of vsl theories , if one analyzes atomic rays at some redshift by usual techniques , one is observing electromagnetic phenomenon using matter rods and clocks , so that no variations of @xmath93 are actually observable . the fact that the ratio of the speed of light to the speed of gravitational waves varies in space and time does not lead to a variation of @xmath93 ( if it is measured in the usual way ) , but simply to a redefinition of the redshift @xmath156 of distant objects . we provided a careful analysis of the meaning of causality in classical field theories . this led us to the conclusion that superluminal behaviors are generically found to be non causal _ only if _ one refers to a _ prior _ chronology on spacetime . this postulate actually states that , locally , some sets of inertial coordinates must be preferred ones , and this appears to be in great conflict with the spirit of general relativity , and more precisely with the spirit of general covariance . on the contrary , we derived in sec . [ causalityandsuperluminalbehav ] , by means of the conditions ( a)(c ) , a formulation of causality in which coordinates are still physically meaningless . note that while referring to a preferred chronology may seem natural in gr since all fields ( both gravity and matter ) propagate along the gravitational metric @xmath7 , it becomes somewhat unnatural whenever spontaneous breaking of lorentz invariance occur , be it driven by quantum polarization ( and in general , the solution of field equations ) , nonlinearities of some new fields , etc . the resulting spacetime is generically endowed with a finite set of lorentzian metrics @xmath16 which may not be conformally related to each others . in that case rods and clocks made of different fields lead to different systems of coordinate that do not transform under the same lorentz group . there is not only one , but at least two invariant speeds in that case . note that it would be misleading to think that there is a preferred lorentz invariance in the theory because of the symmetry of the action . actually , in a gr - like context , the action is not lorentz invariant but diffeomorphism invariant . this notably means that all local system of coordinates are equivalent . on the contrary , the existence of a preferred chronology means that ( as far as causality is concerned ) there exists some preferred class of inertial coordinates , or equivalently some preferred rods and clocks . this seems to be physically unacceptable . as a consequence , one _ can not _ refer to causality in order to assert that nothing can travel faster than one of these speeds . accordingly , causality does not require one of these metric to define a wider cone than the others everywhere on spacetime . bimetric theories of gravity greatly enlightened this point , since both the gravitational and the matter metric could be used to define a `` natural '' chronology on spacetime . depending on the choice made , one then finds that nothing can travel faster than gravity or light . these two opposite requirements can be found in the literature . on the contrary , our definition of causality in which no prior chronology is assumed ( by means of our `` mixed '' chronology , see sec.[causalityandsuperluminalbehav ] ) , enables a causal theory to involve `` superluminal '' propagations . actually the very notion of superluminal behavior is no more meaningful in that framework , and the only point is that some degrees of freedom can propagate faster than other ones . moreover the causal cones may even tip over each other depending on the location on spacetime . as an application , in k - essences theories , causality does _ not _ require the sign of @xmath157 to be fixed ( and notably negative ) , and , in bimetric theory causality does not require either the gravitational light - cone to be wider than the matter one ( nor conversely ) . what actually requires causality is first that the equations of motion have a well - posed cauchy problem . this strongly depends on the precise form of the dynamics , and throughout this paper we used classical results @xcite on that subject . we also discussed in detail the fact that global properties of spacetimes may break causality . we pointed out that this already occurs in standard gr _ so that it can not be related to some intrinsic disease of superluminal propagation_. we discuss three very similar cases : gott s cosmic strings @xcite , the two bubbles of non trivial vacua in k - essences theories @xcite and the two casimir experiments of @xcite . what thus became clear was that such non trivial global properties may be actually suppressed by subtle boundaries effects . this would be an illustration of the chronology protection conjecture . r. c. tolman _ the theory of relativity of motion _ ( university of california press , berkeley , 1917 ) g. a. benford , d. l. book and w. a. newcomb , phys . rev . d**2 * * , 263 ( 1970 ) s. liberati , s. sonego and m. visser , annals phys.*298 * , 167 ( 2002 ) a. d. dolgov and i. d. novikov , phys . lett . b**442 * * , 82 ( 1998 ) m. yu konstantinov , russ . j. * 45 * , 23 ( 2002 ) ; g - qc/9810019 y. aharonov , a. komar and l. susskind , phys . rev . * 182 * , 1400 ( 1969 ) a. adams , n. arkani - hamed , s. dubovsky , a. nicolis and r. rattazzi , _ causality , analyticity and an ir obstruction to uv completion _ , hep - th/0602178 e. a. lim , phys . d**71 * * , 063504 ( 2005 ) c. bonvin , c. caprini , r. durrer _ a no - go theorem for k - essences dark energy _ , astro - ph/0606584 j. w. elliott , g. d. moore and h. stoica , jhep 0508 , 066 ( 2005 ) j. d. bekenstein , phys . d**48 * * , 3641 ( 1992 ) j. d. bekenstein , astrophys . j. * 429 * , 480 ( 1994 ) j. d. bekenstein , phys . d**70**,083509 ( 2004 ) i. t. drummond and s. j. hathrell , phys . d**22**,343 ( 1980 ) ; k. scharnhorst , phys . lett . b**236 * * , 354 ( 1990 ) ; j. i. lattore , p. pascual and r. tarrach , nucl . phys . b**437 * * , 60 ( 1995 ) ; w. dittrich and h. gies , phys . d**58 * * , 025004 ( 1998 ) g. m. shore , venice 2002 time and matter , 45 - 66 ; gr - qc/0302116 k. scharnhorst , annalen phys . * 7 * , 700 ( 1998 ) j. w. moffat , found . * 23 * , 411 ( 1993 ) s. w. hawking and g. f. r. ellis , _ the large scale structure of spacetime _ , cambridge university press ( 1973 ) r. m. wald , _ general relativity _ , the university of chicago press ( 1984 ) m. a. clayton and j. w. moffat , int . phys . d**11 * * , 187 ( 2002 ) j. d. bekenstein and m. milgrom , astrophys.j . * 286 * , 7 ( 1984 ) t. chiba , t. okabe and m. yamaguchi , phys . rev d**62 * * , 023511 ( 2000 ) ; c. armendriz - picn , v. mukhanov and p.j . steinhardt , phys . rev d**63 * * , 103510 ( 2001 ) c. armendriz - picn , t. damour and v. mukhanov , phys . lett . b**458 * * , 219 ( 1999 ) m. milgrom , astrophys . j. * 270 * , 365 ( 1983 ) r. h. sanders , astrophys . j. * 480 * , 492 ( 1997 ) i. t. drummond , _ variable light - cone theory of gravity _ , gr - qc/9908058 ; i. t. drummond , phys . rev . d**63**,043503 ( 2001 ) m. a. clayton and j. w. moffat , phys . lett . b**460 * * , 263 ( 1999 ) ; m. a. clayton and j. w. moffat , phys . lett . b**477 * * , 269 ( 2000 ) b. a. bassett , s. liberati , c. molina - pars and m. visser , phys . rev . d**62 * * , 103518 ( 2000 ) j. d. barrow and j. magueijo , phys . lett . b**443 * * , 104 ( 1998 ) ; j. d. barrow , phys . rev . d**59 * * , 043515 ( 1999 ) ; a. albrecht and j. magueijo , phys . rev . d**59 * * , 043516 ( 1999 ) ; j. d. barrow and j. magueijo , class . * 16 * , 1435 ( 1999 ) g. f. r. ellis and j .- p . uzan , am . * 73 * , 240 ( 2005 ) c. w. misner , k. s. thorne and j. a. wheeler , _ gravitation _ ( freeman , san francisco , 1973 ) . s. w. hawking , phys . d**46 * * , 603 ( 1992 ) e. bois and e. trelut , _ limpact de la structure chrono - gomtrique de lespace - temps sur la causalit . _ , astro - ph/0304520 j. z. simon , phys . rev . d**41 * * , 3720 ( 1990 ) ; r. p. woodard _ avoiding dark energy with 1/r modifications of gravity _ , astro - ph/0601672 r. courant and d. hilbert , _ methods of mathematical physics , vol ii _ , interscience , new york ( 1962 ) j. r. gott , phys . . lett . * 66 * , 1126 ( 1991 ) t. jacobson and d. mattingly , phys . rev . d**64 * * , 024028 ( 2001 ) e. j. copeland , m. sami and s. tsujikawa , _ dynamics of dark energy _ , hep - th/0603057 c. armendariz - picon and e. a. lim , `` haloes of k - essence '' jcap 0508 ( 2005 ) j .- p . bruneton and g. esposito - farse , _ field - theoretical formulations of mond - like gravity _ , to be published . r. r. caldwell , phys . b**545 * * , 23 - 29 ( 2002 ) s. m. carroll , m. hoffman and m. trodden , phys . d**68 * * , 023509 ( 2003 ) j. m. cline , s. jeon and g. d. moore , phys . rev . d**70 * * , 043543 ( 2004 ) n. arkani - hamed , h - c . cheng , m. a. luty and s. mukohyama , jhep 0405 , 074 ( 2004 ) t. buchert , j. larena and j - m . alimi , _ correspondence between kinematical backreaction and scalar field cosmologies - the ` morphon field ' _ , gr - qc/0606020 r. h. sanders ans s. s. mcgaugh , ann . . astrophys . * 40 * , 263 ( 2002 ) j. d. anderson , p. a. laing , e. l. lau , a. s. liu , m. m. nieto and s. g. turyshev , phys . rev . d**65 * * , 082004 ( 2002 ) ; v. t. toht and s. g. turyshev _ the pioneer anomaly : seeking an explanation in newly recovered data _ gr - qc/0603016 c. talmadge , j .- p . berthias , r. w. hellings and e. m. standish , phys . 61 * , 1159 ( 1988 ) k. tangen , _ could the pioneer anomaly have a gravitational origin ? _ , gr - qc/0602089 m .- t . jaekel and s. reynaud , mod . * * , 1047 ( 2005 ) ; m .- t . jaekel and s. reynaud , class . * 23 * , 777 ( 2006 ) n. kaloper , phys . b**583 * * , 1 - 13 ( 2004 ) j. k. webb , v. v. flambaum , c. w. churchill , m. j. drinkwater and j. d. barrow , phys . lett . * 82 * , 884 ( 1999 ) a. v. ivanchik , e. rodriguez , p. petitjean and d.a . varshalovich , astron . lett . * 28 * , 423 ( 2002 )

there has been important progress in the question of local energy decay for the solution of the wave equation in black hole type space - times other the last years . the best results are now known in the schwarzschild space - time . we refer to the papers of blue soffer @xcite , blue sterbenz @xcite , and dafermos rodnianski @xcite and references therein for an overview . see also the paper of finster kamran smoller yau for the kerr space - time @xcite . results on the decay of local energy are believed to be a prerequisite for a possible proof of the global nonlinear stability of these space - times . today global nonlinear stability is only known for the minkowski space - time ( see @xcite ) . from our point of view one of the most efficient approaches to the question of local energy decay is the theory of resonances . resonances correspond to the frequencies and rates of dumping of signals emitted by the black hole in the presence of perturbations ( see ( * ? ? ? * chapter 4.35 ) ) . on the one hand these resonances are today an important hope of effectively detecting the presence of a black hole as we are theoretically able to measure the corresponding gravitational waves . on the other hand , the distance of the resonances to the real axis reflects the stability of the system under the perturbation : larger distances correspond to more stability . in particular the knowledge of the localization of resonances permits to answer the question if there is decay of local energy and at which rate . the aim of the present paper is to show how this method applies to the simplest model of a black hole : the de sitter schwarzschild black hole . in the euclidean space , such results are already known , especially in the non trapping geometry . the first result is due to lax and phillips ( see their book ( * ? ? ? * theorem iii.5.4 ) ) . they have proved that the cut - off propagator associated to the wave equation outside an obstacle in odd dimension @xmath1 ( more precisely the lax phillips semi - group @xmath2 ) has an expansion in terms of resonances if @xmath3 is compact for one @xmath4 . in particular , the local energy decays exponentially uniformly . from melrose sjstrand @xcite , this assumption is true for non trapping obstacles . vanberg @xcite has obtained such results for general , non trapping , differential operators using different techniques . in the trapping case , we know , by the work of ralston @xcite , that it is not possible to obtain uniform decay estimates without loss of derivatives . in the exterior domain of two strictly convex obstacles , the local energy decays exponentially with a loss of derivatives , by the work of ikawa @xcite . this situation is close to the one treated in this paper . we also mention the works tang zworski @xcite and burq zworski @xcite concerning the resonances close to the real line . thanks to the work of s barreto and zworski ( @xcite ) we have a very good knowledge of the localization of resonances for the wave equation in the de sitter schwarzschild metric . using their results we can describe an expansion of the solution of the wave equation in the de sitter schwarzschild metric in terms of resonances . the main term in the expansion is due to a zero resonance . the error term decays polynomially if we permit a logarithmic derivative loss in the angular directions and exponentially if we permit an @xmath0 derivative loss in the angular directions . for initial data in the complement of a one - dimensional space the local energy is integrable if we permit a @xmath5 derivative loss with @xmath6 . this estimate is almost optimal in the sense that it becomes false for @xmath7 . the method presented in this paper does not directly apply to the schwarzschild case . this is not linked to the difficulty of the photon sphere which we treat in this paper , but to the possible accumulation of resonances at the origin in the schwarzschild case . the exterior of the de sitter schwarzschild black hole is given by @xmath8 r_{- } , r_{+ } [ _ { r } \times \s^{2}_{\omega } \label{a1 } \\ g = \alpha^{2 } d t^{2 } - \alpha^{-2 } d r^{2 } - r^{2 } \vert d \omega \vert^{2 } , \quad \alpha = \big ( 1 - \frac{2 m}{r } - \frac{1}{3 } \lambda r^{2 } \big)^{1/2 } , \label{a2}\end{gathered}\ ] ] where @xmath9 is the mass of the black holes and @xmath10 is the cosmological constant . @xmath11 and @xmath12 are the two positive roots of @xmath13 . we also denoted by @xmath14 the standard metric on @xmath15 . the corresponding dalembertien is @xmath16 where @xmath17 and @xmath18 is the positive laplacian on @xmath15 . we also denote @xmath19 the operator on @xmath20 which governs the situation on @xmath21 . we define @xmath22 on @xmath23 , and , in the coordinates @xmath24 , we have @xmath25 we introduce the regge wheeler coordinate given by @xmath26 in the coordinates @xmath27 , the operator @xmath28 is given by @xmath29 on @xmath30 . let @xmath31 and @xmath32 be the potentials appearing in the previous operator . as stated in proposition 2.1 of @xcite , the work of mazzeo melrose @xcite implies that for @xmath33 @xmath34 has a meromorphic extension from the upper half plane to @xmath35 . the poles @xmath36 of this meromorphic extension are called resonances . we recall the principal result of @xcite : _ [ a52 ] there exists @xmath37 and @xmath38 such that for any @xmath39 there exists an injective map , @xmath40 , from the set of pseudo - poles @xmath41 into the set of poles of the meromorphic continuation of @xmath42 such that all the poles in @xmath43 are in the image of @xmath40 and for @xmath44 , @xmath45 if @xmath46 , @xmath47 , @xmath48 , then the corresponding pole , @xmath49 , has multiplicity @xmath50 . _ ( 0,0 ) near the real axis.,title="fig : " ] # 1#2#3#4#5 @font ( 12044,3419)(-21,-3158 ) ( 6976,-361)(0,0)[lb ] the natural energy space @xmath51 for the wave equation is given by the completion of @xmath52 in the norm @xmath53 it turns out that this is not a space of distributions . the problem is very similar to the problem for the wave equation in dimension @xmath54 . we therefore introduce another energy space @xmath55 defined as the completion of @xmath52 in the norm @xmath56 note that for any @xmath57 the norms @xmath58 and @xmath59 are equivalent . we will therefore only work with the space @xmath59 in the future and note it from now on @xmath60 . let us write the wave equation as a first order equation in the following way : @xmath61 let @xmath62 be the scale of sobolev spaces associated to @xmath28 . we note @xmath63 the completion of @xmath64 in the norm @xmath65 then @xmath66 is selfadjoint on @xmath51 . we note @xmath67 the scale of sobolev spaces associated to @xmath68 . note that because of @xmath69 we can define a meromorphic extension of the cut - off resolvent of @xmath68 by using the meromorphic extension of the cut - off resolvent of @xmath28 and the resonances of @xmath68 coincide with the resonances of @xmath28 . recall that @xmath70 is a selfadjoint operator with compact resolvent . its eigenvalues are the @xmath71 with multiplicity @xmath72 . we denote @xmath73 the operator restricted to @xmath74 where @xmath75 is the eigenspace to the eigenvalue @xmath76 . in the following , @xmath77 will be identify with the operator on @xmath78 given by . the spaces @xmath79 are defined in an analogous way to the spaces @xmath80 . let @xmath81 be the orthogonal projector on @xmath82 . for @xmath83 , the space @xmath84 and @xmath85 are the same and the norms are uniformly equivalent with respect to @xmath86 . using proposition ii.2 of bachelot and motet - bachelot @xcite , the group @xmath87 preserves the space @xmath60 and there exists @xmath88 such that @xmath89 from the previous discussion , the same estimate holds with @xmath90 uniformly in @xmath83 . in particular , @xmath91 is bounded on @xmath60 for @xmath92 , and we note @xmath93 for @xmath48 . we first need a result on @xmath28 : _ [ a50 ] for @xmath83 , the operator @xmath94 has no resonance and no eigenvalue on the real axis . _ for @xmath95 , @xmath96 has no eigenvalue in @xmath97 and no resonance in @xmath98 . but , @xmath99 is a simple resonance of @xmath96 , and , for @xmath100 closed to @xmath99 , we have @xmath101 where @xmath1020 , + \infty [ $ ] and @xmath103 is a holomorphic ( bounded ) operator near @xmath99 . equation is an equality between operators from @xmath104 to @xmath105 . the proof of proposition [ a50 ] is given in section [ a51 ] . for @xmath33 we put in the following : @xmath106 for a resonance @xmath107 we define @xmath108 by the laurent expansion of the cut - off resolvent near @xmath107 : @xmath109 we also define @xmath110 by @xmath111 the main result of this paper is the following : _ [ mainth ] let @xmath33 . _ @xmath112 let @xmath113 such that there is no resonance with @xmath114 . then there exists @xmath9 with the following property . let @xmath115 such that @xmath116 . then we have : @xmath117 with @xmath118 and the sum is absolutely convergent in the sense that @xmath119 @xmath120 there exists @xmath121 with the following property . let @xmath122 , positive , strictly decreasing with @xmath123 for @xmath124 large . let @xmath125 be such that @xmath126 . then we have @xmath127 with @xmath128 [ a69 ] a ) by the results of s barreto and zworski we know that there exists @xmath129 such that @xmath99 is the only resonance in @xmath130 . choosing this @xmath131 in @xmath112 the sum on the right hand side contains a single element which is @xmath132 \b ) again by the paper of s barreto and zworski we know that @xmath133 for all the @xmath107 s outside a compact set ( see theorem [ a52 ] ) . for such @xmath134 , we have @xmath135 and @xmath136 is an operator of rank @xmath72 . \c ) let @xmath137 . by part @xmath120 of the theorem , for @xmath138 , the local energy is integrable if @xmath139 , for some @xmath140 , and decays exponentially if @xmath141 for some @xmath142 . \d ) in fact , we can replace @xmath143 by @xmath144 in the first part of the theorem . and , by an interpolation argument , we can obtain the following estimate : for all @xmath142 , there exists @xmath145 such that @xmath146 with @xmath147 in the schwarzschild case the potential @xmath148 is only polynomially decreasing at infinity and we can not apply the result of mazzeo melrose . therefore we can not exclude a possible accumulation of resonances at @xmath99 . this difficulty has nothing to do with the presence of the photon sphere which is treated by the method presented in this paper . let @xmath149 such that @xmath150 for some @xmath6 . then we have from part @xmath120 of the theorem , for @xmath151 , @xmath152 this estimate is almost optimal since it becomes false for @xmath7 . indeed we have ( @xmath151 ) : @xmath153 thus from ( [ lee ] ) we obtain the resolvent estimate @xmath154 it is easy to see that this entails the resolvent estimate @xmath155 we introduce the semi - classical parameter @xmath156 and @xmath157 as in section [ a65 ] . then , for @xmath158 , the above estimate gives the semi - classical estimate : @xmath159 for @xmath160 ( see and ) . such estimate is known to be false for @xmath7 and @xmath161 , the maximum value of the potential @xmath148 ( see ( * ? ? ? * proposition 2.2 ) ) . the proof of the theorem is based on resolvent estimates . using ( [ rl ] ) we see that it is sufficient to prove resolvent estimates for @xmath162 . this is the purpose of the next section . * acknowledgments : * we would like to thank a. bachelot for fruitful discussions during the preparation of this article . this work was partially supported by the anr project jc0546063 `` equations hyperboliques dans les espaces temps de la relativit gnrale : diffusion et rsonances '' . in this section , we obtain estimates for the cut - off resolvent of @xmath94 , the operator @xmath28 restricted to the spherical harmonic @xmath86 . we will use the description of the resonances given in s barreto zworski @xcite . recall that @xmath163 has a meromorphic extension from the upper half plane to @xmath35 . the resonances of @xmath28 are defined as the poles of this extension . we treat only the case @xmath164 since we can obtain the same type of estimates for @xmath165 using @xmath166 . _ [ a44 ] let @xmath167 be fixed . the operators @xmath168 satisfy the following estimates uniformly in @xmath86 . _ \i ) for all @xmath158 , the number of resonances of @xmath28 is bounded in @xmath169 . moreover , there exists @xmath39 such that @xmath170 for all @xmath171 . \ii ) for @xmath172 large enough , @xmath94 has no resonance in @xmath173 + i [ - c_{0 } , 0]$ ] . moreover , there exists @xmath39 such that @xmath174 for @xmath175 + i [ - c_{0 } , c_{0 } ] $ ] . \iii ) let @xmath172 be fixed . for @xmath86 large enough , the resonances of @xmath94 in @xmath176 + i[-c_{0 } , 0]$ ] are the @xmath177 given in theorem [ a52 ] ( in particular their number is bounded uniformly in @xmath86 ) . moreover , there exists @xmath39 such that @xmath178 for @xmath179 + i [ - c_{0 } , c_{0 } ] $ ] . furthermore , @xmath94 has no resonance in @xmath180 + i [ - \varepsilon , 0]$ ] , for some @xmath142 , and we have @xmath181 for @xmath179 + i [ - \varepsilon , 0]$ ] . \iv ) let @xmath182 be fixed . for @xmath172 large enough , @xmath94 has no resonance in @xmath183 . moreover , there exists @xmath39 such that @xmath184 for @xmath185 and @xmath186 . the results concerning the localization of the resonances in this theorem are proved in @xcite and @xcite , the following figure summaries the different estimates of the resolvent . ( 0,0).,title="fig : " ] # 1#2#3#4#5 @font ( 12591,6655)(418,-6769 ) ( 9601,-5686)(0,0)[lb ] ( 976,-2911)(0,0)[lb ] ( 4276,-2911)(0,0)[lb ] ( 7876,-2911)(0,0)[lb ] ( 10351,-2911)(0,0)[lb ] ( 3151,-5011)(0,0)[lb ] ( 2026,-3211)(0,0)[lb ] ( 3001,-1861)(0,0)[lb ] ( 5476,-1786)(0,0)[lb ] ( 8626,-1786)(0,0)[lb ] ( 6601,-1936)(0,0)[lb ] in zone _ i _ which is compact , the result of mazzeo melrose @xcite gives a bound uniform with respect to @xmath86 ( outside of the possible resonances ) . in particular , part _ i ) _ of theorem [ a44 ] is a direct consequence of this work . in zone _ ii _ , the result of zworski @xcite gives us a good ( uniform with respect to @xmath86 ) estimate of the resolvent . here , we use the exponential decay of the potential at @xmath187 and @xmath188 . by comparison , the equivalent potential for the schwarzschild metric does not decay exponentially , and our present work can not be extended to this setting . please note that this problem concerns only the zones _ i _ and _ ii _ , but the zones _ iii _ and _ iv _ can be treated the same way . in zone _ iii _ , we have to deal with the so called `` photon sphere '' . the estimate follows from a general polynomial bound of the resolvent in dimension @xmath54 ( see @xcite ) . in zone _ iv _ , the potentials @xmath189 and @xmath190 are very small in comparison to @xmath191 . so they do nt play any role , and we obtain the same estimate as in the free case @xmath192 ( or as in the non trapping geometry ) . @xmath193 [ a51 ] this part is devoted to the proof of proposition [ a50 ] and of part _ i ) _ of theorem [ a44 ] . since @xmath194 has a meromorphic extension to @xmath35 , the number of resonance in @xmath195 is always bounded and point _ i ) _ of theorem [ a44 ] is clear . it is a classical result ( see theorem xiii.58 in @xcite ) that @xmath94 has no eigenvalue in @xmath98 . on the other hand , from proposition ii.1 of the work of bachelot and motet - bachelot @xcite , @xmath99 is not an eigenvalue of the operators @xmath94 . moreover , from the limiting absorption principle @xcite , @xmath196 for @xmath197 and any @xmath198 , we know that @xmath94 has no resonance in @xmath98 . we now study the resonance @xmath99 using a technique specific to the one dimensional case . we start with recalling some facts about outgoing jost solutions . let @xmath199 be a schrdinger operator with @xmath200 decaying exponentially at infinity . for @xmath201 , there exists a unique couple of functions @xmath202 such that @xmath203 the function @xmath204 is called the outgoing jost solution at @xmath205 . since @xmath206 decays exponentially at infinity , the functions @xmath204 can be extended , as a @xmath207 function in @xmath124 , analytically in a strip @xmath208 , for some @xmath142 . moreover , in such a strip , they satisfy @xmath209 for some @xmath145 . all these properties can be found in theorem xi.57 of @xcite . using these jost solutions , the kernel of @xmath210 , for @xmath201 takes the form @xmath211 where @xmath212 is the heaviside function @xmath213 and @xmath214 is the wronskian between @xmath215 and @xmath216 ( the right hand side of does not depend on @xmath124 ) . in particular , @xmath217 is an analytic function on @xmath35 . since the @xmath204 are always non - zero thanks to , the resonances are the zeros of @xmath218 . such a discussion can be found in the preprint of tang zworski @xcite . remark that @xmath94 is of the form . if @xmath99 is a resonance of one of the @xmath94 s with @xmath83 , the jost solutions @xmath219 are collinear . in particular , from and , the @xmath220 function @xmath221 converge to two non zero limits at @xmath205 and @xmath222 goes to @xmath99 as @xmath223 . since @xmath224 we get , by an integration by parts , @xmath225_{-r}^{r } . \label{b1}\end{aligned}\ ] ] since @xmath226 , the last term in goes to @xmath99 as @xmath172 goes to @xmath187 . thus , if @xmath83 , gives @xmath227 and @xmath99 is not a resonance of @xmath94 . we study now the case @xmath95 . if @xmath228 satisfies @xmath229 , we get from @xmath230 where @xmath231 is a constant . then @xmath232 where @xmath233 are constants . note that @xmath234 as @xmath223 . since @xmath235 are @xmath220 functions bounded at @xmath205 from which satisfy @xmath229 , the two functions @xmath235 are collinear to @xmath236 and then @xmath237 which means that @xmath99 is a resonance of @xmath96 . the resolvent of @xmath96 has thus the form @xmath238 where @xmath239 is an analytic family of bounded operators near @xmath99 and @xmath240 . for all @xmath241 with @xmath142 , we have @xmath242 from the functional calculus . this inequality implies that @xmath243 and @xmath244 if @xmath245 is in the range of @xmath246 , we have @xmath247 and @xmath248 . then , @xmath249 for all @xmath250 and @xmath251 is an eigenvector of @xmath96 for the eigenvalue @xmath99 . this point is impossible because @xmath96 has no eigenvalue . thus @xmath252 and @xmath253 . so @xmath218 has a zero of order @xmath54 at @xmath254 . since @xmath255 , implies that the kernel of @xmath256 is given by @xmath257 finally , since @xmath258 as @xmath259 and since @xmath260 is a strictly positive operator , we get @xmath261 for all @xmath262 . in particular , @xmath263 and then @xmath102 0 , + \infty [ $ ] . @xmath193 in this section , we give an upper bound for the cut - off resolvent for @xmath264 + i [ - c_{0 } , c_{0 } ] $ ] . we assume that @xmath265 + i [ -c_{0 } , c_{0 } ] $ ] with @xmath266 $ ] , and define a new semi - classical parameter @xmath267 , a new spectral parameter @xmath268 + i[- 4 c_{0 } h , 4 c_{0 } h]$ ] and @xmath269 with these notations , we have @xmath270 we remark that @xmath271 in our window of parameters . the potentials @xmath272 and @xmath190 have a holomorphic extension in a sector @xmath273 for some @xmath274 . from the form of @xmath275 ( see ) , there exist @xmath276 and functions @xmath277)$ ] , @xmath278 , analytic in @xmath279 such that @xmath280 for @xmath281 and @xmath282 . moreover , @xmath283 have a ( non zero ) limit for @xmath223 , @xmath281 . under these hypotheses , and following proposition 4.4 of @xcite , we can use the specific estimate developed by zworski in @xcite for operators like with @xmath272 satisfying . in the beginning of section 4 of @xcite , zworski defines a subtle contour @xmath284 briefly described in the following figure . ( 0,0 ) and the contour @xmath284.,title="fig : " ] # 1#2#3#4#5 @font ( 13844,6066)(-21,-5794 ) ( 8251,-1861)(0,0)[lb ] ( 9601,-3361)(0,0)[lb ] ( 10426,-3361)(0,0)[lb ] ( 2101,-2386)(0,0)[lb ] ( 1276,-2386)(0,0)[lb ] ( 976,-1711)(0,0)[lb ] ( 9526,-4036)(0,0)[lb ] ( 6301,-211)(0,0)[lb ] ( 7801,-1336 ) ( 12076,-2311)(0,0)[lb ] ( 7951,-3211)(0,0)[lb ] ( 12301,-961)(0,0)[lb ] recall that the distorted operator @xmath285 is defined by @xmath286 for all @xmath287 analytic in @xmath279 and then extended as a differential operator on @xmath288 by means of almost analytic functions . the resonances of @xmath289 in the sector @xmath2900 , + \infty [ \ } = e^{2 i ] - \theta , 0 ] } ] 0 , + \infty [ $ ] are then the eigenvalues of @xmath291 in that set . for the general theory of resonances , see the paper of sjstrand @xcite or his book @xcite . for @xmath292 large enough , proposition 4.1 of @xcite proves that @xmath289 has no resonances in @xmath293 + i[- 4 c_{0 } h , 4 c_{0 } h]$ ] . moreover , for @xmath100 in that set , this proposition gives the uniform estimate @xmath294 since @xmath284 coincides with @xmath97 for @xmath295 , we have @xmath296 from lemma 3.5 of @xcite . using , we immediately obtain @xmath297 which is exactly . @xmath193 [ a65 ] in this part , we study the cut - off resolvent for the energy @xmath298 + i [ - c_{0 } , c_{0 } ] $ ] . in that zone , we have to deal with the photon sphere . we define the new semi - classical parameter @xmath299 and @xmath300 as previously , we have @xmath301 where @xmath302 + i [ - 3 r c_{0 } h , 0 ] = [ a , b ] + i [ -c , c ] , \ ] ] with @xmath303 and @xmath304 . note that @xmath272 is of the form : ( 0,0).,title="fig : " ] # 1#2#3#4#5 @font ( 10844,4244)(1179,-3983 ) ( 3751,-361)(0,0)[lb ] ( 8551,-1411)(0,0)[lb ] ( 7051,-3136)(0,0)[lb ] ( 11776,-2536)(0,0)[lb ] in particular , @xmath272 has a non - degenerate maximum at energy @xmath305 . on the other hand , for @xmath306 , @xmath307 , the energy level @xmath100 is non trapping for @xmath308 , the principal semi - classical symbol of @xmath289 . we defined @xmath291 by standard distortion ( see sjstrand @xcite ) and can apply the following general upper bound on the cut - off resolvent in dimension one . _ [ a29 ] we assume that @xmath309 and that the critical points of @xmath310 on the energy level @xmath311 are non - degenerate ( i.e. the points @xmath312 such that @xmath313 satisfy @xmath314 is invertible ) . then , there exists @xmath142 such that , for @xmath315 $ ] and @xmath316 with @xmath317 large enough , @xmath318 for @xmath319 and some @xmath9 which depends on @xmath320 . _ note that there is a slight error in the statement of the lemma in @xcite . indeed , @xmath321 depends on @xmath320 , and in the proof of this lemma , the right hand side of ( 6.18 ) , @xmath322 , must be replaced by @xmath323 . recall that , from proposition 4.3 @xcite , which is close to the work of sjstrand @xcite on the the resonances associated to a critical point , there exists an injective map @xmath324 from @xmath325 into the set of resonances of @xmath289 such that @xmath326 and such that all the resonances in @xmath327 + i [ - c , c ] $ ] are in the image of @xmath324 . in particular , the number of resonances of @xmath289 is bounded in @xmath327 + i [ - c , c]$ ] . furthermore , the operator @xmath289 has no resonance in @xmath328 + i [ - ( \mu_{0 } - \varepsilon h ) , ( \mu_{0 } - \varepsilon h ) ] , \ ] ] for any @xmath142 and @xmath329 small enough . using a compactness argument , we get for all @xmath330 + i [ - c , c]$ ] . thus , from , , @xmath331 , the estimate @xmath332 for @xmath298 + i [ - c_{0 } , 0]$ ] , lemma [ a29 ] and the previous discussion , we get @xmath333 for @xmath298 + i [ - c_{0 } , c_{0}]$ ] and follows . on the other hand , @xmath289 has no resonance in @xmath334 and in that set @xmath335 we can now applied the following version of the so - called `` semi - classical maximum principle '' introduced by tang zworski @xcite . _ [ a35 ] suppose that @xmath336 is a family of holomorphic functions defined for @xmath337 in a neighborhood of @xmath338 + i [ - c h , c h ] , \ ] ] with @xmath339 and @xmath304 , such that @xmath340 then , there exists @xmath341 such that , for any @xmath342 , @xmath343 for @xmath344+i [ -c h , 0]$ ] . _ this lemma is strictly analogous to lemma 4.7 of the paper of burq @xcite . combining , , @xmath345 with this lemma , we obtain @xmath346 for @xmath298 + i [ - \varepsilon , 0]$ ] , for some @xmath142 . @xmath193 here , we study the resolvent for @xmath347 . more precisely , we assume that @xmath348 + i [ - c \ln n , c_{0 } ] , \ ] ] for some @xmath39 fixed and @xmath349 . we define the new semi - classical parameter @xmath267 and @xmath350 + i h^{2 } [ - 4 c n \ln n , 4 c_{0 } n^{-1 } ] \subset [ a , b ] + i [ - c h \vert \ln h \vert , c h ] , \ ] ] for some @xmath339 and @xmath304 . then , @xmath94 can be written @xmath351 where @xmath352 with @xmath353 , @xmath354 . for @xmath349 , the coefficients @xmath355 are small , and the operator @xmath289 is uniformly non trapping for @xmath344 $ ] . we can expect a uniform bound of the cut - off resolvent in @xmath356 + i [ - c h \vert \ln h \vert , 0]$ ] . such a result is proved in the following lemma . _ [ a21 ] for all @xmath357 , there exist @xmath358 such that , for all @xmath359 , @xmath360 and @xmath361 , @xmath289 has no resonance in @xmath356 + i [ c h \ln h , c h]$ ] . moreover @xmath362 for all @xmath363 + i [ - c h \vert \ln h \vert , c h]$ ] . _ assume first lemma [ a21 ] . for @xmath364 + i [ - c \ln n , c_{0}]$ ] , we have @xmath365 and the estimate follows . for @xmath131 and @xmath366 small and fixed , the estimate is already know . the proof can be found in the book of vanberg @xcite in the classical case and in the paper of nakamura stefanov zworski @xcite in our semi - classical setting . to obtain lemma [ a21 ] , we only have to check the uniformity ( with respect to @xmath131 and @xmath366 ) in the proof of ( * ? ? ? * proposition 3.1 ) . _ @xmath367 limiting absorption principle . _ the point is to note that @xmath368 is a conjugate operator for all @xmath369 . let @xmath370 ; [ 0,1])$ ] be equal to @xmath54 near @xmath371 $ ] . the operator @xmath372 is well defined on @xmath373 , and its closure , @xmath374 , is self - adjoint . the operator @xmath289 is of class @xmath375 . recall that @xmath289 is of class @xmath376 if there exists @xmath377 such that @xmath378 is @xmath379 for the strong topology of @xmath380 ( see ( * ? ? ? * section 6.2 ) for more details ) . we have @xmath381 = 4 \widetilde{p } - 4 \mu v - 4 \nu w - 2 \mu x v ' - 2 \nu x w ' .\ ] ] in particular , for @xmath131 and @xmath366 small enough , we easily obtain @xmath382 } ( \widetilde{p } ) i [ \widetilde{p } , \ca ] 1_{[a/2 , 2b ] } ( \widetilde{p } ) \geq a h 1_{[a/2 , 2b ] } ( \widetilde{p } ) . \label{a26}\ ] ] note that this mourre estimate is uniform with respect to @xmath355 . it is also easy to check that @xmath383 \vert \leq c h \\ \vert ( \widetilde{p } + i)^{-1 } [ [ \widetilde{p } , a ] , a ] \vert \leq c h^{2 } \\ \vert ( \widetilde{p } + i)^{-1 } [ \widetilde{p } , [ \widetilde{p } , a ] ] \vert \leq c h^{2 } \\ \vert ( \widetilde{p } + i)^{-1 } a [ \widetilde{p } , [ \widetilde{p } , a ] ] \vert \leq c h^{2 } , \end{gathered}\ ] ] uniformly in @xmath355 . the regularity @xmath384 , the mourre estimate and the upper bound are the key assumptions for the limiting absorption principle . in particular , from , by example , the proof of proposition 3.2 in @xcite which is an adaptation of the theorem of mourre @xcite , we obtain the following estimate : for @xmath385 , there exist @xmath386 , such that @xmath387 for all @xmath359 , @xmath360 , @xmath361 and @xmath388 + i ] 0 , c h ] $ ] . in particular , @xmath389 for @xmath388 + i [ 0 , c h ] $ ] . _ @xmath367 polynomial estimate in the complex . _ the second point of the proof is to obtain a polynomial bound of the distorted resolvent . to obtain such bounds , we use the paper of martinez @xcite . in this article , the author studies the resonance of @xmath390 where @xmath391 is a @xmath392 function which can be extended analytically in a domain like @xmath279 ( see ) and decays in this domain . if the energy level @xmath393 is non trapped for the symbol @xmath394 , the operator @xmath395 has no resonance in @xmath396 + i [ a h \ln h , 0]$ ] for a @xmath397 small enough and any @xmath398 . moreover , @xmath399 for @xmath400 + i [ a h \ln h , 0]$ ] . here @xmath401 denotes the distorted operator outside of a large ball of angle @xmath402 , with @xmath403 . of course , @xmath289 satisfies the previous assumption on @xmath395 , for @xmath131 and @xmath366 fixed small enough . but , following the proof of in ( * ? ? ? * section 4 ) from line to line , one can prove that is uniformly true for @xmath369 . this means that there exist @xmath386 such that @xmath404 for all @xmath359 , @xmath360 , @xmath361 and @xmath405 + i [ c h \ln h , 0]$ ] . _ @xmath367 semi - classical maximum principle . _ to finish the proof , we use a version of the semi - classical maximum principle . this argument can be found in ( * ? ? ? * proposition 3.1 ) , but we give it for the convenience of the reader . we can construct a holomorphic function @xmath406 with the following properties : @xmath407 + i [ c h \ln h , 0 ] , \\ & \vert f \vert \geq 1 \quad \text { for } z \in [ a , b ] + i [ c h \ln h , 0 ] , \\ & \vert f \vert \leq h^{m } \quad \text { for } z \in [ a/2 , 2b ] \setminus [ 2a/3 , 3b/2 ] + i [ c h \ln h , 0 ] , \end{aligned}\ ] ] where @xmath321 is the constant @xmath408 given in . we can then apply the maximum principle in @xmath409 + i [ c h \ln h , 0]$ ] to the subharmonic function @xmath410 proving the lemma with and . using theorem [ a44 ] , , the equivalence of the norms @xmath414 as well as the fact that we can always replace @xmath287 by @xmath415 we see that it is sufficient to show : @xmath416 using complex interpolation we see that it is sufficient to show : @xmath417 we start with ( [ 2.4.6 ] ) which follows from @xmath418}u.\end{aligned}\ ] ] let us now observe that @xmath419 } ( p_{\ell } - z^2)^{-1 } \chi+\chi ( p_{\ell } - z^2)^{-1 } p_{\ell}\chi u\\ = & \widetilde{\chi}(p_{\ell}+i)^{-1}{[}p_{\ell},{[}p_{\ell},\chi { ] } { ] } ( p_{\ell } - z^2)^{-1 } \chi u\\ & + \widetilde{\chi}(p_{\ell}+i)^{-1}{[}p_{\ell},\chi { ] } ( p_{\ell } - z^2)^{-1 } ( p_{\ell}+i)\chi u+\chi ( p_{\ell } - z^2)^{-1 } p_{\ell}\chi u.\end{aligned}\ ] ] from this identity we obtain ( [ 2.4.4 ] ) and ( [ 2.4.5 ] ) using ( [ 2.4.6 ] ) ( for ( [ 2.4.4 ] ) ) and that @xmath420}{]}$ ] is uniformly bounded . for the proof of the main theorem we follow closely the ideas of vanberg ( * ? ? ? * chapter x.3 ) . if @xmath421 is a hilbert space we will note @xmath422 the space of all functions @xmath423 with values in @xmath421 such that @xmath424 . let @xmath425 and @xmath426 then @xmath427 for all @xmath428 . we can define @xmath429 as an element of @xmath51 for all @xmath430 with @xmath431 . the function @xmath432 depends analytically on @xmath430 when @xmath431 . also , on the line @xmath433 the function belongs to @xmath434 . we have the inversion formula : @xmath435 and the integral converges in @xmath436 for all @xmath428 . from the functional calculus we know that @xmath437 for all @xmath430 with @xmath431 . we therefore obtain for all @xmath438 : @xmath439 where the integral is convergent in @xmath436 . in the following , we denote @xmath440 the meromorphic extension of @xmath441 . ( 13319,5969)(-621,-4883 ) ( 3601,-4561)(0,0)[lb ] ( 5551,-661)(0,0)[lb ] ( 5326,-2236)(0,0)[lb ] ( 10726,-2461)(0,0)[lb ] ( 751,-2461)(0,0)[lb ] ( 6676,-2911)(0,0)[lb ] ( 8851,239)(0,0)[lb ] ( 2551,-136)(0,0)[lb ] ( 2401,-3811)(0,0)[lb ] ( 9076,-3811)(0,0)[lb ] ( 11476,-1336)(0,0)[lb ] ( -149,-1336)(0,0)[lb ] ( 3676,-1336)(0,0)[lb ] ( 7576,-1261)(0,0)[lb ] let @xmath470 . we have , for @xmath471 large enough , @xmath472 we now take the limit @xmath20 goes to @xmath187 in the @xmath468 sens in . the integrals @xmath473 and @xmath474 go to @xmath99 thanks to and , in the integrals @xmath475 and @xmath476 , the paths @xmath477 are replaced by paths which extend @xmath477 in a natural way and which go to @xmath478 . we note them again @xmath477 . we remark that @xmath479 where the integral is absolutely convergent in @xmath468 . on the other hand , we have the estimate , for @xmath471 large enough , @xmath480 and a similar estimate holds for @xmath481 . since all these estimates hold in @xmath482 , and give the estimate of the rest with @xmath483 . the estimate follows from , theorem [ a44 ] _ iii ) _ and proposition [ prop2.4 ] . let us now show part @xmath120 of the theorem . we choose @xmath484 and the integration path as in part @xmath112 of the theorem . we first suppose @xmath485 for some @xmath486 to be chosen later . then the estimate for @xmath487 can be replaced by @xmath488 let us now suppose @xmath489 . on the one hand we have the inequality : @xmath490 since the norms on @xmath84 and on @xmath85 are uniformly equivalent for @xmath83 . on the other hand by the hypotheses on @xmath491 we have @xmath492 for @xmath493 and @xmath471 large enough . it follows : @xmath494 this finishes the proof of the theorem if we choose @xmath495 sufficiently small and put @xmath496 . we note that for @xmath497 , we have @xmath498 and then @xmath499 estimate can be written @xmath500 with @xmath501 and @xmath502 . let @xmath503 , @xmath504 with @xmath505 for @xmath506 . remark that there exists @xmath39 such that @xmath507 on the support of @xmath508 . using the radial decomposition @xmath509 , we get @xmath510 finally , for the interpolation argument , we use the fact that @xmath511 for @xmath83 . j. sjstrand , _ a trace formula and review of some estimates for resonances _ , microlocal analysis and spectral theory ( lucca , 1996 ) , nato adv . 490 , kluwer acad . publ . , 1997 , pp . 377437 .

the cosmic star - formation rate peaks between @xmath6 and @xmath7 and marks a critical period for galaxy assembly . in the classic model of galaxy formation , collisional baryons cool and dissipate into the center of dark matter halos while conserving primordial angular momentum , leading to the formation of a thin rotating disk galaxy @xcite . although this gas is expected to collapse to a disk in a few crossing times , external processes can disrupt or destroy disks on timescales shorter than the rotational period . in an emerging new picture , clumpy and smooth accretion from the cosmic web can efficiently carry external angular momentum inside the halo viral radius and directly onto the central galaxy @xcite , disrupting the velocity field @xcite . additionally , ordered disks can be disturbed or destroyed through major mergers @xcite , violent disk instabilities @xcite and stellar feedback @xcite . these processes are prominent at @xmath0 , coincident with the assembly of stellar mass in galaxies and may play a stronger role in the shallow potential wells that host low mass galaxies . in general , the formation and development of a galaxy depends strongly on its mass . massive galaxies tend to assemble their stellar mass first ( downsizing " , e.g. , @xcite ) and develop disks earlier ( e.g. , @xcite ) . in contrast , low mass ( @xmath8 ) galaxies tend to exhibit more disturbed and irregular morphologies at both low and high redshift ( e.g. , @xcite ) , indicating that they lack mature structure . recent evidence from imaging suggests that a significant fraction of low mass galaxies at @xmath91.5 appear to have elongated stellar distributions @xcite . a similar phenomenon has been seen in hydrodynamic simulations of galaxy formation @xcite which indicate that oblate ( stellar ) disks form at late times ( @xmath10 ) , when the central potential becomes baryon - dominated . the star - forming gas in these simulated galaxies starts off in a highly perturbed and triaxial state , eventually collapsing to form thick oblate disks @xcite . from @xmath11 to now , star - forming systems appear to have evolved through kinematic downsizing in a process known as disk settling " @xcite . in this picture , galaxies hierarchically settle from turbulent systems with high dispersion to regularly rotating disks , with more massive galaxies forming disks earlier . locally : the majority of high mass galaxies are ordered rotating disks , with kinematically disordered galaxies appearing only below a stellar mass of @xmath12 @xmath13 @xmath14 9.5 @xcite . however , it is unclear whether the kinematic downsizing picture extends beyond @xmath11 and into the peak of cosmic star - formation . with recent advancements in near infrared detectors and instrumental multiplexing capabilities , space and ground based instruments have provided a wealth of knowledge regarding the physical properties of galaxies during this period , the cosmic noon . ground based kinematic data for high redshift ( @xmath151 ) galaxies is rapidly growing , as reviewed in @xcite , with several recent and ongoing surveys : e.g. , sins @xcite , wigglez @xcite , massiv , mosdef @xcite , kmos3d @xcite and kross @xcite . emerging from these surveys is the general picture that star - forming galaxies at @xmath0 are _ unlike _ the thin disk galaxies that make up the massive end up the local hubble sequence . high redshift galaxies tend to be thick and gas - rich with velocity dispersions that are factors of a few higher than local thin disks . the typical velocity dispersion tends to increase with increasing redshift ( e.g. , ) and may reflect a similar rise in gas fractions @xcite . even still , rotational support appears to be relatively common in the more massive galaxies ( @xmath16 ) at @xmath0 , with @xmath17 of the population exhibiting rotational motions comparable to or greater than their velocity dispersion @xcite . only very recently , kinematic data for low mass galaxies ( @xmath8 ) at high redshift is becoming available in gravitationally lensed systems ( e.g. , @xcite ) and in larger numbers in the ongoing surveys with multi - object single slits ( mosdef ; @xcite ) and integral field spectrographs ( kmos-3d ; @xcite ) . while the imaging data at this epoch indicates that massive galaxies tend to be disk - like and the low mass galaxies tend to be irregular , similar conclusions have not yet been drawn from kinematic measurements , and it is unclear whether kinematic downsizing is in effect at @xmath0 . in this paper we aim to take advantage of the multiplexing capabilities and the high sensitivity of keck / mosfire multi - object near - infrared spectrograph @xcite as well as the high resolution and sensitivity of hubble space telescope ( hst)/wfc3 h - band imaging to explore the kinematic state of star - forming galaxies at @xmath0 . this survey is named sigma , survey in the near - infrared of galaxies with multiple angles . the sample is morphologically diverse and is selected along the star - formation main sequence in two redshift bins from @xmath18 and @xmath19 . in 2 we discuss the observations and sample selection . in 3 we describe measurements of the galaxy physical properties from multi - wavelength imaging . in 4 we detail the measurements of the gas - phase kinematics of our sample , and discuss multi - pa observations . in 5 we present the @xmath0 tully - fisher relation . in 6 we examine trends between the kinematics and physical properties , and in 7 we present a comparison of our galaxies with a model for marginally stable gas - rich disks . in 8 we present our conclusions and in the appendix we present simulations to determine the effects of spatial resolution on the measurement of kinematics . we adopt a @xmath20cdm cosmology defined with ( h , @xmath21 , @xmath22 ) = ( 0.7 , 0.3 , 0.7 ) . the sigma sample is drawn from three of the fields ( goods - s , goods - n and uds ) of the hst / wfc3 candels survey @xcite . all of the spectra in goods - n were taken as a part of the tkrs-2 survey @xcite and a subset of the spectra in goods - s and uds have been presented in previous papers @xcite . the sigma galaxy sample is shown in a star formation rate - stellar mass ( sfr - @xmath23 ) diagram in figure [ fig : m_sfr ] . this sample is a part of the well studied goods - n , goods - s , and uds extragalactic fields , and so extensive multi - wavelength imaging is available for all of the galaxies in this sample . galaxies in sigma are selected from the candels wfc3/f160w catalogs @xcite as follows . selection priority was given to galaxies with spectroscopic redshifts from hst / wfc3 grism observations @xcite , taken as a part of the 3d - hst ( @xcite , pi van dokkum ) and aghast ( pi weiner ) surveys . the majority of the tkrs-2 galaxies were selected on previous spectroscopically confirmed redshifts . photometric redshifts were derived from the broadband seds with eazy @xcite and were used to select the remaining galaxies in our sample . star - forming galaxies were selected down to a sfr of @xmath243 m@xmath25 yr@xmath26 at z @xmath24 1.5 and @xmath2410 m@xmath25 yr@xmath26 at z @xmath24 2.2 . there is no selection cut on morphology . as such , both regular and irregular galaxies are included in the sample . there is a priority weight on galaxy size , half - light v - band diameters of @xmath27 , in order to select galaxies with emission extents that are larger than the typical seeing . the full sigma sample for which we have significant detections ( s / n @xmath28 3 ) in h@xmath29 and/or @xmath305007 contains 97 galaxies . a portion of these were observed at multiple slit position angles ( pa ) : 23 with 2 pas and 11 with 3 + pas . we select galaxies with at least one kinematic measurement coincident with the photometric major axis ( within 45@xmath31 ) , removing 35 of the 97 galaxies . we make a further cut on emission line extent for each spectrum of each galaxy , as described in 3.3 and the appendix , such that the effective radius of the emission extent is at least 0.8 times the size of the seeing fwhm . this selection is shown in figure [ fig : cont_size_em_size ] and removes an additional 9 galaxies from the sample . an additional 4 galaxies are removed due to the presence of a strong overlapping skyline . the final sample used in this paper contains 49 galaxies , 28 in the redshift range 1.3 @xmath14 z @xmath14 1.8 and 21 at 2.0 @xmath14 z @xmath14 2.5 . the properties of the full sample are presented in table [ tab : table1 ] of the appendix . at z @xmath24 1.5 and @xmath32 at z @xmath24 2.3 . the grey dots represent galaxies drawn from the h - band selected goods - s candels catalog in the same redshift range . a small amount of noise ( 0.05 dex ) was added to the grey points to remove striations due to the discreet grid structure searched by the model fits to @xmath23 and star - formation rate ( sfr ) . in the bottom panel , we compare the mass range of our sample , indicated by the red and blue arrows , with the blue galaxy stellar mass functions at the same redshifts , taken from @xcite . ] the strong optical nebular emission lines ( e.g. , h@xmath29 , @xmath305007 ) from which we measure kinematics shift into the near infrared ( nir ) at z @xmath28 1.3 . the multi - object spectrograph , mosfire @xcite on keck - i , is one of the premiere instruments to study spectral features in the nir . the spectra in this sample were taken in the h - band ( 1.46 - 1.81 @xmath33 m ) and k - band ( 1.93 - 2.45 @xmath33 m ) . h@xmath29 can be detected in the h - band from 1.3 @xmath14 z @xmath14 1.8 and in the k - band from 2.1 @xmath14 @xmath34 @xmath14 2.8 . additionally , @xmath305007 falls in the h - band from 2.0 @xmath14 z @xmath14 2.8 . the mosfire slit width was set to 0.7@xmath35 and the instrumental spectral resolution is r@xmath363630 in both bands ( @xmath37 km s@xmath26 ) . the masks were dithered along the slit between symmetric sky positions around the initial pointing to facilitate sky and instrument noise subtraction . the on - source exposure times were between 90 and 120 minutes with individual exposure times of 120 seconds . the 2d spectra were sky - subtracted , wavelength - calibrated and rectified using the mosfiredrp reduction pipeline . spectroscopic redshifts were determined via specpro @xcite . observations were taken in natural seeing conditions and the size of the seeing is estimated from unresolved calibration stars , where available , and marginally resolved compact `` blue nugget '' galaxies @xcite that were observed on the same mask as the galaxies in our sample . the nir seeing ranged between 0.45 and 0.85@xmath35 . example cutouts of the 2d spectra around the h@xmath29 emission line are shown in figure [ fig : kin_example ] . throughout this paper we use two structural parameters measured from the hst / wfc3 images , namely the photometric position angle ( pa , i.e. , the direction of the photometric major axis ) and the photometric axis ratio ( b / a ) . for a well - behaved disk galaxy , the photometric pa will align with the kinematic pa ( the direction of the largest velocity gradient ) and the axis ratio can be used to correct the derived rotation velocity for the inclination of the galaxy to the line of sight . the photometric pa and axis ratio were measured with galfit @xcite by @xcite on the h - band ( f160w ) , j - band ( f125w ) and y - band ( f105w ) for all of the h - band selected candels galaxies in our fields . we refer to these papers for measurement details . for this paper , we adopt the h - band measurements ( rest i at @xmath38 , rest v at @xmath39 ) . the available broadband measurements span from the uv to the nir with ancillary spitzer / mips 24/70@xmath40 photometry and herschel far - ir photometry from the goods - herschel and pacs evolutionary probe surveys . as described in @xcite , integrated stellar masses ( @xmath23 ) were derived from the broadband sed of each galaxy using the fast fitting code @xcite assuming a @xcite initial mass function , @xcite stellar population synthesis models and a @xcite extinction law . the errors on the stellar masses are @xmath410.3 dex @xcite . dust - corrected star - formation rates were derived following the sfr - ladder @xcite . for galaxies with spitzer and herschel far - ir detections , the sfr is calculated from both the obscured ( ir ) and unobscured ( uv ) components ( following ) . for non - detections in the mid - far ir , the star - formation rates are derived from the uv and are extinction - corrected following the best fit attenuation from the sed modeling . in figure [ fig : m_sfr ] we show the star - formation rate vs stellar mass diagrams for our sample , the star - formation main sequence " ( e.g. @xcite ) . galaxies drawn from the full candels catalog in goods - s are shown by black dots in the background . our sample is generally representative of the sf main sequence down to a stellar mass @xmath42 in the 1.3 @xmath14 z @xmath14 1.8 redshift bin and 9.6 in the 1.8 @xmath14 z @xmath14 2.8 bin . we make a comparison with the blue galaxy stellar mass functions at 1.5 @xmath14 z @xmath14 2.0 and 2.0 @xmath14 z @xmath14 2.5 @xcite and note that our sample spans close to the knee of the schechter function , @xmath43 , in both redshift intervals . ) of the emission versus the continuum for each slit on every galaxy used in this paper . resolved observations are marked with red filled circles and the discarded unresolved observations are marked as black x s . the range of seeing conditions during the observations is shown as a grey shaded swath . emission line sizes are seeing corrected from the mosfire 1d emission profile . continuum sizes are measured from sersic fits to the h - band hubble images . as shown in the appendix , kinematic structure can be resolved down to a limiting resolution of d@xmath44 = 0.8@xmath45 seeing@xmath46 , where d@xmath44 is the 1d seeing - corrected emission line size . ] one of the main challenges to measuring rotation with ground based data comes from the effects of beam smearing @xcite . seeing smooths out velocity gradients and artificially boosts the velocity dispersion in the central parts of rotating galaxies . this effect can generally be modeled , except in the cases where the emission extent is comparable to or smaller than the seeing . galaxies with small emission extents can appear dispersion - dominated in seeing - limited data but reveal rotation at higher resolution @xcite . in order to examine the effects of beam - smearing on our data , we produce a suite of simulated mosfire / sigma spectra in the appendix . we conclude that , for the typical signal to noise of the spectra in our sample , we can reliably recover intrinsic kinematics ( @xmath1 , @xmath2 ) down to d@xmath44 @xmath47 0.8 @xmath45 seeing@xmath46 , where d@xmath44 is the 1d half - light diameter measured from the emission lines and seeing@xmath46 is the full width at half maximum of the seeing . we make size selections using the emission rather than the continuum for a couple of reasons . first , the 2d half - light radii of the emission typically extend beyond the continua @xcite . secondly , seeing tends to blur emission that is not coincident with the slit placement into the slit @xcite . as such , an emission size determined from a projection of the slit onto the hst image may underestimate the extent of the emission measured with the slit . by measuring sizes directly from the seeing - corrected emission lines , we bypass both of these uncertainties . in figure [ fig : cont_size_em_size ] we show the half - light diameter sizes , in both emission and continuum , for the galaxies in our sample . for each slit , the emission line size is measured directly from the spectrally - collapsed emission line . due to the convolution of the intrinsic profile with the seeing , the spatial profiles of the spectrally - collapsed emission lines are generally well described by a 1d gaussian . the seeing is subtracted from the observed profile in quadrature ( @xmath48 ) and the half - light size is measured from the recovered intrinsic profile . continuum radii are measured from the hst / wfc3 h - band image using the galfit software package in @xcite . for the galaxies in our sample , the half - light radii of the intrinsic emission profiles tend to be slightly larger than the continuum sizes . this could be the case for two reasons . first , the half - light radius that is measured by galfit is a 2d quantity measured on concentric ellipses while the emission size is being measured directly from the 1d emission lines , an along - the - slit measurement . since the 1d measurement essentially integrates the light across the slit , these two sizes will not directly scale to one another in general . the difference may also be physical , in the sense that the emission profile can extend beyond that of the continuum ( @xcite ) . kinematics are measured from strong nebular emission lines ( h@xmath29 , @xmath305007 ] ) in the spectra . these lines generally trace the hot ( t@xmath3610@xmath49 k ) ionized gas in hii regions surrounding ongoing star - formation . the kinematics are measured using the code rotcurve @xcite . the instrumental resolution of our spectra ( r@xmath363630 ) allow us to measure the velocity dispersion down to @xmath4120 km s@xmath26 and the rotation velocity uncorrected for inclination ( @xmath50 ) to @xmath4115 km s@xmath26 . the kinematic modeling performed by rotcurve has been described in detail in previous papers , so we will briefly outline it here and refer to @xcite for further details . rotcurve models the spatial profile of the emission line as a 1d gaussian . in each row along the slit , we recover a velocity and velocity dispersion using an uncertainty - weighted least squares fit to a 1d gaussian . nearby , but not overlapping , skylines are manually masked to prevent noise peaks from driving the fit . the velocity structure of the line is modeled with two components : a rotation curve and a dispersion term . due to our limited spatial resolution , the velocity dispersion is kept constant with radius in the model . the rotation velocity is modeled with an arctan function : @xmath51 where r@xmath52 is the knee radius for the turnover of the rotation curve and @xmath53 is the velocity of the flat part of the rotation curve . we note that the flat part of the rotation curve is not reached in a fraction of our galaxy sample ( see appendix ) and in these cases @xmath53 is driven by the scale of the velocity difference reached by the data and the modeled turnover radius . due to the seeing blur , r@xmath52 is not well constrained and so we fix the turnover radius in the model at a fiducial value of 0.2@xmath35 ( 1.7 kpc at @xmath0 ) and adjust to 0.3@xmath35 or 0.4@xmath35 for galaxies that are not well fit initially . the turnover radius and the rotation velocity are moderately covariant with a @xmath540.1change in r@xmath52 resulting in a @xmath540.1 dex change in @xmath53 . the model parameters of @xmath53 and @xmath2 are explored on a grid of 5 km s@xmath26 . the modeled emission line is then spatially convolved with a fixed seeing kernel ( 0.6 fwhm ) , which is the typical seeing of our observations in the nir . the best fit model is determined through a @xmath55 minimization with the data over this grid . the velocity dispersion term that we measure ( @xmath2 ) is unlike a pressure supported stellar velocity dispersion . the hot gas which @xmath2 is tracing can collisionally radiate and so a high dispersion system can not remain in equilibrium after a crossing time . due to the seeing blur , the velocity dispersion is a measure of the small scale velocity gradients below the seeing limit . the velocity dispersion measured from nebular emission lines will be a combination of : ( . ) thermal broadening of the gas ( @xmath2410 km s@xmath26 for h at 10@xmath56 k ) , ( . ) internal turbulence in hii regions ( @xmath24 20 km s@xmath26 for local hii regions ; ) , ( . ) relative motions between hii regions . the inclination is measured as @xmath57 , where @xmath29 is the correction for the thickness of the disk ( @xmath29 = 0.2 ) and b / a is the axis ratio of the galaxy , measured from the hst h - band image . the rotation velocity is then corrected as @xmath58 . as indicated by @xcite , distinguishing late - stage mergers and rotating disks is difficult in low resolution kinematics observations . at high redshift , where galaxy mergers are expected to be common @xcite , disk classifications made from kinematic signatures alone may be biased high . in 2 galaxies in our sample ( i d : 21897 and 4476 ) we identify an interacting neighbor in both the hst image ( @xmath59 ) and the mosfire spectrum ( @xmath60 km s@xmath26 ) . we mask the emission line of the neighboring source in both of these cases . we note that late - stage interactions , for which two galaxies are identified as a single source , may be included in this sample . a joint analysis of the hst imaging and the kinematic data is necessary to disentangle disks from late - stage mergers in our sample . for all subsequent analyses , we interpret the measured velocity gradient of each galaxy as rotation . in figure [ fig : kin_example ] , we show example kinematic fits for three galaxies in our sample : a strongly rotating galaxy , a galaxy with comparable velocity and velocity dispersion , and a dispersion dominated galaxy . even in the dispersion dominated case , we note the presence of a velocity gradient , although it is small compared to the observed dispersion . the rotationally supported system is a well - ordered disk galaxy , similar to disk galaxies today , with the presence of spiral structure and a bar in the hst image . in its spectrum , we find a strong rotation signature ( @xmath24 175 km s@xmath26 ) and a mild velocity dispersion ( @xmath24 35 km s@xmath26 ) . beam smearing tends to blur together spatially separate regions of emission , leading to a flattening of nearby velocity gradients and an elevation in the local velocity dispersion . in the observed line , this effect leads to a classic peak in the velocity dispersion at the center of the galaxy , where the velocity gradient is the steepest . this can be seen in the example kinematic fits in figure [ fig : kin_example ] , particularly for the strongly rotating galaxy . rotcurve models the seeing and accounts for this artificial beam smearing when constructing its model . the best - fit kinematic parameters of the full sample are presented in table [ tab : table2 ] of the appendix . single - slit spectroscopy is a reliable method for recovering rotation velocity and velocity dispersion in observations where the position angle of the slit is aligned with the kinematic major axis of the galaxy . in ordered disk galaxies , the kinematic and photometric major axes are aligned and so the hst / wfc3 image can be used to determine the kinematic major axis . however , the morphologies of high redshift star - forming galaxies can be more severely disturbed than local or low redshift samples ( e.g. , @xcite ) , implying that the photometric and kinematic major axes might be misaligned . in a sample that combines galaxies at both @xmath61 and @xmath0 , @xcite report that kinematic and photometric axes are at least moderately misaligned ( @xmath62 ) in @xmath2440@xmath63 of galaxies and are largely misaligned ( @xmath64 ) in @xmath2420@xmath63 . this fraction declines to @xmath65 in local galaxies . for seeing - limited observations of high redshift galaxies , slit placement is much less stringent than for local galaxies or for higher resolution observations . to recover rotation , the alignment between the slit and the kinematic major axis should be within 45@xmath31 ( see figure 13 of @xcite ) . a total of 26 galaxies in sigma have observations at multiple position angles ( @xmath66 , see figure [ fig : vel_map_combined ] ) . due to the non - uniform coverage of our maps , we do not simultaneously model all of the observed slits on a galaxy . instead we examine the kinematic measurements along each slit separately and adopt measurements for the slit with the highest rotational signature . we require that every galaxy in sigma have at least one slit that is aligned within 45@xmath31 of the photometric major axis ( @xmath67 ) . this requirement ensures that rotation will be recovered in disk galaxies , where the photometric and kinematic axes are aligned . of the 26 galaxies with kinematic measurements at more than one pa , 17 have at least one slit aligned and one slit misaligned with @xmath68 . of these systems , we find the largest velocity gradient in a misaligned slit in 23@xmath63 of the cases ( 4 galaxies ) and in an aligned slit in 77@xmath63 of the cases ( 13 galaxies ) . this result is consistent with @xcite . the 4 galaxies in our sample for which the strongest velocity shear is found in a misaligned slit all display complex morphology in both the hst / acs v - band and hst / wfc3 h - band images . for these 4 galaxies , we adopt the line of sight velocity shear ( @xmath69 ) for our analysis and do not correct the velocity for inclination . these galaxies are marked in all of the figures . as shown in an example in figure [ fig : vel_map_combined ] , these systems tend to have complex velocity profiles . in some of these cases , velocity shear is found in the aligned slit as well , implying that the velocity gradient ( either tracing global rotation or more complex motions ) may lie between both slits . the tully - fisher relation ( tf ; ) is an empirical scaling relating the dynamics and the luminosity , or more recently , stellar mass of galaxies . in the local universe , the relation is relatively tight for massive disk galaxies ( e.g. , @xcite ) but begins to show small residuals to low mass for gas - rich galaxies @xcite and large residuals to low rotation velocity for irregular low mass galaxies ( @xmath70)@xcite . numerous studies have extended the analysis of the tf relation to both intermediate , 0.1 @xmath14 z @xmath14 1.3 ( e.g. , ) , and high redshift , 1.3 @xmath14 z @xmath14 3.0 ( e.g. , @xcite ) . and @xcite were the first to note that intermediate redshift galaxies with perturbed / complex kinematics and irregular morphologies tend to exhibit large scatter from the tf relation , with subsequent studies confirming such findings ( e.g. , ) . recent lensing work at high redshift has also found evidence for scatter to low rotation velocity from the tf relation for a handful of low mass galaxies @xcite . however , due to a combination of sensitivity limitations and survey selection , the low mass regime of the high - redshift tf relation has not been explored with large numbers of galaxies . we will now explore the low mass tf relation at @xmath0 with sigma galaxies . in the top panel of figure [ fig : kinematics ] we plot the stellar mass tully - fisher relation for the galaxies in sigma . the points are color coded to indicate the ratio of ordered ( @xmath1 ) to disordered ( @xmath2 ) motions in each galaxy , which we will refer to as kinematic order " . only one pa , the pa with the strongest velocity gradient , is shown for each galaxy . as described in 4.1 , we do not apply an inclination correction for the 4 galaxies whose largest velocity gradients are measured at a @xmath71 pa @xmath28 45@xmath31 . a few galaxies have a best - fit rotation velocity which is below the nominal resolution limit of our spectra , shown by the dashed line . we set these galaxies at an upper limit of @xmath1 @xmath45 @xmath72 @xmath14 15 km s@xmath26 in subsequent figures . the most massive galaxies in sigma ( @xmath73 ) lie on or above the tf ridge line , which we define using the local relation from @xcite . these galaxies exhibit values of @xmath74 , factors of a few lower than local galaxies at the same mass . we perform a simple linear regression to the stellar mass tf relation for these 12 massive ( @xmath73 ) rotationally - supported galaxies ( @xmath75 ) . we fix the slope to the value from local relation and report an offset to lower @xmath23 of @xmath76 @xmath77 . if these systems are gas - rich , then the baryonic ( @xmath23+@xmath78 ) tf relation is the more applicable scaling relation . although measurements of gas mass are unavailable for the galaxies in our sample , we note that an offset in mass of @xmath79 implies that we may be missing @xmath80 of the baryonic mass . this is consistent with estimates of the molecular gas reservoirs in massive galaxies at high redshift ( e.g. , @xcite ) . as sigma includes both rotationally - supported and dispersion - dominated galaxies , we make no further attempt to characterize an evolution in the slope , scatter or intercept of the stellar mass tf relation for disks , which would require morphological pruning . at low stellar stellar mass ( @xmath81 ) we find significant scatter from the ridge line to low rotation velocity . a large fraction of these galaxies show marginal rotation support , with values of @xmath82 . in the middle panel of figure [ fig : kinematics ] we show the velocity dispersion as a function of stellar mass . as discussed in 3 , the velocity dispersion that we measure is corrected for seeing and modeled as a constant term across the galaxy . due to the finite width of the slit ( 0.7@xmath35 ) and the effects of seeing ( @xmath24 0.6@xmath35 ) , @xmath2 is an integration of small scale velocity gradients , thermal broadening and internal turbulence . the velocity dispersion is only mildly correlated with stellar mass , with more massive galaxies displaying marginally higher velocity dispersions . the mean and rms scatter of @xmath2 for the low mass ( @xmath83 ) and high mass ( @xmath84 ) samples are 57@xmath8519 km s@xmath26 and 64@xmath8529 km s@xmath26 , respectively . the galaxies which fall from the tf ridgeline to low @xmath1 display characteristically higher @xmath2 at fixed stellar mass . in other words , the lower @xmath1/@xmath2 in these galaxies is due to both lower rotation velocity and higher velocity dispersion . star - forming galaxies at z @xmath24 2 tend to have much stronger kinematic contributions from @xmath2 than their local counterparts . for the bulk of our sample there appears to be a floor in velocity dispersion of @xmath2 = 30 km s@xmath26 . this floor is slightly higher than the typical gas velocity dispersion in local disk galaxies , @xmath86 km s@xmath26 @xcite , and higher than the spectral resolution limit of our observations , @xmath24 20 km s@xmath26 . to combine the quantities @xmath1 and @xmath2 into a more fundamental quantity , many authors ( e.g. , ) adopt a joint kinematic parameter , @xmath87 @xcite . a common choice is made for @xmath88 , motivated by virial arguments for a spherically symmetric tracer distribution ( see @xcite ) . the @xmath89 quantity more reliably traces the overall potential of a galaxy system , independent of morphology , recent disturbance or kinematic state @xcite . in the bottom panel of figure [ fig : kinematics ] we show the @xmath90 relation for our sample , which re - establishes the tf relation for all galaxies . this has been found previously at @xmath0 @xcite . we note that our low mass galaxies are relatively consistent with the @xmath91 relation found by @xcite , while the high mass galaxies tend to be offset to higher @xmath89 . @xcite report a similar offset in the @xmath0 @xmath89 relation over their full mass range ( @xmath92 ) . @xcite have argued that it is necessary for the @xmath93 to evolve to higher @xmath89 at fixed mass in order to correctly predict the cosmic stellar mass density . as discussed before , this offset might reflect a rise in @xmath94 at higher redshift . from the local tf ridge line , defined by @xcite . the largest scatter is found at low stellar mass ( @xmath95 ) . middle : the relation between integrated gas velocity dispersion ( @xmath2 ) and @xmath23 is shown . the slowly rotating galaxies that fall from the tf relation have elevated velocity dispersions . bottom : the s@xmath96 tf relation is shown along with the fit and scatter from the @xmath91 deep2 sample @xcite and the local faber - jackson relation , defined by @xcite . the handful of galaxies in which the largest velocity gradient was measured in an off - axis slit are not inclination corrected and are shown as open circles . the resolution limits for measuring rotation and dispersion in sigma is given by the horizontal dashed lines in the top and middle plots . ] we perform a least - squares linear fit to the @xmath89 tf relation , @xmath12 @xmath89 @xmath97 km s@xmath98 = @xmath99 , for the full sample . we report coefficients of @xmath100 and @xmath101 with an rms scatter @xmath102 = 0.19 . even with the apparent dichotomy between the low mass and high mass regions noted above , we find that the slope , amplitude and scatter of the full sample is relatively consistent with the @xmath91 relation ( @xmath103 , @xmath104 , @xmath102 = 0.14 ) found in @xcite and the local faber - jackson relation ( @xmath29 = 0.286 , @xmath105 = 1.965)@xcite . as discussed in 5 , the low mass galaxies in our sample show the strongest residuals from the tf ridge line and the lowest values of @xmath4 . in the settling of disk galaxies , massive galaxies are expected to form ordered disks first , followed later by lower mass galaxies ( kinematic downsizing " ; @xcite ) . to probe the drivers of kinematic disturbance at @xmath0 , we examine the relation between the kinematic order of sigma galaxies ( @xmath1/@xmath2 ) and three of their physical properties : star - formation rate , effective radius and stellar mass . to separate out the dependance of size and sfr on stellar mass , we examine deviations from the size - mass and sfr - mass relations at @xmath0 . we find that , at fixed stellar mass , @xmath1/@xmath2 is only weakly correlated with sfr and mildly correlated with half - light radius . the strongest correlation is found to be with stellar mass , wherein more massive galaxies tend to be the most rotationally supported . we describe these relations in more detail below . we perform a ranked spearman correlation test to quantify the association in each relation . the spearman coefficient ( @xmath106 ) measures the strength of a correlation and we use the common interpretation : @xmath107 ( very weak correlation ) , @xmath108 ( weak / moderate correlation ) , @xmath109 ( moderate / strong correlation ) . the p - value ( @xmath110 ) indicates the probability of uncorrelated data producing the given spearman coefficient . in the left panel of figure [ fig : vsig_others ] we examine the correlation of @xmath4 with stellar mass following k12 for @xmath1111.2 . we note that the trend is the strongest of the three we investigate , with an increasing trend of @xmath1/@xmath2 with stellar mass and a spearman coefficient , @xmath112 ( @xmath113 ) . if we only include the galaxies with detected rotation and inclination corrections ( i.e. , filled circles ) , the trend with stellar mass strengthens to @xmath114 ( @xmath115 ) . next , we examine the departure of our galaxies from the size - mass relation at @xmath0 , as parameterized for late type galaxies in table 1 of @xcite . we use the @xmath116 and @xmath117 size - mass relations for our low and high redshift bins , respectively . the half - light size used for these relations is calibrated to rest - frame 5000 (1.5 @xmath33 m at z@xmath1182 ) and so we adopt the h - band half - light radii for comparison . in the middle panel of figure [ fig : vsig_others ] , we present the differences between the half - light radii of our galaxies and the size - mass relation ( @xmath119r@xmath120 = r@xmath121 - r@xmath122 . ) . the intrinsic scatter of the relation is 0.18 and 0.19 dex for both the low and high redshift bins , respectively , consistent with the rms scatter in our sample of 0.19 . we note that there is a weak / mild correlation between @xmath1/@xmath2 and @xmath71 r@xmath120 , as quantified by the spearman coefficient @xmath123 ( @xmath124 ) . finally , in the right panel of figure [ fig : vsig_others ] , we examine the correlation of kinematic order with a galaxy s departure from the 1.5 @xmath14 z @xmath14 2.6 star - formation main sequence ( @xmath125 ) . if the dispersion - dominated galaxies are being disrupted by internal processes associated with star - formation , we would expect a correlation between @xmath4 and the position of a galaxy on the star - formation main sequence . we adopt the main sequence fit of @xcite as parameterized in @xcite and adjust to a chabrier imf to match with our sample . this relation is valid over @xmath126 , spanning our full mass interval . we find no strong trend with @xmath127 sfr ( defined as @xmath12 sfr@xmath128 - @xmath12 sfr@xmath129 ) and @xmath1/@xmath2 . in other words , at fixed stellar mass the position of a galaxy above or below the star - formation main sequence is not related to its internal gas kinematics . the spearman coefficient between these two values confirms the lack of a trend , @xmath130 ( with @xmath131 ) . combining kinematic measurements at @xmath0 from both the kmos-3d @xcite and mosdef surveys @xcite , @xcite investigated the trend of v/@xmath2 with specific star formation rate ( @xmath132 ) and stellar mass , finding a declining and weakly increasing trend respectively . we find that , when we fix for stellar mass , there is no dependance on sfr . instead , the strongest correlation in the sigma sample is with stellar mass alone . if we consider that the gas mass scales with sfr @xcite , these results would indicate that the kinematic order of a galaxy is more strongly linked with its mass than its gas fraction . in this section we compare our measurements of @xmath1/@xmath2 with a model for gas rich marginally stable disks using the refined toomre analysis given in @xcite . we find that high mass galaxies in our sample are in general agreement with this model , indicating that they have already assembled primitive disks . galaxies which have already settled to become disks by @xmath0 are gas rich ( e.g. , @xcite ) and are expected to be in a marginally stable ( to collapse ) equilibrium ( e.g. , @xcite ) , regulated by ubiquitous inflows , outflows and star - formation @xcite . this regulator model successfully describes the elevated velocity dispersions in high redshift gas - rich disks @xcite . following the work of @xcite and @xcite , @xmath4 of the ionized gas in a marginally stable disk can be estimated from the toomre criterion : @xmath133 where @xmath134 is a constant which is set to @xmath135 for a flat rotation curve , as discussed in @xcite . the quantity @xmath136 is the critical toomre parameter which , considering the range of realistic disks , can vary between 0.67 ( single - component thick disk ; @xcite ) and 2.0 ( two - component thin disk ; @xcite ) . the gas fraction ( @xmath137 ) can be estimated from the gas depletion timescale ( @xmath138 ) and the specific star - formation rate ( @xmath139 ) as @xmath140 . the gas depletion timescale is relatively constant with stellar mass at @xmath0 and we fix it to @xmath141 yr ( @xcite ) . for our analysis we adopt the ssfr - m@xmath142 relation at z = 2 from @xcite using the fit in @xcite . equation 2 is derived assuming that the gravitational force exerted by @xmath143 is supported by rotation , which is not generally true for our dispersion dominated systems . this stability analysis would not directly apply to these galaxies , but we can still compare @xmath4 in these systems with our disk model . in figure [ fig : vsig_m ] we show @xmath1/@xmath2 as a function of stellar mass for the galaxies in our sample and additional publicly available measurements for the following surveys : sins @xcite , massiv , and lensed galaxies from @xcite and @xcite . the sins sample is the high signal - to - noise adaptive optics assisted subset and is representative of the most massive and star - forming galaxies at @xmath0 . the lensed sample and massiv sample are more generally representative of star - forming galaxies at these epochs . we use these samples to extend our mass baseline to @xmath144 and to illustrate the consistency of our measurements with those in the literature . the expected distribution of @xmath1/@xmath2 for marginally stable disks at @xmath0 is shown in figure [ fig : vsig_m ] and follows from the model described above . this range includes a characteristic break above @xmath145 , due to a kink in the @xmath0 stellar mass ssfr relation . we find that high mass galaxies ( @xmath146 ) are in general agreement with the marginally stable disk model , i.e. , high mass galaxies have already assembled their disks by @xmath0 . in general these disks are expected to contain large gas reservoirs which drive violent disk instabilities and maintain elevated velocity dispersions . these galaxies are unlike local disks of the same mass , which have values of @xmath147 as opposed to @xmath148 . as their gas reservoirs deplete , they are expected to stabilize and form well - ordered disks @xcite . below a stellar mass of @xmath149 we find the common occurrence of systems which are dispersion dominated and/or marginally rotationally supported ( @xmath1/@xmath2 @xmath150 1 ) . for comparison , the most dispersion dominated system expected from the disk model is @xmath1/@xmath2 = 0.7 ( f@xmath151 = 1 , q@xmath152 = 2 in eq 2 ) . galaxies with lower @xmath1/@xmath2 would be considered globally stable to gas collapse ( q @xmath28 q@xmath152 ) , even as they are obviously forming stars . although gas - rich disks with high velocity dispersion are expected at @xmath0 , a large fraction of our low mass sample can not be described by this simple model for marginally stable disks . we present the first results of sigma ( survey in the near - infrared of galaxies with multiple position angles ) , a kinematic survey of 49 star - forming galaxies over 1.3 @xmath14 z @xmath14 3.0 in three of the candels fields . sigma is representative of the @xmath0 star - formation main sequence to @xmath153 = 9.2 at @xmath154 and @xmath153 = 9.6 at @xmath155 , and includes galaxies with both disk and irregular morphologies . internal kinematics ( rotation velocity @xmath1 and gas velocity dispersion @xmath2 ) are measured from resolved rest - optical nebular emission lines ( h@xmath29 , @xmath30 5007 ) in keck / mosfire nir spectra . we find that high mass galaxies ( @xmath156 ) at @xmath0 are generally rotationally supported , fall on the tf relation , and have formed primitive disks with high @xmath2 . these marginally stable disks are expected to decline in @xmath2 with time as they deplete their gas supply @xcite . in contrast , a large fraction of low mass galaxies ( @xmath157 ) at this epoch are in the early phases of assembling their disks . they have much lower kinematic order " ( @xmath158 ) and tend to fall to low @xmath1 from the tully - fisher relation . for perspective , these results imply that a milky - way progenitor at @xmath0 ( @xmath159 ; @xcite ) is likely still in the process of forming a disk . combining both @xmath1 and @xmath2 in @xmath89 , all of the galaxies in sigma fall on the @xmath89 tully - fisher relation , indicating that many galaxies at low mass have strong contributions of kinematic support in the form of @xmath2 . following @xcite at @xmath1111 , the tendency for high mass galaxies to develop disks first is referred to as kinematic downsizing " , and these results imply that it is in place at @xmath0 . to gain insight into the underlying drivers of kinematic disruption , we explore the dependance of @xmath1/@xmath2 on three physical parameters of galaxies : size , star - formation rate and stellar mass . we find that the strongest trend is with stellar mass , wherein high mass galaxies have the strongest rotational support ( i.e. , highest values of @xmath4 ) . we examine the relation between @xmath4 and residuals from the star - formation main sequence and we find no correlation , to within the uncertainties of our measurements . given that the star - formation rate is likely tracing the available gas mass @xcite , our results imply that mass may be more fundamental than gas fraction in determining kinematic order . finally , we find a moderate correlation between @xmath4 and size , likely associated with the increased angular momentum of the rotationally supported systems . the strong trend between @xmath4 and mass and the lack of a correlation with sfr at fixed mass implies that the processes that maintain @xmath2 in galaxies should be most efficient at low stellar mass and should be independent on gas fraction . several processes which disrupt disk galaxies are active at @xmath0 , associated with elevated star - formation and rising merger rates @xcite . several of these mechanisms may play a role in preferentially stalling disk formation in low mass galaxies . feedback through stellar winds and supernovae can effectively blow out gas and dynamically heat stars in low mass dwarf galaxies ( @xmath160 ; e.g , @xcite ) . we note that we find no trend between @xmath4 and @xmath161 at fixed mass among the galaxies studied in this sample ( @xmath162 ) , indicating that such feedback mechanisms may only assume a secondary role in disturbing the gas kinematics in these galaxies . mergers play a complicated role in shaping the angular momentum content of disk galaxies . models ( e.g. , @xcite ) and observations ( e.g. , ) indicate that gas - rich disks can survive major mergers , albeit with thick disks and perturbed kinematics . examining mock observations from a suite of simulated galaxy mergers , @xcite find that merger remnants tend to scatter to low @xmath1 from the tfr , similar to the low mass galaxies in our sample . although several channels exist for stalling ( and promoting ) disk formation , it is as of yet unclear what the dominate processes are in maintaining the elevated kinematic disturbance we observe in low mass galaxies at this epoch . in summary , we find that kinematic downsizing is active as far back in time as @xmath0 . while massive star - forming galaxies ( @xmath163 ) appear to have assembled primitive rotating disks by this time ( consistent with e.g. , @xcite ) , low mass galaxies tend to be more kinematically disordered and still in the early process of assembling their disks . the peak of cosmic star - formation is coincident with the epoch of disk assembly for low mass galaxies and the settling of primitive disks in high mass galaxies . the authors would like to thank the anonymous referee who provided useful suggestions that improved this paper . rcs would like to thank sandra faber , joel primack , avishai dekel and the candels - baltimore group for useful comments and discussions which have also improved the paper . rcs and sak gratefully acknowledge the support from a grant through the stsci jdf . we thank g. wirth and the tkrs-2 team for their enormous effort observing , reducing and releasing the mosfire spectra which are used for the sigma galaxies in goods - n . we wish to extend thanks to those of hawaiian ancestry on whose sacred mountain we are privileged guests . this work has made use of the rainbow cosmological surveys database , which is operated by the universidad complutense de madrid ( ucm ) , partnered with the university of california observatories at santa cruz ( uco / lick , ucsc ) . jrt acknowledges support from nasa through hubble fellowship grant @xmath16451330 awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa under contract nas 5 - 26555 . cp is supported by an appointment to the nasa postdoctoral program at nasa s goddard space flight center . agertz , o. , & kravtsov , a. v. 2015 , arxiv:1509.00853 aumer , m. , burkert , a. , johansson , p. h. , & genzel , r. 2010 , , 719 , 1230 barnes , j. e. , & hernquist , l. 1996 , , 471 , 115 barrera - ballesteros , j. k. , garca - lorenzo , b. , falcn - barroso , j. , et al . 2015 , , 582 , a21 barro , g. , trump , j. r. , koo , d. c. , et al . 2014 , , 795 , 145 begeman , k. g. 1987 , ph.d . thesis , bell , e. f. , & de jong , r. s. 2001 , , 550 , 212 bouch , n. , dekel , a. , genzel , r. , et al . 2010 , , 718 , 1001 bradford , j. d. , geha , m. c. , & van den bosch , f. c. 2016 , arxiv:1602.02757 brammer , g. b. , van dokkum , p. g. , & coppi , p. 2008 , , 686 , 1503 brammer , g. b. , van dokkum , p. g. , franx , m. , et al . 2012 , , 200 , 13 brinchmann , j. , & ellis , r. s. 2000 , , 536 , l77 brooks , a. m. , governato , f. , quinn , t. , brook , c. b. , & wadsley , j. 2009 , , 694 , 396 bruzual , g. , & charlot , s. 2003 , , 344 , 1000 calzetti , d. , armus , l. , bohlin , r. c. , et al . 2000 , , 533 , 682 ceverino , d. , dekel , a. , & bournaud , f. 2010 , , 404 , 2151 ceverino , d. , primack , j. , & dekel , a. 2015 , , 453 , 408 chabrier , g. 2003 , , 115 , 763 conselice , c. j. , bundy , k. , ellis , r. s. , et al . 2005 , , 628 , 160 contini , t. , garilli , b. , le fvre , o. , et al . 2012 , , 539 , a91 contini , t. , epinat , b. , bouch , n. , et al . 2015 , arxiv:1512.00246 cowie , l. l. , songaila , a. , hu , e. m. , & cohen , j. g. 1996 , , 112 , 839 courteau , s. , dutton , a. a. , van den bosch , f. c. , et al . 2007 , , 671 , 203 covington , m. d. , kassin , s. a. , dutton , a. a. , et al . 2010 , , 710 , 279 cresci , g. , hicks , e. k. s. , genzel , r. , et al . 2009 , , 697 , 115 daddi , e. , elbaz , d. , walter , f. , et al . 2010 , , 714 , l118 daddi , e. , bournaud , f. , walter , f. , et al . 2010 , , 713 , 686 dahlen , t. , mobasher , b. , faber , s. m. , et al . 2013 , , 775 , 93 danovich , m. , dekel , a. , hahn , o. , ceverino , d. , & primack , j. 2015 , , 449 , 2087 dav , r. , finlator , k. , & oppenheimer , b. d. 2012 , , 421 , 98 dekel , a. , sari , r. , & ceverino , d. 2009 , , 703 , 785 el - badry , k. , wetzel , a. , geha , m. , et al . 2016 , , 820 , 131 elbaz , d. , dickinson , m. , hwang , h. s. , et al . 2011 , , 533 , a119 elmegreen , b. g. , & burkert , a. 2010 , , 712 , 294 epinat , b. , amram , p. , balkowski , c. , & marcelin , m. 2010 , , 401 , 2113 fall , s. m. , & efstathiou , g. 1980 , , 193 , 189 flores , h. , hammer , f. , puech , m. , amram , p. , & balkowski , c. 2006 , , 455 , 107 frster schreiber , n. m. , genzel , r. , lehnert , m. d. , et al . 2006 , , 645 , 1062 frster schreiber , n. m. , genzel , r. , bouch , n. , et al . 2009 , , 706 , 1364 galametz , a. , grazian , a. , fontana , a. , et al . 2013 , , 206 , 10 gallazzi , a. , charlot , s. , brinchmann , j. , & white , s. d. m. 2006 , , 370 , 1106 genel , s. , dekel , a. , & cacciato , m. 2012 , , 425 , 788 genzel , r. , tacconi , l. j. , eisenhauer , f. , et al . 2006 , , 442 , 786 genzel , r. , burkert , a. , bouch , n. , et al . 2008 , , 687 , 59 genzel , r. , newman , s. , jones , t. , et al . 2011 , , 733 , 101 genzel , r. , tacconi , l. j. , lutz , d. , et al . 2015 , , 800 , 20 glazebrook , k. 2013 , pasa , 30 , e056 grogin , n. a. , kocevski , d. d. , faber , s. m. , et al . 2011 , , 197 , 35 governato , f. , brook , c. b. , brooks , a. m. , et al . 2009 , , 398 , 312 guo , y. , ferguson , h. c. , giavalisco , m. , et al . 2013 , , 207 , 24 hammer , f. , flores , h. , elbaz , d. , et al . 2005 , , 430 , 115 hammer , f. , flores , h. , puech , m. , et al . 2009 , , 507 , 1313 hopkins , p. f. , cox , t. j. , younger , j. d. , & hernquist , l. 2009 , , 691 , 1168 hung , c .- l . , rich , j. a. , yuan , t. , et al . 2015 , , 803 , 62 hung , c .- hayward , c. c. , smith , h. a. , et al . 2016 , , 816 , 99 inoue , s. , dekel , a. , mandelker , n. , et al . 2016 , , 456 , 2052 jones , t. a. , swinbank , a. m. , ellis , r. s. , richard , j. , & stark , d. p. 2010 , , 404 , 1247 juneau , s. , glazebrook , k. , crampton , d. , et al . 2005 , , 619 , l135 kassin , s. a. , de jong , r. s. , & weiner , b. j. 2006 , , 643 , 804 kassin , s. a. , weiner , b. j. , faber , s. m. , et al . 2007 , , 660 , l35 kassin , s. a. , weiner , b. j. , faber , s. m. , et al . 2012 , , 758 , 106 kennicutt , r. c. , jr . 1998 , , 36 , 189 kennicutt , r. c. , jr . 1998 , , 498 , 541 kimm , t. , devriendt , j. , slyz , a. , et al . 2011 , arxiv:1106.0538 koekemoer , a. m. , faber , s. m. , ferguson , h. c. , et al . 2011 , , 197 , 36 kriek , m. , van dokkum , p. g. , labb , i. , et al . 2009 , , 700 , 221 kriek , m. , shapley , a. e. , reddy , n. a. , et al . 2015 , , 218 , 15 law , d. r. , shapley , a. e. , steidel , c. c. , et al . 2012 , , 487 , 338 lehnert , m. d. , nesvadba , n. p. h. , le tiran , l. , et al . 2009 , , 699 , 1660 livermore , r. c. , jones , t. a. , richard , j. , et al . 2015 , , 450 , 1812 madau , p. , & dickinson , m. 2014 , , 52 , 415 mac low , m .- m . , & ferrara , a. 1999 , , 513 , 142 magnelli , b. , popesso , p. , berta , s. , et al . 2013 , , 553 , a132 mandelker , n. , dekel , a. , ceverino , d. , et al . 2015 , arxiv:1512.08791 masters , k. l. , springob , c. m. , & huchra , j. p. 2008 , , 135 , 1738 masters , d. , & capak , p. 2011 , , 123 , 638 mcgaugh , s. s. 2005 , , 632 , 859 mclean , i. s. , steidel , c. c. , epps , h. w. , et al . 2012 , , 8446 , 84460j miller , s. h. , bundy , k. , sullivan , m. , ellis , r. s. , & treu , t. 2011 , , 741 , 115 miller , s. h. , ellis , r. s. , sullivan , m. , et al . 2012 , , 753 , 74 miller , s. h. , ellis , r. s. , newman , a. b. , & benson , a. 2014 , , 782 , 115 mobasher , b. , dahlen , t. , ferguson , h. c. , et al . 2015 , , 808 , 101 mo , h. j. , mao , s. , & white , s. d. m. 1998 , , 295 , 319 momcheva , i. g. , brammer , g. b. , van dokkum , p. g. , et al . 2015 , arxiv:1510.02106 morris , a. m. , kocevski , d. d. , trump , j. r. , et al . 2015 , , 149 , 178 mortlock , a. , conselice , c. j. , hartley , w. g. , et al . 2013 , , 433 , 1185 mortlock , a. , conselice , c. j. , hartley , w. g. , et al . 2015 , , 447 , 2 nelson , e. j. , van dokkum , p. g. , frster schreiber , n. m. , et al . 2015 , arxiv:1507.03999 newman , s. f. , genzel , r. , frster schreiber , n. m. , et al . 2013 , , 767 , 104 noeske , k. g. , weiner , b. j. , faber , s. m. , et al . 2007 , , 660 , l43 papovich , c. , labb , i. , quadri , r. , et al . 2015 , , 803 , 26 peng , c. y. , ho , l. c. , impey , c. d. , & rix , h .- w . 2010 , , 139 , 2097 pizagno , j. , prada , f. , weinberg , d. h. , et al . 2005 , , 633 , 844 price , s. h. , kriek , m. , shapley , a. e. , et al . 2016 , , 819 , 80 puech , m. , flores , h. , hammer , f. , et al . 2008 , , 484 , 173 reddy , n. a. , pettini , m. , steidel , c. c. , et al . 2012 , , 754 , 25 reyes , r. , mandelbaum , r. , gunn , j. e. , pizagno , j. , & lackner , c. n. 2011 , , 417 , 2347 rodriguez - gomez , v. , genel , s. , vogelsberger , m. , et al . 2015 , , 449 , 49 rodriguez - puebla , a. , behroozi , p. , primack , j. , et al . 2016 , arxiv:1602.04813 shapiro , k. l. , genzel , r. , frster schreiber , n. m. , et al . 2008 , , 682 , 231 shields , g. a. 1990 , , 28 , 525 simons , r. c. , kassin , s. a. , weiner , b. j. , et al . 2015 , , 452 , 986 speagle , j. s. , steinhardt , c. l. , capak , p. l. , & silverman , j. d. 2014 , , 214 , 15 stewart , k. r. , brooks , a. m. , bullock , j. s. , et al . 2013 , , 769 , 74 stott , j. p. , swinbank , a. m. , johnson , h. l. , et al . 2016 , , 457 , 1888 swinbank , a. m. , papadopoulos , p. p. , cox , p. , et al . 2011 , , 742 , 11 swinbank , a. m. , sobral , d. , smail , i. , et al . 2012 , , 426 , 935 tacconi , l. j. , neri , r. , genzel , r. , et al . 2013 , , 768 , 74 tomassetti , m. , dekel , a. , mandelker , n. , et al . 2016 , , 458 , 4477 torres - flores , s. , epinat , b. , amram , p. , plana , h. , & mendes de oliveira , c. 2011 , , 416 , 1936 trump , j. r. , konidaris , n. p. , barro , g. , et al . 2013 , , 763 , l6 tully , r. b. , & fisher , j. r. 1977 , , 54 , 661 van der wel , a. , bell , e. f. , hussler , b. , et al . 2012 , , 203 , 24 van der wel , a. , franx , m. , van dokkum , p. g. , et al . 2014 , , 788 , 28 van der wel , a. , chang , y .- y . , bell , e. f. , et al . 2014 , , 792 , l6 vergani , d. , epinat , b. , contini , t. , et al . 2012 , , 546 , a118 verheijen , m. a. w. 2001 , , 563 , 694 weiner , b. j. , willmer , c. n. a. , faber , s. m. , et al . 2006 , , 653 , 1027 weiner , b. j. , willmer , c. n. a. , faber , s. m. , et al . 2006 , , 653 , 1049 wheeler , c. , pace , a. b. , bullock , j. s. , et al . 2015 , arxiv:1511.01095 whitaker , k. e. , franx , m. , leja , j. , et al . 2014 , , 795 , 104 wirth , g. d. , trump , j. r. , barro , g. , et al . 2015 , , 150 , 153 wisnioski , e. , glazebrook , k. , blake , c. , et al . 2011 , , 417 , 2601 wisnioski , e. , frster schreiber , n. m. , wuyts , s. , et al . 2015 , , 799 , 209 wuyts , s. , frster schreiber , n. m. , lutz , d. , et al . 2011 , , 738 , 106 zolotov , a. , dekel , a. , mandelker , n. , et al . 2015 , , 450 , 2327 in this appendix , we perform mock observations of simulated sigma spectra in order to determine the limiting spatial scale from which we can recover rotation velocity , if present , in our observations . we find that we can recover kinematics down to a limit of d@xmath44/seeing@xmath46 = 0.8 . in 2 , we apply this limit to our data set . the spectra presented in this paper were observed in seeing limited conditions , and as such the effects of beam smearing @xcite are important to understand . seeing tends to blur together spatially distinct velocities , leading to a decreased rotation signature and an elevated velocity dispersion . to explore the effective resolution set by beam smearing in sigma , we create a set of model spectra with various sizes and fixed kinematic parameters . as described below , we perform mock observations of the kinematics for these spectra and explore how the results vary with size . our model for a single mock sigma / mosfire spectrum consists of first creating a 1d emission line with a sersic intensity profile : where r@xmath120 is the half - light radius , @xmath166 is the intensity at the half - light radius , @xmath167 is the sersic index and @xmath168 is a numerical constant that depends on @xmath167 ( @xmath169 ) . we set @xmath170 , characteristic of the emission line intensity profile for galaxies at this redshift @xcite . we then draw from an appropriate error spectrum to set the central signal - to - noise per pixel to 15 , typical of sigma spectra . the final integrated s / n of these simulated spectra are well - matched to our sample . next , we apply an arctan velocity profile ( see eq 1 ) and a constant velocity dispersion to the emission line . we fix the rotation velocity to 100 km s@xmath26 and the velocity dispersion to 50 km s@xmath26 , characteristic of @xmath0 disk galaxies . we then apply a seeing kernel of 0.7 " in the spatial direction and mosfire instrumental broadening ( @xmath171 = 40 km s@xmath26 ) in the spectral direction . following the seeing convolution , the emission profile is well described by a gaussian ( @xmath172 ) . we repeat this process 100 times , redrawing from the error spectrum each time to create independent realizations with the same fixed parameters . we then vary the effective radius of the emission line from 0.1 - 0.7 @xmath35 in 30 even steps of 0.02@xmath35/step . at each step we again simulate 100 realizations with the same fixed properties . the final dataset contains 3000 simulated sigma spectra . to measure the kinematics from the mock spectra , we run the same fitting technique used for our observations ( rotcurve ) for each spatial scale modeled , we measure 100 values of @xmath1 and @xmath2 and examine their mean and variance . the results of this study are shown in figure [ fig : model_size ] . the rotation velocity is well recovered with an intrinsic error of @xmath24 15 km s@xmath26 . for simulated spectra with effective diameter at or greater than the seeing , we accurately recover the input rotation and dispersion . below a value of d@xmath44/seeing@xmath46 = 0.8 , the limiting spatial resolution drastically affects the recovered values of @xmath1 and @xmath2 . there is both a larger uncertainty in the measurements and a systematic decline in the measured rotation velocity . based on the results of these models , we determine that the spatial scale for recovering kinematics in our observations is diameter d@xmath44 @xmath28 0.8 @xmath45 seeing@xmath46 . above this value we can reliably recover rotational motion , if present . ccllccccc 5492 & & 03 32 41.78 & @xmath17327 51 35.28 & 1.67 & 22.51 & 10.39 & 1.95 & 0.57 + 14602 & & 03 32 14.66 & @xmath17327 47 02.70 & 1.73 & 22.37 & 10.18 & 1.44 & 0.91 + 16209 & & 03 32 15.82 & @xmath17327 46 22.36 & 1.33 & 23.47 & 9.36 & 0.23 & 0.34 + 16600 & & 03 32 30.71 & @xmath17327 46 17.17 & 1.31 & 21.20 & 10.51 & 1.59 & 0.70 + 16905 & & 03 32 15.75 & @xmath17327 46 04.32 & 1.54 & 22.15 & 10.13 & 2.05 & 0.32 + 17488 & & 03 32 20.18 & @xmath17327 45 49.31 & 1.61 & 23.16 & 9.46 & 1.15 & 0.41 + 17673 & & 03 32 33.86 & @xmath17327 45 42.65 & 1.62 & 22.69 & 10.05 & 1.68 & 0.85 + 18177 & & 03 32 12.48 & @xmath17327 45 30.42 & 1.38 & 22.03 & 10.19 & 1.57 & 0.41 + 19818 & & 03 32 18.07 & @xmath17327 44 33.44 & 1.38 & 22.43 & 9.23 & 1.31 & 0.81 + 20495 & & 03 32 11.86 & @xmath17327 44 13.38 & 1.33 & 21.30 & 10.43 & 1.69 & 0.78 + 20569 & & 03 32 29.19 & @xmath17327 44 13.21 & 1.47 & 23.89 & 9.24 & 1.44 & 0.49 + 20593 & & 03 32 12.77 & @xmath17327 44 07.77 & 1.61 & 23.13 & 10.26 & 1.49 & 0.29 + 20883 & & 03 32 26.77 & @xmath17327 43 58.22 & 1.61 & 23.43 & 9.88 & 1.14 & 0.51 + 21007 & & 03 32 31.83 & @xmath17327 43 56.26 & 1.55 & 21.80 & 10.67 & 1.55 & 0.81 + 21162 & & 03 32 23.95 & @xmath17327 43 49.09 & 1.31 & 22.16 & 9.79 & 1.02 & 0.87 + 21258 & & 03 32 30.75 & @xmath17327 43 45.34 & 1.43 & 23.17 & 9.37 & 0.93 & 0.33 + 21737 & & 03 32 30.14 & @xmath17327 43 35.44 & 1.60 & 22.55 & 10.20 & 1.08 & 0.69 + 21897 & & 03 32 34.08 & @xmath17327 43 28.43 & 1.60 & 22.44 & 9.82 & 1.51 & 0.42 + 22236 & & 03 32 26.19 & @xmath17327 43 17.38 & 1.61 & 23.09 & 9.61 & 1.03 & 0.49 + 22354 & & 03 32 13.66 & @xmath17327 43 13.08 & 1.47 & 22.45 & 9.67 & 1.28 & 0.91 + 22783 & & 03 32 26.30 & @xmath17327 43 01.89 & 1.55 & 23.11 & 9.83 & 0.71 & 0.46 + 23095 & & 03 32 38.50 & @xmath17327 42 27.89 & 1.61 & 21.93 & 10.39 & 1.65 & 0.76 + 23632 & & 03 32 28.13 & @xmath17327 42 48.27 & 1.61 & 23.35 & 9.45 & 1.01 & 0.33 + 25715 & & 03 32 22.72 & @xmath17327 41 40.74 & 1.61 & 22.12 & 10.34 & 1.60 & 0.96 + 25973 & & 03 32 26.08 & @xmath17327 41 38.49 & 1.54 & 22.74 & 9.84 & 0.92 & 0.62 + 26050 & & 03 32 13.89 & @xmath17327 41 58.29 & 1.32 & 22.57 & 9.64 & 1.78 & 0.24 + 26177 & & 03 32 27.28 & @xmath17327 42 05.20 & 1.61 & 21.11 & 10.65 & 1.73 & 0.91 + 26736 & & 03 32 26.41 & @xmath17327 42 28.30 & 1.61 & 22.21 & 10.18 & 2.09 & 0.49 + 1826 & & 12 36 18.78 & @xmath17462 08 39.70 & 2.30 & 23.03 & 9.79 & 1.12 & 0.32 + 2815 & & 12 36 35.56 & @xmath17462 09 20.33 & 2.36 & 23.26 & 9.71 & 1.11 & 0.66 + 4106 & & 12 36 39.37 & @xmath17462 10 06.60 & 2.35 & 22.40 & 10.65 & 1.58 & 0.31 + 4476 & & 12 36 13.13 & @xmath17462 10 21.11 & 2.24 & 23.50 & 9.89 & 0.95 & 0.27 + 4714 & & 12 36 36.08 & @xmath17462 10 29.90 & 2.31 & 23.01 & 10.17 & 1.03 & 0.83 + 4925 & & 12 36 11.51 & @xmath17462 10 33.72 & 2.25 & 22.52 & 10.92 & 1.73 & 0.72 + 4962 & & 12 36 18.77 & @xmath17462 10 37.28 & 2.27 & 23.56 & 9.68 & 1.12 & 0.59 + 9157 & & 12 36 13.56 & @xmath17462 12 21.58 & 2.44 & 23.13 & 9.90 & 1.75 & 0.48 + 9190 & & 12 36 18.27 & @xmath17462 12 22.18 & 2.44 & 23.38 & 9.85 & 1.13 & 0.89 + 10230 & & 12 37 06.71 & @xmath17462 12 44.61 & 2.21 & 23.23 & 10.21 & 0.94 & 0.17 + 11525 & & 12 36 26.94 & @xmath17462 13 17.86 & 2.40 & 23.91 & 9.48 & 2.11 & 0.42 + 13230 & & 12 36 09.05 & @xmath17462 13 59.03 & 2.05 & 24.83 & 9.96 & 1.19 & 0.37 + 13678 & & 12 36 44.08 & @xmath17462 14 10.10 & 2.27 & 26.53 & 9.66 & 0.82 & 0.41 + 14428 & & 12 36 35.59 & @xmath17462 14 24.01 & 2.02 & 25.47 & 11.79 & 3.61 & 0.86 + 15497 & & 12 36 21.74 & @xmath17462 14 52.85 & 2.21 & 26.28 & 10.06 & 1.14 & 0.76 + 16028 & & 12 36 51.82 & @xmath17462 15 04.72 & 2.19 & 26.38 & 10.76 & 1.00 & 0.76 + 16260 & & 12 37 00.46 & @xmath17462 15 08.86 & 2.33 & 24.82 & 10.92 & 1.14 & 0.77 + 10404 & & 02 17 35.93 & @xmath17305 13 05.20 & 2.31 & 22.92 & 10.04 & 1.55 & 0.51 + 12138 & & 02 17 48.89 & @xmath17305 12 30.38 & 2.25 & 23.26 & 9.98 & 1.24 & 0.28 + 14004 & & 02 17 30.72 & @xmath17305 11 56.22 & 2.39 & 23.25 & 10.46 & 1.54 & 0.57 + 14042 & & 02 17 32.78 & @xmath17305 11 55.93 & 2.15 & 22.65 & 9.97 & 1.52 & 0.27 + [ tab : table1 ] ccccccccc 5492 & 1 & 65.0 @xmath175 0.7 & 34.5 & 30.5 & 0.2 & 10 @xmath175 30 & 11 @xmath175 35 & 45 @xmath175 14 + 14602 & 1 & @xmath17311.7 @xmath175 0.7 & 19.0 & 30.7 & 0.2 & 25 @xmath175 37 & 59 @xmath175 87 & 85 @xmath175 23 + 16209 & 2 & 9.4 @xmath175 1.3 & 19.0 & 9.6 & 0.2 & 65 @xmath175 28 & 67 @xmath175 29 & 86 @xmath175 22 + 16600 & 2 & 7.0 @xmath175 0.2 & 10.0 & 3.0 & 0.2 & 171 @xmath175 23 & 234 @xmath175 31 & 35 @xmath175 35 + 16905 & 3 & @xmath17387.0 @xmath175 0.3 & -56.0 & 31.0 & 0.4 & 131 @xmath175 63 & 135 @xmath175 65 & 80 @xmath175 24 + 17488 & 2 & 17.4 @xmath175 1.8 & 19.0 & 1.6 & 0.2 & 5 @xmath175 14 & 5 @xmath175 15 & 45 @xmath175 5 + 17673 & 1 & @xmath1732.8 @xmath175 0.6 & 8.0 & 10.8 & 0.2 & 44 @xmath175 20 & 81 @xmath175 37 & 35 @xmath175 15 + 18177 & 2 & 83.1 @xmath175 0.4 & 56.0 & 27.1 & 0.4 & 161 @xmath175 50 & 172 @xmath175 53 & 90 @xmath175 26 + 19818 & 2 & @xmath17386.6 @xmath175 1.4 & 56.0 & 37.4 & 0.3 & 24 @xmath175 21 & 40 @xmath175 35 & 80 @xmath175 13 + 20495 & 3 & 74.1 @xmath175 0.3 & 56.0 & 18.1 & 0.4 & 120 @xmath175 27 & 187 @xmath175 42 & 65 @xmath175 22 + 20569 & 1 & 22.5 @xmath175 1.8 & 8.0 & 14.5 & 0.2 & 31 @xmath175 27 & 34 @xmath175 30 & 36 @xmath175 15 + 20593 & 2 & 34.1 @xmath175 0.9 & 19.0 & 15.1 & 0.4 & 130 @xmath175 44 & 133 @xmath175 45 & 49 @xmath175 26 + 20883 & 1 & 29.8 @xmath175 2.1 & 8.0 & 21.8 & 0.2 & 51 @xmath175 36 & 58 @xmath175 41 & 89 @xmath175 14 + 21007 & 5 & @xmath17359.5 @xmath175 0.4 & 117.0 & 3.5 & 0.2 & 136 @xmath175 24 & 227 @xmath175 40 & 39 @xmath175 40 + 21162 & 3 & @xmath17382.7 @xmath175 0.7 & 19.0 & 78.3 & 0.2 & 52 @xmath175 19 & & 80 @xmath175 6 + 21258 & 3 & @xmath17352.7 @xmath175 1.4 & 117.0 & 10.3 & 0.4 & 101 @xmath175 78 & 104 @xmath175 80 & 55 @xmath175 38 + 21737 & 1 & @xmath17319.6 @xmath175 0.6 & 8.0 & 27.6 & 0.2 & 157 @xmath175 29 & 212 @xmath175 39 & 30 @xmath175 31 + 21897 & 2 & @xmath17324.5 @xmath175 0.7 & 120.5 & 35.0 & 0.2 & 136 @xmath175 20 & 146 @xmath175 21 & 55 @xmath175 17 + 22236 & 2 & @xmath17315.7 @xmath175 1.5 & 19.0 & 34.7 & 0.2 & 110 @xmath175 32 & 123 @xmath175 35 & 50 @xmath175 23 + 22354 & 2 & 36.0 @xmath175 0.7 & 19.0 & 17.0 & 0.2 & 56 @xmath175 20 & 132 @xmath175 47 & 30 @xmath175 20 + 22783 & 1 & 32.1 @xmath175 1.5 & 10.0 & 22.1 & 0.2 & 5 @xmath175 38 & 5 @xmath175 41 & 40 @xmath175 16 + 23095 & 2 & @xmath17370.9 @xmath175 0.5 & 120.5 & 11.4 & 0.2 & 200 @xmath175 42 & 301 @xmath175 63 & 34 @xmath175 36 + 23632 & 2 & 88.3 @xmath175 1.2 & 78.5 & 9.8 & 0.2 & 34 @xmath175 18 & 35 @xmath175 18 & 39 @xmath175 8 + 25715 & 1 & @xmath17339.7 @xmath175 0.9 & 120.5 & 19.8 & 0.2 & 75 @xmath175 43 & 262 @xmath175 150 & 109 @xmath175 23 + 25973 & 3 & 33.0 @xmath175 0.8 & 19.0 & 14.0 & 0.2 & 79 @xmath175 21 & 98 @xmath175 26 & 40 @xmath175 15 + 26050 & 1 & @xmath17353.0 @xmath175 0.6 & 117.0 & 10.0 & 0.3 & 222 @xmath175 33 & 224 @xmath175 33 & 50 @xmath175 24 + 26177 & 2 & @xmath17320.1 @xmath175 0.2 & 120.5 & 39.4 & 0.2 & 74 @xmath175 18 & 174 @xmath175 42 & 45 @xmath175 13 + 26736 & 2 & @xmath17335.2 @xmath175 0.5 & -10.0 & 25.2 & 0.2 & 242 @xmath175 35 & 272 @xmath175 39 & 75 @xmath175 42 + 1826 & 1 & @xmath17311.7 @xmath175 1.2 & 32.0 & 43.7 & 0.3 & 30 @xmath175 30 & 31 @xmath175 31 & 49 @xmath175 12 + 2815 & 2 & @xmath17312.8 @xmath175 3.4 & 32.0 & 44.8 & 0.3 & 29 @xmath175 40 & 37 @xmath175 52 & 64 @xmath175 12 + 4106 & 2 & @xmath17329.5 @xmath175 0.4 & 32.0 & 61.5 & 0.3 & 82 @xmath175 64 & & 129 @xmath175 25 + 4476 & 1 & 68.0 @xmath175 2.4 & 106.9 & 38.9 & 0.3 & 90 @xmath175 37 & 91 @xmath175 37 & 60 @xmath175 25 + 4714 & 1 & @xmath17347.9 @xmath175 1.2 & 106.9 & 25.2 & 0.3 & 68 @xmath175 38 & 119 @xmath175 66 & 84 @xmath175 30 + 4925 & 3 & @xmath17329.0 @xmath175 1.2 & -21.5 & 7.5 & 0.2 & 318 @xmath175 35 & 448 @xmath175 49 & 25 @xmath175 61 + 4962 & 3 & 24.9 @xmath175 3.9 & 32.0 & 7.1 & 0.2 & 124 @xmath175 25 & 150 @xmath175 30 & 55 @xmath175 20 + 9157 & 3 & 14.6 @xmath175 0.8 & 106.9 & 87.7 & 0.3 & 64 @xmath175 22 & & 85 @xmath175 15 + 9190 & 3 & @xmath17384.2 @xmath175 1.6 & 32.0 & 63.8 & 0.2 & 21 @xmath175 28 & & 36 @xmath175 18 + 10230 & 1 & @xmath1738.3 @xmath175 7.7 & -53.0 & 44.7 & 0.3 & 50 @xmath175 26 & 50 @xmath175 26 & 45 @xmath175 20 + 11525 & 2 & 44.7 @xmath175 2.7 & 32.0 & 12.7 & 0.2 & 256 @xmath175 32 & 276 @xmath175 34 & 10 @xmath175 52 + 13230 & 1 & 22.1 @xmath175 18.5 & -21.5 & 43.6 & 0.3 & 10 @xmath175 22 & 10 @xmath175 23 & 75 @xmath175 7 + 13678 & 1 & @xmath17334.9 @xmath175 0.2 & -53.0 & 18.1 & 0.3 & 60 @xmath175 25 & 64 @xmath175 26 & 61 @xmath175 8 + 14428 & 1 & @xmath17368.3 @xmath175 5.6 & -53.0 & 15.3 & 0.2 & 82 @xmath175 24 & 157 @xmath175 46 & 71 @xmath175 14 + 15497 & 1 & 7.9 @xmath175 21.9 & 32.0 & 24.1 & 0.2 & 49 @xmath175 31 & 73 @xmath175 46 & 54 @xmath175 20 + 16028 & 1 & @xmath17371.4 @xmath175 5.8 & -53.0 & 18.4 & 0.2 & 203 @xmath175 36 & 306 @xmath175 54 & 110 @xmath175 36 + 16260 & 1 & @xmath17363.0 @xmath175 10.3 & -53.0 & 10.0 & 0.2 & 200 @xmath175 48 & 307 @xmath175 73 & 85 @xmath175 51 + 10404 & 1 & 52.2 @xmath175 1.8 & -297.0 & 10.8 & 0.2 & 100 @xmath175 33 & 114 @xmath175 37 & 74 @xmath175 14 + 12138 & 1 & 67.3 @xmath175 1.9 & -297.0 & 4.3 & 0.1 & 46 @xmath175 17 & 46 @xmath175 17 & 55 @xmath175 10 + 14004 & 1 & 40.6 @xmath175 1.3 & -297.0 & 22.4 & 0.2 & 248 @xmath175 68 & 296 @xmath175 81 & 15 @xmath175 83 + 14042 & 1 & 34.9 @xmath175 0.9 & -297.0 & 28.1 & 0.3 & 50 @xmath175 26 & 50 @xmath175 26 & 70 @xmath175 8 + [ tab : table2 ]

biological networks transform multiple input signals into outputs that have functional significance for the organism . one path to understanding these transformations is to `` read out , '' or decode this relevant information directly from the network activity . in neural networks , for example , features of the organism s sensory inputs and motor outputs have been decoded from observed action potential sequences , sometimes with very high accuracy @xcite . decoding provides an explicit test for hypotheses about how biologically meaningful information is represented in the network and which computations are needed to recover it . here we address these questions in a small genetic network , taking advantage of experimental methods that allow us to measure , quantitatively , the simultaneous expression levels of multiple genes . the gap genes involved in patterning the early fruit fly embryo provide a particularly accessible example of the decoding problem @xcite . as schematized in fig [ f1]a , the gap genes form an interconnected layer in an otherwise feed - forward network that takes inputs from the primary maternal morphogens to drive the expression of the pair - rule genes . pair - rule expression occurs in stripes that are precisely and reproducibly positioned within the embryo , forming an outline for the segmented body plan of the fully developed organism . in this system , states of the network are the expression levels of the gap genes , and the functional information being encoded is the position of cells along the anterior posterior axis ( fig [ f1]b ) . we will use the gap gene network to distinguish between two very different views of network function . in one view , the control of gene expression is noisy , network interactions are essential to achieve robustness against uncontrolled variations in the input signals , and many layers are needed to funnel the states of cells toward reliable choices among discrete fates ( canalization ) . in this view , information available to individual cells from gap gene expression levels should be very limited , and the accuracy associated with a reproducible body plan should emerge only in subsequent layers of processing . in another view , noise levels are as low as possible given the limited number of molecules involved @xcite , the reproducibility of developmental patterning can be traced back to reproducible maternal inputs @xcite , and network interactions are selected to extract the maximum amount of information from these inputs @xcite . in support of this latter , precisionist view , we have shown that the expression levels of the four gap genes , taken together at one moment in time , provide enough information to specify the location of each cell along the anterior posterior axis of the embryo with @xmath0 precision @xcite , comparable to the precision with which pair - rule stripes and other morphological markers are specified . showing that precise information is available is not enough : to say that the system operates in a precisionist mode we must show that this information is used by the embryo to guide subsequent developmental events . to do this , we will decode the information carried by gap gene expression levels , and manipulate these signals through mutations in the maternal inputs to the network . if the organism really uses the information that we have found encoded in the gap gene network , then our decoding of this information should make quantitative predictions for the functional consequences of mutations at the input . in the fly embryo , information encoded in the levels of gap gene expression is decoded by the complex regulatory logic of the pair rule gene enhancers @xcite , and one might think that decoding requires us to make a model of these molecular events . but if the embryo makes optimal use of the available information @xcite , then the algorithm for decoding positional information is determined , mathematically , by the pattern of gene expression as a function of position and by the distribution of fluctuations around this mean . exploiting our ability to make precise , quantitative measurements on the expression of all four gap genes simultaneously @xcite , we can characterize these fluctuations and construct the optimal decoder with no free parameters ; this is not a model of the decoder , but rather a theoretical prediction of what the decoder should be if the embryo performs optimally . applied to gap gene expression levels in wild - type embryos , this optimal decoder generates unambiguous estimates of position , which fluctuate only slightly from the true position . in mutant embryos deficient for various primary maternal inputs , gap gene expression changes at each point in the embryo , and the decoder generates distorted `` maps '' of implied vs true position ; these maps in turn can be used to predict the locations of pair - rule stripes . we find that these predictions are in detailed , quantitative agreement with experiment , again with no adjustable parameters . these results strongly support the idea that biologically relevant information is carried by precisely controlled gene expression levels , so that the function of this genetic network is fundamentally quantitative , down to @xmath0 accuracy in the encoded variable . the expression levels of the gap genes vary with position , but even at a single position there are variations around the mean expression level . thus , at each point @xmath1 along the anterior posterior axis of the embryo , there is a probability distribution , @xmath2 , where @xmath3 are the expression levels of the genes , in our case the gap genes _ hunchback , _ _ krppel , _ _ knirps , _ and _ giant_. the problem facing a cell , however , is not to predict the expression level given knowledge of its position . on the contrary , the cell must respond to the expression level by taking actions appropriate to its position . mathematically , the possible positions of the cell are drawn out of the conditional distribution @xmath4 which we construct using bayes rule ; we use @xmath5 to remind us that this is the position _ implied _ by the expression levels , rather than the actual position of the cell . @xmath6 is the probability that a cell will be at position @xmath5 , independent of gene expression level , and in the simplest case the cells are uniformly distributed along the axis @xmath7 , so that @xmath8 . @xmath9 is the probability that we will find a cell with expression levels @xmath10 , independent of its position , @xmath11 . the probability distribution @xmath12 contains everything that a cell could `` know '' about its position based on the expression levels @xmath10 . if the cell responds to the expression levels in a way that matches the structure of this distribution , then that response is part of an optimal decoding . to be more precise , if the probability distribution @xmath12 has a single , reasonably sharp peak at @xmath13 , then we can translate expression levels back into positions , unambiguously , using a dictionary @xmath14 , and the width of the distribution @xmath12 tells us how much error or uncertainty there will be in this estimate of position . we could formalize this by asking for the value of @xmath5 that maximizes @xmath12 , which is called the maximum a posteriori estimator , or for the value of @xmath5 such that we make smallest mean - square error relative to the true position , and these are essentially the same if there is a single sharp peak in the distribution . but it also is possible that @xmath12 has multiple peaks , in which case there are genuine ambiguities in decoding , and hence the gene expression levels we are analyzing in fact are not sufficient to determine the position and hence the fate of the cell . since we do nt know in advance that decoding will be successful , it is useful to keep track of the entire distribution . indeed , since the information that we are discussing is information about position , it is useful to express the results of decoding as a _ map_. consider a particular embryo @xmath15 , in which we find expression levels @xmath16 in the cell at actual position @xmath17 . instead of thinking of the distribution of possible positions in eq ( [ decode1gene ] ) as depending on all of the gene expression levels , we can think of it as depending on the actual position of the cell , forming a map of possible or inferred positions vs actual positions , @xmath18 if the genes we are considering provide enough information to specify position accurately and unambiguously , then @xmath19 will be a narrow ridge of density surrounding the line such that the implied position is equal to the true position , @xmath20 . to implement these ideas , we begin by measuring expression of all four gap genes , simultaneously , as described in detail in appendix [ app_exp ] . we sort the emrbyos by time , and focus on a window 38@xmath2148 min into nuclear cycle fourteen , during which gap gene expression provides the most information about position @xcite . we approximate the distribution @xmath2 as being gaussian @xcite , which means that we are estimating the mean expression level at each position , across the ensemble of embryos , and the covariance matrix ( fig [ sf12 ] ) that describes fluctuations around this mean . given these features of the data , the construction of the decoding map proceeds with no adjustable parameters ; for details see appendix [ methods+thy ] and figs [ sf0 ] through [ sf4 ] . in figure [ f2 ] we show the decoding maps based on the information carried by one , two , three , or all four gap genes . for most locations in the embryo , decoding based on a single gene provides little information , as for kr in fig [ f2]a and [ sf0 ] ; other examples are in fig [ sf2 ] . in small regions of the embryo , decoding can be more precise , but substantial ambiguities remain , where one expression level is equally consistent with two different implied positions . decoding based on two ( figs [ f2]b and [ sf3 ] ) or three ( figs [ f2]c and [ sf4 ] ) genes results in less ambiguity and more precision ; we can follow this sharpening of the distribution by the increasing height of its peak , since narrower distributions have higher peaks to maintain normalization , or by estimating the median width of the distribution @xmath22 ( fig [ sf11]d ) . finally , with all four genes , the distribution @xmath22 is approximately gaussian , with a width @xmath23 , for nearly all points @xmath1 . it has been conventional to think about a single morphogen gradient coding for position @xcite , or about `` expression domains '' of single gap genes pointing to large regions of the embryo . figure [ f2 ] shows how multiple expression levels combine , quantitatively , to synthesize an unambiguous code for position that reaches extraordinary precision . this one percent precision in the encoding of position is also the precision with which subsequent developmental markers , including the pair - rule gene stripes and the cephalic furrow , are generated @xcite . the fact that the four gap genes carry precise , unambiguous information about position does not mean that the embryo uses this information . to test whether this is the case , we perturbed the maternal signals bicoid ( _ bcd _ ) , nanos ( _ nos _ ) , and torso - like ( _ tsl _ ) , which strongly affect the output of the gap gene network ( fig [ sf1 ] ) . if our optimal decoding strategy is used by the embryo , our decoder should generate meaningful position estimates in the mutants . in particular , we will use the stripes of pair - rule gene expression as a measure of the embryo s readout of positional information , and compare this with our own . we have analyzed embryos from lines in which we delete _ bcd _ , _ tsl _ , and _ osk _ ( which controls the localization of the _ nos _ signal ) , singly and in pairs . for each of these six possibilities , we have measured expression levels for all four gap genes simultaneously , in samples that also include wild - type embryos ; results are summarized in fig [ sf1 ] . in every case , we construct the dictionary for decoding gap gene expression levels from the wild - type embryos , and then apply this dictionary to the mutants measured in the same batch , avoiding any concerns about systematic variations in staining , imaging , etc . across batches . the results of these analyses are a series of decoding maps , shown in fig [ f3 ] , which should be compared to the map for wild - type embryos in fig [ f2]d . many features of the decoding maps in fig [ f3 ] are expected from previous , qualitative characterizations of these mutants . thus , when we delete _ tsl _ the distortions are largely at the two ends of the embryo ( fig [ f3]a ) , since expression of _ tsl _ is confined to the poles ; when we delete _ bcd _ there are major distortions in the anterior portion of the map ( fig [ f3]b ) , where the concentration of bcd protein is highest ; and when we delete _ osk _ ( fig [ f3]c ) , we see major distortions in the posterior , consistent with _ nos _ being a posterior determinant . more striking are the quantitative predictions that we obtain by decoding . when we delete _ bcd _ , quantitative distortions of the map extend into the posterior half of the embryo , so that , for example , the expression levels found at @xmath24 generate a distribution @xmath25 in which the most likely value is @xmath26 , and at @xmath27 the most likely estimate is @xmath28 . thus , even in the posterior half of the embryo , the map is shifted , and the plot of @xmath5 vs @xmath1 ( following the ridge of high probability in the map ) does not have unit slope . but in the wild - type embryo , the positions @xmath29 and @xmath30 are associated with peaks in the stripes of expression for the pair - rule gene _ , as shown at left in fig [ f3 ] . if the machinery for interpreting gap gene expression is using the same dictionary that we have constructed here , then we predict that the _ bcd _ deletion mutants should shift these _ eve _ stripes to @xmath31 and @xmath32 , and this is what we see ( fig [ f3]b ) . even more dramatically , expression levels at @xmath33 in the _ bcd _ mutant are decoded as @xmath26 with high probability , and correspondingly there is an _ eve _ expression stripe at this anomalously anterior location . this is predicted to be not a displacement of the first ( nearest ) _ eve _ stripe , but rather a duplication of the seventh stripe , which is consistent with classical observations on cuticle morphology in these mutants @xcite , and with recent rnai / reporter experiments @xcite . the quantitative agreement between the decoding maps and the locations of the _ eve _ stripes extends to all six examples of single and double maternal mutants shown in fig [ f3 ] . notably , there is good agreement both when the shifts are small , as with the deletion of _ tsl _ ( fig [ f3]a ) , and when the shifts are much larger , resulting in the deletion of several stripes , as with the _ bcd osk _ and _ osk tsl _ double mutants ( figs [ f3]d and e ) . in cases where the implied position of a stripe crosses a diffuse band of probability density in the map , as with _ eve _ stripe iii in the _ osk tsl _ mutant ( fig [ f3]e ) , we might expect that there would be expression of _ eve _ but not a sharp stripe , and this is what we see ; similar effects occur in the anterior of the _ bcd tsl _ mutant ( fig [ f3]f ) . for simplicity fig [ f3 ] shows decoding maps that are averaged over all embryos for each mutant line . if we focus on decoding maps for individual embryos instead , their variability predicts the embryo - to - embryo variability in pair - rule gene expression . in particular , for _ bcd tsl _ mutants the positions that map to the wild - type locations of _ eve _ stripes four and five ( @xmath34 and @xmath35 ) vary substantially in the window @xmath36 . if we look at the _ eve _ expression patterns in individual embryos ( thin lines at bottom of fig [ f3]f ; for detailed analysis see fig [ sf14 ] ) , we see two peaks with variable positions , as predicted . for the _ bcd _ mutant , the average decoding map again has density at @xmath34 and @xmath35 ( fig [ f3]b and fig [ sf13 ] ) , but when we decode the gap gene expression patterns from individual mutant embryos we find that these features vary not only in their position but even in their presence or absence , so that individual embryos are predicted to have a variable number of _ eve _ stripes , and this is what we see . we have repeated the analysis of fig [ f3 ] , comparing our decoding maps with the expression profiles of the pair - rule genes _ paired _ ( _ prd _ ) and _ runt _ ( _ run _ ) , as summarized in figs [ sf6 ] and [ sf7 ] . in most cases we see relatively few stripes , which agrees with the theory since there are few intersections between expected stripe positions and the ridge of maximal probability . in the case of the _ tsl _ mutant , however , the three different pair - rule stripes provide a rather dense collection of markers through the range @xmath37 , as shown in detail in fig [ f3]a . these markers trace the predicted ridge of maximal probability with very high accuracy . the results for all the _ eve _ , _ run _ , and _ prd _ stripes in all six mutants are summarized in fig [ summary ] . for almost all the 57 stripe positions shown , the predicted position agrees with the measured position within the error bar defined by the standard deviation of these positions across embryos in our sample . more subtly , when we decode the positional signals from multiple embryos , we find variations in the resulting maps , as noted above , and the standard deviation of predicted stripe positions are in most cases close to the observed variations ( vertical and horizontal error bars , respectively ) . we do note a small number of errors in our predictions . in the _ osk _ mutants we observe a posterior eve stripe where none is predicted ( fig [ f3]c ) , and in _ bcd osk _ mutants , a variable number of prd stripes ( fig [ sf6]d ) . we predict a run stripe at @xmath38 where none is observed ( fig [ sf7]c ) ; we have no prediction for the very blurred band of run expression at @xmath39 ( fig [ sf7]c ) . in addition , in the _ bcd , tsl _ mutant we predict a run stripe at @xmath40 where none is observed ( fig [ sf7]f ) . this last failure occurs at a rare point where the combinations of gap gene expression are outside the range that we have sampled in the wild - type embryos ( fig [ sf5 ] ) , and thus we may be simply extrapolating the probability distributions too far . the errors in the _ osk _ mutants cluster around a point where the predicted maps have a discontinuity , and thus are especially sensitive to our assumption that decoding is done in a purely local fashion . a small amount of spatial averaging , perhaps to reduce noise introduced in the expression of the pair - rule genes , would result in very different predictions . the expression levels of genes carry information about signals that matter in the life of the organism . we have tried to answer several questions : how much information is available @xcite ? what features of the expression pattern are crucial in conveying this information ? how can the organism `` read out '' what is relevant ? in the developing embryo , one biologically relevant signal is position . in trying to understand how positional information is represented , we have focused on gene expression levels at a single moment in time , and on the levels observable by a single cell . it is possible that the cells could extract more reliable information by averaging expression levels over time , although since protein concentrations accumulate the snapshot that we consider already reflects some degree of temporal averaging , and further averaging may be of limited impact . similarly , individual cells might be able to extract more information by communicating with their neighbors @xcite , but we know that fluctuations in gap gene expression are correlated over substantial distances @xcite , limiting the benefits of such spatial averaging . importantly , once we focus on the information available locally in space and time , our experimental tools make it possible to give an essentially complete characterization of the available information and the nature of its encoding , as summarized by the probability distributions @xmath2 . positional information encoded by the gap gene expression levels is effectively decoded by a broad array of molecular mechanisms , in particular the complex rules for activation of the multiple enhancers that control the expression of pair - rule genes @xcite . one might think that connecting the patterns of gap expression to the positioning of pair - rule stripes would require us to make a model of these mechanisms . we have taken an alternative view , in which we assume that the embryo makes optimal use of the available information . once we make this hypothesis , decoding becomes a well defined mathematical problem , with no free parameters . the result is a dictionary that translates the expression levels of gap genes into inferred positions , and applying this dictionary to the expression profiles from a single embryo generates a map of inferred position vs actual position . for wild - type embryos , once we include all four gap genes this map is precise and nearly unambiguous , indicating positions along the anterior posterior axis with an accuracy of @xmath0 , sufficient to distinguish each cell from its neighbors @xcite . for mutant embryos , we find maps that are distorted , imprecise , and in some cases ambiguous . using these maps to predict the position of pair - rule expression stripes gives results that are in excellent agreement with experiment , both qualitatively and quantitatively . the strongly supports the hypothesis of optimality on which our decoding scheme is based . we draw attention to several aspects of the connection between theory and experiment . first , we decode positions based on graded expression levels of the gap genes @xcite . if we think of the gap genes as forming expression domains that are on or off , then maps are ambiguous even in wild - type embryos ( fig [ sf11 ] ) , and we certainly could not make meaningful predictions for stripe positions in the mutants . second , we analyze gene expression in absolute units ; the only normalization is global , which is equivalent to choosing units in which to measure concentration . contrary to this approach , it often is thought that biological functions must be robust against presumably uncontrollable variations in absolute concentration . in fact we find that significant components of the difference , for example , between the wild - type embryo and the _ tsl _ mutant are @xmath41 differences in the absolute concentrations of gap gene products ( fig [ sf10 ] ) , and these differences propagate through our decoding to generate distorted maps that correctly predict the shifts in pair - rule stripes : quantitative variations in absolute expression have precise functional consequences . third , we find that maps in the mutants are sufficiently variable that they predict variability in the _ number _ of pair - rule gene stripes , not just their locations ; such stripe number variations essentially never occur in wild - type embryos . one might have thought that reproducibility of stripe numbers emerges only as a property of the network of interactions among the different pair - rule genes , and even among neighboring cells . instead , we see that stripe numbers can be reproducible in wild - type embryos because the positional signal carried by gap gene expression levels is extremely precise , with so little noise that a direct local readout of these signals generates near zero probability of dropping or adding a stripe . deleting one or two of the maternal inputs to the gap gene network pushes the system into a regime where noise levels are higher , and applying our dictionary to these noisier expression profiles correctly predicts the occurrence of stripe variability . this is consistent with the idea that the naturally occurring inputs are matched to the signal and noise characteristics of the network @xcite . in many ways our maps of implied position as a function of actual position provide a quantitative , probabilistic version of the older idea that one can plot cell fate vs position a fate map even in mutants ; see , for example , ref @xcite . in its original form , this notion of a map depends on the fact that what we see in the mutant are rearrangements , deletions , and duplications , but no new pattern elements . it usually is assumed that this arises from canalization @xcite : although the early stages of pattern formation must generate new and different signals in response to the mutation , subsequent stages of processing force these signals back into a limited set of possibilities set . what we see here is that even rather early signals , those which are responding immediately to the primary maternal inputs , can be decoded to recapitulate the patterns seen in the wild - type . there is no need for subsequent steps to drive the pattern back to something built from wild - type elements , since it already is in this form . the theoretical framework we have developed here makes additional , testable predictions . simultaneous measurements of pair - rule expression with all of the gap genes would allow us to test directly whether , for example , the predicted variations in stripe number are correct , embryo by embryo , rather than just in aggregate . more subtly , since there are spatial correlations in the fluctuations of gap gene expression levels @xcite , our decoding predicts that there should be correlations in the small positional errors that occur even in wild - type embryos , and hence the fluctuations in position of the pair - rule stripes must also be correlated . these correlations should be different in the mutants , in ways that can be predicted quantitatively . most fundamentally , the molecular mechanisms that lead from gap gene product concentrations to pair - rule expression must , effectively , implement the dictionary that we have developed . thus , we should be able to predict the functional logic of these developmental enhancers by asking that they provide an optimal decoding of positional information , rather than fitting to data . perhaps the most important qualitative conclusion from our results is that precision matters . we are struck by the ability of embryo to generate a body plan that is reproducible on the scale of single cells , corresponding to positional variations @xmath0 of the length of the egg . as with other examples of extreme precision in biological function , from molecule counting in bacterial chemotaxis to photon counting in human vision , we suspect that this developmental precision is a fundamental observation , and to the extent that precision approaches basic physical limits it can even provide the starting point for a theory of how the system works @xcite . but precision in the final result of development could arise from many paths . we have a theoretical framework that suggests how such precision could arise from the very earliest stages in the control of gene expression , if this control itself is very precise , and this has motivated experiments to measure gene expression levels with correspondingly high experimental precision . what we have done here is to bring theory and experiment together , providing a parameter - free prediction of how quantitative variations in gap gene expression levels should influence the developmental process on the hypothesis that the embryo makes optimal use of the available information , in effect maximizing precision at every step . genetics then gives us a powerful tool to test these predictions , manipulating maternal inputs and observing pair - rule outputs . these rich data are in detailed agreement with theory , providing strong support for the precisionist view . we thank jo dubuis and r samanta for help with the experiments , and m biggin and n patel for sharing antibodies used in the pair - rule gene measurements . this work was supported , in part , by us national science foundation grants phy1305525 and ccf0939370 ; by national institutes of health grants p50gm071508 , r01gm077599 , and r01gm097275 , by austrian science fund grant fwf p28844 , and by hhmi ( to efw and through an international fellowship to mdp ) . 99 f rieke , d warland , r de ruyter van steveninck , and w bialek , _ spikes : exploring the neural code . _ ( mit press , cambridge , 1997 ) . ng hatsopolous and jp donoghue , the science of neural interface systems . _ annu rev neurosci _ * 32 , * 249266 ( 2009 ) . o marre , v botella soler , kd simmons , t mora , g tkaik , and mj berry ii , high accuracy decoding of dynamical motion from a large retinal population . _ plos comput biol _ * 11 , * e1004304 ( 2015 ) . c nsslein vollhard and ef wieschaus , mutations affecting segment number and polarity in _ drosophila_. _ nature _ * 287 , * 795801 ( 1980 ) . jp gergen , d coulter , and ef wieschaus , segmental pattern and blastoderm cell identities . in _ gametogenesis and the early embryo _ , jg gall , ed , pp 195220 ( liss , new york , 1986 ) . pa lawrence , _ the making of a fly : the genetics of animal design _ ( blackwell scientific , oxford , 1992 ) . j jaeger , the gap gene network . _ cell mol life sci _ * 68 , * 243274 ( 2011 ) . j briscoe and s small , morphogen rules : design principles of gradient - mediated embryo patterning . _ development _ * 142 , * 39964009 ( 2015 ) . t gregor , dw tank , ef wieschaus , and w bialek , probing the limits to positional information . _ cell _ * 130 , * 153164 ( 2007 ) . md petkova , sc little , f liu , and t gregor , maternal origins of developmental reproducibility . _ curr biol _ * 24 , * 12831288 ( 2014 ) . g tkaik , cg callan jr , and w bialek , information flow and optimization in transcriptional regulation . _ proc natl acad sci ( usa ) _ * 105 , * 1226512270 ( 2008 ) . g tkaik , am walczak , and w bialek , optimizing information flow in small genetic networks . _ phys rev e _ * 80 , * 031920 ( 2009 ) . am walczak , g tkaik , and w bialek , optimizing information flow in small genetic networks . ii : feed forward interaction . _ phys rev e _ * 81 , * 041905 ( 2010 ) . g tkaik , am walczak , and w bialek , optimizing information flow in small genetic networks . iii . a self interacting gene . _ phys rev e _ * 85 , * 041903 ( 2012 ) . tr sokolowski and g tkaik , optimizing information flow in small genetic networks . iv . spatial coupling . _ phys rev e _ * 91 , * 062710 ( 2015 ) . tr sokolowski , am walczak , w bialek , and g tkaik , extending the dynamic range of transcription factor action by translational regulation . _ phys rev e _ * 93 , * 022404 ( 2016 ) . jo dubuis , g tkaik , ef wieschaus , t gregor , and w bialek , positional information , in bits . _ proc natl acad sci ( usa ) _ * 110 , * 1630116308 ( 2013 ) . g tkaik , jo dubuis , md petkova , and t gregor , positional information , positional error , and read out precision in morphogenesis : a mathematical framework . _ genetics _ * 199 , * 3959 ( 2015 ) . s small , r kraut , t hoey , r warrior , and m levine , transcriptional regulation of a pair - rule stripe in _ drosophila_. _ genes dev _ * 5 , * 827839 ( 1991 ) . m levine , transcriptional enhancers in animal development and evolution . _ curr biol _ * 20 , * r754r763 ( 2010 ) . g tkaik , cg callan jr , and w bialek , information flow and optimization in transcriptional regulation . _ proc natl acad sci ( usa ) _ * 105 , * 1226512270 ( 2008 ) . jo dubuis , r samanta , and t gregor , accurate measurements of dynamics and reproducibility in small genetic networks . _ mol sys biol _ * 9 , * 639 ( 2013 ) . l wolpert , positional information and the spatial pattern of cellular differentiation . _ j theor biol _ * 25 , * 147 ( 1969 ) . w driever and c nsslein - volhard , the bicoid protein determines position in the _ drosophila _ embryo in a concentration - dependent manner . _ cell _ * 54 , * 95104 ( 1988 ) . mv staller , cc fowlkes , mdj bragdon , z wunderlich , j estrada , and ah depace , a gene expression atlas of a _ bicoid_depleted _ drosophila _ embryo reveals early canalization of cell fate . _ development _ * 142 , * 587596 ( 2015 ) . t erdmann , m howard , and pr ten wolde , role of spatial averaging in the precision of gene expression patterns . _ phys rev lett _ * 103 , * 258101 ( 2009 ) . d krotov , jo dubuis , t gregor , and w bialek , morphogenesis at criticality ? _ proc natl acad sci ( usa ) _ * 111 , * 36833688 ( 2014 ) . u gaul and h jackle , analysis of maternal effect mutant combinations elucidates regulation and function of the overlap of _ hunchback _ and _ krppel _ gene expression in the _ drosophila _ blastoderm embryo . _ development _ * 107 , * 651662 ( 1989 ) . t schpbach and e wieschaus , maternal - effect mutations altering the anterior - posterior pattern of the _ drosophila _ embryo . _ roux s arch dev biol _ * 195 , * 302317 ( 1986 ) . ch waddington , canalization of development and the inheritance of acquired characters . _ nature _ * 150 , * 563565 ( 1942 ) . ch waddington , canalization of development and genetic assimilation of acquired characters . _ nature _ * 183 , * 16541655 ( 1959 ) . w bialek , _ biophysics : searching for principles _ ( princeton university press , princeton nj , 2012 ) . c wang , lk dickinson , and r lehmann , genetics of nanos localization in _ drosophila_. _ dev dyn _ * 199*(2 ) , 103115 ( 1994 ) . embryos lacking single maternal patterning systems were obtained from females homozygous for @xmath42 , @xmath43 or @xmath44 . for embryos with positional information only from the osk patterning system , we used females homozygous for @xmath42 @xmath44 . for embryos with only information from bcd , we used _ p[_w+ bcd - gfp _ ] ; @xmath45 @xmath43 @xmath44 females . p[_w+ bcd - gfp _ ] encodes a fully rescuing gfp tagged bcd protein @xcite . to obtain embryos with input only from the torso patterning system , we used @xmath45 @xmath43 females for gap gene measurements and @xmath42 @xmath46 females for pair - rule embryos . the segmentation phenotypes of @xmath47and @xmath46 are equivalent @xcite . embryos lacking all maternal patterning systems were obtained from triply mutant @xmath42 @xmath46 @xmath44 females . all stocks were balanced with tm3 , sb . gap protein levels were measured as described by dubuis et al . @xcite . to quantify mutant gap protein levels in units of wild - type protein levels , mutants and wild - type embryos were stained together , and imaged alongside on the same microscope slide in a single acquisition cycle . we draw attention to the discussion of experimental errors in ref @xcite , because this is especially important for our analysis . as described there , we focus on a narrow time window , 38@xmath2148 min into nuclear cycle 14 . expression levels were normalized such that the mean expression levels of wild - type embryos ranged between 0 ( assigned to the minimal value across the ap axis of the mean spatial profile , separately for each gap gene ) and 1 ( similarly assigned to the maximal value across the ap axis ) . in detail , gene expression profile @xmath48 of any embryo @xmath15 was calculated as : @xmath49 where @xmath50 and @xmath51 are the lowest and highest raw fluorescence intensity values of the mean wild - type embryo fluorescence profiles ; @xmath52 is the raw fluorescence profile of the particular embryo , which can be either mutant or wild - type . note that this normalization simply assigns a conventional unit of measurement to gap gene concentrations ; no per - embryo profile `` alignment '' is used to reduce embryo - to - embryo variance . furthermore , mutant embryos were normalized to their wild - type reference for each gap gene , so absolute changes in gap gene concentrations not only changes in the shape of the gap gene spatial profiles were retained in all analyses . a summary of results on the mutant gap gene expression profiles ( mean @xmath53 sd across embryos ) is given in fig [ sf1 ] . to image pair - rule proteins , we used guinea pig anti - runt , and rabbit anti - eve ( gift from mark biggin ) polyclonal antibodies , and monoclonal mouse anti - pax3/7(dp312 ) antibody ( gift from nipam patel ) . secondary antibodies are , respectively , conjugated with alexa-594 ( guinea pig ) , alexa-568 ( rabbit ) , and alexa-647 ( mouse ) from invitrogen , grand island , ny . embryo fixation , antibody staining , imaging and profile extraction were performed as described by dubuis et al . @xcite . our goal was to predict features of pair - rule protein concentration profiles , such as the locations of expression peaks , for which comparisons between wild - type and mutant expression levels of pair - rule genes were not essential . pair - rule protein profiles were measured in mutant embryos in time widows of 45- to 55-min into nuclear cycle 14 ; for consistency with gap gene analyses and convenience we normalized such that the mean expression levels for each gene in each batch of embryos ranged between 0 and 1 ; individual profiles were scaled to minimize @xmath54 , as in refs @xcite . as an exception , we report pair - rule expression levels in triple maternal mutants ( _ bcd nos tsl _ ) in wild - type units , because the pair - rule genes are expressed uniformly and therefore lack positional features . to construct decoding maps and subsequently predict pair - rule expression stripes , eqs ( [ decode1gene ] ) and ( [ map1 ] ) require us to estimate the distribution of gap gene expression levels at each position , @xmath2 , from data . direct sampling is unfeasible even for just a single gene . instead , we approximated the embryo - to - embryo fluctuations in gene expression as gaussian with mean and ( co)variance that vary with position . in previous work we tested this approximation ; while we can see deviations from gaussianity @xcite , the gaussian approximation gives very accurate estimates of the positional information carried by the expression levels of individual genes @xcite , which is most relevant for the decoding that we attempt here . the gaussian approximation for a single gene corresponds to @xmath55 where @xmath56 measures the similarity of the gene expression level to the mean , @xmath57 , at position @xmath1 , @xmath58 and @xmath59 is the standard deviation in expression levels at point @xmath1 ( e.g. , shown by the shading in fig [ sf1 ] ) . given measurements of gene expression vs position in a large set of embryos , we can compute the mean and variance in the standard way , so that eqs ( [ s1gauss]-[s2gauss ] ) can be applied directly to the data . the generalization of the gaussian approximation to the case where coding and decoding are based on a combination of @xmath60 genes simultaneously is given by @xmath61}}e^{- \chi_k^2\left ( \{g_{\rm i } \ } , x \right)/2 } \label{mapk}\ ] ] where @xmath62 measures the similarity of the gene expression pattern to the mean _ pattern _ expected at @xmath1 , @xmath63 and @xmath64 is the covariance matrix of fluctuations in the expression of the different genes at point @xmath1 . figures [ sf12]a - f shows the estimation of @xmath65 covariance matrix of gap gene fluctuations across embryos , @xmath66 ( where @xmath67 denotes averaging over embryos @xmath15 ) . note that the covariance matrix , as well as the mean profiles themselves , are a function of position along the ap axis . comparing the covariance matrix estimates across replicates of wild - type datasets , fig [ sf12]g shows that our data are sufficient for us to have control over estimation errors , so that eqs ( [ mapk]-[chisq ] ) can likewise be applied directly to the data . figure [ sf0 ] shows a step - by - step procedure for constructing a decoding dictionary " based on a single gap gene , kr , from measured data , and a decoding map " for a single wild - type embryo ; the decoding map in fig [ f2]a is an average over 38 such individual decoding maps . similarly , top panels of fig [ sf2]a - d show the profiles of all four individual gap genes in the wild - type embryos , while the bottom panels show the corresponding decoding maps . as with the case of krppel in fig [ sf0 ] , all of these maps show substantial ambiguities , where the signal at one point in the embryo is consistent with a wide range of possible positions . ambiguity arises whenever a vertical slice through these density plots encounters multiple peaks , but in the case of decoding based on single genes these ambiguities are so common that they result in either vast swaths of grey or in intricate folded patterns . in particular locations specifically , at the flanks of mean expression profiles where the slope of the profile is high the distributions @xmath68 become highly concentrated , indicating that the quantitative expression levels of individual genes provide the ingredients for precise inferences of position , as suggested in refs @xcite . figure [ sf2 ] shows that combining two genes always reduces ambiguity relative to the single gene case , but does not eliminate it entirely , and a similar trend is observed in fig [ sf3 ] with triplets of gap genes . the four - gene case shown in fig [ f2]c sharpens the decoding maps further ( cf . the scale of distributions @xmath68 ) , achieving a low positional error of @xmath69 across the majority of locations along the ap axis . we can quantify this sharpening by computing the standard deviation of the distribution @xmath68 , and then finding the median over @xmath1 ; a summary of these results is given in fig [ sf11]d . figure [ sf11 ] further shows that the traditional interpretation of gap genes as generating binary domains of expression separated by sharp boundaries significantly blurs the decoding map , irrespective of whether the gap gene thresholds are selected simply at the midpoint of the expression range ( at @xmath70 ) or are adjusted separately for each gap gene to optimize the decoding map . this shows that quantitative , graded levels of gap gene expression are essential for the precise specification of position . it is worth emphasizing that the notion of a threshold that determines gene expression ( or cell fate ) boundaries , which is well defined for a single signal being thresholded , is more ambiguous in the case of multiple concurrent signals that drive patterning , as is the case here with gap genes . the idea of putting independent , and possibly different , thresholds on each of the inputs separately may appear as a natural extension of a single - gene case , but it is important to realize that this idea already entails a drastic ( and untested ) independence assumption . it would be equally possible that the relevant patterning thresholds act on some unknown , even nonlinear , combination of the four gap gene signals . in particular , in biophysical models of enhancer function where the gene expression is controlled by the concentrations of multiple inputs ( and where the threshold is determined by the sigmoid activation function of the enhancer ) , the interpretation of thresholds applying to nonlinear combinations of inputs is more realistic than the interpretation of different thresholds independently applying to each of the inputs . furthermore , the picture of independent thresholds acting on individual gap genes leaves completely unanswered the question of how binarized gap gene profiles can be read out in a biophysically realistic fashion to combinatorially drive the expression of their target genes . we analyzed patterns of gap gene expression in six mutant lines of flies , deficient in one or two of the three maternal inputs to the gap gene network , as summarized in fig [ sf1 ] . to construct decoding maps for the mutants , as in fig [ f3 ] , we first computed posterior distributions @xmath71 as prescribed by eq ( [ decode1gene ] ) from wild - type embryo data , and evaluated these distributions at gap gene expression levels measured in mutant embryos . but the wild - type expression levels fill only a very small region of the full four dimensional space of possibilities ; if the expression levels in mutant embryos fell largely outside this region , then we would be extrapolating too far from the wild - type measurements and could not make reliable inferences . to test whether this could be the case , we computed @xmath54 [ eq ( [ chisq ] ) ] between the observed combinations of expression levels and the mean expression levels expected at each position in the wild - type , and compared that to the @xmath54 values for mutant embryos . figure [ sf5 ] shows the cumulative distribution of @xmath54 across the entire population of wild - type embryos , from all six experiments . normalized per gene , the mean of @xmath54 is one , but the distribution has a tail extending to @xmath72 this value . to construct a comparable distribution for the mutants , we first note that the gene expression values at one point @xmath1 in the embryo can be decoded to a position @xmath73 that is very far from @xmath1 , as shown by fig [ f3 ] and explained in the corresponding discussion in the main text . consequently , in the mutants we looked for the point @xmath73 in the wild - type that achieved the minimum of @xmath74 over all possible @xmath73 ( which is the location that the mutant gap gene profiles decode to ) and then look at the cumulative distribution of @xmath54 at these decoded locations . as expected , @xmath54 values in the mutant are larger than in the wild - type , but there is a surprising degree of overlap between the two distributions : the largest value of @xmath54 that we observe in the wild - type embryos is larger than @xmath75 of the values that we see in the mutants . although mutant background induces huge changes in the inputs of the gap gene network and in the gap gene profiles themselves , the gap gene network responds in a way that is not so far outside the distribution of possible responses under natural conditions . this fact is what makes decoding positional information in the mutants feasible . decoding maps make parameter free predictions for the locations of positional markers in mutant embryos . to test these predictions , we compare to the locations of expression peaks for the pair - rule genes.if a cell at position @xmath1 in the mutant embryo has expression levels for the gap genes that lead to a high probability of inferring a position @xmath76 , where @xmath77 is the position of a pair - rule stripe in the wild - type , then we expect that there will be a peak in pair - rule gene expression at the point @xmath1 in the mutant . mathematically , this process ( shown graphically in fig [ f3 ] ) proceeds as follows : we construct @xmath78 and look at the line @xmath76 ; this gives us a ( non normalized ) density @xmath79 , and there should be pair - rule stripes at the local maxima of this density . because stripes in the wild - type are driven by different enhancers and are thus not identical , it is important that our calculation should predict the occurrence of a particular identified stripe @xmath80 at @xmath1 . the construction of the density @xmath81 is shown in fig [ sf15 ] for each stripe of eve , prd , and run , and for each individual wild - type embryo . there is an excellent correspondence between the average pair - rule gene expression profile and the set of individual embryo densities for all stripes ; the sole discrepancy appears to be a small ambiguity in the decoding map that hints at two weak `` echoes '' of pair - rule stripes 1 and 2 in the far anterior ( for @xmath82 ) , which we did not detect in the data . these may be missing because of influences from other gap genes that are active in the far anterior . for ease of visualization , we often look just at the _ average _ decoding map , as in fig [ f3 ] as well as figs [ sf6 ] and [ sf7 ] , below . this average , @xmath83 , which can be easily plotted as a single map , and then decoded analogously to the procedure outlined above : we looked for the position @xmath1 where the decoding map peaks if the inferred position @xmath5 were equal to a known pair - rule stripe location in the wild - type . decoding the `` mean pair - rule stripe position '' in this manner does not differ from decoding single embryos to predict the pair - rule stripe positions individually , and then taking the average prediction , but now we can also predict fluctuations in stripe locations , a fact we used in making fig [ f4 ] . decoding in individual embryos permits us to compare not only the predicted mean stripe locations to the measurements , but also to compare the variability in stripe position , shape , and in the total number of observed stripes . figures [ sf14 ] and [ sf13 ] show examples of individual eve profiles where some of the eve stripes 3 , 4 , 5 were either missing or had a broad , poorly localized `` diffuse '' profile in mutant backgrounds . we predicted these phenomena for exactly the same mutant backgrounds and the same stripes from individual embryo decoding maps . a detailed description of individual embryo pair - rule stripe predictions in mutant backgrounds , analogous to fig [ sf15 ] for the wild - type , is shown in fig [ sf16eve ] . in these panels , we denote separately diffuse stripes , as well as a small number of observed - but - not - predicted and predicted - but - unobserved stripes . all non - diffuse predictions across the three pair - rule genes and all mutants are shown in the summary fig [ f4 ] in the main text . we invested substantial experimental effort to measure gap gene expression levels in mutant embryos side - by - side with the wild - type controls , so that absolute concentrations can contribute to the decoding . but do they ? in fig [ sf10 ] we show the effect of the absolute level on the decoding map , and consequently on the pair - rule stripe prediction performance . in the _ bcd _ mutant background ( fig [ sf10]a , b ) , gap gene expression levels are strongly perturbed in shape but also suppressed in magnitude by @xmath84 . decoding these profiles gives predictions of pair - rule stripes that agree very closely with data ( fig [ sf10]c , black symbols ) . in contrast , when mutant profiles are individually normalized so that they span the range of expressions between 0 and 1in essence , keeping the profile shape but undoing the magnitude effect leads to much worse predictions of pair - rule stripes ( fig [ sf10]c , red ) . in the _ tsl _ mutant background , the effect of absolute concentrations is more subtle . in these mutants , kr and kni are overexpressed by @xmath85 relative to the wild - type , which leads to a slight deformation in the decoding map in the posterior ( @xmath86 ) , and this effect disappears if we normalize to keep only relative expression levels . while the effect is smaller than in the _ bcd _ background , pair - rule stripes at @xmath87 are consistently predicted better using absolute gap gene concentrations . in sum , both for large scale and precision effects on our pair - rule predictions , being able to measure gap gene concentrations relative to the wild - type is crucial . this suggests as well that the embryo itself responds to precisely determined , absolute concentrations of signaling molecules . when we delete all three of the maternal inputs , positional information is abolished ( fig [ flatline ] ) . there is a low uniform expression of eve in mutant embryos , consistent with the predictions from our decoding map .

the study of far - infrared ( fir ) emission from elliptical galaxies is particularly interesting because the internal rates that dust is formed and destroyed can both be estimated . dust is produced by mass loss from evolving red giant stars in the old stellar population typical of these galaxies . this mass loss is directly observed in the cumulative mid - infrared 1024@xmath1 m emission from circumstellar regions around giant stars , causing the surface brightness profiles at 1024@xmath1 m to resemble those at optical wavelengths ( knapp et al . 1989 ; athey et al . the dust destruction rate by sputtering can be estimated from the frequency of thermal impacts from ions in the hot interstellar gas which for many ellipticals can be estimated from x - ray observations . during their sputtering lifetimes , dust grains ejected from giant stars are heated by starlight and thermal electron impacts to temperatures that radiate in the fir ( e.g. tsai & mathews 1995 , 1996 ; temi et al . this radiation has been observed both by the _ infrared astronomical satellite , iras _ ( knapp et al . 1989 ) and by the _ infrared space observatory , iso _ ( temi et al . in addition , small optically obscuring clouds and disks of dusty gas have been detected in about half of the bright ellipticals studied with the _ hubble space telescope , hst _ ( van dokkum & franx 1995 ; ferrari et al . 1999 ; lauer et al . 2005 ) . these clouds probably formed from dusty gas ejected from stars within the central @xmath10kpc in these elliptical galaxies ( mathews & brighenti 2003 ) . galaxy mergers are another obvious possible source of dust in elliptical galaxies , indeed it is widely thought that ellipticals are formed by mergers . elliptical - looking galaxies can be produced after a few gyrs following the merger of two disk galaxies ( e.g. barnes & hernquist 1992 ) , but this simple recipe may not easily explain all structural and kinematic properties of these galaxies ( naab , jesseit & burkert 2006 ) . many of the mergers that produced the ripples , shells and counter - rotating cores in elliptical galaxies probably occurred a very long time ago . it is likely that massive boxy ellipticals , similar to many of those with strong fir emission , have experienced mostly `` dry '' gas and dust - free mergers during the last few gyrs ( bell et al . 2004 ; faber et al . nevertheless , mergers do occasionally occur between low - redshift ellipticals and gas - rich galaxies and these post - merger ellipticals often selectively appear in the observing lists of infrared observers . our previous study of fir emission from elliptical galaxies was based on a sample chosen from the _ iso _ data archive ( temi et al . 2004 ) that included a significant number of galaxies with abnormally large masses of dust and cold gas which are uncharacteristic of elliptical galaxies in general . the large range of fir luminosities that we found in the _ iso _ sample was therefore not unexpected , and it was reasonable to suppose that mergers with dusty galaxies explained the absence of correlations between optical and fir emission . the failure to find correlations between the fir and optical luminosities of elliptical galaxies has been used in the past to argue that the cold dust is acquired in a stochastic fashion by mergers ( forbes 1991 ) . to understand the relative role of internal and external sources of dust in elliptical galaxies , we describe here new _ spitzer _ observations at 24 , 70 and 160@xmath1 m . our full sample consists of 46 elliptical galaxies , formed by combining our own proprietary _ spitzer _ observations with those of other observers . the improved sensitivity and angular resolution of _ spitzer _ provides clear evidence that the 70@xmath1 m emission is spatially extended in many ellipticals . in particular , we find extended fir emission in elliptical galaxies that do not contain uncharacteristically large masses of neutral , molecular or optically visible cold gas . the diffuse nature of the dust in these otherwise normal galaxies , combined with the relatively short ( ion sputtering ) lifetime for dust destruction in the hot gas , favors an internal source of dust rather than dust - rich mergers which are probably too infrequent . the luminosity of extended fir dust emission in most ellipticals greatly exceeds that expected from normal mass loss from local stars . however , small , optically thick nuclear dust clouds observed in many elliptical galaxies may provide an alternative , more plausible source for the extended fir dust emission . consequently , we argue that dust accumulated in the central regions of elliptical galaxies from stellar sources in the core is transported out to larger radii by buoyantly rising gas intermittently heated by low - luminosity active galactic nuclei ( agn ) in the galactic cores . if this hypothesis can be confirmed in more detail , the presence of extended cold dust can be used to trace the radial circulation of hot gas in these galaxy / groups . such circulation flows are known to be consistent with x - ray observations in which very little hot gas cools to low temperatures ( mathews , brighenti & buote 2004 ) . for this study we construct a sample of 46 elliptical galaxies by combining far - infrared _ spitzer _ observations from our own _ spitzer _ data with additional galaxies in the public _ spitzer _ archive . our proprietary far - infrared _ spitzer _ observations of elliptical galaxies ( pid 20171 , pi p.temi ) consists of 11 non - peculiar , bright ( @xmath11 @xmath12 ) elliptical galaxies , most having high - quality x - ray observations . galactic proximity and spatial scale were also important factors in choosing this sample since high angular resolution is needed to separate emission from dusty cores and diffuse emission from truly interstellar dust spread over the galactic volume . furthermore , we avoided sources with strong radio sources and agn that have detectable signals in the far infrared . ellipticals with strong infrared milky way cirrus or bright spiral galaxies in the vicinity were also avoided . since one of our goals is to study the interaction between dust and hot gas , we included many galaxies for which the x - ray emission is well - studied by either _ chandra _ or _ xmm_. in addition to our original sample we selected 35 additional ellipticals from data currently available in the public _ spitzer _ archive targeted by several pis during their general observation in cycles 1 and 2 , guaranteed time observations , and the legacy program sings ( kennicutt et al . all the galaxies added to our original sample were chosen using selection criteria tailored to our specific scientific goals . thus , the 46 galaxies studied here form a rather homogeneous sample and is large enough to be statistically meaningful . tables 1 and 2 summarize the relevant properties of the sample galaxies including the far - infrared photometric measurements . optical data are taken from the leda catalog ( paturel et al . 1997 ) and the rsa catalog ( sandage & tamman 1981 ) , while x - ray data are from ellis and osullivan et al . luminosities and distances are calculated with @xmath13 km s@xmath14 mpc@xmath14 and are derived from the flux densities in table 1 using far - infrared ( fir ) passbands of @xmath15 5.3 , 19 and 34@xmath1 m for emission at 24 , 70 and 160@xmath1 m respectively . the pis of the original spitzer program from which the galaxies have been selected are shown in column two of table 1 . all galaxies are classified as ellipticals in the ned catalog , and the morphological type t is taken from the leda catalog . nevertheless , we shall find that our sample contains some s0 galaxies . most of the sample galaxies have known x - ray luminosities @xmath16 , which are a measure of the hot gas present in each galaxy . of the 46 galaxies in the sample three are dwarf elliptical galaxies ( ngc855 , ngc1377 , and ngc3265 ) with @xmath17 l@xmath18 . although they are listed in table 1 , they are not included in our discussion . the data presented here were acquired with the multiband imager photometer ( mips ) ( rieke et al . 2004 ) on the _ spitzer space telescope _ ( werner et al . 2004 ) in three wavebands centered at 24 , 70 and 160@xmath19 . since the galaxy sample has been constructed by collecting archival data from several spitzer observing programs , the image data were recorded using observing modes and integrations times tailored for their original scientific goals . as a consequence , although the maps represent the deepest fir images acquired to date , some data achieve sensitivities that are not always optimized for targets that are intrinsically faint at far - infrared wavelengths . for example , three of the brightest ellipticals in the virgo cluster ( ngc4472 , ngc4636 , and ngc4649 ) recorded under a guaranteed time program ( pid 69 , pi g. fazio ) reach a relatively low sensitivity of 0.5mjy / sr and 1.1mjy / sr ( 1@xmath20 ) at 70 and 160@xmath19 , respectively , while other galaxies ( i.e. pid 20171 , pi p. temi ) have deeper maps at a sensitivity level of only 0.12mjy / sr and 0.3mjy / sr for the same two wavebands . the sings data are recorded in mips scan mode , covering a very large sky area ( @xmath21 ) , incorporating two separated passes at each source location . sings images correspond to maps with intermediate sensitivity . details on the observing strategies , field coverage and integration times for the sings program are described by kennicutt et al . apart from the sings observations , data have been acquired in mips photometry mode , allowing appropriate coverage of the sources and their extended emission . at all wavelengths , sufficiently large field sizes were chosen to reliably measure the background emission . we used the basic calibrated data ( bcd ) products from the spitzer science pipeline ( version 13.2 ) to construct mosaic images for all objects . pipeline reduction and post - bcd processing using the mopex software package ( makovoz et al . 2006 ) provide all necessary steps to process individual frames : dark subtraction , flat - fielding , mux - bleed correction , flux calibration , correction of focal plane geometrical distortion , and cosmic ray rejection . foreground stars and background galaxies were present in the final mosaiced images at all bands . these were identified by eye and cross - checked using surveys at other wavelengths ( digital sky survey and 2mass ) . they were then masked out from each mips image before flux extraction was performed . also , a few asteroids were identified and removed from the affected maps . aperture photometry was performed by measuring fluxes in suitable circular apertures around the centers . flux densities were extracted from apertures extending out to the entire optical disk of each galaxy ( @xmath22 ) . sky subtraction was performed by averaging values from multiple apertures placed around the target , avoiding any overlap with the faint extended emission from the galaxy . uncertainties in flux densities due to sky subtraction are generally less than 1% , but can be of the order of tens of percent for faint sources . the measured flux densities are listed in table 1 . most of the galaxies in the sample have been detected by mips , and among them a large fraction show far - infrared emission well beyond the sensitivity capabilities of previous infrared space observatories . although most of the galaxies presented in this paper are unpublished either because they are our own proprietary data or because the analysis is still in progress by the original pis , results for a few galaxies are already present in the literature ( i.e. the sings galaxies ) . we find that our flux densities are in good agreement with previously published data . the only exception is the measured flux density at 160@xmath1 m reported by dale et al . ( 2005 ) for the sings galaxy ngc584 . very likely this discrepancy is due to a misidentification of the target in their very large 160@xmath1 m scan map . corrections for extended emission were applied to the fluxes as described in the spitzer observer s manual . photometric errors listed in table 1 refer only to the statistical uncertainties . total errors should include systematic uncertainties due to flux calibration and to sky subtraction errors . the uncertainties on the final absolute calibration are estimated at 10% for the 24@xmath1 m data , and 20% at 70 and 160@xmath19 . figure 1 shows gray - scale images of a representative sub - sample of galaxies . the column of panels on the left is the 24@xmath1 m data , while 70 and 160@xmath1 m data are shown in the center and right columns , respectively . galaxies are ordered by ascending ir flux in a common field of view of @xmath23 . some very faint galaxies are shown , including a non detection at 160@xmath1 m and few sources with extended far - infrared emission . detailed images of the diffuse extended emission in ngc4636 and ngc5044 will be presented in a companion paper currently in preparation . in figure 2 we compare fluxes at 24 , 70 and 160@xmath1 m with optical b - band fluxes and a related plot of fir and b - band luminosities is shown in figure 3 . both the 24@xmath1 m and b - band fluxes and luminosities are clearly correlated , @xmath24 and @xmath25 the approximate consistency with unit slope @xmath26 indicates that the emission at 24@xmath1 m , like that at mid - ir wavelengths ( knapp et al . 1989 ; athey et al . 2002 ) , is dominated by circumstellar emission that closely traces optical starlight . however , the huge scatter seen in the 70@xmath1m bband panels in figures 2 and 3 suggest that this colder dust is unassociated with stars and is truly interstellar . the absence of any clear correlation between fir and optical emission from ellipticals has been discussed by forbes ( 1991 ) , goudfrooij & de jong ( 1995 ) , trinchieri & goudfrooij ( 2002 ) and temi et al . ( 2004 ) . in principle cold , fir - emitting dust could either be diffusely distributed in the hot gas or confined to dense central clouds similar to those visible in optical absorption ( e.g. lauer et al . we discuss below our preference for the diffuse component in producing most of the fir - emission . it is remarkable that @xmath3 and @xmath4 range over more than two orders of magnitude for galaxies with similar @xmath2 this is a significant new result that must be explained . in our previous survey of archival data from _ infrared space observatory _ ( iso ) we found a similar spread in fir luminosities at 60 - 200@xmath1 m , but the fir luminosities in the iso data spanned only one order of magnitude for fixed @xmath2 ( temi et al . 2004 ) . at the time we attributed much of this scatter to the unusual nature of the iso sample which contained many ( post - merger ) elliptical galaxies with massive disks of cold , dusty gas . now with this _ spitzer _ sample of ( mostly ) normal elliptical galaxies we find that the range of fir luminosities is even larger , although we show in appendix a that some of the fir - brightest sample galaxies can not be regarded as normal . in figure 4 we show that there is no obvious trend of dust color @xmath27 ( i.e. mean dust temperature ) with @xmath3 except perhaps a tendency for cooler dust in more fir - luminous galaxies . a similar conclusion results if we replace @xmath28 on the horizontal axis with @xmath29 , @xmath3 or @xmath4 . figures 5 and 6 show no discernable correlation of either @xmath6 or @xmath7 with stellar balmer line ages or metallicity . however , @xmath30 may show a slight tendency to decrease with age @xmath31 or increase with metallicity @xmath32.\ ] ] a negative correlation of @xmath30 with age would be expected since the rate of mass loss from evolving stars decreases with stellar age ( piovan et al . 2003 ) , although this expected trend is not supported by our study of elliptical galaxies at mid - ir wavelengths ( temi et al . 2005 ) . it must be noted however that balmer line ages for the same galaxy often show considerable variation when determined by different authors , so the anticorrelation of @xmath30 with age needs further confirmation . a positive correlation of @xmath30 with metallicity in figure 6 is expected if more interstellar grains are produced by mass - losing stars of higher metallicity . however , this is inconsistent with the negative correlation at 70@xmath1 m in figure 6 , @xmath33.\ ] ] no correlations with x - ray luminosity with any of the fir luminosities are visible in figure 7 . decreasing fir luminosities with increasing hot gas density might be expected since interstellar grains are destroyed by sputtering which is proportional to the density of the hot gas . however , the relationship between @xmath16 and gas density is unclear . for example , there is no correlation between @xmath16 and the hot gas density at @xmath34 kpc in the elliptical galaxies observed with _ chandra _ by humphrey et al . this disconnect between gas density and @xmath16 may arise because the hot gas density on kpc - galactic scales is influenced by intermittent energy supplied by the central black holes , while @xmath16 is largely determined by more extensive gas that fills the dark halos that surround the elliptical galaxies ( mathews et al . consequently , the absence of correlations in figure 7 is not surprising . finally , we note that all qualitative features in figure 7 are preserved if luminosities are replaced with fluxes . in figure 8 we compare fir fluxes with optical emission line fluxes determined by three different observers : goudfrooij et al . ( 1994a , b ) , macchetto et al . ( 1996 ) and sarzi et al . the optical observations require difficult background subtraction and the procedures used vary among the observers . for this reason , we compare fir fluxes in figure 8 with each set of optical emission line fluxes independently . we believe that both @xmath35 and @xmath36 correlate weakly with observed optical line fluxes , but that @xmath37 does not . the recent sauron data for @xmath38 ( sarzi et al . 2006 ) show the tightest relationship . the goudfrooij et al . data ( left column of panels in figure 8) are marginally correlated , but the significance increases if it is combined with the macchetto et al . data ( shown in the central column ) . in figure 9 we compare the fluxes in the optical emission lines with b - band fluxes , using only data for non - peculiar e or e / s0 galaxies , excluding peculiar and known s0 galaxies . the top and bottom panels from macchetto et al . ( 1996 ) and sarzi et al . ( 2006 ) show that the line flux is weakly correlated with the stellar continuum ( see also phillips et al . 1986 ; macchetto et al . 1996 ; goudfrooij 1999 ) . a correlation between extended line emission and the optical continuum suggests that the warm gas with @xmath39 k has a stellar origin . but the quality of these difficult optical observations is poor . there are very few objects in common among the goudfrooij , macchetto and sarzi data sets , but the flux - flux plots of these common galaxies show large scatter with almost no clear monotonic relationship . five of the six galaxies in common between goudfrooij and macchetto correlate _ negatively _ in a flux - flux plot . the correlation in figure 9 , if real , may simply indicate that the warm gas and the dust associated with it both have a stellar origin more stars of a similar age and metallicity eject more gas . however , the optical emission line flux depends sensitively on the density of the warm gas and therefore on the pressure in the local hot interstellar gas . at present there are too few accurate observations to interpret figure 9 with confidence . all ellipticals in our sample show evidence of extended emission at 24@xmath1 m similar to that of the optical stellar images details of this will be discussed in a separate publication . however , spatial extension is also evident at 70@xmath1 m as shown in table 1 , but the emission at 160@xmath1 m is not spatially resolved due to decreasing resolution at larger wavelengths . figure 10 shows the surface brightness at 70@xmath1 m for eight galaxies together with the point response function ( prf ) for the mips detector ( including the first airy ring ) . all contours in figure 10 are arbitrarily normalized to 50 at the center . the ( logarithmic ) extension in arseconds beyond the prf is best measured at 0.08 of the central fluxes , shown with arrows in figure 10 . the full width at 4/50 = 0.08 maximum ( subsequently referred to as fw0.1 m ; last two columns in table 1 ) of the prf is 37@xmath40 . twelve galaxies in our full sample have fw0.1 m greater than 40@xmath40 , indicating that these galaxies are spatially resolved at 70@xmath1 m . when fw0.1m/2@xmath41 is plotted against @xmath27 , there is no clear evidence that galaxies with more extended @xmath28 have colder dust . if @xmath50 be the number of grains @xmath51 with radius between @xmath48 and @xmath52 at galactic radius @xmath53 at time @xmath54 , the grain population evolves according to @xmath55 = s(r , a),\ ] ] where @xmath56 is the rate at which grains with radii @xmath48 are expelled from stars per @xmath57 ( tsai & mathews 1995 ) . we assume @xmath58 for @xmath59 , with @xmath60 ( mathis et al . 1977 ) , so the coefficient @xmath61 depends on the rate that the old stellar population ejects mass @xmath62 s@xmath14 ( mathews , 1989 ) , the stellar density @xmath63 , the initial dust to gas mass ratio scaled to the stellar metal abundance in solar units @xmath64 , and the density of ( amorphous ) silicate grains , @xmath65 gm @xmath51 . the solution of eq . ( 7 ) is @xmath66 the grain temperature @xmath67 is determined by the balance of heating ( by both absorption of stellar radiation and by electron - grain collisions ) and radiative cooling : @xmath68 @xmath69 in view of the high electron - grain collision frequency , we neglect stochastic heating due to single impacts . we assume that all grains have properties similar to astronomical silicates ( laor & draine 1993 ) using the following approximations for the absorption coefficient @xmath70 , and for the planck - averaged values @xmath71 ( see draine & lee 1984 ) . the mean intensity of starlight is found by integrating over a de vaucouleurs stellar profile using the appropriate effective radius @xmath41 and stellar luminosity for each galaxy . the collisional heating term in the equation above includes a correction @xmath72 for small grains that do not completely stop the incident thermal electrons ( dwek 1986 ) and @xmath73 is an average over a maxwellian distribution . the electron density and temperature of the hot gas are calculated using deconvoluted profiles from x - ray observations . the fir emissivity at each wavelength is calculated from @xmath74 @xmath75 in figure 11 we compare the expected fir spectral energy distribution for three galaxies in our sample that have well - known hot gas density and temperature profiles from x - ray observations : ngc 1399 , 1404 and 4472 . none of these galaxies are spatially resolved at 70@xmath1 m . the x - ray data for these galaxies has been taken from the following sources : ngc 1404 & 1399 : ( paolillo et al 2002 ; jones et al . 1997 ) ; ngc 4472 : ( mathews & brighenti 2003 ; brighenti & mathews 1997 ) . the profiles in figure 11 show the expected fir spectral energy distributions computed with the equations above . the solid and dashed lines correspond respectively to emission from grains with original maximum sizes @xmath76 m and 1.0@xmath1 m . the observed fluxes for ngc 1399 and ngc 4472 at 70@xmath1 m and 160@xmath1 m are in excellent agreement with our simple model . the fit to ngc 1404 is also satisfactory , given the many assumptions in our model . while the flux at 24@xmath1 m is also shown in figure 11 , this circumstellar emission is not included in our model , but it supports our assumption that much of the diffuse interstellar dust in these three galaxies has a stellar origin . it is of particular interest to determine fir spectral energy distributions for galaxies that have accurate interstellar gas density and temperature profiles and are also spatially resolved at 70@xmath1 m . seven galaxies in table 1 have fw0.1 m greater than 50@xmath40 but only six of these are in the ellis and osullivan ( 2006 ) list of x - ray galaxies detected with rosat : ngc 2768 , 2974 , 4125 , 4636 , 4696 , and 5044 . of these six galaxies , only three ngc 4636 , 4696 and 5044 have x - ray luminosities ( @xmath77 ) large enough for an accurate determination of the gas density and temperature profiles . however , the optically peculiar elliptical ngc 4696 is the central galaxy in the centaurus cluster with powerful fri - type radio emission , a large mass of dusty neutral gas and highly disturbed hot cluster gas ( fabian et al . 2006 ) . in view of these complexities , ngc 4696 is an inappropriate galaxy to learn about the nature of fir emission from normal elliptical galaxies . we find that the observed fluxes for the two remaining fir - extended galaxies ngc 5044 and ngc 4636 are very much larger than predicted by our steady state model for the fir spectral energy distribution . we believe that the dust in these galaxies has an entirely different origin . the large and extended fir emission from ngc 5044 and ngc 4636 , which is related to the enormous range of fir luminosities in figure 3 , will be discussed in a separate forthcoming publication . the first important conclusion to be drawn from figures 2 and 3 is that warm circumstellar dust and colder interstellar dust behave very differently . the relatively small scatter in @xmath37 in figure 2 is slightly larger than our estimated observational errors . nevertheless , we have been unable to find correlations between stellar age or metallicity and the residuals in @xmath37 from the mean correlation . we conclude that the warm dust emitting at 24@xmath1 m is largely circumstellar and this is confirmed by the similarity of optical and 24@xmath1 m surface brightness profiles which will be discussed in detail elsewhere . the second and most remarkable result in figures 2 and 3 is that the emission from colder dust at 70 and 160@xmath1 m varies by a huge factor relative to that at 24@xmath1 m and optical wavelengths . in the previous section and in figure 11 we show that 70 and 160@xmath1 m fluxes from three galaxies having well - observed interstellar gas density profiles can be fit reasonably well with our simple dust heating model without varying any of the dust parameters from those we assumed at the outset . however , notice that all three galaxies in figure 11 that agree with our model sed ngc 1399 , 1404 and 4472 have values of @xmath3 and @xmath4 in table 2 that lie near the bottom of the observed distributions in figure 3 . we think it is impossible to explain the huge dispersion of @xmath3 and @xmath4 in figure 3 by simply varying the dust parameters in the sed models used for the three galaxies in figure 11 , i.e. , the stellar mass loss rate , dust composition , dust size distribution , etc . are unlikely to be in error by an order of magnitude or more . our sed model clearly fails to fit most of the fir observations . some of the large spread of @xmath3 and @xmath4 in figure 3 can be understood because our sample includes very fir - luminous elliptical galaxies that contain abnormal masses of interstellar dust and cold gas , probably acquired by a recent merger . in appendix a we review relevant observations of 12 sample galaxies having the highest 70@xmath1 m luminosities in order of descending @xmath3 . we conclude from this limited information that most ellipticals with the largest @xmath3 and @xmath4 in figure 3 can not be regarded as typical or normal . of the 12 galaxies in appendix a , we regard the following 8 as either likely s0s ( which often have extended , rotationally supported , star - forming disks of cold dusty gas ) or having massive hi or molecular lanes or disks probably acquired by mergers : ic 3370 , ngc 5018 , 2974 , 4125 , 5322 , 2768 , 3962 , and 5077 . these atypical or mis - classified elliptical galaxies are the brightest galaxies in our sample . however , we regard the following galaxies in appendix a with slightly lower @xmath3 as normal or quasi - normal ellipticals : ngc 3557 , 5044 , 4589 , and ( possibly ) 4697 . if we disregard the eight galaxies with atypical dust content , the range in @xmath3 and @xmath4 occupied by normal ellipticals is reduced from @xmath5 ( shown in figure 3 ) to about @xmath79 , but this is still larger than can be explained by varying dust parameters in our sed model . these normal ellipticals with excess fir are the most interesting galaxies in our sample . we assume that galaxies with @xmath78 significantly above that predicted by our model seds in figure 11 have dust contents that exceed those that can be understood by a steady state balance between dust creation by mass - losing stars and dust destruction by sputtering . galaxies in figure 3 with @xmath80 must contain additional cold dust from another source . this additional cold dust is either ( 1 ) spread diffusely throughout a large region of the galactic hot interstellar gas or ( 2 ) located in dense optically thick clouds similar to those seen in absorption by optical observers . in appendix b we describe a simple argument that supports the first option above , diffuse fir emission . we suppose in appendix b that the fir - emitting dust is in a small spherical cloud of radius @xmath81 located at the center of an elliptical galaxy . the dust is heated both by internal stars and by starlight incident from the entire galaxy . the dust cloud is assumed to be optically thick to most stellar radiation . as the cloud radius @xmath81 is increased , the estimated 70@xmath1 m luminosity also increases because the cloud encloses more stars and intercepts more light from external stars . however , this simple dusty cloud model can not account for the luminosity range in figure 3 unless @xmath81 is so large ( @xmath82 pc ) that it would be easily apparent in moderate resolution optical images . the uv - ionized surfaces of these clouds may also be visible . while agn heating within optically thick dusty clouds could easily provide the needed fir energy , this seems unlikely since agns would need to be bolometrically brighter than the stars within @xmath81 and would therefore be seen in those ellipticals in which the dusty clouds are in disks or not present at all . such luminous agn would almost certainly be visible at other frequencies . comparisons of the most fir - luminous ( normal ) ellipticals in our sample with recent hst images ( e.g. van dokkum & franx 1995 ; lauer et al . 2005 ) , show a correspondence between the absence of nuclear dust and low @xmath78 . for example , of the 19 galaxies in common between our sample and that of lauer et al . ( 2005 ) , all 11 of the ellipticals with @xmath83 erg s@xmath14 have detectable nuclear dust , while all but one of the 9 galaxies with @xmath84 have no nuclear dust detected . nevertheless , even for these galaxies with lower @xmath3 we believe that most of the @xmath78 comes from a diffuse component . in addition to the argument in appendix b , our preference for extended excess dust is supported by noting that all six galaxies with extended fir emission in table 1 ( e.g. fw0.1 m @xmath85 ) have excess fir emission , i.e. @xmath83 erg s@xmath14 . if the fir emission in excess of that expected from our simple model sed in figure 11 is from cold dust diffusely distributed throughout the hot interstellar gas , where has it come from ? it is often supposed that optically - observed dust in elliptical galaxies has an external origin , resulting from a recent merger with a small dust - rich galaxy that is otherwise difficult to observe . we adopted this explanation previously in our discussion of the excess dust in ngc 4636 ( temi , et al . 2003 ) , which we interpreted as a dusty merger that occurred within the last @xmath86 yrs , the estimated sputtering time in the hot gas in this galaxy . however , it has since been determined that the mean stellar age in ngc 4636 is very old ( 10.3 gyrs , sanchez - blazquez et al . 2006 ) , and this strongly argues against a merger @xmath86 yrs ago with a dusty , star - forming galaxy . moreover , we observe that the additional dust in ngc 4636 is extended ( fw0.1 m = 54.4@xmath40 in table 1 ) and appears to be distributed rather evenly about the center unlike the irregular distribution that might be expected to result from a recent merger undergoing dynamical friction with accompanying ram - stripping and tidal disruption . so our previous interpretation of ngc 4636 needs to be reevaluated . the dust sputtering time for grains of initial radius @xmath48 in the hot gas of an elliptical galaxy is @xmath87 yrs ( where @xmath88 is in microns ) provided the ambient gas temperature is @xmath89 k. ( draine & salpeter 1979 ; tsai & mathews 1995 ) . even if the dust is mixed with warm gas at @xmath8 k , where sputtering is negligible , its overall sputtering destruction time @xmath90 can not be significantly reduced since , as discussed below , we generally expect that the warm gas is thermally heated and joins the ( spatially extended ) hot gas phase on a time scale @xmath43 yrs that is very much less than @xmath90 . therefore it is reasonable to assume that the diffuse fir - emitting dust in our sample galaxies is in direct contact with the hot gas during most of its sputtering lifetime . the dynamical ( freefall ) time of a merging dusty galaxy across the optical image of a large elliptical is several @xmath9 years , comparable to @xmath90 . therefore , on the merger hypothesis , diffusely distributed cold dust must be resupplied by mergers in a regular fashion at least every @xmath9 yrs . therefore , an entourage of dusty merging galaxies should be commonly observed in the outer regions of the majority of elliptical galaxies having fir excess emission but they are not . in fact , mergers between elliptical and ( small ) dust rich galaxies are rarely observed at zero redshift . while we believe that mergers do occur and that some of the most fir luminous galaxies in our sample can only be explained by mergers , the similarity of the sputtering and dynamical time scales makes it difficult to accept the proposition that mergers are the dominant or only source of ( extended ) dust in most luminous elliptical galaxies . this point of view is supported by figure 5 , showing that the incidence of young stars in elliptical galaxies is uncorrelated with the fir excess luminosities seen in figure 3 . before exploring the origin of the additional dust required to produce fir luminosities in excess of @xmath91 erg s@xmath14 , it is useful to review the evolution of cold dusty gas that is introduced into the hot , x - ray emitting interstellar gas by mass - losing red giant stars . our sed model is based on an internal stellar source of dust which is confirmed by mid - ir observations of ( apparently outflowing ) circumstellar dust ( knapp et al . 1989 ; athey et al . 2002 ; bressan et al . 2006 ; bregman et al . we think that the fate of this dusty gas varies with galactic radius . throughout most of the volume occupied by galactic stars beyond @xmath92 kpc , we assume that both the dust and gas ejected from red giant stars ultimately merge with the hot gas which is the dominant interstellar phase . we expect that gas recently ejected from stars can be maintained in pressure equilibrium at warm temperatures ( @xmath93 k ) by photoionization by uv emitted by post - agb stars . emission from warm gas at @xmath94 kpc is faint and has not been observed . eventually this warm gas thermally mixes with nearby hot gas . when the dust in this gas confronts the hot gas it will be destroyed by sputtering in @xmath9 yrs . however , within @xmath101 kpc of the centers of elliptical galaxies the evolution of dusty stellar ejecta is different . extended warm gas in this region is commonly observed , but its velocities differ from those of nearby stars ( e.g. caon , macchetto & pastoriza 2000 ) . evidently the hot and warm gas exchanged momentum when the warm gas had a much lower density . after dusty stellar envelopes and winds are ejected from orbiting red giant stars , the gas expands until its density approaches ( but still exceeds ) the local hot gas density . at this time the ejected gas , which is still moving with the stellar orbital velocity , is disrupted by hydrodynamic instabilities and shocked to a few million degrees where its density is comparable to that of the hot interstellar gas . during this period of interaction between the stellar ejecta and the hot gas , momentum is exchanged and the ejected gas acquires some of the velocity of the ambient hot gas . even if the gas ejected from stars is ( transiently ) heated completely to the temperature of the local hot gas ( @xmath95 k ) , its dust content will cool the gas rapidly back to @xmath96 where it is stabilzed by uv from post - agb stars . this rapid dust - assisted cooling occurs because of very frequent inelastic collisions between thermal electrons in the hot gas and the dust grains . mathews & brighenti ( 2003 ) showed that efficient dust cooling of the hot gas is also faster than the sputtering lifetime of the dust , so the gas retains some dust after it cools to @xmath97 k. since the cooling time to @xmath97 k within 1 kpc is less than the dynamical time in the local ( stellar ) potential , these dense warm clouds freefall toward the galactic center and accumulate there . as dusty warm gas collects into a disk near the center , similar to those observed by van dokkum & franx ( 1995 ) and lauer et al . ( 2005 ) , it eventually becomes optically thick to the stellar uv radiation . because of its residual dust , we expect molecular formation and eventual cooling far below @xmath44 k. dense , cold nuclear dust disks and clouds similar to those observed can form naturally from stellar mass loss in elliptical galaxy cores ( mathews & brighenti 2003 ) . dusty galaxy mergers are not required . we speculate below that this central dust reservoir may be the source of extended diffuse fir emission in many ellipticals . central dusty clouds and disks in ellipticals are not merely an amusing curiosity , but are likely to be key to understanding the energetics of agn feedback in regulating cooling flows in the hot gas within and around elliptical galaxies . indeed , simoes lopes et al . ( 2006 ) find a correlation between nuclear dust and agn activity in elliptical galaxies . consider the black hole accretion luminosity expected in a large group - centered elliptical galaxy by dusty mass loss just within the central kpc where rapid dust - assisted cooling is expected . accretion is expected to quickly follow cooling . the agn luminosity is @xmath98 ( \epsilon / 0.1)$ ] erg yr@xmath14 where @xmath99 is the total rate of stellar mass loss within the central kpc . since a @xmath100 year old stellar population of mass @xmath101 ejects @xmath102 @xmath103 yr@xmath14 , the stellar mass within 1 kpc , @xmath104 @xmath103 , generates @xmath105 @xmath103 yr@xmath14 . when accreted by the central massive black hole , this translates into a luminosity @xmath106 ergs s@xmath14 that exceeds the entire x - ray luminosity of the most luminous galaxy groups , @xmath107 ergs s@xmath14 . consequently , dusty stellar mass loss within the central kpc of bright ellipticals can generate enough accretion energy to balance the entire bolometric x - ray luminosity of a large galaxy group . the heating of the hot gas as a consequence of this accretion also provides enough energy to drive buoyant mass circulation in the hot gas out to 5 - 10 kpc . the evolution of dusty disks in galactic cores that we just described can not continue very long . if small core disks accrete dusty gas at @xmath105 @xmath103 yr@xmath14 , for @xmath108 yrs , and allowing for a larger mass - loss rate in the past , currently observed disks should have a typical mass of @xmath108 @xmath103 and a dust mass of @xmath109 , both far in excess of those observed . ( small stellar disks with masses @xmath110 @xmath103 are not commonly observed in the cores of most ellipticals , so normal star - formation is ruled out . ) a large minority of bright ellipticals show no evidence of central dusty disks / clouds . where has this dusty gas gone ? the excess diffuse dust observed in fir - emission may have been exploded out from the centers of elliptical galaxies by the release of agn energy . we estimate from our model sed that the total mass of excess dust required to explain the fir luminosities in figure 3 is @xmath111 @xmath103 . this is comparable to the estimated dust masses in the central dusty disks or lanes ( e.g. goudfrooij & de jong 1995 ) . if the ( very uncertain ) dust to gas mass ratio is @xmath112 , then the mass of gas that initially contained this dust is @xmath113 @xmath103 . if cold dusty gas is ballistically ejected from the core out to a few kpc , velocities of @xmath114 km s@xmath14 are required . the kinetic energy needed to explosively transport dusty gas from the galactic core to @xmath115 kpc is @xmath116 ergs . although very uncertain , the energy involved is certainly larger than that of a single supernova . since multiple starburst activity in elliptical galaxies is very rare , agn energy is required for dust ejection from the core . if the agn dust ejection process is intermittent , this would explain those elliptical galaxies that are not observed to have optically obscuring dust in their cores or extended fir excess emission . the fir variance expected from intermittent agn outbursts obviates the argument of forbes ( 1991 ) and simoes lopes et al . ( 2006 ) that the failure of an @xmath78-@xmath2 correlation necessarily requires an external source of dust by mergers . some explosive character to the agn - dust cloud interaction is implied by the chaotic nature of the central dust observed in many hst images ( lauer et al . this chaotic , denser dusty gas is highly transient and must have been ejected from the core . these chaotic clouds can not be understood as infalling stellar ejecta ( as suggested by lauer et al . ) , formed by the coalescence of many dusty stellar envelopes produced by individual stars , since the free fall time ( @xmath43 yrs ) to the galaxy center is too short for such accumulations to occur . however , explosive ejections of central dust to @xmath117 kpc with energies @xmath118 ergs would produce kpc - sized central cavities , strong shocks and heated cores that are rarely observed in x - ray images while extended excess fir emission is common . this may be the best argument against explosive dust transport . density irregularities and multi - temperature x - ray spectra of the hot interstellar gas in elliptical galaxies reveal buoyant activity , presumably related to a less explosive heating of the hot gas with agn energy ( buote , lewis , brighenti & mathews 2003a , b ) . in view of these x - ray observations , it is plausible that central dust can be buoyantly transported out to @xmath119 kpc , producing the extended fir emission that we observe . the buoyant rise time @xmath120 of heated gas is expected to be comparable to or somewhat less than the sputtering destruction time in the buoyant gas . for example in the galaxy / group ngc 5044 the freefall time from radius @xmath121 to the galactic center is @xmath122 yrs , based on a mass model that combines a de vaucouleurs stellar profile ( @xmath123 , @xmath103 ; @xmath124 kpc ) with an navarro - frenk - white dark halo ( mass @xmath125 @xmath103 ; concentration @xmath126 ) . the buoyancy rise time @xmath127 is greater than @xmath128 by some factor @xmath129 . > from draine & salpeter ( 1979 ) the approximate grain sputtering time is @xmath130 where @xmath88 is the grain radius in microns . adopting @xmath131 @xmath51 for the observed hot gas density in ngc 5044 within 30 kpc , the local sputtering time is nearly proportional to @xmath128 , @xmath132 . however , the buoyant gas will have a density that is @xmath133 times lower than the observed hot gas density . therefore , we see that the ratio of the local sputtering time in the buoyant gas to the buoyancy rise time is @xmath134 , which may well exceed unity considering the many uncertainties involved . the exact nature of the agn energy release that initiates buoyant activity is unclear at present . in about half of the ellipticals with optically visible dust , the images of the obscuring dust are morphologically `` chaotic '' and distributed in an asymmetric fashion away from the center of the galaxy ( lauer et al . 2005 ) . evidently the agn energy is sufficient to disrupt the dense central dusty disks and lanes which must reform again later by local stellar mass loss . it is interesting that the radii of the dusty disks observed by lauer et al . ( 2005 ) do not correlate with the stellar rotation , also suggesting short disk lifetimes and intermittent disruption of the dusty disks . regardless of the mode of outward dust transport , explosive or buoyant , when the dust rises to 5 - 10 kpc it may cause the hot gas to cool back to warm gas temperatures @xmath97 k. the correlations in figure 8 between @xmath28 and @xmath29 and the optical line fluxes , if real , indicate a connection between the warm gas and ( excess ) dust . similar morphologies of gas and dust images have been discussed by optical observers ( e.g. goudfrooij et al . although optical line fluxes appear to correlate with the fir and b - band fluxes , figure 2 shows that the fir and b - band fluxes do not mutually correlate . this can be understood if most of the fir - emitting dust comes not from recent stellar mass loss , but is transported out from a reservoir in the galactic core , which contains a large mass of dust unrelated to the the global stellar b - band flux or luminosity . our main conclusions are : .1 in ( 1 ) the optical luminosities @xmath2 of elliptical galaxies correlate with luminosities at 24@xmath1 m ( @xmath135 ) but not with @xmath3 or @xmath4 . this indicates that the transition between circumstellar and interstellar dust occurs for dust grains with emission that peaks somewhere between 24@xmath1 and 70@xmath1 m . .1 in ( 2 ) elliptical galaxies with similar @xmath2 have @xmath3 and @xmath4 that vary over a factor @xmath5 . this variation is reduced to @xmath79 if abnormally dust - rich ellipticals are excluded from the sample . .1 in ( 3 ) neither @xmath6 nor @xmath7 correlates with the stellar age or metallicity of the galaxies . .1 in ( 4 ) weak correlations appear between optical line emission fluxes from warm ( @xmath97 k ) gas and @xmath28 and @xmath29 . this suggests a relationship between dust and gas that may have cooled from the hot phase . .1 in ( 5 ) seven galaxies in our sample are clearly extended at 70@xmath1 with fw0.1 m exceeding 50@xmath40 corresponding to 510 kpc . .1 in ( 6 ) the luminosities and spectral energy distributions of elliptical galaxies with the lowest @xmath3 and @xmath4 are consistent with a simple steady state dust model in which dust expelled from stars is eroded by sputtering in the hot interstellar gas . .1 in ( 7 ) we show in appendix b that the observed emission at 70 and 160@xmath1 m can not come from small optically visible clouds in the galactic cores similar to those seen in hst images . .1 in ( 8) in view of the common occurrance of spatially extended emission at 70@xmath1 m in many ellipticals , it is unlikely that this dust can be supplied by successive mergers with gas - rich galaxies , each occurring during the relatively short sputtering lifetime of the diffuse dust , @xmath9 yrs . addiiontal supporting evidence for such mergers is lacking . .1 in ( 9 ) the appearance of extended fir - emitting dust in otherwise normal elliptical galaxies can be understood by the disruption and buoyant transport of centrally accumulated dust by intermittent agn activity . the buoyant rise time out to 5 - 10 kpc , @xmath136 yrs , is less than the sputtering lifetime of the dust , @xmath9 yrs . this work is based on observations made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology , under nasa contract 1407 . support for this work was provided by nasa through spitzer guest observer grant rsa 1276023 . studies of the evolution of hot gas in elliptical galaxies at uc santa cruz are supported by nasa grants nag 5 - 8409 & atp02 - 0122 - 0079 and nsf grant ast-0098351 for which we are very grateful . we thank justin harker for his help with the photometric properties of old stellar populations . athey , a. , bregman , j. n. , bergman , j. d. , temi , p. & sauvage , m. 2002 , apj , 571 , 272 barnes , j. e. & hernquist , l. 1992 , ara&a , 30 , 705 bell , e. f. et al . 2004 , apj , 608 , 752 bertola , f. , buson , l. m. & zeilinger , w. w. 1992 , apj , 410 , l79 birkinshaw , m. & davies , r. l. 1985 , apj , 291 , 32 bregman , j. n. , temi , p. & bregman , j. d. 2006 , 647 , 265 bressan , g. et al . 2006 , apj , 639 , 55 brighenti , f. & mathews , w. g. 1997 , apj , 486 , l83 buote , d. a. , lewis , a. d. , brighenti , f. & mathews , w. g. 2003a , apj , 594 , 741 buote , d. a. , lewis , a. d. , brighenti , f. & mathews , w. g. 2003b , apj , 595 , 151 buson , l. m. et al . 1993 , a&a , 280 , 409 caon , n. , macchetto , d. & pastoriza , m. 2000 , apjs 127 , 39 carollo , c. m. & danziger , i. j. 1994 , mnrs , 270 , 743 carollo , c. m. , franx , m. , illingworth , g. d. , forbes , d. a. 1997 , apj , 481 , 710 dale d. a. et al . 2005 , apj , 633 , 857 denicolo . 2005 , mnras , 358 , 813 draine , b. t. & salpeter , e. e. 1979 , apj , 231 , 77 draine , b. t. & lee , h. m. 1984 , apj , 285 , 89 dwek , e. 1986 , apj , 302 , 363 emsellem , e. et al . 2004 , mnras , 352 , 721 ellis , s. c. & osullivan , e. 2006 , mnras , 367 , 627 faber , s. m. 2006 , apj ( submitted ) ( astro - ph/0506044 ) fabian a. c. , sanders , j. s. , taylor , g. b. & allen , s. w. 2006 , mnras , ( in press ) ( astro - ph/0503154 ) fazio , g. g. et al . 2004 , apjs , 154 , 10 ferrari , f. , et al . 1999 , a&as , 136 , 269 filho , m. e. et al . 2002 , apjs , 142 , 223 forbes , d. a. 1991 , mnras , 249 , 779 fried , j. w. & illingworth , g. d. 1994 , aj , 107 , 992 funes , j. g. et al . 2002 , a&a , 388 , 50 goudfrooij , p. 1994 thesis , university of amsterdam , the netherlands goudfrooij , p. et al . 1994a , a&as , 104 , 179 goudfrooij , p. et al . 1994b , a&as , 105 , 341 goudfrooij , p. & de jong , t. 1995 a&a , 298 , 784 goudfrooij , p. 1999 , in _ star formation in early - type galaxies _ eds . j. cepa & p. carral , asp conf . ser . v. 163 , 55 humphrey , p. et al , 2006 , apj , 646 , 899 jarvis , b. j. 1987 , iau symposium _ structure and dynamics of elliptical galaxies _ , ed . t. de zeeuw , 411 jones , et al . 1997 , apj , 482 , 143 kennicutt , r. c. et al . 2003 , pasp , 115 , 928 kim , d .- w . 1988 , apj , 330 , 684 knapp , g. r. , guhathakurta , p. , kim , d .- w . , & jura , m. a. 1989 , apjs , 70 , 329 koprolin , w. & zeilinger , w. w. 2000 , a&as , 145 , 71 laor , a. & draine , b. t. 1993 , apj , 402 , 441 lauer , t. r. et al . 2005 , aj , 129 , 2138 leonardi , a. j. & worthey , g. 2000 , apj , 534 , 650 macchetto , f. et al . 1996 , a&as , 120 , 463 makovoz , d. roby , t. , khan , i. , booth , h. 2006 , spie , 6274 , 10 mathews , w. g. et al . 2006 , apj , 652 , l17 mathews , w. g. 1989 , aj 97 , 42 mathews , w. g. & brighenti , f. 1999 , apj 526 , 114 mathews , w. g. & brighenti , f. 2003 , apj 599 , 992 mathews , w. g. , brighenti , f. & buote , d. 2004 , apj , 615 , 662 mathis , j. s. , rumpl , w. & nordsieck , k. h. 1977 , apj , 217 , 425 michard , r. 1999 , a&as , 137 , 245 michard , r. 2006 , a&a , 449 , 519 mollenhoff , c. & bender , r. 1989 , a&a , 214 , 61 moellenhoff , c. , hummel , e. & bender , r. 1992 , a&a , 255 , 35 naab , t. , jesseit , r. & burkert , a. 2006 , mnras , 328,461 paolillo et al . 2002 , apj , 565 , 883 paturel , g. , andernach , h. , bottinelli , l. , di nella , h. , et al . , 1997 , a&as , 124 , 109 phillips , m. m. et al . 1986 , aj , 91 , 1062 piovan , l. , tantalo , r. , chiosi , c. , 2003 , a&a , 408 , 559 proctor , r. n. & sansom , a. e. 2002 , mnras , 333 , 517 rampazzo , r. et al . 2005 , a&a , 433 , 497 richter , o .- g . , sackett , p. d. , & sparke , l. s. 1994 , aj , 107 , 99 rieke , g. h. 2004 , apjs , 154 , 25 samurovic , s. & danziger , i. j. 2005 , mnras , 363 , 769 sanchez - blazquez , p. , gorgas , j. , cardiel , n. & gonzalez , j.j . 2006 , a&a , 457 , 809 sandage , a. & tamman , g. a. , 1981 , carnegie inst . of washington , 635 sandage , a. & bedke , j. 1994 , _ the carnegie atlas of galaxies _ , vol 1 sarzi , m. et al . 2006 , mnras , 366 , 1151 shen , s. , et al . 2003 , mnras , 343 , 978 simoes lopes , r. d. , storchi - bergmann , t. , de fatima o. saraiva , m. & martini , p. 2006 , apj ( in press ) ( astro - ph/0610380 ) sofue , y. & wakamatsu , k. i. 2003 , pasj , 45 , 529 tantalo , r. & chiosi , c. 2004 , mnras , 353 , 405 temi , p. , mathews , w. g. , brighenti , f. & bregman , j. d. 2003 , apj , 585 , 121 temi , p. , brighenti , f. , mathews , w. g. & bregman , j. d. 2004 , apjs , 151 , 237 temi , p. , mathews , w. g. & brighenti , f. 2005 , apj , 622 , 235 trinchieri , g. & goudfrooij , p. 2002 , a&a , 386 , 472 tsai , j. c. & mathews , w. g. 1995 , apj , 448 , 84 tsai , j. c. & mathews , w. g. 1996 , apj , 468 , 571 van dokkum , p. g. & franx , m. 1995 , aj , 110 , 2027 verdoes , kleijn , g. a. , & de zeeuw , p. t. 2005 , a&a , 435 , 43 werner , m. w. , et al . 2004 , apjs , 154 , 1 wicklind , t. , combes , f. & henkel , c. 1995 , a&a , 297 , 643 xilouris , e. m. et al . 2004 , a&a , 416 , 41 zeilinger , w. w. et al . 1996 , a&as , 120 , 257 lccrrrrrrrrrr ngc 507 & temi & -3.3 & & @xmath137 & & 71.99 & 77.1 & & & & & + ngc 533 & temi & -4.8 & 7.1@xmath1383.1 & 23.7@xmath1384.9 & 26.1@xmath1385.6 & 68.23 & 47.5 & & & & & + ngc 584 & kennicutt & -4.6&47.6@xmath1387.6 & 49.8@xmath1386.9 & @xmath139 & 23.76 & 27.4 & & & & & + ngc 596 & surace & -4.3&15.9@xmath1384.6 & 21.4@xmath1385.1 & 16.7@xmath1387.3 & 23.87 & 30.0 & & & & & + ngc 636 & surace & -4.8&10.3@xmath1383.6 & @xmath140 & @xmath141 & 22.28 & 18.9 & & & & & + ngc 720 & temi & -4.8&25.6@xmath1385.1 & 18.2@xmath1384.9 & @xmath142 & 22.29 & 39.5 & & & & & + ngc 855 & kennicutt & -4.6&72.3@xmath1388.5 & 1335@xmath13836 & 2086@xmath13846 & 9.33 & 18.9 & 22.5 & 1.0 & & 58.2 & 7.6 + ngc 1377 & kennicutt & -2.1&1603@xmath13840 & 4988@xmath13871 & 2375@xmath13848 & 22.20 & & & & & & + ngc 1395 & kaneda & -4.8&45.4@xmath1386.5 & 129@xmath13811 & 194@xmath13814 & 21.98 & 45.4 & & & & & + ngc 1399 & temi & -4.5&59.9@xmath1387.3 & 15.9@xmath1387.6 & 24.1@xmath1388.6 & 19.40 & 42.4 & & & & & + ngc 1404 & kennicutt & -4.7&56.6@xmath1386.3 & 32.6@xmath1386.3 & @xmath143 & 19.40 & 26.7 & & & & & + ngc 1407 & kaneda & -4.5&42.5@xmath1386.4 & & @xmath144 & 22.08 & 71.8 & & & & & + ngc 1700 & surace & -4.7&18.7@xmath1384.9 & 29.3@xmath1385.3 & 40.3@xmath1387.2 & 38.04 & 13.7 & & & & & + ngc 2325 & temi & -4.7&27.0@xmath1386.4 & 36.5@xmath1386.8 & 21.7@xmath1386.4 & 31.92 & 86.5 & & & & & + ngc 2768 & temi & -4.3&46.6@xmath1388.2 & 694 @xmath13826 & 377 @xmath13829 & 22.38 & 53.5 & 21.4 & 2.3 & & 56.4 & 6.1 + ngc 2974 & kaneda & -4.7&62.5@xmath13812.4&682 @xmath13818 & 1979@xmath13854 & 30.33 & 36.9 & 22.8 & 3.4 & & 55.4 & 8.1 + ngc 2986 & temi & -4.7&13.7@xmath1384.8 & @xmath137 & @xmath145 & 32.51 & 41.4 & & & & & + ngc 3265 & kennicutt & -4.8&287 @xmath13817 & 1952@xmath13844 & 2312@xmath13849 & 21.33 & & & & & & + ngc 3379 & fazio & -4.8&65.5@xmath1388.8 & 60.5@xmath1388.0 & 59.1@xmath1388.2 & 10.71 & 35.2 & & & & & + ngc 3557 & temi & -4.8&30.5@xmath1384.5 & 263 @xmath13816 & 247 @xmath13818 & 34.51 & 37.8 & 20.0 & 3.3 & & 38.8 & 6.5 + ngc 3610 & surace & -4.2&18.1@xmath1384.5 & 19.1@xmath1387.3 & 32.8@xmath13816.5 & 29.24 & 12.8 & & & & & + ngc 3923 & temi & -4.6&42.9@xmath1386.7 & 22.7@xmath1385.4 & 44.0@xmath1387.8 & 19.14 & 53.3 & & & & & + ngc 3962 & kaneda & -4.7&21.5@xmath1385.6 & 374 @xmath13818 & 495 @xmath13824 & 23.23 & 34.4 & 22.6 & 2.5 & & 44.0 & 5.0 + ngc 4125 & kennicutt & -4.8&53.7@xmath1386.9 & 773 @xmath13828 & 1395@xmath13837 & 27.79 & 59.9 & 21.6 & 2.9 & & 56.8 & 7.7 + ngc 4365 & cote & -4.8&21.7@xmath1384.7 & 63.8@xmath1388.0 & 52.9@xmath1387.3 & 17.06 & 57.2 & & & & & + ngc 4472 & fazio & -4.8&73.3@xmath1388.6 & 58.1@xmath1387.6 & 60.4@xmath1387.8 & 17.06&104.0 & & & & & + ngc 4552 & kennicutt & -4.6&57.3@xmath1387.8 & 92.1@xmath13810.2 & 171 @xmath13816 & 17.06 & 30.0 & & & & & + ngc 4589 & kaneda & -4.8&14.9@xmath1384.5 & 262 @xmath13816 & 360 @xmath13821 & 26.30 & 41.4 & 22.6 & 2.9 & & 45.6 & 5.8 + ngc 4621 & cote & -4.8&34.2@xmath1386.3 & 31.7@xmath1385.7 & 43.4@xmath1386.6 & 17.06 & 46.5 & & & & & + ngc 4636 & fazio & -4.8&31.2@xmath1385.6 & 188 @xmath13812 & 169 @xmath13814 & 17.06&101.7 & 22.3 & 1.8 & & 54.4 & 4.5 + ngc 4649 & fazio & -4.6&106 @xmath13810 & 46.3@xmath1386.8 & @xmath146 & 17.06 & 73.6 & & & & & + ngc 4660 & cote & -4.7&15.2@xmath1384.3 & 36.4@xmath1386.2 & 53.7@xmath1388.2 & 17.06 & 12.8 & 20.0 & 1.7 & & 39.0 & 3.2 + ngc 4696 & sparks & -3.7&22.9@xmath1384.7 & 127 @xmath13813 & 269 @xmath13821 & 39.65&212.4 & 22.6 & 4.4 & & 52.2 & 10.1 + ngc 4697 & surace & -4.8&43.8@xmath1386.7 & 589 @xmath13825 & 755 @xmath13828 & 16.22 & 75.3 & & & & & + ngc 4915 & surace & -4.5&11.5@xmath1385.5 & 29.2@xmath1384.8 & 54.2@xmath1388.9 & 46.98 & 7.9 & & & & & + ngc 5018 & surace & -4.6&60.5@xmath1389.1 & 1119@xmath13834 & 1687@xmath13841 & 32.36 & 24.9 & 20.4 & 3.2 & & 39.6 & 6.21 + ngc 5044 & temi & -4.7&23.0@xmath1385.3 & 230@xmath13814 & 242@xmath13818 & 32.36 & 78.6 & 23.1 & 3.6 & & 53.6 & 8.4 + ngc 5077 & fazio & -4.8&32.4@xmath1385.7 & 127 @xmath13811 & 141 @xmath13813 & 32.36 & 25.0 & 20.2 & 3.2 & & 49.5 & 7.8 + ngc 5322 & fazio & -4.8&41.4@xmath1386.4 & 455 @xmath13823 & 625 @xmath13827 & 29.79 & 35.2 & & & & & + ngc 5557 & surace & -4.8&13.2@xmath1385.0 & 25.1@xmath13810.5 & @xmath147 & 49.02 & 28.6 & & & & & + ngc 5813 & fazio & -4.8&14.9@xmath1384.9 & 58.4@xmath1387.6 & 34.8@xmath1385.9 & 30.78 & 48.6 & 19.6 & 2.9 & & 49.8 & 7.4 + ngc 5846 & fazio & -4.7&38.5@xmath1386.2 & 102 @xmath13810 & 117 @xmath13812 & 24.55 & 82.6 & 20.4 & 2.4 & & 43.1 & 5.1 + ngc 5982 & surace & -4.8&13.5@xmath1385.1 & 37.2@xmath1387.2 & 59.9@xmath13811.7 & 40.18 & 29.4 & & & & & + ngc 6703 & fazio & -2.8&19.9@xmath1384.5 & 41.4@xmath1386.4 & 17.9@xmath1385.1 & 32.06 & 23.8 & & & & & + ngc 7619 & temi & -4.7 & 9.9@xmath1384.7 & @xmath148 & @xmath145 & 42.85 & 32.1 & & & & & + ic 3370 & kaneda & -4.7&43.8@xmath1386.7 & 649 @xmath13825 & 873 @xmath13830 & 45.90 & 18.5 & 20.5 & 4.5 & & 39.7 & 8.8 + lrrrrrrr ngc 507 & & @xmath149 & & 11.02 & 42.76 & 8.1@xmath1381.9 & 0.261@xmath1380.071 + ngc 533 & 41.04 & 41.19 & 40.76 & 10.96 & 42.29 & & + ngc 584 & 40.95 & 40.59 & @xmath150 & 10.42 & @xmath151 & 2.8@xmath1380.3 & 0.478@xmath1380.046 + ngc 596 & 40.48 & 40.23 & 39.65 & 10.27 & @xmath150 & 2.8@xmath1380.4 & 0.360@xmath1380.035 + ngc 636 & 40.29 & @xmath152 & @xmath153 & 10.06 & @xmath154 & 4.4@xmath1380.6 & 0.376@xmath1380.028 + ngc 720 & 40.62 & 40.10 & @xmath155 & 10.44 & 40.67 & 5.4@xmath1382.4 & 0.485@xmath1380.083 + ngc 855 & 40.38 & 41.27 & 41.00 & 8.95 & @xmath156 & & + ngc 1377 & 42.48 & 42.59 & 41.81 & 9.60 & & & + ngc 1395 & 40.86 & 40.94 & 40.65 & 10.50 & 40.95 & 7.6@xmath1381.4 & 0.439@xmath1380.033 + ngc 1399 & 40.87 & 39.92 & 39.64 & 10.58 & 41.69 & & + ngc 1404 & 40.85 & 40.56 & @xmath157 & 10.41 & 41.25 & & + ngc 1407 & 40.83 & @xmath158 & @xmath159 & 10.64 & 41.06 & 7.4@xmath1381.8 & 0.379@xmath1380.044 + ngc 1700 & 40.95 & 40.77 & 40.44 & 10.55 & 40.66 & 2.6@xmath1380.3 & 0.500@xmath1380.049 + ngc 2325 & 40.96 & 40.71 & 40.01 & 10.66 & 40.76 & & + ngc 2768 & 40.88 & 41.68 & 40.95 & 10.63 & 40.44 & 10.0@xmath1387.0 & 0.14@xmath1380.25 + ngc 2974 & 41.27 & 41.94 & 41.94 & 10.56 & 40.64 & & + ngc 2986 & 40.68 & @xmath160 & @xmath161 & 10.57 & 41.02 & 11.2@xmath1381.9 & 0.347@xmath1380.045 + ngc 3265 & 41.69 & 42.15 & 41.76 & 9.46 & & & + ngc 3379 & 40.39 & 39.98 & 39.51 & 10.12 & @xmath162 & 10.0@xmath1381.1 & 0.299@xmath1380.036 + ngc 3557 & 41.08 & 41.64 & 41.15 & 10.82 & 40.64 & & + ngc 3610 & 40.71 & 40.36 & 40.13 & 10.46 & 39.89 & 1.7@xmath1380.1 & 0.76@xmath1380.16 + ngc 3923 & 40.71 & 40.06 & 39.88 & 10.58 & 40.72 & 3.3@xmath1380.8 & 0.623@xmath1380.070 + ngc 3962 & 40.58 & 41.45 & 41.10 & 10.34 & @xmath163 & & + ngc 4125 & 41.14 & 41.92 & 41.71 & 10.86 & 41.0 & & + ngc 4365 & 40.32 & 40.41 & 39.86 & 10.40 & 40.31 & 7.9@xmath1381.2 & 0.344@xmath1380.065 + ngc 4472 & 40.85 & 40.37 & 39.92 & 10.96 & 41.49 & 9.6@xmath1381.4 & 0.342@xmath1380.046 + ngc 4552 & 40.74 & 40.57 & 40.37 & 10.35 & 40.77 & 12.4@xmath1381.5 & 0.356@xmath1380.034 + ngc 4589 & 40.53 & 41.40 & 41.07 & 10.39 & 40.42 & & + ngc 4621 & 40.52 & 40.11 & 39.78 & 10.38 & 40.08 & 8.8@xmath1381.1 & 0.299@xmath1380.050 + ngc 4636 & 40.48 & 40.88 & 40.37 & 10.57 & 41.65 & 10.3@xmath1381.3 & 0.181@xmath1380.104 + ngc 4649 & 41.01 & 40.27 & @xmath164 & 10.79 & 41.34 & 14.1@xmath1381.5 & 0.362@xmath1380.029 + ngc 4660 & 40.16 & 40.17 & 39.87 & 9.80 & @xmath165 & & + ngc 4696 & 41.07 & 41.44 & 41.30 & 10.99 & 43.23 & & + ngc 4697 & 40.58 & 41.33 & 40.98 & 10.61 & 40.18 & 8.3@xmath1381.4 & 0.148@xmath1380.043 + ngc 4915 & 40.92 & 40.95 & 40.76 & 10.34 & @xmath166 & & + ngc 5018 & 41.32 & 42.21 & 41.92 & 10.63 & @xmath167 & & + ngc 5044 & 40.90 & 41.52 & 41.08 & 10.76 & 42.80 & & + ngc 5077 & 41.05 & 41.27 & 40.85 & 10.32 & 40.54 & & + ngc 5322 & 41.08 & 41.75 & 41.42 & 10.73 & 40.27 & & + ngc 5557 & 41.02 & 40.92 & @xmath168 & 10.90 & & & + ngc 5813 & 40.67 & 40.89 & 40.20 & 10.70 & & 16.6@xmath1382.2 & 0.059@xmath1380.042 + ngc 5846 & 40.88 & 40.93 & 40.53 & 10.72 & 41.71 & 14.2@xmath1382.2 & 0.226@xmath1380.051 + ngc 5982 & 40.86 & 40.92 & 40.66 & 10.59 & 41.22 & & + ngc 6703 & 40.83 & 40.77 & 39.93 & 10.43 & @xmath169 & 4.8@xmath1380.8 & 0.354@xmath1380.034 + ngc 7619 & 40.78 & @xmath170 & @xmath171 & 10.64 & 41.69 & 15.4@xmath1381.4 & 0.291@xmath1380.031 + ic 3370 & 41.48 & 42.28 & 41.94 & 10.82 & & & + m ( left column ) , 70 @xmath1 m ( center column ) , and 160 @xmath1 m ( right column ) . galaxies are shown in ascending order from lowest flux density to highest flux density , including a non detection and sources with extended far - infrared emission . the contour levels decrease by factors of 2 from the central peak value . ] 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in 2.in .7 in in this appendix we briefly review other observations of galaxies in figure 3 and table 2 having the most luminous fir emission . our objective is to determine if elliptical galaxies with the most luminous fir emission can be regarded as unusual or abnormal and , if so , what is the maximum fir luminosity of sample galaxies that can be regarded as normal . in the following we emphasize observations that are consistent with large fir emission evidence of large ( @xmath34 kpc ) extended dusty disks , detections of large masses of cold gas or recent star formation , designations as an s0 galaxy which often contain cold , rotationally supported gaseous disks , strong radio emission , galaxies with nearby dusty galaxies or stars that may have contaminated some observations , evidence of a recent merger , etc . but we do not mention attributes that are found in many typical ellipticals : extended optical emission lines perhaps with a liner spectrum , small central ( @xmath172 kpc ) dust clouds that obscure starlight , weak central radio sources , x - ray emission @xmath16 which is highly variable among ellipticals , evidence of old mergers , etc . galaxies are listed in descending order of @xmath3 . .15 in * ic 3370 : * classified as an elliptical in rc2 an rsa , but regarded by jarvis ( 1987 ) and samurovic & danziger ( 2005 ) as an s0 or s0pec with an extremely boxy profile . a bright iras source in fir . richter et al . ( 1994 ) detect an hi mass of @xmath173 @xmath103 . contains a prominent dust lane in the inner regions with associated h@xmath174 + [ nii ] emission and possible evidence of ongoing star formation ( michard 2006 ) . leonardi & worthey ( 2000 ) and tantalo & chiosi ( 2004 ) suggest an intermediate age population for the central regions . .15 in * ngc 5018 : * forms a pair with ngc 5022 7.2@xmath175 away . carollo & danziger ( 1994 ) describe a very strong dust lane that obscures the innermost regions that may contain recent star formation . they also suggest that the galaxy is very flat ( about e6 ) . detected by iras at 60 and 100@xmath1 m . xilouris et al . ( 2004 ) find excess mir emission at 6.7 and 15@xmath1 m , perhaps suggesting heating by recent star formation . .15 in * ngc 2974 : * a rapidly rotating e4 galaxy ( emsellem et al . 2004 ) . strong iras emission . kim et al . ( 1988 ) observes a massive hi disk , @xmath176 @xmath103 ( with @xmath177 mpc ) . sarzi et al . ( 2006 ) observe a nuclear bar with two spiral arms emerging . a bright star is seen in ned images 0.7@xmath175 to sw . .15 in * ngc 4125 : * filho et al . ( 2002 ) regard this as an e6pec galaxy . wickland et al . ( 1995 ) detect tentative molecular hydrogen emission with mass @xmath178 @xmath103 . verdoes kleijn & de zeeuw ( 2005 ) observe a central dust lane of size 160 pc ( d = 26 mpc ) . .15 in * ngc 5322 : * carollo et al . ( 1997 ) describe a completely obscuring dust lane ( perpendicular to the radio jet ) that hides the center of the galaxy . iras detection at 60 and 100@xmath1 m . proctor & sansom ( 2002 ) and denicolo et al . ( 2005 ) find evidence of a @xmath179 gyr old young stellar population . .15 in * ngc 2768 : * referred to as an e6 * in rc3 , this is almost certainly an s0 galaxy ( funes et al . 2002 ) with faint dust patches surrounding the galactic bulge . significant dust extinction is visible along the minor axis ( michard 1999 ) . bertola et al . ( 1992 ) notes that the ionized gas is dynamically decoupled from the stars , which may be related to gas orbiting in a polar ring ( mollenhoff et al . 1992 ; fried & illingworth 1994 ) . wiklind et al . ( 1995 ) observed co emission with an estimated molecular mass of @xmath180 @xmath103 . .15 in * ngc 3557 : * an e3 galaxy containing an fri radio source having a double - tailed radio source with a jet and central knot ( birkinshae & davies 1985 ) . detected by iras at 60 and 100@xmath1 m . rampazzo et al . ( 2005 ) report a possible dust ring near the center and goudfrooij ( 1994b ) finds an unusual optical emission line image associated with the minor axis . .15 in * ngc 5044 : * detected by iras at 60 and 100@xmath1 m . very bright and asymmetrically extended optical emission ( ferrari , et al . 1999 ; goudfrooij et al . 1994a ) . .15 in * ngc 3962 : * the h@xmath174+[nii ] image shows two sub - systems : an elongated ( dusty ) central component strongly misaligned with the stellar isophotes extending to @xmath181 kpc with regular kinematics and a more extended arm - like structure associated with bluer stars ( buson et al . 1993 ; goudfrooij , et al . 1994b ; zeilinger et al . 1996 ) . detected by iras at 60 and 100@xmath1 m . .15 in * ngc 4589 : * prominent dust lane near minor axis extending to 20@xmath40 ( goudfrooij , et al . 1994a , b ) complex gas and stellar kinematics suggest a merger remnant ( mollenhoff & bender 1989 ) . detected by iras at 60 and 100@xmath1 m . .15 in * ngc 4697 : * an e6 galaxy , but may be an s0 ( koprolin & zeilinger 2000 ) . sofue and wakamatsu ( 2003 ) detect a molecular mass of @xmath182 @xmath103 . .15 in * ngc 5077 : * the e3 designation in rc2 is wrong , this is an s0 galaxy ( sandage & bedke 1994 ) . in this appendix we determine if a large fraction of the fir emission from elliptical galaxies can come from small , dense , dusty clouds near the centers of the galaxy , or , as we have generally assumed , comes from fir - emitting dust diffusely distributed throughout the inner galaxy . specifically , can reprocessed starlight from internal and external stars heat small dusty regions sufficiently to account for the large fir luminosities shown in figure 3 ? we describe a simple calculation to estimate the maximum luminosity at 70@xmath1 m @xmath3 expected from a small optically thick dust cloud , including dust heating by both internal and external starlight . consider a spherical cloud of dusty cold gas with radius @xmath81 placed at the center of an elliptical galaxy having a de vaucouleurs luminosity profile . since the stellar density is greatest at the center , placing the cloud there maximizes the energy from internal starlight that can be absorbed and reemitted at infrared wavelengths . the mean intensity of starlight in an ( optically thin ) elliptical galaxy also is greatest at the galactic center , where the stellar flux incident on the dusty cloud from the outside is also maximal . without investigating in detail how the dust is heated and reemitted in the infrared from small , dense clouds , we assume that this type of reprocessing can occur . to estimate the fraction of the the fir that appears in the _ spitzer _ 70@xmath1 m pass band , we assume that the overall infrared sed can be approximated with a modified blackbody , with emissivity @xmath183 where @xmath184 depends on the properties of the dust and @xmath67 is the dust temperature . we assume @xmath185 and @xmath186k , the mean values in our sed fits to iso ( _ infrared space observatory _ ) fir data for elliptical galaxies ( temi et al . 2004 ) . the fraction of fir emission that emerges in the _ spitzer _ 70@xmath1 m passband ( width @xmath187 m ) is @xmath188 where @xmath189 . to be specific we consider a typical elliptical in figure 3 with @xmath190 . based on the galaxy - size observations of shen et al . ( 2003 ) , we assume a typical effective radius for this galaxy , @xmath191 kpc . to determine the total b - band luminosity of stars inside the dusty cloud , we integrate the de vaucouleurs profile from the center to radius @xmath81 . to convert the b - band luminosity to bolometric luminosity of an old single stellar population ( ssp ) , we assume solar abundance for the stars , a kroupa imf and an age of @xmath100 yrs , resulting in @xmath192 ( private communication from justin harker ) . with this information we can estimate the total stellar luminosity produced within a dusty cloud of any chosen radius @xmath81 and the fraction of this energy that is observed in the _ spitzer _ 70@xmath1 m passband , @xmath193 . the dusty cloud can also be heated externally by stellar radiation incident on its outer surface . to estimate the flux of stellar radiation incident on the dusty cloud we first determine the mean b - band mean intensity within 10 - 100 parsecs of the galactic center , @xmath194 erg @xmath195 s@xmath14 ster@xmath14 , found by integrating over the de vaucouleurs emissivity profile corresponding to @xmath196 @xmath197 and @xmath191 kpc . the b - band flux on the surface of the cloud is then @xmath198 . the bolometric flux is larger by a factor @xmath199 so the rate that radiative energy from external galactic stars is absorbed by the dusty cloud is @xmath200 where @xmath201 is the assumed albedo . figure b1 shows how the 70@xmath1 m luminosities @xmath3(int ) and @xmath3(ext ) vary with the radius @xmath81 of the dusty cloud where we see that @xmath202 . but the important result is that clouds of size @xmath203 pc are required to produce luminosities @xmath204 erg s@xmath14 typical of those plotted in the central panel of figure 3 . in many of the elliptical galaxies in figure 3 , a @xmath5 parsec dark cloud would be easily visible even without hst resolution . we conclude that it is very unlikely that the large excess fir emission in figure 3 above that predicted by our simple model for recently produced diffuse dust , can be explained by emission from small dense clouds even under the most favorable circumstances . more importantly , the approximation of spherical clouds is seriously in error . the gas temperatures associated with these dusty clouds ( @xmath5k ? ) are much less than the virial temperature of the stars , @xmath205 k required to support a spherical cloud as we hypothesized here . the gas temperature within the dusty cloud is insufficient to provide pressure support against the stellar potential due to embedded stars or the external hot gas pressure . consequently , we expect that most of the nuclear dust clouds observed in ellipticals are highly flattened , nearly two - dimensional pancakes or disks . in this case we imagine that the @xmath3(ext ) luminosities in figure b1 are more relevant , further reducing the predicted @xmath206 m emission . we conclude that most of the fir emission observed at 70@xmath1 m and 160@xmath1 is diffusely distributed , not concentrated in dense clouds similar to those seen in absorption by many optical observers .

if @xmath13cdm is correct , then all dark matter halos hosting galaxies from those hosting dwarfs to those hosting giant clusters should be filled with substructure @xcite . dark - matter only simulations over a vast range of particle masses and physical scales show that substructure persists down to the resolution limit of @xmath13cdm simulations @xcite . the basic expectation is that the mass function of subhalos rises steadily to masses well below the molecular cooling limit of @xmath14 @xcite , with thousands of sites for potentially star - forming satellites . observations of our own milky way , on the other hand , have revealed the presence of only @xmath15 confirmed satellite galaxies @xcite , the faintest of which were all discovered by the sloan digital sky survey ( sdss ) @xcite . however , the sdss covers only a fraction of the sky and is incomplete to the most distant and faintest satellites of the milky way . any solution to the mismatch between the predicted abundance of subhalos and the observed counts of satellite galaxies around our own milky way will likely involve the discovery of ultra - faint satellites in unprobed regions of the sky , at large distances , and with surface brightnesses low enough that they lie just outside current detection limits @xcite . over the last several years , searches by panstarrs and vst atlas have failed to lend critical support to this idea finding far fewer dwarf galaxy satellites than expected @xcite . however , the recent discovery of up to nine new ultra - faint satellites in the southern sky by the dark energy survey ( des , * ? ? ? * ; * ? ? ? * ) , the detection of three more faint dwarf satellite candidates @xcite , and ongoing efforts to discover dwarfs at large distances @xcite provide an exciting glimpse into the near future and point to a much larger population of as - yet - undiscovered dwarf galaxies in the local volume . interestingly , most of the newest milky way satellite candidates were all discovered near the large magellanic cloud ( lmc ) . while this could be nothing more than a location - based selection effect ( the lmc happens to lie next to the region probed by the year one data release of des ) , @xcite show that several of the candidates could very well be satellites of the lmc , and their discovery highlights the potential for discovering ultra - faint satellites of other more massive dwarf satellites , or of isolated dwarf galaxies in the local group . the scale - free nature of the subhalo mass function in @xmath13cdm suggests that groups of subhalos should be common @xcite . because low - mass halos form earlier , are denser , and fall into smaller hosts before larger ones @xcite , it is likely that satellites of satellites or of lower mass isolated halos may have survived longer than their counterparts that fell directly into the milky way @xcite . this suggests that one way to search for ultra - faint galaxies might be as satellites of known dwarf galaxies . associations and pairs of satellite galaxies in the local group have been observed for some time @xcite , and several milky way satellites are suspected to host their own satellites @xcite . once a group of satellites falls into the milky way , however , tidal forces will eventually disassemble the group and wipe out evidence of coherent structure @xcite . isolated massive dwarfs in the local group may serve as complementary targets in the hunt for ever - fainter dwarfs ( which presumably probe the lowest mass dark matter subhalos ) . @xcite suggest that even already - merged satellites of isolated dwarfs could be detected as ghost halos " of ancient stellar populations surrounding their hosts . despite the obvious challenge of detecting ultra - faints at large distances , isolated dwarfs also make efficient use of telescope pointings , removing some of the chance inherent in any random ultra - faint search without losing any of the search volume . in the near future , large - area , deep - sky surveys such as lsst , des , panstarrs , and skymapper will push current detection limits to lower surface brightness and may be able to see ultra - faint satellites of isolated dwarf galaxies if they exist @xcite . dynamical measurements of most milky way dwarf spheroidals ( dsphs ) show that , despite having luminosities that vary over nearly five orders of magnitude , they have almost the same central densities , comparable to dark matter halos with @xmath16 @xcite . this is surprising , not only because this is well above the mass where physical processes are believed to prevent star formation , but because the mass function of potentially star - forming subhalos should be heavily populated by smaller halos . segue 2 may be the first of these long - searched - for small subhalos to be identified @xcite , as it appears to have a total dark matter mass @xmath17 . however , based on its metallicity , @xcite hypothesize that segue 2 is a bare remnant of a much larger galaxy that has been severely tidally stripped . the common central density for observed dsphs might signify a low - mass cutoff in galaxy formation . halos with slightly lower central densities may have been unable to shield themselves from the reionizing background , perhaps due to having a lower baryon fractions than their denser counterparts @xcite . alternatively , ( * ? ? ? * hereafter b10 ) suggest that the apparent common mass scale is a selection effect , and that a large population of unobserved stealth galaxies " may reside in halos with masses just below those that host observed dwarfs . these low - mass halos have shallow potential wells , and therefore the galaxies that form within them have larger effective radii and lower surface brightnesses , allowing them to more easily avoid detection @xcite . the number of these stealth galaxies expected to exist is sensitive to the presence of a low - mass cutoff in galaxy formation , as the lowest mass halos are expected to host the puffiest " galaxies in this picture . the heating of accreted gas or the prevention of gas accretion by the ambient ionizing uv background created when the first galaxies formed can arrest or prevent star formation in the smallest dark matter halos , suggesting a threshold in halo mass below which all halos remain dark @xcite . it is possible , however , that some very low mass halos were able to form some population ii stars _ before _ reionization @xcite . if star formation ceased after this time , it would cause them to have uniformly ancient stellar populations @xcite with extremely low metallicity @xcite . remarkably , this appears to be the case with the known ultra faint dwarfs of the milky way @xcite . several authors have shown that photoheating prevents gas from condensing in halos with masses that lie just below or at the common halo mass scale for milky way satellites @xcite . however , the specific timing of the onset and end of reionization , the spectrum of the background radiation , the ability of gas to temporarily shield itself from these high energy photons , and particularly the mass resolution , can have a large effect on the number and minimum halo mass of galaxies in any simulation @xcite . moreover , prescriptions for how stars and supernovae return energy back to the interstellar medium ( ism ) will affect the formation history of small galaxies in the presence of an ionizing background . in what follows , we present a series of simulations of galaxy formation within small dark matter halos run with the ` gizmo ` code @xcite in psph - mode " @xcite . ` gizmo ` implements the feedback in realistic environments ( fire ) @xcite feedback scheme for converting gas into stars and capturing the energy fed back from those stars into the surrounding medium . using these simulations , we find that the subhalos of the halos that surround @xmath18 dwarf galaxies should form stars fairly abundantly , and produce potentially observable ultra - faint satellite galaxies of known classical dwarfs in the local group . further , the ultra - faint dwarfs in our simulations with @xmath19 that form in isolation or as satellites within subhalos of @xmath20 all have completely ancient stellar populations , as is seen for the known population of the milky way . they also have low - surface brightnesses , as expected for stealth galaxies " forming in halos of these low masses , but may be within the detection capabilities of future surveys . this paper is organized as follows : in section [ sec : sims ] , we give the details of our simulations , including halo finding and merger - tree analysis . section [ sec : results ] outlines our main results , including the @xmath21 relation for our galaxies and their star formation histories . in section [ sec : det ] , we use the exploring the local volume in simulations ( elvis ) suite of dark - matter - only local group simulations @xcite to determine how common these objects should be and the likelihood of their detection . we compare to several recent works in section [ sec : compare ] , and summarize our results and conclude in section [ sec : dis ] . we present six cosmological zoom - in simulations of four isolated dwarf galaxy halos using two implementations of subgrid physics . the four host halos are @xmath22 and @xmath23 , which have @xmath24 virial masses of @xmath25 ; and @xmath26 and @xmath27 , which have @xmath28 ( see table 1 for precise numbers ) . the @xmath23 halo was simulated three times two runs varied the specific subgrid feedback implementation , while the third used a different choice of gravitational softening ( see below ) , resulting in changes in the star formation histories of the main galaxies ( see @xcite for details ) . we refer to these three cases as @xmath29 ( run with the fiducial parameters the same parameters as @xmath22 , @xmath26 , and @xmath27 ) , @xmath30 , and @xmath31 . the subscripts refer to when the halos formed their stars . every run has a dark matter force softening of @xmath32 to @xmath33 pc and a dark matter particle mass of @xmath34 except @xmath27 , which uses @xmath35 . the @xmath26 and three @xmath23 simulations are identical to the ultrafaint " and dwarf runs presented by @xcite . and @xmath26 were also previously presented in @xcite as @xmath36 and @xmath37 respectively . ] the initial conditions were generated using ` music ` @xcite , and selected from a cosmological box run at low resolution to @xmath24 . @xmath22 , @xmath23 , and @xmath26 were all selected from @xmath38 boxes to have typical values of spin parameter @xmath39 , concentration , and formation time for their mass range , and also to have small lagrangian volumes @xcite . @xmath27 was selected from a @xmath40 box and required to have no other halos of @xmath41 or more of its mass within @xmath42 at @xmath43 and a small lagrangian volume . all dwarfs in this work are isolated . once the halos are identified , the particles are traced back in time and an enclosing lagrangian volume is chosen and re - simulated at higher resolution with dark matter and gas , buffered by dark matter - only regions of increasing particle mass . this process is done according to the zoom - in techniques outlined by @xcite and with the goal of minimizing low resolution particles within the halos of interest . all simulations begin at a redshift of @xmath44 and have @xmath45 contamination from low resolution particles within a distance of @xmath46 at all redshifts . all simulations were run using the fully conservative cosmological hydrodynamic code ` gizmo ` @xcite in psph - mode " . this code adopts the lagrangian pressure - entropy " formulation of the smooth particle hydrodynamic ( sph ) equations , ameliorating previous difficulties sph codes had in modeling multiphase fluids @xcite . for feedback and star formation we use the feedback in realistic environments ( fire ) scheme @xcite . fire tracks momentum imparted locally from stellar radiation pressure , radiation pressure on larger scales via the light that escapes star - forming regions , hii photoionization heating , supernovae ( type i and ii ) heating , momentum and mass loss and stellar winds from o - type and agb stars . each star particle has an age determined by its formation time , an initial mass function ( imf ) taken from @xcite , and a metallicity inherited from its parent gas particle . the star particle then loses mass and creates metals according to the ` starburst99 ` stellar population synthesis model @xcite and the input imf . our simulations use an ambient ionizing uv background from @xcite , which starts at @xmath47 and completes reionization by @xmath48 . the fire implementation has as few free parameters as possible , with star formation in the multiphase ism naturally self - regulating . giant molecular clouds ( gmcs ) form , and then are subsequently heated and disrupted by the stars that form within the self - gravitating , molecular gas . in order to ensure that the gas within the gmcs reaches a density high enough to form stars ( @xmath49 ) , the code requires extremely high mass and spatial resolution . every run uses a gas particle mass of @xmath50 except for @xmath27 , which uses @xmath51 . the gas force resolution varies from @xmath52 in all runs except one version of @xmath23 ( see below ) . the @xmath29 run uses the fiducial subgrid parameters . in @xmath30 and @xmath31 , mass , momentum , and energy are deposited to particles within the sph kernel according to a mass - weighting scheme , rather than volume - weighting as in the fiducial runs . in addition to this change , @xmath30 was run to test equal softening lengths for the dark matter and the gas ( 25 pc ) . all three runs of @xmath23 end up with stellar masses in the central galaxies that differ by only @xmath53 , suggesting that the major differences between the runs are more a result of stochasticity than the changes to the feedback implementation and softening length ( see section [ sec : sample ] ) . we use the amiga halo finder ( ` ahf ` , * ? ? ? * ) to identify gravitationally bound dark matter , stellar and gas particles in each snapshot of the simulation . because pure particle - matching algorithms generally underperform relative to more sophisticated methods in terms of detection and removal of spurious interlopers between snapshots @xcite , we track particle overdensities between simulation snapshots using the ` consistent - trees ` software @xcite . this is particularly important given our focus on subhalos . because the current stable version of ` consistent - trees ` which makes use of the ` rockstar ` @xcite halo finder does not track baryonic particles , we build a pipeline between ` ahf ` and ` rockstar ` that allows for a seamless transition between the ` ahf ` halo catalogs and the merger trees this pipeline uses the halo phase space information combined with the dark matter particle ids to match each halo in the catalogue to a ` consistent - trees ` merger tree . ` rockstar ` and ` consistent - trees ` were used to determine the peak virial mass and infall times for subhalos . all other halo , galaxy and stellar quantities were obtained from the ` ahf ` halo catalogs or the raw particle files . all runs were initialized and simulated assuming the wmap-7 cosmology @xmath54 , @xmath55 , @xmath56 , @xmath57 , @xmath58 and @xmath59 @xcite . further detail on the fire/`gizmo ` code can be found in @xcite . [ cols="<,^,^,^,^,^,^",options="header " , ] [ tab : tablebig ] figure [ fig : dm_hists ] shows the dark matter distributions of all six runs in greyscale with the stellar mass density overlaid in color . the physical scales of the panels are identical : the white line in the lower left corner of each visualization indicates @xmath60 . for our detailed analysis , we focus on objects with @xmath61 in order to avoid effects from spurious star formation at our resolution limit . the gas density required to form stars in our simulations is quite high and can only occur in dense , molecular , self - gravitating regions . this onerous minimal criteria suggests that the formation of this minimum stellar mass represents a physical star formation episode . requirement have at least 16 star particles and had @xmath62 dark matter particles at peak mass . all but one had @xmath63 dark matter particles at peak mass , corresponding to typical initial baryonic particle numbers of @xmath64 before feedback dispelled most of these . ] furthermore , we find that below a subhalo mass of @xmath65 our results become far less stable against the stochasticity in star formation ( see below ) . we therefore consider only as robust those objects that form within halos that have a minimum dark matter particle number @xmath66 and refer to satellites that meet both this requirement and the stellar mass cut as massive satellites " . @xmath22 has two satellite galaxies that meet both of these requirements . only one subhalo the most massive in each of the runs of @xmath23 forms a massive satellite " . in @xmath30 there is a second satellite that meets the stellar mass but not the dark matter particle number cut . because all three runs of @xmath23 use identical initial conditions , each subhalo in @xmath30 has a corresponding subhalo in the other runs that shares a majority of the dark matter particles . the subhalos corresponding to this second satellite host galaxies below the stellar mass cut in @xmath29 and @xmath31 . in figure [ fig : dm_hists ] , stars are shown in this particular subhalo for all three runs of @xmath23 . importantly , although the three subhalos that host the second satellites of @xmath23 are effectively identical , they produce galaxies that differ by a factor of @xmath67 in stellar mass due to only minor changes in parameters between the runs . this motivates the fairly large dark matter particle number we require to consider a system well resolved . @xmath68 and @xmath69 for the central galaxy ( central ) , most - massive satellite ( 1st satellite ) , and second - most - massive satellite ( 2nd satellite ) and @xmath31 , the 2nd satellite is the second most massive satellite in stellar mass . in these two runs only , the 2nd satellites are selected as the satellites that correspond to the high stellar mass outlier in run @xmath30 . only the 1st satellites in @xmath23 runs qualify as massive satellites " . ] in all six runs can be found in table 1 . in the tables , figures and throughout the text , @xmath68 is @xmath70 for centrals and @xmath71 for satellites . values in parentheses indicate satellite systems that we regard as too poorly resolved to trust for detailed analysis . they are included here for completeness . the 2nd satellites of all @xmath23 runs fall into this category , as well as all satellites of the isolated ultra - faints @xmath26 and @xmath27 . in figure [ fig : dm_hists ] , the dark matter distribution is shown for all dark matter particles located within @xmath72 of the virial radius of the halo listed in each panel . star particles are only shown if they reside within any halo that currently has @xmath73 in stars in the inner @xmath74 of a central ( satellite ) galaxy or are in a corresponding halo in the other two runs of @xmath23 . many of the other subhalos also contain star particles ( see figure 2 ) , but they are not shown so that we may focus on systems that are reasonably well resolved . the radial extent of the galaxies was chosen by inspecting their stellar mass profiles ( each central galaxy also has a _ very _ diffuse stellar halo that we aim to avoid ) . the galaxy stellar masses are all computed using these radial limits . in all six runs , the dark matter halos hosting the central galaxies contain significant substructure . the leftmost panel of figure [ fig:3panel ] shows the current subhalo mass function for all six runs . we plot all subhalos with @xmath75 , which corresponds to @xmath76 . @xmath22 , with the highest virial mass , has both the most and the most massive subhalos . the subhalo mass functions for the three runs of @xmath23 lie predictably on top of each other . as expected , the two isolated ultra - faints " ( @xmath77 ) form far fewer subhalos ( at fixed mass ) than their more massive counterparts . the satellite stellar mass functions are shown in the middle panel of figure [ fig:3panel ] . they are intriguingly steep , rising to as many as fifty tiny satellites in the case of @xmath22 . however , the majority of these star - forming satellites form only a single star particle , and this potentially exciting behavior will need to be confirmed at much higher resolution . in order to make this clear , the dark green shaded region shows the average stellar mass of a single star particle , while the light green shaded region shows the mass range for all other objects that form with @xmath78 ( fewer than 16 star particles ) . in all six simulations , a total of @xmath79 subhalos form at least one star particle , but a mere six of them form galaxies with @xmath61 ( and only five of these also meet the @xmath80 requirement ) . it is these , the most massive satellites , that will be the focus of our analysis ( along with the centrals themselves ) . the @xmath81 relation for the five resulting massive satellites " ( filled triangles ) , their less - well - resolved counterparts ( open triangles ) , and all central galaxies ( filled circles ) , is shown in the right panel of figure [ fig:3panel ] . plotted are the six central galaxies along with the 1st and 2nd satellites of each host ( detailed in table 1 ) for each run . although 18 symbols are shown , we only regard the 11 filled points as reasonably well - resolved . all of the well - resolved systems in this work lie only slightly offset from the extrapolated abundance matching relationship from @xcite , which has been constrained down to a stellar mass of @xmath82 . this abundance matching relation , obtained by updating the @xcite relation with a faint - end slope derived from the measurements of the gama survey @xcite , accurately reproduces the stellar mass functions of the milky way and andromeda satellites , as well as the stellar mass function of the local field , when applied to the elvis simulations @xcite . we do not find the bending " of the @xmath21 relation reported by @xcite at @xmath83 . instead , the relation maintains a fairly steady power - law down to @xmath84 . we do see a qualitatively similar break to that seen by @xcite but at a significantly lower halo mass of @xmath85 . below this point it is clear that we are witnessing a large amount of stochasticity , the onset of which corresponds to the scale of extremely faint galaxies ( @xmath86 ) , as opposed to the results of @xcite , where stochasticity sets in near the masses of classical dsphs ( @xmath87 ) . however , it is in this regime where our resolution is severely limited these galaxies have @xmath88 star particles . at this time we are unable to determine whether the apparent break and associated stochasticity in our relation at small mass are physical or simply artifacts of resolution , although it is intriguing that the break witnessed by @xcite occurs at the mass of @xmath89 of their baryonic particles as well . future work at higher resolution will allow us to explore this question more fully . the most massive satellite of @xmath22 represents an interesting statistical rarity ( see section [ sec : elvis ] ) in that the mass of the subhalo hosting this satellite is just over @xmath90 of the mass of its host , and the satellite itself has @xmath91 of the stellar mass of its central galaxy . @xmath22 is clearly undergoing a major merger ( see figure [ fig : dm_hists ] ) and the associated large satellite has properties that are distinct from the other satellites described in this work . most notably , its stellar mass is two orders of magnitude greater than the other massive satellites " , and it s stellar and dark matter masses are both larger than those of the central hosts of @xmath26 and @xmath27 . for identification only ( with no relation to the structural properties of the satellite ) , we refer when necessary to this most massive satellite as dsph " , and all other satellites as ultra - faint " satellites . the broader term ultra - faints " is reserved for all galaxies , centrals and satellites , that have @xmath19 in our simulations . the top panel of figure [ fig : sfh_h12 ] shows the cumulative fractional star formation histories ( sfh ) for @xmath22 ( solid line ) and its two massive satellites " . the sfh of its large , dsph " satellite ( @xmath92 ) is shown by the dotted line . its second , ultra - faint " satellite ( @xmath93 ) is shown by the dashed line . the shaded band corresponds to the epoch of reionization in our simulations . the bottom panel presents similar histories for the three @xmath23 runs along with the single most massive ultra - faint " satellite ( @xmath94 ) that forms in each run ( dashed lines ) . in both panels , the short vertical lines at the top mark the first virial crossing of the satellites . the massive dsph " satellite of @xmath22 demonstrates a particularly interesting sfh . it stops forming stars shortly after the epoch of reionization for almost @xmath95 ( from @xmath96 to @xmath97 ) . however , it manages to keep some of its gas on hand in a state that is just below the threshold for star formation . upon infall into the virial radius of @xmath22 at a lookback time of @xmath98 , followed by pericentric passage @xmath99 later , its ism is compressed enough to create a new burst of stars , all of which are concentrated at the center of the satellite and were formed in situ . the central galaxy of @xmath22 itself also appears to experience a related ( albeit mild ) burst . this type of bimodal star formation history was suggested by @xcite as a signature of massive reionization fossils " . our dsph " satellite has a slightly higher @xmath100 than the range given in that work , and its rebirth of star - formation was triggered by a merger rather than by a change in its halo concentration ( as originally suggested by @xcite ) ; it is nevertheless similar in its qualitative nature . in our simulations , halos with @xmath101 appear to robustly maintain star formation to @xmath24 . this massive satellite at @xmath102 seems to mark a transition point where reionization begins to favor uniformly ancient star formation . the influence of its impending merger was enough to trigger a rebirth . the ultra - faint " satellites shown in figure [ fig : sfh_h12 ] all form in halos smaller than this , with @xmath103 . they all form their stars entirely before @xmath104 and are quenched over @xmath105 before their infall onto the larger hosts , indicating that their star formation is shut down by reionization rather than environmental processes . importantly , these subsequent mergers do not trigger fresh star formation , as was seen in the more massive satellite . these small halos are either quenched during reionization or shortly thereafter , running out of fuel after the fresh gas supply was shut off by the ambient uv field . this tendency for our ultra - faint galaxies in @xmath103 halos to quench early is illustrated more clearly in figure [ fig : sfh_ufds ] , where we have plotted the cumulative fractional star formation histories for all seven ultra - faints " formed in our simulations , e.g. all ultra - faint " satellites from figure [ fig : sfh_h12 ] ( dashed lines ) , along with @xmath26 and @xmath27 ( solid lines ) . we also include the unresolved 2nd satellite of @xmath30 ( dotted line ) due to it s high stellar mass and to illustrate its similar sfh . from figure [ fig : sfh_ufds ] it is clear that , in every case , reionization plays a significant role in the shutting down of star formation in our ultra - faints " . not only were @xmath106 of all stars in all seven of these objects formed before @xmath107 , four of the objects form over @xmath108 of their stars before the completion of reionization at @xmath109 . ) . short vertical lines along the top axes indicate infall times of the satellites . top : @xmath110 central ( solid ) along with its two satellites . the massive dsph " satellite is shown by the dotted line and it displays a late - time burst in conjunction with its recent infall . the ultra - faint " satellite ( dashed ) is quenched early , during reionization ( shaded band ) , well before it is accreted . bottom : @xmath111 centrals ( solid ) and the most massive ( ultra - faint ) satellites ( dashed ) in each run . [ fig : sfh_h12 ] ] ) . the sfh of the unresolved outlier that passes the stellar mass cut ( see section [ sec : sample ] for details ) is also shown ( dotted line ) . all seven of these systems have ancient stellar populations , similar to those observed for the ultra - faint galaxies of the milky way . [ fig : sfh_ufds ] ] it is beyond the scope of this paper to investigate a statistically significant sample of hydrodynamic simulations of @xmath25 dwarfs in order to estimate the frequency with which they will host satellites above a given mass . however , if we assume an @xmath69 - @xmath71 relation similar to that presented in figure 2 , we can make an estimate using dark - matter - only simulations . we do so using the exploring the local volume in simulations ( elvis ) suite of collisionless zoom - in simulations of local group - like environments @xcite . we select all isolated ( @xmath112 for all more massive halos ) dwarf - size dark matter halos ( @xmath113 ) in the 12 local group - like pairs and the 24 isolated elvis simulations , and determine the fraction of those hosts that have @xmath114 subhalos with @xmath115 . this is the halo mass that corresponds to @xmath116 according to figure 2 . forms a well - resolved satellite . although we focus here on satellites of isolated dwarfs , we note that this criteria would suggest the existence of @xmath117 undetected ultra - faint stealth galaxies " within @xmath118 of the milky way . ] we similarly compute the fraction of dwarf halos that have a very large subhalo with @xmath119 , set by the most massive satellite of @xmath120 ) . figure [ fig : ngtn ] shows the probability for an isolated dwarf halo in elvis to have @xmath121 satellites of at least these two peak virial masses within a projected distance of @xmath122 ( the typical virial radius for our dwarfs ) as a function of @xmath123 . according to figure [ fig : ngtn ] , isolated halos with @xmath124 will have one or more subhalos that could host @xmath125 satellites about @xmath126 of the time . the likelihood that an isolated dwarf in the same mass range hosts a satellite as massive @xmath127 , however , is just under @xmath128 . using both sdss and the semi - analytic models of @xcite applied to the ms - ii simulation @xcite , @xcite show that the probability for a central galaxy to host a satellite a given fraction of its stellar mass , @xmath129 , decreases as a strong function of @xmath130 until @xmath131 , becoming independent of @xmath130 for centrals below the stellar mass of the milky way . although the stellar masses of the central galaxies in our simulations lie far below their stellar mass range ( @xmath132 ) , this decoupling of the probability from @xmath130 for satellites of dwarf galaxies allows us to make a comparison with the probabilities they find . their figure 2 suggests that for all central galaxies with @xmath133 , there is a @xmath134 chance that they host a satellite with stellar mass @xmath135 similar to our ultra - faints " . this is roughly consistent with the @xmath136 probability we find . they also find that the likelihood for a satellite as massive in proportion to its central as dsph " ( @xmath137 ) , is vanishingly small . again , according to their figure 2 and assuming that the lack of dependence of this probability on @xmath130 for dwarfs , there is only about a @xmath138 chance for a central to have such a massive satellite , which is again roughly consistent with the results shown in figure [ fig : ngtn ] . range @xmath139 with n or more subhalos with @xmath140 ( thick line ) and @xmath141 ( thin line ) that are found within @xmath122 of their host galaxy . in all of our hydrodynamic runs , all subhalos with @xmath142 have formed a galaxy with @xmath61 , while the most massive satellite , dsph " , forms in a subhalo with @xmath143 . according to these dark - matter only simulations , @xmath144 of isolated dwarfs should have a massive satellite " within their virial radii , but only @xmath145 should host a satellite as massive as dsph " . [ fig : ngtn ] ] our ultra - faint galaxies with @xmath146 all reside within low mass halos ( @xmath77 ) . they also have extremely diffuse stellar distributions , as expected for the ultra - faint , low dark - matter mass , stealth galaxies " discussed in b10 . figure [ fig : rhalf_v_l ] shows the projected ( 2d ) half stellar - mass radii , @xmath147 , of our simulated galaxies vs their total stellar mass . the central galaxies are shown as colored circles and the satellites as triangles ( the single unresolved satellite with @xmath61 is shown as an open triangle ) . the open black circles show observed milky way dwarfs @xcite . the black points with error bars are the half - mass radii and stellar masses of the eight recently reported ultra - faint satellite candidates from @xcite , and the single blue square with error bars is hydra ii from @xcite . the surface brightness detection limit for sdss is shown as a solid black line in the figure . it represents a constant peak central surface brightness for a plummer profile @xmath148 and corresponds to a surface brightness of @xmath149 for solar absolute magnitude @xmath150 and assuming a stellar mass - to - light ratio of @xmath151 . except for @xmath26 , all of our simulated ultra - faint " dwarfs have surface brightnesses fainter than @xmath152 , and would qualify as stealth galaxies " . ( see @xcite ) . while this suggests that we are dynamically well resolved , the effect on the baryonic component within this radius is harder to determine . in lower resolution runs , the satellites half - mass radii do vary by about a factor of two , but in both directions ( some are larger , some smaller ) . higher resolution simulations would be particularly useful for solidifying the expectations for the sizes of these galaxies . ] the dashed line in figure [ fig : rhalf_v_l ] shows a surface brightness of @xmath153 , a limit that will likely be achieved by upcoming surveys such as the large synoptic survey telescope ( lsst ) . once the full co - added lsst data is collected , satellites this faint should be able to detected out to @xmath154 over half the sky @xcite . , vs @xmath69 for all resolved galaxies in our simulations ( filled circles and triangles ; colors are the same as in figure [ fig:3panel ] ) , the one unresolved satellite with @xmath155 ( open triangle ) , as well as for observed milky way dwarfs ( * ? ? ? * open circles ) , the newly discovered des ultra - faint dwarf candidates ( * ? ? ? * black points with error - bars ) , and hydra ii ( * ? ? ? * blue square ) . the solid line represents a surface brightness limit of @xmath152 while the dashed line shows @xmath153 . both lines and the hydra ii data point assume a stellar mass - to - light ratio of 1 . satellites as massive as dsph " , which occur around @xmath156 of isolated dwarfs ( see figure [ fig : ngtn ] ) , should be currently visible , but most of the satellites lie just out of reach , and will visible only with future surveys such as lsst or potentially even des . [ fig : rhalf_v_l ] ] in the more immediate future , the dark energy survey ( des ) is already online . des will only be able to detect ultra - faint objects of this kind out to about @xmath118 , and is only slated to cover @xmath157 of the sky over its first 5 years . however , des contains within its footprint the dwarf galaxy phoenix . phoenix has a stellar mass of @xmath158 and is about @xmath159 away . it is falling into the milky way with a velocity of @xmath160 @xcite and is likely on first approach , meaning that it is a good analog to the isolated systems simulated in this work . this suggests that phoenix may be an excellent dwarf galaxy candidate to host an ultra - faint satellite detectable in the immediate future . it also has an advantage over more distant dwarfs like cetus or aquarius in that it is close enough that low luminosity satellites such as the ultra - faints " predicted in this work have a greater chance of being visible . although , according to figure [ fig : ngtn ] , there is only a @xmath144 chance that phoenix will host a subhalo massive enough to form a @xmath161 satellite , once the volume that lies between us and phoenix has been factored in , we find that the probability for finding an ultrafaint in its field increases to anywhere between @xmath162 according to the elvis simulations and depending on the specifics of abundance matching for @xmath163 satellites . in general , for future telescopes that have a limited number of pointings available , targeting the @xmath164 region around isolated dwarf galaxies should prove to be a much more efficient strategy than pointing into blank sky , as it should increase the chances of observing an ultra - faint satellite by @xmath8 . @xcite run a series of 12 zoom - in simulations of local group - like environments at three levels of resolution , both with baryons and with dark matter only . they run each simulation with and without a cosmic uv background that turns on sharply at @xmath165 , and argue that the onset of the ionizing background radiation sets the mass scale at which all halos become dark . their simulations shut down star formation in over @xmath166 of halos with present - day virial masses less than @xmath167 . their figure 2 does seem to show convergence in their highest resolution simulation run with reionization , but the gas particle mass in their highest resolution run ( @xmath168 ) is an order of magnitude higher than the total stellar mass formed in most of the the satellites that form in our simulations , indicating that they would fail to detect these galaxies . furthermore , their reionization turns on at a higher redshift than does ours ( @xmath47 ) , which may lead to a higher halo mass cutoff for star formation . @xcite run a zoom - in simulation of a small group of seven dwarf galaxies with a range of virial masses ( @xmath169 @xmath170 ) . they find that galaxies only form in the halos with present day @xmath171 , and of the halos that form galaxies , the two with @xmath172 only form stars long after the end of reionization . although they do not model @xmath173 cooling nor self - shielding in their simulations , they do approximate the effects of @xmath173 cooling and run a parallel simulation without a cosmic uv background . in their reionization - free simulation , all dwarfs form stars , but they show that the gas in the three low mass halos never reaches the column density required for self shielding . their baryonic particle mass is @xmath174 , the same particle mass we achieve in the low resolution runs of @xmath23 ( see @xcite for details ) . in our low resolution runs , each of the most massive subhalos forms an object with stellar mass similar to its high resolution counterpart , but at this resolution each satellite object consists of only @xmath175 star particles . while it is difficult to say conclusively that resolution is the main driver of their apparent halo mass limit for star formation , in general it has been the case that every time resolution is increased in simulations , the minimum virial mass for a halo that can form stars has approached lower values @xcite . in addition to enabling the formation of galaxies with @xmath69 lower than the low resolution baryonic particle mass , higher resolution simulations allow for a more accurate description of shielding , which will also affect the minimum halo mass that can form stars . furthermore , resolving the formation of dense substructures that collapse under self - gravity and become self - shielding requires resolving the jeans / toomre mass of the galaxies , which can be as low as @xmath176 ( and corresponds to a required force / gravitational softening of at least @xmath177 ) in baryon - poor dwarfs . our simulations do not escape the challenges of imperfect resolution , but because our particle mass is an order of magnitude lower than most simulations that attempt to form galaxies in low mass halos , we can push down the predicted low mass limit for the halos that can form ultra - faint satellites . rather than attempt to make a prediction for the low__est _ _ mass halo that can form a galaxy , we predict that ultra - faint dwarf galaxies _ can _ form in dark matter subhalos as low mass as @xmath65 . while we can not state with certainty that these galaxies are fully converged in our high resolution runs , the stellar masses of the satellites do show a better convergence between low and high resolution runs than do the centrals . ( see @xcite for a discussion of the convergence of the centrals ) . additionally , at these low masses , we still see sensitivity to spatial resolution and feedback ( e.g. the second satellite of @xmath30 ) . more and higher resolution runs of dwarfs with @xmath178 that vary feedback prescriptions , reionization onset and spectrum , and particle mass are needed to fully probe the low mass end of the simulated stellar mass function and the @xmath81 relation . in the @xmath13cdm paradigm , all dark matter halos , from those around giant galaxy clusters to those hosting ultra - faint galaxies , should be filled with subhalos . we have used ultra - high resolution ( @xmath179 ) simulations with the psph version of ` gizmo ` @xcite and the feedback in realistic environments ( fire ) prescriptions @xcite to predict that subhalos of isolated dwarf halos should form galaxies . most of the subhalos around dwarf galaxies are expected to be of low - mass ( @xmath180 ) and we predict that some should host ultra - faint galaxies with @xmath181 . if these tiny satellites are observed , it would provide evidence that dark matter substructure persists to very small scales , as predicted in the standard paradigm . using the dark - matter - only simulations of the elvis suite , we show that each isolated field dwarf ( @xmath182 ) galaxy in the local group has about a @xmath126 chance of hosting at least one satellite with @xmath116 . the extended @xmath164 regions around known field dwarfs in the local group should prove to be fruitful search areas for ultra - faint satellites , as each pointing towards a dwarf also contains all of the volume of the milky way dark matter halo along the line of sight to that region . the phoenix dwarf galaxy in particular is an excellent target due to its proximity to the milky way and the high probability that it is on first infall . although we consider only isolated dwarfs in this work , it is worth noting that some satellites of the milky way may also have their own satellites . @xcite show , using the elvis simulations , that approximately @xmath183 of milky way satellites ( but as many as @xmath184 depending on the infall time and mass of the group ) , fell in as a part of lmc - size groups , and that several of the recent des satellites are likely satellites of the lmc . a simple calculation for a rigid satellite assuming the common halo mass scale for milky way dwarf spheroidals ( @xmath185 ) orbiting a milky way - size host ( @xmath186 ) shows that fornax could potentially hold on to ultra - faint satellites residing within its inner @xmath187 . leo i and leo ii could each host ultra - faints out to @xmath188 , and sculptor out to @xmath189 , assuming each satellite is on first infall . this suggests that these galaxies might also be suitable candidates to host satellites of satellites " . we do not see a sharp cut - off or break in the @xmath81 relation , at least for @xmath140 . we do , however , see a sharp cut - off in halos that host galaxies with uniformly ancient stellar populations . the ultra - faint dwarfs in our simulations form most of their stars in the first billion years after the big bang and are subsequently deprived of the cold gas required for star formation due to the ionizing background radiation , and not by infall into a more massive dark matter halo . we predict that below a critical mass threshold ( @xmath190 , @xmath191 ) all galaxies are ubiquitously ancient and , unlike more massive galaxies ( @xmath192 ) that are nearly uniformly star forming in the field @xcite , both central and satellite ultra - faint galaxies should all be quenched . for satellites with @xmath193 , it is likely that infall time is one of the primary factors in quenching their star formation . massive satellites ( @xmath194 ) have been shown to have extremely long quenching timescales , probably due to the cutting off of the fresh gas supply after infall @xcite . slightly less massive galaxies ( @xmath195 ) are likely quenched over much shorter timescales by the tidal or ram - pressure forces they experience upon falling in as satellites @xcite . however , it is likely that the quenching of ultra - faints is completely independent of their infall time , and set by the timing of the onset of and the mass they had at reionization . the precise stellar masses of the ultra - faint " dwarfs in our simulations are likely also sensitive to the reionization redshift ( with an earlier onset of reionization likely to quench star formation in higher mass halos ) and may also be sensitive to star formation physics at very low metallicities . we plan to explore these dependencies in future work , but here we point out that it is possible to form ultra - faint satellites of regular dwarf galaxies and highlight that they are quenched by reionization rather than infall . the recent discovery of up to nine milky way dwarf satellite galaxy candidates in the southern sky by des @xcite has important consequences for our understanding of star formation in low mass galaxies . of particular interest is des j0344.3 - 4331 ( eridanus ii ) , because of its large distance from the milky way ( @xmath196 ) @xcite . at this distance , whether the satellite is on first infall or even headed out after a pericentric passage , it was likely accreted within the last @xmath197 . given that the initial age estimate for eridanus ii shows that it likely has an ancient stellar population ( @xmath198 ; * ? ? ? * ) , this would mean that it was quenched long before infall . if confirmed in follow - up observations , eridanus ii would be the first known ultra - faint galaxy shown to be quenched in the field , supporting our findings that , below a critical mass scale @xmath199 , all galaxies host ancient stellar populations quenched by reionization - related feedback , and not by environmental processes . this would be a clear example of a reionization fossil " , as first discussed by @xcite . we thank m. c. cooper for helpful discussions . this work used computational resources granted by nasa advanced supercomputing ( nas ) division , nasa center for climate simulation , teragrid and by the extreme science and engineering discovery environment ( xsede ) , which is supported by national science foundation grant number oci-1053575 and aci-1053575 , the latter through allocation ast140080 ( pi : boylan - kolchin ) . support for this work was also provided by nasa through _ hubble space telescope _ theory grants ( programs ar-12836 and ar-13888 ) from the space telescope science institute ( stsci ) , which is operated by the association of universities for research in astronomy ( aura ) , inc . , under nasa contract nas5 - 26555 . cw and oe were supported by _ hubble space telescope _ grants hst - ar-13921.002-a and hst - ar-13888.003-a , and sgk by the national science foundation grant ast-1009999 . support for pfh was provided by the gordon and betty moore foundation through grant # 776 to the caltech moore center for theoretical cosmology and physics , an alfred p. sloan research fellowship , nasa atp grant nnx14ah35 g , and nsf collaborative research grant # 1411920 . dk received support from national science foundation grant number ast-1412153 , funds from the university of california , san diego and xsede allocation tg - ast-120025 . n. et al . , 2002 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4836 , survey and other telescope technologies and discoveries , tyson j. a. , wolff s. , eds . , pp . 154164 s. , bessell m. , schmidt b. , francis p. , 2007 , in astronomical society of the pacific conference series , vol . 364 , the future of photometric , spectrophotometric and polarimetric standardization , sterken c. , ed .

observed non - thermal line widths in molecular clouds ( mcs ) , where all star formation in the galaxy takes place , point to the presence of supersonic turbulence in such regions ( falgarone & philips 1990 ) . the properties of the turbulent medium , such as mach number and magnetic field strength , may determine the products of the star formation process . as an important source of heating within molecular clouds ( stone et al . 1998 , hereafter s98 ) , turbulent energy dissipation may also play a role . much effort has been directed toward numerically simulating turbulent media in order to better understand the link between turbulence and star formation ( see elmegreen & scalo 2004 , maclow & klessen 2004 , and mckee & ostriker 2007 and references therein ) . as we have not yet identified the turbulent driving mechanism , there remain many unanswered questions about the evolution of molecular clouds . is the turbulence periodically re - energized , or does it simply decay away ? how much impact do magnetic fields have on the properties of the turbulence ? crutcher ( 1999 ) , using observations of zeeman splitting , found magnetic fields in some clouds strong enough that one can not safely neglect their effects . although it has been shown that magnetic fields do not appreciably lengthen the turbulent decay time ( s98 ; mac low 1999 ) , they do create anisotropy within the clouds ( e.g. vestuto et al . 2003 , hereafter v03 ; esquivel et al . 2003 ) , which may have important observational and evolutionary consequences . for example , molecular clouds are often observed to be filamentary ( e.g. mizuno et al . 1995 ; churchwell et al . 2004 ) . in this paper , we will investigate the energy dissipation properties of supersonic hydrodynamic and mhd turbulence with a variety of magnetic field strengths using athena , a new higher - order godunov code . an important goal of this study is to investigate the effect of the assumed driving mechanism on the properties of the resulting turbulence , such as power spectra and intermittency indicators . our analysis utilizes data from high - resolution numerical simulations with twenty - five different parameter sets . in recent years , a variety of results have been reported on the properties of supersonic mhd turbulence , including energetics ( e.g. s98 ; mac low 1999 ; ostriker et al . 2001 ) , power spectra ( e.g. cho & lazarian 2003 , 2005 ; v03 ; padoan et al . 2007 , hereafter p07 ) , and probability distribution functions ( e.g. padoan et al . 1997 ; passot & vazquez - semadeni 1998 ; ostriker et al . 2001 ; kowal et al . 2007 ; kritsuk et al . 2007 , hereafter k07 ) . where possible , we identify differences in our methods and results as compared to those of other groups . in a separate letter ( lemaster & stone 2008 , hereafter paper i ) , we have reported the results of an investigation of the variation of the probability distribution function ( pdf ) of density with mach number . our primary result in that paper was that the intermittent behavior of turbulence could be responsible for the large cloud - to - cloud variation in the observed star formation rate per solar mass . the primary tool available to investigate the properties of highly nonlinear , supersonic mhd turbulence is direct numerical simulation . to date , most results have been computed using a few numerical algorithms , such as zeus ( e.g. s98 ; v03 ; mac low 1999 ; ostriker et al . 1999 ; ostriker et al . 2001 ) , the pencil code ( e.g. haugen & brandenburg 2004 ) , the stagger code ( e.g. p07 ) , and eno methods ( e.g. cho & lazarian 2003 ) . there has been concern expressed in the literature that previous results may be strongly affected by numerical dissipation . thus , it is worth re - examining the problem with more accurate methods . this study represents one of the first applications of higher - order godunov methods to the study of supersonic mhd turbulence . without knowing the driving mechanism , we are left with many possible methods of generating turbulence in simulations . the hope is that the choice of method for the simulated driving will have an insignificant effect on the results . federrath , klessen , & schmidt ( in prep . ) , however , have shown that compressive and solenoidal forcing produce dramatically different turbulent states . for some diagnostics , such as intermittency , the time - variability of the turbulent state is critical . even if an array of driving methods produce the same mean state , do the instantaneous states have the same distribution about the mean ? we will investigate the dependence on driving scale of various diagnostics , given our particular driving method , which is very similar to that employed by , e.g. , s98 and v03 . for power spectra of simulated turbulence to be valuable , the resolution needs to be high enough that the driving and numerical dissipation scales are well separated , allowing the inertial range to be studied . when magnetic fields are taken into account , simulating turbulence at these resolutions can be prohibitively expensive . another goal of this study is to investigate whether properties of the power spectra and other diagnostics are converged at the numerical resolutions feasible at the moment , up to @xmath0 . we begin in [ sec : methods ] by describing our numerical methods in detail . we proceed , in [ sec : results ] , to present our results . [ sec : satresults ] includes a convergence study and mach number scaling study of saturation amplitudes , [ sec : specresults ] includes a power spectrum analysis , [ sec : sonicresults ] analyzes the sonic scale in @xmath0 runs , and [ sec : mittresults ] considers time - variability and temperature intermittency in the turbulence . finally , we summarize our results in [ sec : concl ] and discuss the implications . the simulations we present were conducted up to a resolution of @xmath0 with the athena code ( gardiner & stone 2005 , 2008 ; stone et al . 2008 ; stone & gardiner 2008 ) on a three - dimensional cartesian grid of length @xmath1 with periodic boundary conditions . athena utilizes a higher - order godunov scheme which exactly conserves mass , momentum , and magnetic flux . we solve the equations of ideal isothermal mhd , @xmath2 @xmath3 and @xmath4 where @xmath5 and @xmath6 are the isothermal sound speed and pressure , respectively . because our method of driving the turbulence , described in the following paragraph , does not involve modifying the equations of mhd , we include no explicit forcing term here . we use an approximate nonlinear riemann solver ( hlld ; miyoshi & kusano 2005 ) for our mhd runs and an exact nonlinear riemann solver for our hydrodynamic runs . both our mhd and hydro simulations are integrated well past the turbulent saturation time using a directionally - unsplit van leer scheme ( stone & gardiner 2008 ) . further details of how we overcame the numerical difficulties associated with utilizing this method for turbulence can be found in appendix [ sec : godunov ] . we initialize a uniform , stationary ambient medium with density @xmath7 and magnetic field parallel to the @xmath8-axis whose amplitude @xmath9 is fixed by the value of @xmath10 . we then apply divergence - free velocity perturbations following a gaussian random distribution with a fourier power spectrum of the form @xmath11 for @xmath12 , where @xmath13 is the resolution and @xmath14 is the wavenumber of peak driving , in all but two of our runs . for the remaining two runs , with @xmath15 and @xmath16 at @xmath0 , we truncate the driving spectrum at @xmath17 . before applying the perturbations to the grid , we shift them such that no net momentum will be added and normalize them to give the desired energy injection rate , @xmath18 , or , in the decaying case , initial kinetic energy . for our driven runs , we choose our energy injection rate to give the desired turbulent mach number , @xmath19 , in the saturated state . here , @xmath20^{1/2}$ ] is the 3d velocity dispersion of the gas and @xmath21 is the 1d velocity dispersion in a given direction . the method we use in this paper to compute the turbulent mach number differs from that used by some other groups ; we utilize a density - weighted velocity dispersion , @xmath22^{1/2}$ ] . note that , due to the zero net momentum of our turbulent medium , the second term on the rhs vanishes on the global scale . if we compare the mach number computed with and without density - weighting the velocity dispersion , we find that the two are fairly correlated , but that the time - average of the latter is @xmath23 larger than that of the former for driven strong - field mhd turbulence with , e.g. , @xmath16 and @xmath15 , or @xmath24 and @xmath25 . the comparison for the latter case is shown in figure [ fig : machmeth ] . we will use a density - weighted velocity dispersion for all but our sonic scale analysis . to drive turbulence in our simulations , we inject energy before each time step , with a new realization of the power spectrum generated at intervals @xmath26 . this differs from that done in s98 and v03 , where the energy was injected only when a new realization of the spectrum was generated . ostriker et al . ( 1999 , 2001 ) and kowal et al . ( 2007 ) used different driving spectra than our own but still constrained their velocity perturbations to be divergence free . our method differs from that of k07 in that they approximate an isothermal equation of state using an adiabatic index of @xmath27 , drive their turbulence on very large scales using an acceleration that is fixed in time , and also allow a substantial fraction of the energy introduced by the forcing to be in compressional modes . p07 also used a fixed acceleration for turbulent driving . it is important to understand if these differences have any significant impact on the results . for our decaying runs , we again begin with a uniform , stationary ambient medium but apply only one driving impulse to the velocity field , with a power spectrum of the same form as is used to initialize our driven runs . we then allow it to evolve undisturbed . we choose the turbulent kinetic energy applied to our medium so as to give the desired initial turbulent mach number . our decaying runs differ from those of sytine et al . ( 2000 ) , who used a different initial driving spectrum and allowed a compressible component in the initial perturbations . they used an adiabatic equation of state with @xmath28 , which we will use only for our decaying simulations . in this case , we solve the total energy equation , @xmath29 = 0,\ ] ] in addition to equations ( [ eq : mhdeqn1])([eq : mhdeqn3 ] ) , using the hllc riemann solver . here , @xmath30 with @xmath31 and @xmath32 , where @xmath33 is the internal energy density and @xmath34 is the ratio of specific heats . the numerical methods in athena conserve total energy exactly . we primarily investigate strong - field mhd turbulence with @xmath35 and hydrodynamic turbulence ( @xmath36 ) , although we present runs with @xmath37 ( moderate - field ) and @xmath38 ( weak - field ) as well . note that , due to a definition of @xmath39 in s98 and v03 that differs from ours by a factor of 2 , our @xmath35 results should be compared to their @xmath40 results , and similarly for other values of @xmath39 . the magnetic fields we use in our simulations correspond to physical values of @xmath41 , where @xmath42 is the temperature and @xmath43 is the number density of molecular hydrogen . our simulations are scale - free , allowing them to be scaled to any set of physical parameters using appropriate choices of @xmath44 , @xmath45 , and @xmath46 . utilizing the same values given in s98 , i.e. @xmath47 , @xmath48 , and @xmath49 , yields energy injection rates of up to @xmath50 , with a magnetic field strength @xmath51 for the strong - field case . figure [ fig : den1024p2a ] shows slices in mass density along the far faces of the computational domain for driven @xmath52 strong - field mhd turbulence with @xmath16 at @xmath0 . also included are magnetic field vectors in a slice normal to the y - axis near the bottom of the cube . due to the strong background magnetic field , the vectors are fairly well aligned . figure [ fig : den1024p2d ] also shows slices in density , but this time for @xmath53 hydrodynamic turbulence with the same driving scale and resolution . for both cases the structure seen is filamentary , with anisotropy in the scale of the structures in the strong - field case as a result of the field this is not to say , however , that the filaments are aligned with the magnetic field . in fact , some appear to be oriented perpendicular to the field . figures [ fig : col1024p2a ] and [ fig : col1024p2d ] show column density along the line of sight parallel to the z - axis for the same runs as in figures [ fig : den1024p2a ] and [ fig : den1024p2d ] , respectively . the projected filamentary structure is visible in these column density images . the column density contrast is higher for the strong - field than for the hydrodynamic case . we begin our quantitative analysis by studying the energy in fluctuations once our driven turbulence runs have reached saturation . since the method with which we drive our turbulence injects energy at a constant rate , at saturation the energy dissipation rate of the turbulence will , on average , equal the energy injection rate . at sufficiently high numerical resolution , the numerical dissipation will become negligible compared to the physically interesting sources of dissipation ( shocks ) , and the energy dissipation properties of the turbulence can given us insight into the heating within and evolution of molecular clouds . in this section , we will first determine the resolution at which the saturation energy has converged , i.e. reached a state such that further increasing the resolution has a negligible effect on the state , for turbulence evolved using athena and compare the turbulent decay rates to those presented in s98 computed using zeus . the kinetic energy associated with the fluctuations in our turbulent medium can be quantified by @xmath54 , where @xmath55 is the density and @xmath56 is the magnitude of the velocity . an integration for this and all similar quantities is performed over the entire domain . similarly , the energy in magnetic field perturbations is @xmath57 , where @xmath58 and @xmath9 are the magnitudes , respectively , of the magnetic field and its initial value . the total energy in perturbations is the sum of the energy in kinetic and magnetic field fluctuations , @xmath59 . to analyze the energy dissipation properties of the turbulence , we compute dissipation timescales using @xmath60 , normalizing them to the flow crossing time , @xmath61 , where @xmath62 is the wavelength of peak driving . we can partition the kinetic energy by breaking down the velocity into its solenoidal and compressive components . the solenoidal component is divergence - free , whereas the compressive component is curl - free . these can easily be computed using @xmath63 \hat{k}$ ] and @xmath64 \times \hat{k}$ ] , respectively . for simulations with comparable parameters ( i.e. @xmath24 and @xmath25 ) , these energies can be compared directly with the values given in s98 and v03 after accounting for the difference in definition of @xmath39 . we average our quantities over at least one dynamical time , often several , beginning after the turbulence has fully saturated . to determine the rate of convergence of a given volume - integrated quantity , @xmath65 , we compute the percent error in the quantity at each resolution relative to the converged value . in the case where the quantity changes monotonically with resolution , we find the converged value by performing a three - parameter fit , @xmath66 , where @xmath13 is the resolution and @xmath67 is the value at that resolution . the results of this fit tell us ( 1 ) the converged value , @xmath68 , ( 2 ) the order of convergence , @xmath69 , and ( 3 ) the coefficient , @xmath70 , that determines the resolution at which our result has converged , i.e. when @xmath71 . to determine the resolution at which our numerical dissipation has become small compared to shock dissipation , we analyze the properties of driven , supersonic hydrodynamic turbulence at resolutions from @xmath72 to @xmath73 . we study two sets of simulations with identical energy injection rate , @xmath25 , but differing driving scales . the set of runs with small - scale driving , @xmath24 , correspond to the hydrodynamic run in s98 , while the other set of runs are driven at twice the scale , @xmath74 . the properties of the @xmath73 run from each set can be found in table [ tab : hydro ] . we have not performed a convergence study of the @xmath16 runs due to the high level of time - variability ( discussed in [ sec : varresults ] ) . cccccccrrr 8 & 1000 & 5.8 & 17 & 3.8 & 13 & 0.77 & @xmath75 & & + 4 & 1000 & 7.2 & 26 & 5.9 & 20 & 0.76 & @xmath76 & @xmath77 & @xmath78 + 4 & 375 & 5.3 & 14 & 3.1 & 11 & 0.78 & @xmath76 & @xmath77 & @xmath78 + 4 & 140 & 3.8 & 7.2 & 1.6 & 5.6 & 0.78 & @xmath76 & @xmath78 & @xmath78 + 4 & 40 & 2.6 & 3.3 & 0.7 & 2.6 & 0.84 & @xmath76 & @xmath79 & @xmath79 + 4 & 3.5 & 1.2 & 0.8 & 0.1 & 0.7 & 1.1 & @xmath75 & @xmath79 & @xmath79 + 2 & 500 & 7.0 & 25 & 5.5 & 19 & 0.7 & @xmath80 & @xmath78 & @xmath81 + 2 & 187.5 & 5.2 & 13 & 3.1 & 10 & 0.73 & @xmath82 & @xmath77 & @xmath81 + 2 & 70 & 3.7 & 6.9 & 1.6 & 5.4 & 0.74 & @xmath82 & @xmath78 & @xmath78 + 2 & 20 & 2.5 & 3.1 & 0.6 & 2.5 & 0.77 & @xmath76 & @xmath83 & @xmath83 + 2 & 1.75 & 1.2 & 0.7 & 0.1 & 0.6 & 0.99 & @xmath82 & @xmath81 & @xmath81 although both sets of runs are driven with energy injection rate @xmath25 , the larger driving scale of the @xmath74 set causes it to converge to @xmath84 , while the @xmath24 set converges to only @xmath85 . the former set reaches convergence by @xmath86 , with higher resolutions having a small scatter about the converged value . the latter set , on the other hand , converges monotonically at order @xmath87 for all resolutions analyzed , reaching within @xmath88 of the converged value by @xmath89 . the kinetic energy in fluctuations of the set with larger driving scale , @xmath74 , converges to a value of @xmath90 by @xmath86 and has some scatter about the converged value for higher resolutions . for the set with smaller driving scale , @xmath24 , the kinetic energy in fluctuations converges at order @xmath91 to only @xmath92 . the value at @xmath89 is within @xmath93 of the converged value , while at @xmath94 it is within a fraction of a percent . although we have an extremely low sampling rate for the fraction of the kinetic energy in solenoidal and compressive modes , they appear to be independent of driving scale . the solenoidal fraction converges to @xmath95 , decreasing with increasing resolution , while the compressive fraction converges to @xmath96 . the solenoidal fraction is within a fraction of a percent of the converged value by @xmath94 ; however , it varies by only a small amount down to low resolution . finally , the ratio of the energy dissipation timescale to the flow crossing time at the driving scale increases with resolution , to @xmath97 for the @xmath74 set and @xmath98 for the @xmath24 set , only a small difference . consistent with our previous results , the @xmath74 set converges at lower resolution than the @xmath24 set , @xmath86 and @xmath94 , respectively . the turbulence dissipates in less than a flow crossing time in all cases . although the converged mach number and saturation energies are higher for @xmath74 than for @xmath24 , the fraction of the energy in the solenoidal or compressive mode , as well as the ratio of the energy dissipation timescale to the flow crossing time at the driving scale , are relatively independent of the driving scale . while most of the quantities of interest converge between @xmath89 and @xmath94 for the @xmath24 hydrodynamic runs , convergence has already been reached for these quantities by @xmath86 for the @xmath74 runs . high resolutions are critical for separating the driving and dissipation scales in the power spectra far enough to study the inertial range ; however , quantities such as energy dissipation rate and turbulent mach number converge at resolutions which are more easily attainable . we now analyze driven , strong - field , supersonic mhd turbulence in the same manner as in the previous section . to determine the resolution at which our numerical dissipation becomes negligible , we study the convergence of two sets of mhd simulations with energy injection rate @xmath25 at resolutions from @xmath72 to @xmath73 . one set has @xmath24 , similar to the strong - field mhd run in s98 , while the other has @xmath74 . the properties of the @xmath73 run from each set can be found in table [ tab : mhd ] . again , we have not performed a convergence study of the @xmath16 runs due to the high level of time - variability ( discussed in [ sec : varresults ] ) . cccccccccrrr 2.0 & 8 & 1000 & 5.3 & 14 & 2.1 & 12 & 5.8 & 0.83 & @xmath75 & & + 0.2 & 8 & 1000 & 5.1 & 13 & 1.6 & 11 & 8.6 & 0.89 & @xmath75 & & + 0.02 & 8 & 1000 & 5.4 & 15 & 1.5 & 13 & 8.2 & 1.0 & @xmath75 & & + 0.02 & 4 & 1000 & 6.8 & 23 & 2.3 & 21 & 13 & 0.98 & @xmath75 & @xmath77 & @xmath78 + 0.02 & 4 & 375 & 5.2 & 14 & 1.1 & 13 & 6.4 & 1.1 & @xmath75 & @xmath79 & @xmath78 + 0.02 & 4 & 140 & 3.8 & 7.3 & 0.5 & 6.7 & 3.1 & 1.1 & @xmath99 & @xmath78 & @xmath79 + 0.02 & 4 & 40 & 2.7 & 3.6 & 0.2 & 3.4 & 1.4 & 1.3 & @xmath75 & @xmath83 & @xmath83 + 0.02 & 4 & 3.5 & 1.3 & 0.8 & 0.04 & 0.8 & 0.3 & 1.7 & @xmath75 & @xmath78 & @xmath78 + 0.02 & 2 & 500 & 7.0 & 24 & 2.2 & 22 & 11 & 0.98 & @xmath80 & @xmath81 & @xmath100 + 0.02 & 2 & 187.5 & 5.0 & 13 & 1.2 & 11 & 5.9 & 0.99 & @xmath99 & @xmath78 & @xmath78 + 0.02 & 2 & 70 & 3.7 & 6.8 & 0.6 & 6.3 & 2.9 & 1.0 & @xmath99 & @xmath81 & @xmath100 + 0.02 & 2 & 20 & 2.5 & 3.1 & 0.2 & 2.9 & 1.3 & 1.1 & @xmath82 & @xmath83 & @xmath81 + 0.02 & 2 & 1.75 & 1.2 & 0.8 & 0.06 & 0.7 & 0.4 & 1.6 & @xmath101 & @xmath102 & @xmath103 as in the hydrodynamic case , the @xmath74 mhd runs converge to a higher mach number , @xmath104 , than the @xmath24 runs ( @xmath105 ) . the former is converged by @xmath94 , while the latter converges at @xmath106 order , only reaching convergence at @xmath73 . the total energy in fluctuations converges at nearly @xmath106 order to @xmath107 for @xmath74 , shown in figure [ fig : mpert4res ] . while convergence has definitely been reached by @xmath73 , the @xmath94 value is less than @xmath93 from the converged value . for @xmath24 , on the other hand , convergence is not reached until @xmath73 , where @xmath108 . convergence is at nearly @xmath106 order for this case as well . for the energy in magnetic field perturbations , convergence is reached at @xmath73 for both the @xmath74 and @xmath24 cases . convergence is approached at roughly 0.7 order for both cases . the fraction of the total energy ( magnetic plus kinetic ) in magnetic field fluctuations increases with resolution , reaching @xmath109 at @xmath73 . when driven at @xmath74 , the kinetic energy in fluctuations converges to @xmath110 by around @xmath94 , while at @xmath24 it converges at order 0.9 to @xmath111 by @xmath73 . the fraction of the kinetic energy fluctuations in solenoidal modes increases with resolution for both the @xmath74 and @xmath24 cases , converging to @xmath112 . the fraction in compressive modes , on the other hand , decreases with resolution to @xmath113 . convergence for these quantities is not reached until @xmath73 . finally , the ratio of the dissipation timescale to the flow crossing time at the driving scale converges to @xmath114 by @xmath94 for @xmath74 , while for @xmath24 it converges to @xmath115 by @xmath73 . even with a strong magnetic field , the turbulence dissipates in a flow crossing time . just as in the hydrodynamic case , the increased driving scale caused a higher turbulent mach number and energy at saturation for the @xmath74 than for the @xmath24 case . the fraction of the kinetic energy in solenoidal or compressive modes , however , was independent of driving scale , just as before . the ratio of the dissipation timescale to flow crossing time was also relatively independent of the driving scale . convergence for strong - field mhd turbulence was reached at higher resolutions , typically @xmath73 , than for hydrodynamic turbulence . because we performed a convergence study using the same turbulence parameters as in s98 , we can directly compare our high - resolution results to those published therein . for driven strong - field mhd turbulence with @xmath24 , we find at @xmath94 that our total energy in fluctuations at saturation is only @xmath116 higher than that found in s98 , due to the lower level of numerical dissipation at this resolution in athena than in zeus . for intermediate- and weak - field mhd , our energies are @xmath117 and @xmath116 higher , respectively . it is likely that neither our results nor the s98 results have converged by this resolution , however . at @xmath73 , there is no obvious difference between the total energy in fluctuations for our strong - field mhd simulation and the s98 result . the values from these two codes converge , even though they utilize completely different numerical methods . the ratio of magnetic to kinetic energy fluctuations , however , is different for the two codes , yielding ratios of dissipation timescale to flow crossing time that differ more substantially . at @xmath94 , this ratio is @xmath118 greater for our strong - field mhd run than that presented in s98 . for intermediate- and weak - field mhd , our ratios are both @xmath119 greater . in all cases , the ratio of timescales remains below unity . for our driven hydro run at @xmath94 with @xmath24 , we find that our total energy in fluctuations at saturation is also only @xmath116 higher than that found in s98 . although our result has converged by this resolution , the zeus result may not yet be converged . the ratio of dissipation timescale to flow crossing time in this case is @xmath119 greater than that found in s98 , but is still below unity . although it has been suggested in the literature that the rapid decay of supersonic turbulence is due to excessive numerical dissipation in zeus , these results , computed with a higher - order godunov scheme , do not support that conclusion . to investigate the effect of the turbulent mach number on energies and dissipation rates in the turbulence , we now analyze two series of five driven , supersonic hydrodynamic turbulence simulations at @xmath73 . one series is driven at @xmath74 , while the other is driven at a larger scale of @xmath16 . the energy injection rates of the latter series , the same that we use in paper i , are half that of the former series , giving roughly equal mach numbers to the corresponding pairs . these mach numbers all fall within the range @xmath120 . the properties of these runs can be found in table [ tab : hydro ] . the kinetic energy in each series of runs spans 1.5 orders of magnitude . we find a power law relationship between the kinetic energy and energy injection rate , i.e. @xmath121^{-1/2 } ( { \ensuremath{\dot{e}/\bar{\rho}l^2c_s^3}})^{0.62}.\ ] ] the relationship between the kinetic energy and mach number is , of course , exactly @xmath122 since our mach numbers are computed from the _ density - weighted _ velocity dispersion and there is no net momentum associated with the turbulent medium as a whole . we find for our driving method that we can estimate the mach number that will result from a given energy injection rate using @xmath123^{-1/4 } ( { \ensuremath{\dot{e}/\bar{\rho}l^2c_s^3}})^{0.31}.\ ] ] while this equation gives an indication of how the dissipation scales with mach number , the exact relations will only apply to our unique driving method . for the runs driven at @xmath74 , the fraction of the kinetic energy in compressive modes varies by only a fraction of a percent among the runs with @xmath124 . these runs have @xmath96 . the fraction drops off only slightly for the @xmath125 run , but then drops substantially , to @xmath126 , for the @xmath127 run . for the runs driven at @xmath16 , the fraction for the @xmath53 and @xmath128 runs is also @xmath96 , while the @xmath129 and @xmath130 runs have a bit larger and smaller fractions , respectively . once again , the fraction for the @xmath127 run is substantially smaller , @xmath131 . it appears as though the fraction of the energy in compressive modes remains roughly the same except when the turbulence is only mildly supersonic , where the fraction is much lower . the ratio of the dissipation timescale to the flow crossing time at the driving scale remains below unity for all except the @xmath127 run at @xmath74 . while the ratio for the runs at @xmath74 with @xmath132 does nt vary much , when we increase the driving scale to @xmath16 the value does change with mach number . the ratio for the @xmath74 runs is always larger for a given mach number than is the ratio for the @xmath16 runs . finally , we analyze two series of five driven , supersonic , strong - field mhd turbulence simulations at @xmath73 to determine the effect of the turbulent mach number on the energies and dissipation rates . as before , one series is the same @xmath16 set of runs used in paper i , while the other is driven at @xmath74 with twice the energy injection rate , yielding pairs of simulations of roughly the same mach number within the range @xmath133 . the properties of these runs can be found in table [ tab : mhd ] . as in the hydro case , the kinetic energy in each strong - field mhd series spans nearly 1.5 orders of magnitude . the total energy in fluctuations , which also includes the energy in magnetic field perturbations , increases by an even larger amount from the lowest to highest mach number runs . the power law relationship between the mach number and energy injection rate is @xmath134^{-1/4 } ( { \ensuremath{\dot{e}/\bar{\rho}l^2c_s^3}})^{0.30},\ ] ] which leads to a slightly narrower range of mach numbers than does the relation for pure hydro . again , although this gives an indication of how energy dissipation scales with mach number , the exact relationship is unique to our driving method . the fraction of the kinetic energy in compressive modes for strong - field mhd is much less than for hydro . when @xmath16 , the fractions increase monotonically with mach number , from only @xmath135 for @xmath136 to @xmath137 for @xmath138 . when @xmath74 , however , the values are much less predictable . in this case , the @xmath128 and @xmath127 runs have nearly identical values when @xmath16 , whereas for hydro the fraction at low mach number was substantially smaller than the values for higher mach numbers . the ratio of the dissipation timescale to the flow crossing time at the driving scale decreases with mach number . while the values for high mach number are below unity , the low mach number values become as large as @xmath139 . the values for the highest mach numbers are roughly the same for both driving scales , but for the smaller driving scale they increase more quickly . the mach number has a strong influence on the dissipation rate of the turbulence , but even for @xmath140 the dissipation timescale does not exceed twice the flow crossing time . we next consider turbulent velocity power spectra . we compute the power spectrum ( ps ) of the velocity , @xmath141 , as well as that of its compressive and solenoidal components , @xmath142 and @xmath143 , respectively , in the same way as was done in v03 for the psd of the specific kinetic energy and its components . to generate the spherically - integrated compensated power spectra that we will present , we average @xmath144 within spherical shells and multiply by the volume within the shell , @xmath145 , to find @xmath146 , where @xmath147 . to determine the degree of anisotropy in mhd turbulence , we will also analyze cylindrically - integrated power spectra . we generate these spectra by averaging @xmath148 over cylindrical shells whose axes are oriented along the mean magnetic field direction . this yields @xmath149 , where @xmath150 is parallel to the mean magnetic field and @xmath151 is in the plane normal to the field . the majority of the power spectra we present in our figures are compensated divided by a power law to produce plots where the inertial range is very roughly horizontal , making small deviations from a power law easier to see . for each of our driven turbulence runs , we compute power spectra from many snapshots taken at regular intervals , averaging them together to minimize the effects of rare events . with the exception of the @xmath0 runs , we average over at least 69 snapshots spanning at least 4.6 dynamical times . while we averaged over 2.7 dynamical times for the @xmath0 hydro run , the mhd case is not sufficiently time - averaged . sytine et al . ( 2000 ) presented the power spectra of decaying , subsonic , adiabatic hydrodynamic turbulence at a range of resolutions . these power spectra demonstrate the formation of a feature known as the bottleneck effect . energy cascades down from larger scales but can not easily be dissipated , causing a build - up of small - scale power . we begin by verifying that athena can reproduce the bottleneck . these are the only set of adiabatic runs we will consider . the initial velocity perturbations in these decaying runs have the same form of power spectrum as in our driven runs with @xmath74 and are normalized to give an initial turbulent mach number of @xmath152 . after this impulse is given to the initially uniform ambient medium , it is allowed to evolve undisturbed until @xmath153 . figure [ fig : hdcpwr1k4res ] shows the uncompensated velocity power spectrum at this time for runs with resolutions from @xmath94 to @xmath0 . as expected , this plot appears very similar to figure 11 of sytine et al . since athena was designed to have low numerical dissipation , the bottleneck is strong in our simulations of mildly - compressible ( subsonic ) turbulence . analyzing the compressive and solenoidal components of the velocity separately , we find that the bottleneck appears quite strong in the ps of the latter while apparently absent in that of the former . this is not surprising as shocks directly couple large and small scales , allowing the compressive energy to bypass the turbulent cascade and be immediately dissipated . the absence of the bottleneck in the compressive component agrees with the results of porter et al . ( 1999 ) for mach 1 driven , adiabatic hydro turbulence . at the very highest wavenumbers in our simulations , the power in the compressive component flattens out . although this could be an artifact of the code related to its treatment of shocks , one should be wary of putting too much stock in the high-@xmath154 region of any power spectrum as effects such as aliasing ( introduced in the calculation of the power spectrum , not the fluid dynamics ) could substantially alter the power in that region ( see , e.g. , press et al . 1992 for a more detailed discussion ) . next we consider driven supersonic ( isothermal ) hydrodynamic turbulence . we compute velocity power spectra for runs with @xmath16 to maximize the separation between the driving and dissipative scales . to obtain a slightly higher turbulent mach number , @xmath155 , than before , we use an energy injection rate of @xmath15 . these are the same runs from the mach number scaling analysis of [ sec : hydmachsatresults ] . figure [ fig : hpwr1k2res ] shows our compensated time - averaged velocity power spectrum for resolutions of @xmath94 through @xmath0 . to align the dissipative range for simulations with different resolutions , we express wavenumber as a fraction of the nyquist value , @xmath156 for a simulation with resolution @xmath157 . at low resolution we see no inertial range , but by @xmath0 we have separated the driving and dissipative scales enough that we may be just starting to see an inertial range . there is a small range , @xmath158 , where we see roughly a @xmath159 power law , but we are still using a small amount of forcing up through @xmath17 so the slope in this range may be affected . we find a much longer power law range over @xmath160 , @xmath161 . although it may be argued that a slope this shallow must be due to a bottleneck , the range of scales with steeper slope is far too limited to conclude this definitively from our data . a higher resolution would be required to determine if this is actually the case . figures [ fig : hpwr1s2res ] and [ fig : hpwr1c2res ] show power spectra of the solenoidal and compressive components of velocity , respectively . for the solenoidal component , we find power laws @xmath162 and @xmath163 over the ranges @xmath158 and @xmath160 , respectively . for the compressive component , however , we find only @xmath164 over the range @xmath165 . in two independent @xmath73 simulations that use different random perturbations to seed and drive the turbulence , we find the slopes of the compressive component to agree with each other as well as that of the @xmath0 simulation to within a percent , suggesting that the length of our time - averaging is sufficient . the ratio of power in the compressive and solenoidal components of velocity is much higher than for subsonic turbulence , having the effect of slightly washing out the shallow feature in the velocity spectrum in the supersonic case . when comparing hydro turbulence with @xmath24 , we find more power at high wavenumbers for turbulence evolved using athena than using zeus . although it was stated in v03 that the bottleneck would not affect finite difference codes , plotting the compensated power spectrum shows some evidence of blending of the driving peak with a secondary bump or shallowing of the spectrum ( much less prominent than in our own ) . if this is indeed a bottleneck , this difference likely results from the greater numerical dissipation in zeus than in athena . analyzing driven strong - field mhd turbulence with @xmath16 and an energy injection rate of @xmath15 , shown in figure [ fig : mpwr1k2res ] , we again find no inertial range at low resolution . in this @xmath155 run , the velocity power spectrum is dominated by the solenoidal component , resulting in a substantial shallowing of the spectrum that is apparent in the higher - resolution runs . at @xmath73 , the slope of the spectrum over the interval @xmath166 is only slightly steeper than @xmath167 for the velocity and its solenoidal ( figure [ fig : mpwr1s2res ] ) component . we find spectral slopes in two independent @xmath73 simulations ( whose power spectra were averaged together to give that plotted in the figures ) that agree to within a few percent , suggesting that our time - averaging to obtain a reliable spectral slope . although we do not have a sufficiently long time - average to find a robust fit to the slope of the @xmath0 run , the slopes appear to be very similar . as before , the slope we have found is much shallower than is typically predicted predicted for the inertial range , while being steeper than that presented in k07 as the bottleneck spectral slope . we note that in the mhd case , the range over which we find a power law seems to increase with resolution , which one would not expect if the feature was due to numerical dissipation ( bottleneck ) . it has been argued by k07 that the quantity @xmath168 should have an inertial range power law of @xmath169 even for supersonic turbulence , but we see over a wide range of wavenumbers a @xmath170 law ( figure [ fig : mpwr1u2res ] ) . kurien et al . ( 2004 ) have shown for subsonic turbulence that , while the @xmath169 law is appropriate when the energy timescale dominates over the helicity timescale , a @xmath170 law is expected when the helicity timescale is non - negligible compared to the energy timescale , even when the relative helicity may seem insignificant . this should be investigated as a possible cause of the bottleneck effect in supersonic turbulence . over the range @xmath171 , the compressive component ( figure [ fig : mpwr1c2res ] ) of the @xmath73 run follows a power law @xmath172 . for @xmath0 run , the power spectrum appears to smoothly change slope , indicating that it is not a power law , but a longer time - average might change the shape . the power spectrum of the magnetic field at @xmath73 follows the power law @xmath173 over the range @xmath174 . at @xmath0 , the slope appears very similar , but with extra noise at the smaller wavenumbers due to the short time - average used . for all resolutions studied , the uncompensated power approaches a constant , non - zero value at the highest wavenumbers . this produces a very prominent upturn in the compensated spectra . a direct comparison for strong - field mhd with @xmath24 shows more power at high wavenumber from athena than from zeus for the velocity and its compressive and solenoidal components , just as was true for the hydro case . if the shallowing of our velocity spectrum is due to the bottleneck effect , then what appeared to be inertial range when v03 was published is likely affected by the bottleneck as well . figure [ fig : mpwrspec2d ] shows our cylindrically - integrated power spectra at @xmath73 . we find anisotropy in the power spectrum of magnetic field perturbations as well as the velocity and its solenoidal component . there is more power perpendicular to the mean magnetic field at a given wavenumber than parallel to it . parallel to the mean field , the power law is roughly @xmath175 , similar to that of the purely hydrodynamic case . perpendicular to the field , however , the slope is much more shallow , roughly @xmath170 . we find the power spectrum of the compressive component of velocity to be nearly isotropic , in contradiction to what was found in v03 . the velocity power spectrum of _ subsonic _ turbulence results from a conservative cascade of energy from large to small scales through interactions local in fourier space . it is commonly believed that the velocity spectrum of supersonic turbulence , because it too is a power law , must also result from such a cascade . it is commonly known , however , that the power spectrum of a discontinuity , or an ensemble of shocks as in burgers turbulence , also have velocity spectra that follow a power law , @xmath159 . in figure [ fig : blastcomprimh ] , we confirm that the compressive component of velocity in our @xmath94 supersonic , strong - field mhd turbulence run with @xmath16 has a spectrum similar to that of velocity in an mhd blast wave that was initially spherical . this calls into question the long - held assumption that the power spectrum seen in supersonic turbulence results from an energy cascade facilitated by interactions local in fourier space . it would seem that other diagnostics , such as structure functions , are necessary in order to determine the cause of the power law spectrum , either a fourier - space cascade as in incompressible turbulence , or an ensemble of shocks as in burgers turbulence . to get a clearer idea of which features in our power spectra are representative of the turbulence and which were introduced in the post - processing , we compare power spectra computed by multiple methods . our control spectrum will be computed as were the others presented in this section averaged over the cells within a shell and then multiplied by the volume of the shell , @xmath145 , which we will refer to as `` shell - averaged '' for the remainder of this section . our comparison will be to another commonly used method ( e.g. p07 ) , where the power is simply summed over the cells falling within the shell ( `` shell - summed '' ) . these methods differ due to the discretization of the grid , i.e. because the fourier - space volume occupied by the cells within a bin is not , in general , equal to the volume of the perfectly spherical shell that the bin represents . while the bins in our control spectrum are centered on integer values of @xmath176 , those in the alternative spectrum instead run _ between _ integer values . we use the time - averaged power spectrum of the driven strong - field mhd turbulence run at @xmath73 for our comparison , although the effect should be independent of the run being analyzed . when overplotting the spectra produced by these two methods ( see figure [ fig : psmethod ] ) , the most obvious difference is at low @xmath154 . although the shell - averaged spectrum is relatively smooth in the range @xmath177 , the other has a jagged shape over this same range resulting from its sensitivity to the number of cells falling within a bin , making the `` slope '' of the power spectrum in this region much less obvious . if we compare the shell - summed spectrum to one computed by the same method but with the bins shifted by half a bin width ( aligning these bins with our own ) , we find that shape of the spectrum at low @xmath154 changes considerably . if , on the other hand , we take the shell - averaged spectrum and compare it to one computed in the same manner but with bins shifted to align with those typically used in the shell - summed method , we see that , although the power in each bin does change , the shape and slope of the spectrum change very little . while neither method is right or wrong , we advise caution when choosing a binning method for power spectra ; differences in slope computed by different means are not necessarily indicative of different turbulent states . although molecular clouds have supersonic velocity dispersions on large scales , as one looks to smaller and smaller scales , turbulent compressions will become progressively weaker , at some point becoming subsonic . the length scale at which the rms velocity dispersion is equal to the sound speed is referred to as the sonic scale ( mckee & ostriker 2007 ) . we now investigate how the velocity dispersion varies on spatial scales larger than the sonic scale , where shocks are most important . this scaling can be determined observationally in the form of linewidth - size relations using many different methods . for example , brunt ( 2003 ) used principal component analysis ( pca ) to determine the linewidth - size relation within individual clouds . to determine the velocity dispersion at a given length scale , @xmath178 , where @xmath179 and @xmath180 is an integer , we divide our computational domain along each axis into @xmath181 sections , yielding @xmath182 sub - volumes . we compute the velocity dispersion in each sub - volume and then average over all sub - volumes at that scale . because crossing the sonic scale represents a change in the physics dominating the flow , i.e. waves steeping to form shocks , for this analysis only we will compute the velocity dispersion without mass - weighting , i.e. @xmath20^{1/2}$ ] , where @xmath183^{1/2}$ ] . figure [ fig : sonichyd ] shows the velocity dispersion versus length scale for the @xmath0 , @xmath184 driven hydrodynamic turbulence run with @xmath16 . since driving may affect the scaling relation , we truncate our driving spectrum at @xmath17 , where the driving has already dropped to only @xmath88 of the peak level , and consider only this and smaller scales . fitting to the data points falling between the sonic scale and the driving cutoff , a factor of eight in length , we find a clear power law of the form @xmath185 . this scaling index falls well within the range of indices found observationally by brunt ( 2003 ) . shown in figure [ fig : sonicmhd ] is the velocity dispersion versus length scale for the @xmath0 , @xmath184 driven strong - field mhd turbulence run with @xmath16 . this time , over the range of scales where the flow is supersonic , we find that the slope is not a power law . unlike in hydrodynamic turbulence , the sonic scale is not the only scale at which the dominant physics changes . in fact , we would expect the multiple wave families of mhd to lead to multiple transitions , and for strong fields , these transitions are widely separated . for example , parallel to the magnetic field , slow waves will travel at the sound speed , whereas fast waves will travel at the alfvn speed . perpendicular to the field , however , the slow waves will have zero velocity and the fast waves will travel with a speed @xmath186 . this is complicated further by the fact that the parallel and perpendicular directions are defined relative to the _ local _ magnetic field , not the mean field . as a result , we have no reason to expect a uniform power law between the sonic scale and driving cutoff . it is not clear to what extent this result is affected by dissipation . up to this point we have averaged quantities from multiple snapshots in order to minimize the effect of rare events on our results . our final set of diagnostics will instead be an analysis of the time - variability and temperature intermittency of the turbulent gas . we begin by analyzing the time - variability of some of the quantities we analyzed in [ sec : satresults ] . for our hydrodynamic runs with the smallest driving scale , @xmath24 , we find the time - variability of the saturation energy to be only half a percent at both @xmath94 and @xmath73 . time - variability of less than a percent agrees with the results in s98 even though they drove their turbulence impulsively . the time - variability quoted in k07 , however , is much higher . the k07 runs were driven at a much larger driving scale than our own . to investigate if this could be the reason for the discrepancy , we will also analyze the time - variability of our runs with larger driving scales . when we increase the driving scale to @xmath74 , we find the time - variability of the energy to be a full percent at both @xmath94 and @xmath73 . as both the driving scale and mach number are varied here as compared to the run we discussed above , this could be due to either . when we increase the driving scale further , to @xmath16 , without changing the mach number , we find the time - variability increases yet again , coming to @xmath93 at @xmath94 and approaching @xmath187 at @xmath73 . it would seem that the driving scale chosen has a significant impact on the time - variability of the energy . the values we find here are still less than that shown in the k07 figure , however . for strong - field mhd , we again find time - variability of about half a percent for the @xmath24 runs . increasing the driving scale to @xmath74 decreases the time - variability at @xmath94 but increases it a small amount at @xmath73 . increasing the driving scale further , to @xmath16 , yields time - variability of @xmath188 at @xmath94 and @xmath187 at @xmath73 , comparable to the hydro runs . when we compare the time - variability of the energy among hydro runs at @xmath73 with varied mach number that are driven at @xmath74 , we find that the relation is non - monotonic . with @xmath16 , however , we do see a monotonic relationship , with the time - variability increasing with mach number . for our strong - field mhd runs , we again find non - monotonicity at @xmath74 , while the time - variability increases with mach number at @xmath16 , with the exception of the lowest mach number run . a feature of turbulence that may have a strong influence on star formation is intermittency , dissipation that is highly localized in space and time ( mckee & ostriker 2007 ) . although such dissipation does not require shocks , it is worthwhile to study the shock contribution in supersonic turbulence . were we not assuming an isothermal equation of state , energy dissipation in these regions would lead to local heating and thus increased temperatures , which would be evident in observations . therefore , to study intermittency due to shocks , we analyze the maximum heating rate per unit mass , @xmath189 , a surrogate for temperature . since we expect the maximum heating rate , @xmath190 , to be strongly influenced by rare events , such as the interaction of multiple shocks , we consider the time - variability of this quantity . due to the discretization of the grid , however , we would expect a component of the time - variability to be due to grid - scale fluctuations . therefore , instead of considering @xmath190 directly , we consider the statistics of the high-@xmath191 tail of the pdf of the heating rate per unit mass . for each simulation that we consider , we compute the heating rate per unit mass , @xmath191 , in each cell over many snapshots . considering only the cells in each snapshot that compose the @xmath88 of material ( by mass or volume ) with the highest cell - averaged @xmath191 , we compute the mean value . following the time - variability of this value should allow us to ignore meaningless grid - scale fluctuations while studying the intermittency due to shocks . for driven hydrodynamic turbulence at @xmath73 , we find the low mach number runs driven at @xmath74 to have less than @xmath88 variability in the peak temperature , while the higher mach number runs with this same driving scale can have variability as large as @xmath93 but without a monotonic dependence on mach number . at the larger driving scale @xmath16 , however , the variability in the peak temperature always exceeds @xmath88 and can be as high as @xmath187 . it would seem that this , like the time - variability of the saturation energies , increases with driving scale . for our runs , the variability of the peak temperature as measured by the top @xmath88 of the volume never exceeds that from the top @xmath88 of the mass by more than a tiny amount . as we have very few data dumps with which to calculate the statistics , however , the quantitative behavior may not be robust . we find that , in @xmath73 strong - field mhd turbulence driven at @xmath16 , the variability of the peak temperature can be as large as @xmath192 . when driven at @xmath74 , however , the variability always decreases , never exceeding @xmath188 . the variability measured by the top @xmath88 of mass is typically larger than that from the top @xmath88 of volume , but there does not appear to be a strong dependence on mach number . more often than not , the variability in the strong - field mhd runs are larger than in the hydro runs at the same mach number . again , however , these statistics are computed from very few data dumps , making them subject to large errors . figure [ fig : mmitt4 ] shows the time - evolution of the peak temperature in the @xmath74 strong - field mhd run with @xmath25 , for which we have better than typical statistics , compared to the @xmath16 run with @xmath15 . the saturation energies and dissipation timescales we find support the conclusion of s98 that supersonic turbulence dissipates rapidly even in the presence of magnetic fields . the results of our godunov code agree with zeus , an operator - split method that relies on artificial viscosity to capture shocks , but reach convergence at slightly lower resolutions . at @xmath73 the difference in the total energy in fluctuations between athena and zeus for strong - field mhd is very small , indicating that these two codes converge to similar turbulent states at sufficiently high resolution . the convergence resolution of our simulations depends both on the driving scale and the presence of a magnetic field . while hydro turbulence driven at @xmath74 converges by @xmath86 , strong - field mhd turbulence driven at @xmath24 does not converge until @xmath73 . although very high resolutions are needed to capture the inertial range of the turbulence , lower resolutions are often adequate for studying the energy dissipation characteristics of the turbulence as well as other volume - integrated quantities . at high mach number , the ratio of the dissipation timescale to the flow crossing time at the driving scale increases with increasing magnetic field strength ; however , it does not exceed unity even for strong - field mhd turbulence . this ratio is independent of driving scale . the fractions of the kinetic energy in solenoidal and compressive modes are also independent of driving scale , with the compressive fraction generally being more than twice as high for hydrodynamic than for strong - field mhd turbulence . our spherically - integrated velocity power spectra for decaying , subsonic hydrodynamic turbulence show evidence of the bottleneck effect in the velocity power spectrum , consistent with the findings of sytine et al . we find more power at high wavenumber in our driven , supersonic hydro and mhd turbulence simulations than was found by v03 , but it is unclear , particularly in the mhd case , whether or not this is due to a bottleneck . resolutions exceeding @xmath0 will be necessary to draw firm conclusions about the slope of the inertial range . the cylindrically - averaged velocity power spectrum for driven mhd turbulence is very anisotropic ; it has a slope that approximates that of the hydrodynamic case parallel to the magnetic field , while perpendicular to the field it is much more shallow . the compressive component of velocity has an isotropic power spectrum , contrary to what was found in v03 . we find the compressive component of velocity in driven , supersonic mhd turbulence to have a power spectrum that is difficult to distinguish from the velocity spectrum of an initially - spherical mhd blast wave . this calls into question the long - held assumption that supersonic turbulence power spectra result from an energy cascade facilitated by interactions local in fourier space . the analysis of structure functions may be useful in determining the source of the power law spectrum in supersonic turbulence , either a fourier - space cascade as in incompressible turbulence , or an ensemble of shocks as in burgers turbulence . for hydrodynamic turbulence , we find a power law scaling of the velocity dispersion with spatial scale , @xmath193 , for scales where the velocity dispersion is supersonic . however , we find no such power law for strong - field mhd turbulence , where the sonic scale is not the only scale of interest . in this case , we find that the velocity dispersion drops off more rapidly than a power law as one approaches smaller scales . we see time - variability in the saturation energies comparable to that of the equivalent runs in s98 despite the impulsive driving employed therein . our time - variability increases when we apply our turbulent driving at larger scales , but remains lower than that shown in k07 even with @xmath16 . it is possible that the acceleration - based driving method of k07 is responsible for the difference . at this large driving scale , the time - variability increases with mach number for both hydro and strong - field mhd . the method used to drive the turbulence appears to have a substantial impact on the resulting turbulent state . further investigation should be conducted to determine how much of the increase in time - variability with driving scale is due to the limited range of scales over which an inverse cascade can occur . if the level of time - variability is determined to be a result of simulation setup and not representative of the physical system we are trying to simulate , better diagnostics or simulation methods need to be developed in order to quantify intermittency in turbulent media . although we have very poor statistics in our temperature intermittency analysis , it appears that this , too , increases with driving scale . considering the difference that the driving method makes on the results , it may be more realistic to study decaying turbulence instead of driving it arbitrarily . for identical turbulent data cubes , we find that the post - processing methods chosen significantly influence the results . the turbulent mach number changes by @xmath23 depending on the method used in its computation . the means of computing the power spectrum also has a strong influence on the power at wavenumbers overlapping with the inertial range . when comparing results published by different groups , one should keep in mind that this , as well as the code , driving method , and initial conditions , can affect the quantities being compared . although the saturation energies and energy dissipation characteristics of the turbulence converge at resolutions within our current computational capabilities , the power spectra appear to require much higher resolutions to provide valuable information . also considering that turbulent power spectra can be approximated by non - turbulent phenomena , it would seem that , for the time being , our focus should be on other diagnostics . in conducting these numerical simulations , we have made many simplifications in order to make the problem more tractable . these assumptions , however , may prove to significantly impact the results . future studies should consider non - ideal mhd in order to model low - ionization molecular clouds , a non - isothermal equation of state in order to study heating and cooling , and self - gravity in order to follow the collapse of the bound clumps that form in the turbulent medium . we thank eve ostriker for very productive discussion and tom gardiner for his invaluable assistance in bringing athena up to the task of successfully simulating supersonic turbulence . we would also like to thank alexei kritsuk , cristoph federrath , and alex lazarian for their feedback . simulations were performed on the ibm blue gene at princeton and on computational facilities supported by nsf grants ast-0216105 and ast-0722479 . the calculations presented in this paper were conducted using athena , a directionally - unsplit , higher - order godunov code . this code exactly conserves mass , momentum , and magnetic flux , as well as energy when assuming an adiabatic equation of state . the code captures shocks well and has a low level of numerical dissipation . although the full details of the algorithms and implementation can be found in the literature ( gardiner & stone 2005 , 2008 ; stone et al . 2008 ; stone & gardiner 2008 ) , we will briefly summarize here , noting any modifications made to the algorithms in order to successfully run the challenging problem of high mach number turbulence . the integration algorithm used in our calculations is a simple predictor - corrector scheme based on the muscl - hancock scheme of van leer ( 2006 ) , combined with the constrained transport method of evans & hawley ( 1998 ) to enforce the divergence - free constraint on the magnetic field ( i.e. the vl+ct algorithm described in detail in stone & gardiner 2008 ) . we find the additional diffusion associated with this scheme as compared to our ctu+ct algorithm ( described in detail in stone et al . 2008 ) to make it more robust to the strong rarefactions that arise in a highly turbulent medium . although the algorithm is formally only second - order accurate , we use third order ( piecewise parabolic ) spatial reconstruction , finding it to provide more accurate solutions in test problems due to smaller truncation error . for our isothermal hydrodynamics runs , we found there to be strong rarefactions within the turbulent medium for which an approximate riemann solver simply was not accurate enough , necessitating the use of an exact nonlinear riemann solver . for adiabatic hydrodynamics as well as isothermal and adiabatic mhd , we were able to use approximate nonlinear riemann solvers , namely hllc for the hydro case and hlld for mhd , more details about which can be found in stone et al . although we used our own adaptation of the adiabatic hlld solver of miyoshi & kusano ( 2005 ) for our isothermal mhd runs , we found it to produce turbulent states extremely similar to those from the isothermal hlld solver of mignone ( 2007 ) . although these riemann solvers are positive definite in 1d , it is not guaranteed that they will be so in multidimensions . in fact , we found that , under extreme conditions , the net mass flux out of a cell in our isothermal 3d turbulent medium occasionally exceeded the cell s initial mass . in the rare circumstance that this occurred , we recomputed the fluxes of all variables through all interfaces bordering such cells using first - order reconstruction . in our mach 7 strong - field mhd run at @xmath73 that uses @xmath16 , this affected only a fraction @xmath194 of the fluxes computed in the corrector step of the integrator . dropping to first order introduced enough diffusion in the immediate vicinity to keep the cell - averaged density positive , while having a negligible effect on the overall system . adding diffusion in this manner instead of enforcing a density floor maintains exact conservation . brunt , c.m . 2003 , apj 584 , 293 cho , j. & lazarian , a. 2003 , mnras 345 , 325 cho , j. & lazarian , a. 2005 , thcfd 19 , 127 churchwell , e. et al . 2004 , apjs 154 , 322 crutcher , r.m . 1999 , apj 520 , 706 elmegreen , b.g . & scalo , j. 2004 , araa 42 , 211 esquivel , a. , lazarian , a. , & pogosyan , d. 2003 , mnras 342 , 325 evans , c.r . & hawley , j.f . 1988 , apj 322 , 659 falgarone , e. , & philips , t.g . 1990 , apj 359 , 344 federrath , c. , klessen , r.s . , & schmidt , w. 2008 , in prep . gardiner , t.a . & stone , j.m . 2005 , jcoph 205 , 509 gardiner , t.a . & stone , j.m . 2008 , jcoph 227 , 4123 haugen , n.e.l . & brandenburg , a. 2004 , phrve 70 , 026405 kowal , g. , lazarian , a. , & beresnyak , a. 2007 , apj 658 , 423 kritsuk , a.g . , norman , m.l . , padoan , p. , & wagner , r. 2007 , apj 665 , 416 ( k07 ) kurien , s. , taylor , m.a . , & matsumoto , t. 2004 , phrve 69 , 066313 lemaster , m.n . & stone , j.m . 2008 , apjl 682 , l97 ( paper i ) mac low , m .- m . 1999 , apj 524 , 169 mac low , m .- & klessen , r.s . 2004 , rvmp 76 , 125 mckee , c.f . & ostriker , e.c . 2007 , araa 45 , 565 mignone , a. 2007 , jcoph 225 , 1427 miyoshi , t. & kusano , k. 2005 , jcoph 315 , 344 mizuno , a. , onishi , t. , yonekura , y. , nagahama , t. , ogawa , h. , & fukui , y. 1995 , apj 445 , l161 ostriker , e.c . , gammie , c.f . , & stone , j.m . 1999 , apj 513 , 259 ostriker , e.c . , stone , j.m . , & gammie , c.f . 2001 , apj 546 , 980 padoan , p. , jones , b.j.t . , & nordlund , a.p . 1997 , apj 474 , 730 padoan , p. , nordlund , a. , kritsuk , a.g . , norman , m.l . , & li , p.s . 2007 , apj 661 , 972 ( p07 ) passot , t. & vazquez - semadeni , e. 1998 , phrve , 58 , 4501 porter , d. , pouquet , a. , sytine , i. , & woodward , p. 1999 , phya 263 , 263 press , w.h . , flannery , b.r . , teukolsky , s.a . , & vetterling , w.t . 1992 , numerical recipes in c ( 2nd ed . ; cambridge : cambridge university press ) stone , j.m . , & gardiner , t.a . 2008 , newa , in press stone , j.m . , ostriker , e.c . , & gammie , c.f . 1998 , apj 508 , l99 ( s98 ) stone , j.m . , gardiner , t.a . , teuben , p. , hawley , j.f . , & simon , j. 2008 , apjs , in press ( arxiv0804.0402 ) sytine , i.v . , porter , d.h . , woodward , p.r . , hodson , s.w . , & winkler , k .- h . 2000 , jcoph 158 , 225 van leer , b. 2006 , comm . phys . 1 , 192 vestuto , j.g . , ostriker , e.c . , & stone , j.m . 2003 , apj 590 , 858 ( v03 )

theoretical calculations show that , in the presence of a perpendicular magnetic field , a two - dimensional electron gas ( 2deg ) should crystallize below a filling factor @xmath0@xcite . several experimental groups have reported transport measurements indicative of this electron crystallization when the filling factor of the lowest landau level is decreased below @xmath1 these measurements include the observation of a strong increase in the diagonal resistivity @xmath2 non - linear @xmath3 characteristics , and broadband noise . all these observations have been interpreted as the pinning and sliding of a wigner crystal ( wc)reviewexperimentwc . moreover , microwave absorption experiments@xcite have also detected a resonance in the real part of the longitudinal conductivity , @xmath4 , that has been attributed to the pinning mode of a disordered wigner crystal . the vanishing of the pinning mode resonance at some critical temperature @xmath5 has been used to derive the phase diagram of the crystal@xcite in the quantum regime where the kinetic energy is frozen by the quantizing magnetic field . similar microwave absorption experiments also showed a pinning resonance at higher filling factors close to @xmath6 where the formation of a wigner solid is expected in very clean sampleschen1,lewis1,lewis2 . finally , in landau levels of index @xmath7 , a study of the evolution of the pinning mode with filling factor reveals several transitions of the 2deg ground state from a wigner crystal at low @xmath8 to bubble crystals with increasing number of electrons per lattice site as @xmath8 is increased , and into a modulated stripe state ( or anisotropic wigner crystal ) near half filling@xcite . in earlier works@xcite , some of us have studied several crystalline states of the 2deg using a combination of hartree - fock ( hfa ) and generalized random - phase approximations ( grpa ) . in these works , the energy and order parameters of the crystal were calculated in a self - consistent hfa while the collective excitations were derived from the poles of density response functions computed in the grpa . this microscopic approach ( hfa + grpa ) works well at zero temperature but is difficult to generalize to consider finite temperature effects or to include quantum fluctuations beyond the grpa . finite - temperature or quantum fluctuations effects ( not already included in the grpa ) , are most easily computed by writing down an elastic action for the system . for a crystaline solid , this requires the knowledge of the dynamical matrix ( dm ) or , equivalently , of the elastic coefficients of the solid . a direct way to obtain these elastic coefficients is to compute the energy required for various static deformations of the crystal . using elasticity theory , each deformation energy @xmath9 can be written in the form @xmath10 where @xmath11 is a parameter characterizing the amplitude of the deformation and @xmath12 is generally a combination of elastic constants . in the limit @xmath13 , one can obtain the elastic coefficients by computing the deformation energy of one or more static deformations and using the known symmetry relations between the elastic constants . alternatively , one can obtain a dm from the grpa density response function much more directly without the need to compute the elastic coefficients @xcite . in this paper , we compare the dm obtained from these two methods ( deformation energy and grpa ) in order to find the range of validity as well as the limitations of the gpra approach . we first consider the simple case of an isotropic ( triangular ) wigner crystal before tackling the more complex anisotropic wigner crystal @xcite or stripe phase that occurs near half - filling in the higher landau levels . we show that although the grpa method gives a good description of the qualitative behavior of the dm as a function of filling factor , its quantitative predictions must be used with caution . as we show below , an averaging procedure must be applied to the method in order to obtain a dm in the grpa that compares favorably with the one obtained by computing the deformation energy . our paper is organized as follows . in section ii , we define the elastic constants needed to build an elastic model for the wigner and stripe crystals . we then explain , in section iii , how these elastic constants can be derived by computing the deformation energy of the crystals in the hfa . in section iv , we summarize the grpa method of obtaining the dynamical matrix . our numerical results for the wc are discussed in section v and those for the stripe crystal in section vi . section vi contains our conclusions . we describe the elastic deformation of a crystal state by a displacement field @xmath14 defined on each lattice site @xmath15 . the fourier transform of this operator is given by : @xmath16where @xmath17 is the number of lattice sites . in two dimensions , the general expression for the deformation energy of a crystal requires the use of @xmath18 elastic coefficients @xmath19 and is given , in the continuum limit , by the following expression@xcite : @xmath20 \label{m4 } \\ & + \frac{1}{2}\int d\mathbf{r}\left [ 2c_{12}e_{x , x}e_{y , y}+4c_{16}e_{x , x}e_{x , y}+c_{22}e_{y , y}^{2}\,\right ] , \notag\end{aligned}\ ] ] where @xmath21 is the symmetric strain tensor . the wigner and bubble crystals have a triangular lattice structure for which the following equation holds : @xmath22 for such a triangular structure , the elastic energy in the long - wavelength limit can be written in a form that contains only two elastic coefficients , namely : @xmath23 . \label{m6}\end{aligned}\ ] ] the anisotropic stripe state can be seen either as a centered rectangular lattice with two electrons per unit cell or as a rhombic lattice with one electron per unit cell with reflection symmetry in both @xmath24 and @xmath25 axis . the deformation energy is given by @xmath26 \notag\\ & + \frac{1}{2}\int d\mathbf{r}\left [ 2c_{12}e_{x , x}e_{y , y}+c_{22}e_{y , y}^{2}\,\right ] . \end{aligned}\ ] ] in this paper , we assume that the stripes are aligned along the @xmath25 axis . the above formulation of elasticity theory assumes short - range forces only . for the electronic crystals that we consider , these forces are of coulombic origin i.e. the hamiltonian of the crystal contains only the coulomb interaction between electrons and the kinetic energy which is frozen by the quantizing magnetic field . both the direct ( hartree ) and exchange ( fock ) terms are considered by the hartree - fock approximation as we explain in the next section . to take into account in the elasticity theory the long - range part of the coulomb interaction present in a crystal of electrons , it is necessary to add to @xmath27 the deformation energy @xmath28 given by@xmath29 where @xmath30 is the area of the crystal , @xmath31 is the fourier transform of the change in the electronic density and @xmath32 is the dielectric constant of the host semiconductor . we consider the positive background of ionized donors as homogeneous and inert so that no linear term in @xmath33 is introduced by the coulomb interaction . to define a dynamical matrix , we assume that the crystal can be viewed as a lattice of electrons with static form factor @xmath34 on each crystal site ( with the normalisation @xmath35 ) . the time - dependent density can then be written as @xmath36and , to first order in the displacement field , we have for a density fluctuation @xmath37 where @xmath38 is a reciprocal lattice vector and @xmath39 a vector in the first brillouin zone of the crystal . it follows that we can write the coulomb energy as@xmath40where @xmath41 is the average electronic density . we pause at this point to remark that the form factor @xmath42 ( with @xmath43 the magnetic length , @xmath44 being the applied magnetic field ) in eq . ( [ defo ] ) renders the summation over the wavectors rapidly convergent . our hartree - fock calculation of the ground - state energy of the electronic crystals as well as our grpa calculation of the dynamical matrix also involve summations over reciprocal lattice vectors @xmath38 of some functions weighted by @xmath45 . if the magnetic field is not too strong , we can perform these summations directly . there is no need to use ewald s summation technique as is the case if one works with a crystal of point electrons . of course , as the filling factor @xmath46 , the magnetic length @xmath47 so the electrons behave more and more like point particles and the convergence is lost . in all cases that we consider , the summations involved are rapidly convergent because we restrict ourselves to filling factors @xmath48 where @xmath49 is sufficiently large for @xmath50 to be small ( @xmath51 being the lattice constant ) . the cutoff in @xmath52 is choosen so that the summations are evaluated with the required degree of accuracy . the total deformation energy , which we now write as @xmath53 now includes the long - range coulomb interaction and can be written in the form : @xmath54 where we have introduced the dynamical matrix : @xmath55for the triangular lattice , a comparison of eqs . ( [ m18 ] ) and ( [ m6 ] ) gives the dynamical matrix ( to order @xmath56 ) as : [ m32 ] @xmath57 , \\ d_{x , y}\left ( \mathbf{k}\right ) & = & n_{0}^{-1}\left ( \widetilde{c}_{12}\left ( k\right ) + c_{66}\right ) k_{x}k_{y } , \\ d_{y , y}\left ( \mathbf{k}\right ) & = & n_{0}^{-1}\left [ \left ( \widetilde{c}% _ { 12}\left ( k\right ) + 2c_{66}\right ) k_{y}^{2}+c_{66}k_{x}^{2}\right ] .\end{aligned}\ ] ] the long - range coulomb interaction renders the elastic coefficient @xmath58 ( but not the shear modulus @xmath59 ) nonlocal , so that @xmath58 contains a diverging term @xmath60 . we shall write : @xmath61 where @xmath58 is the weakly dispersive part of the elastic coefficient , and where the plasmonic ( first ) term on the _ rhs _ is due to the long - range nature of the coulomb interaction . for the stripe state , eq . ( [ triangulaire ] ) is no longer valid . in addition , all three elastic coefficients @xmath62 become nonlocal . we have in this case : @xmath63 \nonumber\\ & + & n_{0}^{-1}kk_{y}^{4 } , \\ d_{x , y}\left ( \mathbf{k}\right ) & = & n_{0}^{-1}\left ( \widetilde{c}_{12}\left ( k\right ) + c_{66}\right ) k_{x}k_{y } , \\ d_{y , y}\left ( \mathbf{k}\right ) & = & n_{0}^{-1}\left [ \widetilde{c}_{22}\left ( k\right ) k_{y}^{2}+c_{66}k_{x}^{2}\right ] , \end{aligned}\ ] ] where @xmath64 , with @xmath65 ; and where we added to @xmath66 a term @xmath67 in order to take into account the bending rigidity of the stripes which , due to the small value of the shear modulus @xmath59 in these systems , is quantitatively important over a sizeable region of the brillouin zone@xcite . using the fact that @xmath68 , we finally obtain : @xmath69 we now want to discuss how one can evaluate the non - dispersive part @xmath19 of the elastic coefficients . this will be the subject of the follwing section . in the hartree - fock approximation , a crystalline phase is described by the fourier components @xmath70 of the average electronic density , where @xmath38 is a reciprocal lattice vector . in the strong magnetic field limit where the hilbert space is restricted to one landau level , it is more convenient to work with the guiding - center density @xmath71 which is related to @xmath70 by : @xmath72 where @xmath73 is the landau - level degeneracy and @xmath74is the form factor of an electron in landau level @xmath75 ( @xmath76 being a generalized laguerre polynomial ) . the magnetic field @xmath77 is perpendicular to the 2deg . the hartree - fock energy per electron in the partially filled landau level is given by@xcite : @xmath78 \left\vert \left\langle \rho \left ( \mathbf{g}\right ) \right\rangle \right\vert ^{2 } , \label{energiehf}\]]where the @xmath79 term in this equation accounts for the neutralizing background of the ionized donors . the parameter @xmath80 is the filling factor of the partially filled level , and we take all filled levels below @xmath75 to be inert . the hartree and fock interactions in landau level @xmath75 are defined by : @xmath81 ^{2 } , \\ x\left ( \mathbf{q}\right ) & = & \left ( \frac{e^{2}}{\kappa \ell } \right ) \sqrt{2% } \int_{0}^{\infty } dx\,e^{-x^{2 } } \\ & & \times \left [ l_{n}^{0}\left ( x^{2}\right ) \right ] ^{2}j_{0}\left ( \sqrt{2}% xq\ell \right ) , \notag\end{aligned}\ ] ] where @xmath82 is the bessel function of the first kind . to compute the @xmath83 , we first write this quantity in second quantization and in the landau gauge @xmath84 as : @xmath85 the average values @xmath86 are obtained by computing the single - particle green s function ( here and in what follows , @xmath87 denotes the time ordering operator ) : @xmath88whose fourier transform we define as : @xmath89so that : @xmath90 we use an iterative scheme to solve numerically@xcite the hartree - fock equation of motion for @xmath91 for the undeformed lattice , we use the basis vectors : @xmath92 where @xmath93 is the aspect ratio and @xmath94 is the angle between the two basis vectors . for the triangular lattice , @xmath95 and @xmath96 . if we apply an elastic deformation @xmath97 to the lattice , the new lattice vectors are given by @xmath98 ( where @xmath99 are integers ) . we can write this expression as @xmath100 if we define the new basis vectors as : @xmath101 the parameters @xmath102 and @xmath103 are functions of the original lattice and of the type of deformation considered . the reciprocal lattice vectors of the deformed lattice are easily computed once these parameters are known . then , the cohesive energy @xmath104 of the deformed lattice can be calculated using the deformed reciprocal lattice vectors and eq . ( [ m7 ] ) . under these circumstances , we find that the deformation energy per electron is given by : @xmath105to find the elastic coefficients for the wigner and stripe crystals , we need to consider the following deformations ( note that the magnetic field and the number of electrons are kept fixed@xcite ) : _ ( i ) _ a shear deformation with @xmath106 and @xmath107 : the strain tensors in this case are given by @xmath108 , and @xmath109 . the area of the system , @xmath30 , is not changed by this deformation and the elastic energy is given by @xmath110 . it then follows that the shear modulus @xmath59 is given by : @xmath111where @xmath112 is the deformation energy per electron . the parameters of the distorted lattice for this shear deformation are given by : @xmath113 _ ( ii ) _ a one - dimensional dilatation along @xmath114 , with @xmath115 and @xmath107 : here , the strain tensors @xmath116 and @xmath117 , and the new area of the system is @xmath118 and @xmath119 . it then follows that the compression constant @xmath120 is given by : @xmath121 while the parameters of the deformed lattice are given by : @xmath122 the surface of the deformed lattice is @xmath123 , so that the filling factor is now given by @xmath124 . _ ( iii ) _ a one - dimensional dilatation along @xmath125 with @xmath126 and @xmath127 : now the strain tensors @xmath128 and @xmath117 . the new area of the system is @xmath118 and @xmath129 the compression constant @xmath130 is therefore given by : @xmath131on the other hand , the parameters of the deformed lattice are given by : @xmath132 the surface of the deformed lattice is @xmath133 , so that the filling factor @xmath134 _ ( iv ) _ a two - dimensional dilatation with @xmath135 and @xmath127 : now , the strain tensors @xmath136 and @xmath117 . the new area of the system is @xmath137 and @xmath138 it follows that the combination @xmath139 is given by : @xmath140 for this case , there is no need to actually compute the energy of the deformed lattice since we can extract @xmath139 from the hartree - fock energy @xmath141 given in eq . ( [ energiehf ] ) in the following manner . the area per electron , @xmath142 , in the deformed lattice is @xmath143 , so that ( @xmath144 here is the area per electron of the undeformed lattice ) : @xmath145 the change in @xmath142 causes a change in the filling factor , which is now given by : @xmath146 writing the hf energy as @xmath147 , we have the relation : @xmath148 . \label{m9}\ ] ] note that the long - wavelength coulomb term @xmath149 must be added to @xmath150 and @xmath130 that we compute in order to get @xmath151 and @xmath152 . 1 shows the expected quadratic behavior of the deformation energy as a function of @xmath153 , eq . ( [ deformation ] ) , for a shear deformation in the small @xmath154 limit . in the one and two - dimensional compressions ( _ i_)-(_iv _ ) , however , eq . ( [ deformation ] ) leads to the addition of a non physical linear term in the dependence of the energies @xmath155 on @xmath11 as can be seen in fig . 2 . in the absence of deformation , the average electronic density is equal to that of the positive background . this neutrality removes the divergence of @xmath156 at @xmath157 in the hartree - fock energy of eq . ( [ energiehf ] ) . when the electron lattice is dilated ( but not the positive background ) , the electronic density no longer matches the density of the positive background and there is a restoring force that arises from this density imbalance . it is easy to show , assuming a density of the form of eq . ( [ m13 ] ) , that no linear term in @xmath11 arises when the interaction with the positive background is properly taken into account , and that the interaction with the background does _ not _ give rise to higher order terms in @xmath11 . our hartree - fock procedure requires that the electronic and background densities be the same even for @xmath158 , which has the immediate consequence that we can not directly compute the deformation energy using eq . ( [ deformation ] ) . for all but the shear deformation , it is thus necessary for us to substract the linear term in eq . ( [ deformation ] ) and to add by hand the long - wavelength coulomb contribution of eq . ( [ defo ] ) in order to get the correct elastic constants . 2 shows that a quadratic behavior for the deformation energy is recovered when the linear term is substracted . note that the deformation energy in fig . 2 does not contain the long - wavelength coulomb contribution of eq . ( [ defo ] ) , so that it can be either positive or negative . our definitions in eqs . ( [ m2 ] , [ m3 ] , [ m4p ] ) of the elastic coefficients are not affected by this procedure of removing the linear term in @xmath11 since they involve the second derivative of the energy with respect to @xmath11 . for a shear deformation : @xmath159 in a triangular wigner crystal in landau level @xmath160 with filling factor @xmath161 . the square symbols are the hfa result while the solid line is a polynomial fit of order @xmath162 ( the linear term is negligible ) . ] for a one - dimensional dilatation of a triangular wigner crystal in landau level @xmath160 with filling factor @xmath163 . the square symbols are the hfa result ( left axis ) while the solid line is a polynomial fit . the dashed line ( right axis ) is the deformation energy with the linear term removed . ] it is instructive at this point to note that , for a triangular wigner crystal of classical electrons , the calculation of bonsall and maradudin@xcite gives the following expression for the quantity @xmath164 : @xmath165 and for the elastic coefficients : @xmath166 if we use eqs . ( [ m9 ] ) ( with the relation ( [ triangulaire ] ) ) and take @xmath167 as given by eq . ( [ m11 ] ) , we find @xmath168 , which is consistent with eq . ( [ m10 ] ) . we now turn our attention to the calculation of the dm of electron crystals in the grpa method . in the strong magnetic field limit where the hilbert space is restricted to one landau level only , the hamiltonian of the system is given by : @xmath169where @xmath170 and @xmath39 is a vector restricted to the first brillouin zone of the crystal . if we define the matsubara displacement green s function by : @xmath171we find , using @xmath172 $ ] and the commutation relation @xmath173 = i\ell ^{2}\delta _ { \mathbf{k},\mathbf{-k}^{\prime } } , $ ] that this green s function is related to the dynamical matrix by : @xmath174 } \label{m20 } \\ & \times \left ( \begin{array}{cc } d_{y , y}\left ( \mathbf{k}\right ) & -\frac{\hslash \omega _ { n}}{\ell ^{2}}% -d_{x , y}\left ( \mathbf{k}\right ) \\ \frac{\hslash \omega _ { n}}{\ell ^{2}}-d_{y , x}\left ( \mathbf{k}\right ) & d_{x , x}\left ( \mathbf{k}\right)% \end{array}% \right ) _ { \alpha \beta } , \notag\end{aligned}\]]where @xmath175 is a bosonic matsubara frequency and @xmath176 } \]]is the magnetophonon dispersion relation . we now define the following density green s function : @xmath177where @xmath178 in the generalised random - phase approximation ( grpa ) , this green s function is found by solving the set of equations@xcite : @xmath179 \chi _ { \mathbf{g}^{\prime \prime } , \mathbf{g}% ^{\prime } } ^{\left ( \rho , \rho \right ) } \left ( \mathbf{k},i\omega_{n}\right ) = b_{\mathbf{g , g}^{\prime } } ( \mathbf{q } ) , \label{m21}\]]with the definitions : @xmath180 \notag \\ & \times \big [ h\left ( \mathbf{g}-\mathbf{g}^{\prime } \right ) -x\left ( \mathbf{g - g}^{\prime } \right ) \nonumber\\ & -h\left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) + x\left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) \big ] , \end{aligned}\ ] ] and : @xmath181 .\end{aligned}\ ] ] diagonalizing the matrix @xmath182 and making the analytic continuation @xmath183 we can write @xmath184 in the form : @xmath185at small @xmath186 the pole @xmath187 with the biggest weight @xmath188 gives the grpa magnetophon mode . we define this pole as @xmath189 and the corresponding weigth as @xmath190 . we now relate the displacement green s function to the density green s function using eq . ( [ m7 ] ) . this last equation , coupled to eq . ( [ m24 ] ) , gives the following relation between the density and displacement response functions ( here @xmath191 is the function defined in eq . ( [ nphi2 ] ) , where for simplicity we now drop the landau level index @xmath75 ) : @xmath192 in deriving eq . ( [ r26 ] ) , we have assumed that @xmath193 , so that a density fluctuation can be linearly related to the displacement @xmath194 by eq . ( [ m7 ] ) . this is equivalent to assuming that the crystal can be described in the harmonic approximation so that only a knowledge of the dynamical matrix is necessary . to get eq . ( [ r26 ] ) , we have also assumed that @xmath195 so that @xmath196is real . we can now use eq . ( [ m20 ] ) and the symmetry relation @xmath197 to relate the density response function to the dynamical matrix : @xmath198 } { \left [ \left ( \omega+i\delta \right ) ^{2}-\omega _ { mp}^{2}\left ( \mathbf{k}\right ) \right ] } \notag\\ & \times & \frac{h(\mathbf{k+g})h(\mathbf{k+g}^{\prime } ) } { f(\mathbf{k+g})f(% \mathbf{k+g}^{\prime } ) } , \end{aligned}\ ] ] where we defined : @xmath199 \cdot \widehat{\mathbf{z } } , \\ & \gamma _ { 2}\left ( \mathbf{k}\right ) = \widehat{\mathbf{z}}\cdot \left [ \left ( \mathbf{k}+\mathbf{g}\right ) \times \left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) \right ] .\end{aligned}\ ] ] for @xmath200 close to the magnetophonon resonance , we can write : @xmath201where we defined the quantity : @xmath202then , equating eq . ( [ m26 ] ) with eq . ( [ m25 ] ) for @xmath200 close to @xmath203 , we obtain : @xmath204 because @xmath205 must be equal to @xmath206 , we can finally write@xmath207or , taking the real and imaginary parts of this equation ( we remind the reader that both functions @xmath208 and @xmath191 are real ) : @xmath209 & = & \frac{\nu } { 2}\left [ \frac{h\left ( \mathbf{k}+\mathbf{g}\right ) h\left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) } { f\left ( \mathbf{k}+\mathbf{g% } \right ) f\left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) } \right ] \notag \\ & & \times \frac{\gamma _ { 1}\left ( \mathbf{k}\right ) \ell ^{4}}{\hslash \omega _ { mp}\left ( \mathbf{k}\right ) } , \\ \im \left [ w_{\mathbf{g , g}^{\prime } } ^{(grpa)}\left ( \mathbf{k}\right ) % \right ] & = & \frac{\nu } { 2}\left [ \frac{h\left ( \mathbf{k}+\mathbf{g}\right ) h\left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) } { f\left ( \mathbf{k}+\mathbf{g% } \right ) f\left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) } \right ] \notag \\ & & \times \gamma _ { 2}\left ( \mathbf{k}\right ) \ell ^{2}.\end{aligned}\ ] ] we can get rid of the unknown form factors @xmath210 if we work with the ratio of the imaginary and real parts of the weights . we thus define : @xmath211 } { % \im \left [ w_{\mathbf{g , g}^{\prime } } ^{(grpa)}\left ( \mathbf{k}\right ) % \right ] } , \label{m27 } \\ & = & \frac{-\ell ^{2}}{\hslash \omega _ { mp}\left ( \mathbf{k}\right ) } \frac{\left ( \mathbf{k}+\mathbf{g}\right ) \times d\left ( \mathbf{k}\right ) \times \left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) } { \left ( \mathbf{k}+\mathbf{g}% \right ) \times \left ( \mathbf{k}+\mathbf{g}^{\prime } \right ) } . \notag\end{aligned}\ ] ] a careful examination shows that , because @xmath212 is given by the determinant of the dynamical matrix @xmath213 , the quantity @xmath214 is unchanged if all the components of the dynamical matrix are multiplied by some constant . ( [ m27 ] ) is thus indeterminate . to avoid this problem , we replace @xmath215 by @xmath216 in eq . ( [ m27 ] ) . our final result is thus:@xmath217 because @xmath218 , we need to choose three pairs of vectors @xmath219 to get the components of the dynamical matrix . to be valid , the dynamical matrix obtained in this way must satisfy the equation:@xmath220 } . \label{m33}\ ] ] eq . ( [ m33 ] ) provides a check on the validity of our calculation . in this section , we illustrate the application of our method by computing the lam coefficients for the triangular wigner crystal in landau levels @xmath160 and @xmath221 . fig . [ fig3 ] shows the first two shells of reciprocal lattice vectors of the triangular lattice with @xmath222 we take the vectors @xmath38 and @xmath223 in eq . ( [ m29 ] ) on these first two shells . not all combinations of vectors satisfy eq . ( [ m33 ] ) . by experimentation , we found that with a combination of the form @xmath224 $ ] with @xmath225 this equation is satisfied in the irreducible brillouin zone shown in fig . 3 to better than @xmath226 for @xmath227 . we will thus stick to this type of combination for the rest of this paper . . ] in the small - wavevector limit , the dynamical matrix of the wigner crystal with a triangular lattice structure is given by eq . ( [ m32 ] ) . using eq . ( [ m29 ] ) , we can extract the elastic coefficients by fitting @xmath228 along the path @xmath229 where : of the dynamical matrix along @xmath230 computed for differents couples of vectors @xmath231 . the numbers in the legend refer to the numerotation of the reciprocal lattice vectors shown in fig . @xmath232 where @xmath58 and @xmath59 are expressed in units of @xmath233 . fig . [ fig4 ] illustrates one limitation of our method : the grpa dynamical matrix is very much dependent on the choice of the couple ( @xmath234 ) . different choices give the same dynamical matrices @xmath235 only in the small wavector limit @xmath236 as shown in fig . 4 and , in this limit , the dynamical matrix element @xmath237 is almost entirely dominated by the long - range coulomb term ( the first term in eq . ( [ m35 ] ) ) . it follows that different choices of ( @xmath238 ) lead to quite different values of the elastic coefficients @xmath58 even though eq . ( m33 ) is satisfied . the coefficient @xmath59 obtained from @xmath239 , however , is not affected by the long - range coulomb interaction and appears to be independent of the choice of ( @xmath240 ) . note that the dynamical matrix given by eq . ( m29 ) does not have the correct transformation symmetries of the triangular lattice . in cases where @xmath241 is needed in all the brillouin zone , it becomes necessary to compute @xmath242 in the irreducible brillouin zone and obtain @xmath243 in the rest of the brillouin zone by symmetry . to give an idea of the variability of the numerical results with @xmath244 , we show in fig . [ fig5 ] ( for @xmath221 ) and fig . [ fig6 ] ( for @xmath160 ) the coefficients @xmath58 and @xmath59 extracted from the grpa dynamical matric of the triangular wigner crystal for different couples of vectors @xmath244 . these coefficients are compared with those computed using the hfa described in sec . we show the hfa results by a full line in figs . for both @xmath160 and @xmath221 , we find that the hartree - fock results for the coefficient @xmath59 are extremely well reproduced by the grpa method and , as we said above , do not depend on the choice of ( @xmath238 ) . this is what we expect since the grpa is the linear response of the crystal about the hfa ground state so that the coefficients @xmath19 obtained from the two methods should be roughly equal , taking into account the various approximations made in deriving the grpa dynamical matrix . the coefficient @xmath59 is easy to obtain , in view of eq . ( [ m37 ] ) , since it is given by a one - parameter fit of the @xmath245 curve . the elastic coefficient @xmath58 ( which is related to the bulk modulus ) is , on the other hand , much more difficult to obtain from eq . ( [ m35 ] ) . indeed , this elastic coefficient turns out to be very sensitive to how accurately the long - wavelength limit @xmath246 of @xmath237 in eq . ( [ m35 ] ) is obtained by the grpa numerical calculation . ( we here note that the grpa dynamical matrix _ does _ contain the long - range coulomb interaction discussed in sec . ii . the latter does not have to be added by hand as was the case for the elastic coefficients computed in the hfa . ) as we see in figs . 5(b ) and 6(b ) , @xmath58 is also very sensitive to the choice of the vectors ( @xmath238 ) with one particular choice ( 2,3 ) reproducing the hfa results almost exactly . the other two choices give very different values for @xmath58 . in the absence of any criteria to choose ( @xmath238 ) a priori , we would say that the grpa dynamical matrix can not be used to make _ quantitative _ predictions . the _ qualitative _ behaviour of the grpa elastic coefficient @xmath58 is consistent with that of @xmath58 computed in the hfa . if we exclude the domain @xmath247 where our numerical results become noisy , we find that the average of the grpa results for the three couples of ( @xmath238 ) , as shown in figs . 5(b ) and 6(b ) , are in very good agreement with the hf calculation . in the absence of any criteria to choose the best couple ( @xmath238 ) , this averaging procedure must be used to get qualitatively and quantitatively reliable results for the grpa dynamical matrix . for @xmath247 , the crystal softens and the quantum fluctuations in @xmath248 are important . there is a transition@xcite into a bubble state with 2 electrons per unit cell at approximately @xmath249 . we do not expect the assumptions underlying our method to be valid in this region . and ( b ) @xmath58 of the triangular wigner crystal for landau level @xmath221 computed using the different approximations listed in the legend . for the grpa , the coefficients are computed using @xmath250 differents couples of reciprocal lattice vectors . the numbers in the legend correspond to the numeration of the vectors given in fig . the empty circles give an average of the 3 grparesults . for @xmath59 the different symbols are superimposed . ] and ( b ) @xmath58 of the triangular wigner crystal in landau level @xmath160 computed in the different approximations indicated in the legend . for the grpa , the coefficients are computed using @xmath250 differents couples of reciprocal lattice vectors . the numbers in the legend correspond to the numeration of the vectors given in fig . the empty circles give the average of the 3 grpa results . for @xmath251 the different symbols are superimposed . ] for the stripe crystal , the dynamical matrix is given by : @xmath252 [ zz ] and the elastic coefficients evaluated in the hfa@xcite for landau level @xmath221 are listed in table [ table1 ] . .elastic coefficients @xmath253 in units of @xmath254 for the stripe crystal at various filling factors and in landau level @xmath221 . [ cols="<,<,<,<,<",options="header " , ] the first 4 shells of reciprocal lattice vectors of the stripe crystal are represented in fig . 7 . from eq . ( [ m29 ] ) , the vectors ( @xmath255 ) must not be parallel otherwise the denominator in this equation vanishes . this forces us to use @xmath38 in the second shell and @xmath223 in the fourth shell of reciprocal lattice vectors to evaluate the dm in the grpa . we show in figs . 8 - 10 the elements @xmath256 and @xmath257 computed at filling factor @xmath258 ( in landau level @xmath221 ) along different directions in @xmath39-space together with the corresponding dm in the hfa element obtained from eqs . ( [ zz ] ) with the coefficients of table [ table1 ] . similar results are obtained at other filling factors . notice that the bending coefficient @xmath259does not contribute to any of these curves . for each curve , eq . ( [ m33 ] ) is perfectly satisfied and the coefficient @xmath59 , which can be extracted from the grpa@xmath260function @xmath261 , is in excellent agreement with the hfa results given in table [ table1 ] . of the grpa and hfa dynamical matrices of the stripe crystal computed along the direction @xmath262 for partial filling factor @xmath263 in landau level @xmath221 , computed using @xmath162 differents couples of reciprocal lattice vectors . the numbers in the legend refer to numbering of the reciprocal lattice vectors in fig . 7 . ] of the grpa and hfa dynamical matrices of the stripe crystal computed along the direction @xmath264 for partial filling factor @xmath263 in landau level @xmath265 , computed using @xmath162 differents couples of reciprocal lattice vectors . the numbers in the legend refer to numbering of the reciprocal lattice vectors in fig . 7 . ] of the grpa and hfa dynamical matrices of the stripe crystal computed along the direction @xmath266 for partial filling factor @xmath263 in landau level @xmath221 , computed using @xmath162 differents couples of reciprocal lattice vectors . the numbers in the legend refer to numbering of the reciprocal lattice vectors in fig . 7 . ] for the grpa , figs . 8 - 10 show results for the couples ( @xmath238 ) that produce the maximum and minimum values of the dm element . in all but the @xmath267 case , the hfa curve lies between these two results . for @xmath267 , one of the grpa curves almost coincides with the hfa result for @xmath268 this is reassuring for the validity of the grpa method , but it also makes it impossible for us to find what part of the difference between the grpa and hfa is numerical and what part is physical ( i.e. due to anharmonicity for example ) . we remark that , in the range @xmath269 , the grparesults are not numerically very different from the small @xmath270 expansion of the dynamical matrix given in eqs . ( [ zz ] ) with the hfa coefficients . to the credit of our grpa method , we add that the evolution of the different @xmath271 s with filling factor is consistent with that of the corresponding elements calculated in the hfa as shown in fig . we thus see that the evaluation of the elastic coefficients other than @xmath59 from the grpa results seems hazardous for the stripe crystal . the curvature of the functions @xmath256 and @xmath257 in figs . 8,9,10 is proportional to @xmath272 and @xmath130 respectively . it is clear that the elastic coefficients extracted from these @xmath271 are much bigger than those obtained from the hfa ( the curvature of the hfa function is barely visible in the figures ) . these coefficients also show very strong variation with the choice of ( @xmath238 ) . an averaging of the grpa results for different couples ( @xmath238 ) would give a result closer to the hfa but the improvement would not be as dramatic as in the triangular lattice case . in fact , in the case of @xmath257 , we find that averaging over different choices of reciprocal lattice vectors does not bring any improvement to the numerical results . of the grpa and hfa dynamical matrices of the stripe crystal computed along the direction @xmath273 for different filling factors in landau level @xmath221 . ] finally , we remark that our grpa results for @xmath274 are dominated by a strong @xmath275 behavior indicating that the bending term @xmath276 is absolutely essential in the elastic description of the stripe crystal in eqs . ( [ zz ] ) . in conclusion , in this paper we have shown that is it possible to derive an effective dynamical matrix for various crystal states of the 2deg in a strong magnetic field by computing the density response function in the grpa . we have compared the dynamical matrix obtained in this way with the one obtained from standard elasticity theory with elastic coefficients computed in the hfa . our comparison was done for crystals with very different elastic properties , namely a triangular wigner crystal and stripe crystal . our motivation for deriving a dynamical matrix using the grpa response consists in the fact that the latter has the advantage of giving the dynamical matrix directly without having to compute the elastic coefficients separately . our comparison with the hartree - fock results showed , however , that the grpa method must be used with care because of the variability of the results with the choice of the couples @xmath277 . the shear modulus @xmath59 computed in the grpa agrees very well with the one computed from the hfa , but the values of the other elastic coefficients @xmath19 which are affected by the long range coulomb interaction depend very much on the choice of the couples @xmath277 . in some cases , as for a triangular wigner crystal , an averaging procedure over different couples @xmath278 improves the numerical accuracy of the method . in the long wavelength @xmath279 limit , however , the grpa dynamical matrix is a good approximation , both qualitatively and quantitatively , and gives reasonable estimates for the elastic constants of the electronic solids that are in agreement with the static hartree - fock calculations . this work was supported by a research grant ( for r. ct ) from the natural sciences and engineering research council of canada ( nserc ) . c. b. doiron acknowledges support from nserc , the fonds qubcois de recherche sur la nature et les technologies ( fqrnt ) , the swiss nsf and nccr nanoscience . computer time was provided by the rseau qubcois de calcul haute performance ( rqchp ) . 99 p. k. lam and s. m. girvin , phys . b * 30 * , 473 ( 1984 ) ; d. levesque , j. j. weis , and a. h. macdonald , phys . rev . b * 30 * , 1056 ( 1984 ) ; k. esfarjani and s. t. chui , phys . rev . b * 42 * , 10758 ( 1990 ) ; k. yang , f. d. m. haldane , and e. h. rezayi , phys . rev . b * 64 * , 081301(r ) ( 2001 ) ; x. zhu and s. g. louie , phys . rev . b * 52 * , 5863 ( 1995 ) . for recent reviews , see physics of the electron solid , edited by s. t. chui ( international press , boston ( 1994 ) and h. fertig and h. shayegan in perspectives in quantum hall effects , edited by s. das sarma and a. pinczuk ( wiley , new york , 1997 ) , chaps . 5 and 9 respectively . r. m. lewis , p. d. ye , l. w. engel , d. c. tsui , l. n. pfeiffer , and k. w. west , phys . lett . * 89 * , 136804 ( 2002 ) ; r. m. lewis , y. chen , l. w. engel , d. c. tsui , p. d. ye , l. n. pfeiffer , and k. w. west phys . 93 * , 176808 ( 2004 ) ; r. m. lewis , yong chen , l. w. engel , p. d. ye , d. c. tsui , l. n. pfeiffer , and k. w. west , physica e**22 * * , 119 ( 2004 ) . r. ct , c. b. doiron , j. bourassa , and h. a. fertig , phys . b * 68 * , 155327 ( 2003 ) ; m .- r . li , h. a. fertig , r. ct , and hangmo yi , phys . lett . * 92 * , 186804 ( 2004 ) ; mei - rong li , h. a. fertig , r. ct , and hangmo yi , phys . b * 71 * , 155312 ( 2005 ) ; r. ct , mei - rong li , a. faribault , and h. a. fertig , phys . rev . b * 72 * , 115344 ( 2005 ) . these coefficients were evaluated previously in refs . . we remark however that there was an error in the evaluation of the coefficient @xmath58 in ref . that we have corrected in the present paper . our previous calculation underestimated the value of @xmath58 by a factor of approximately @xmath250 . note also that in the present paper the stripes are aligned along the @xmath25 axis while they were aligned along @xmath24 in our previous calculation so that @xmath280 .

graphical models are used in many application domains , running the gamut from computer vision and civil engineering to political science and epidemiology . in many applications , estimating the edge structure of an underlying graphical model is of significant interest . for instance , a graphical model may be used to represent friendships between people in a social network @xcite or links between organisms with the propensity to spread an infectious disease @xcite . it is a classical corollary of the hammersley clifford theorem @xcite that zeros in the inverse covariance matrix of a multivariate gaussian distribution indicate absent edges in the corresponding graphical model . this fact , combined with various types of statistical estimators suited to high dimensions , has been leveraged by many authors to recover the structure of a gaussian graphical model when the edge set is sparse ( see the papers @xcite and the references therein ) . recently , liu et al . @xcite and liu , lafferty and wasserman @xcite introduced the notion of a nonparanormal distribution , which generalizes the gaussian distribution by allowing for monotonic univariate transformations , and argued that the same structural properties of the inverse covariance matrix carry over to the nonparanormal ; see also the related work of xue and zou @xcite on copula transformations . however , for non - gaussian graphical models , the question of whether a general relationship exists between conditional independence and the structure of the inverse covariance matrix remains unresolved . in this paper , we establish a number of interesting links between covariance matrices and the edge structure of an underlying graph in the case of discrete - valued random variables . ( although we specialize our treatment to multinomial random variables due to their widespread applicability , several of our results have straightforward generalizations to other types of exponential families . ) instead of only analyzing the standard covariance matrix , we show that it is often fruitful to augment the usual covariance matrix with higher - order interaction terms . our main result has an interesting corollary for tree - structured graphs : for such models , the inverse of a generalized covariance matrix is always ( block ) graph - structured . in particular , for binary variables , the inverse of the usual covariance matrix may be used to recover the edge structure of the tree . we also establish more general results that apply to arbitrary ( nontree ) graphs , specified in terms of graph triangulations . this more general correspondence exploits ideas from the geometry of exponential families @xcite , as well as the junction tree framework @xcite . as we illustrate , these population - level results have a number of corollaries for graph selection methods . graph selection methods for gaussian data include neighborhood regression @xcite and the graphical lasso @xcite , which corresponds to maximizing an @xmath0-regularized version of the gaussian likelihood . alternative methods for selection of discrete graphical models include the classical chow liu algorithm for trees @xcite ; techniques based on conditional entropy or mutual information @xcite ; and nodewise logistic regression for discrete graphical models with pairwise interactions our population - level results imply that minor variants of the graphical lasso and neighborhood regression methods , though originally developed for gaussian data , remain consistent for trees and the broader class of graphical models with singleton separator sets . they also convey a cautionary message , in that these methods will be inconsistent ( generically ) for other types of graphs . we also describe a new method for neighborhood selection in an arbitrary sparse graph , based on linear regression over subsets of variables . this method is most useful for bounded - degree graphs with correlation decay , but less computationally tractable for larger graphs . in addition , we show that our methods for graph selection may be adapted to handle noisy or missing data in a seamless manner . naively applying nodewise logistic regression when observations are systematically corrupted yields estimates that are biased even in the limit of infinite data . there are various corrections available , such as multiple imputation @xcite and the expectation - maximization ( em ) algorithm @xcite , but , in general , these methods are not guaranteed to be statistically consistent due to local optima . to the best of our knowledge , our work provides the first method that is provably consistent under high - dimensional scaling for estimating the structure of discrete graphical models with corrupted observations . further background on corrupted data methods for low - dimensional logistic regression may be found in carroll , ruppert and stefanski @xcite and ibrahim et al . @xcite . the remainder of the paper is organized as follows . in section [ secbackground ] , we provide brief background and notation on graphical models and describe the classes of augmented covariance matrices we will consider . in section [ secresults ] , we state our main population - level result ( theorem [ thmfull ] ) on the relationship between the support of generalized inverse covariance matrices and the edge structure of a discrete graphical model , and then develop a number of corollaries . the proof of theorem [ thmfull ] is provided in section [ secproofthmfull ] , with proofs of corollaries and more technical results deferred to the supplementary material @xcite . in section [ secsample ] , we develop consequences of our population - level results in the context of specific methods for graphical model selection . we provide simulation results in section [ secsims ] in order to confirm the accuracy of our theoretically - predicted scaling laws , dictating how many samples are required ( as a function of graph size and maximum degree ) to recover the graph correctly . in this section , we provide background on graphical models and exponential families . we then present a simple example illustrating the phenomena and methodology underlying this paper . an _ undirected graphical model _ or _ markov random field _ ( mrf ) is a family of probability distributions respecting the structure of a fixed graph . we begin with some basic graph - theoretic terminology . an undirected graph @xmath1 consists of a collection of vertices @xmath2 and a collection of unordered and the edge @xmath3 . in this paper , we forbid graphs with self - loops , meaning that @xmath4 for all @xmath5 . ] vertex pairs @xmath6 . a _ vertex cutset _ is a subset @xmath7 of vertices whose removal breaks the graph into two or more nonempty components [ see figure [ figgraphprops](a ) ] . clique _ is a subset @xmath8 such that @xmath9 for all distinct @xmath10 . the cliques in figure [ figgraphprops](b ) are all _ maximal _ , meaning they are not properly contained within any other clique . for @xmath11 , we define the neighborhood @xmath12 to be the set of vertices connected to @xmath13 by an edge . for an undirected graph @xmath14 , we associate to each vertex @xmath15 a random variable @xmath16 taking values in a space @xmath17 . for any subset @xmath18 , we define @xmath19 , and for three subsets of vertices , @xmath20 , @xmath21 and @xmath7 , we write @xmath22 to mean that the random vector @xmath23 is conditionally independent of @xmath24 given @xmath25 . the notion of a markov random field may be defined in terms of certain _ markov properties _ indexed by vertex cutsets or in terms of a _ factorization property _ described by the graph cliques . [ cols="^,^ " , ] still focusing on the single - cycle graph in panel ( b ) , suppose that instead of considering the ordinary covariance matrix , we compute the covariance matrix of the _ augmented _ random vector @xmath26 , where the extra term @xmath27 is represented by the dotted edge shown in panel ( c ) . the @xmath28 inverse of this generalized covariance matrix takes the form @xmath29.\ ] ] this matrix safely separates nodes @xmath30 and @xmath31 , but the entry corresponding to the nonedge @xmath32 is _ not _ equal to zero . indeed , we would observe a similar phenomenon if we chose to augment the graph by including the edge @xmath33 rather than @xmath32 . this example shows that the usual inverse covariance matrix is not always graph - structured , but inverses of augmented matrices involving higher - order interaction terms may reveal graph structure . now let us consider a more general graphical model that adds the 3-clique interaction terms shown in panel ( d ) to the usual ising terms . we compute the covariance matrix of the augmented vector @xmath34 empirically , one may show that the @xmath35 inverse @xmath36)^{-1}$ ] respects aspects of the graph structure : there are zeros in position @xmath37 , corresponding to the associated functions @xmath38 and @xmath39 , whenever @xmath40 and @xmath41 do not lie within the same maximal clique . [ e.g. , this applies to the pairs @xmath42 and @xmath43 . ] the goal of this paper is to understand when certain inverse covariances do ( and _ do not _ ) capture the structure of a graphical model . at its root is the principle that the augmented inverse covariance matrix @xmath44 , suitably defined , is _ always _ graph - structured with respect to a graph triangulation . in some cases [ e.g. , the dino graph in figure [ figgraphs](e ) ] , we may leverage the block - matrix inversion formula @xcite , namely , @xmath45 to conclude that the inverse of a sub - block of the augmented matrix ( e.g. , the ordinary covariance matrix ) is still graph - structured . this relation holds whenever @xmath20 and @xmath21 are chosen in such a way that the second term in equation ( [ eqnchai ] ) continues to respect the edge structure of the graph . these ideas will be made rigorous in theorem [ thmfull ] and its corollaries in the next section . we now state our main results on the relationship between the zero pattern of generalized ( augmented ) inverse covariance matrices and graph structure . in section [ secsample ] to follow , we develop some consequences of these results for data - dependent estimators used in structure estimation . we begin with some notation for defining generalized covariance matrices , stated in terms of the sufficient statistics previously defined ( [ eqnsuff ] ) . recall that a clique @xmath46 is associated with the collection @xmath47 of binary - valued sufficient statistics . let @xmath48 , and define the random vector @xmath49 consisting of all the sufficient statistics indexed by elements of @xmath50 . as in the previous section , the set @xmath51 contains both maximal and nonmaximal cliques . we will often be interested in situations where @xmath50 contains all subsets of a given set . for a subset @xmath18 , let @xmath52 denote the collection of all @xmath53 nonempty subsets of @xmath20 . we extend this notation to @xmath50 by defining @xmath54 our first main result concerns a connection between the inverses of generalized inverse covariance matrices associated with the model ( [ eqngeneralmultinomial ] ) and any triangulation of the underlying graph @xmath55 . the notion of a triangulation is defined in terms of chordless cycles , which are sequences of distinct vertices @xmath56 such that : * @xmath57 for all @xmath58 , and also @xmath59 ; * no other nodes in the cycle are connected by an edge . as an illustration , the @xmath31-cycle in figure [ figgraphs](b ) is a chordless cycle . given an undirected graph @xmath1 , a _ triangulation _ is an augmented graph @xmath60 that contains no chordless cycles of length greater than 3 . note that a tree is trivially triangulated , since it contains no cycles . on the other hand , the chordless @xmath31-cycle in figure [ figgraphs](b ) is the simplest example of a nontriangulated graph . by adding the single edge @xmath32 to form the augmented edge set @xmath61 , we obtain the triangulated graph @xmath62 shown in panel ( c ) . one may check that the more complicated graph shown in figure [ figgraphs](e ) is triangulated as well . our first result concerns the inverse @xmath63 of the matrix @xmath64 , where @xmath65 is the set of all cliques arising from some triangulation @xmath66 of @xmath55 . for any two subsets @xmath67 , we write @xmath68 to denote the sub - block of @xmath63 indexed by all indicator statistics on @xmath20 and @xmath21 , respectively . ( note that we are working with respect to the exponential family representation over the triangulated graph @xmath66 . ) given our previously - defined sufficient statistics ( [ eqnsuff ] ) , the sub - block @xmath68 has dimensions @xmath69 , where @xmath70 for example , when @xmath71 and @xmath72 , the submatrix @xmath68 has dimension @xmath73 . with this notation , we have the following result : [ thmfull ] consider an arbitrary discrete graphical model of the form ( [ eqngeneralmultinomial ] ) , and let @xmath74 be the set of all cliques in any triangulation of @xmath55 . then the generalized covariance matrix @xmath64 is invertible , and its inverse @xmath63 is _ block graph - structured _ : a. for any two subsets @xmath75 that are not subsets of the same maximal clique , the block @xmath76 is identically zero . b. for almost all parameters @xmath77 , the entire block @xmath68 is nonzero whenever @xmath20 and @xmath21 belong to a common maximal clique . in part ( b ) , `` almost all '' refers to all parameters @xmath77 apart from a set of lebesgue measure zero . the proof of theorem [ thmfull ] , which we provide in section [ secproofthmfull ] , relies on the geometry of exponential families @xcite and certain aspects of convex analysis @xcite , involving the log partition function @xmath78 and its fenchel legendre dual @xmath79 . although we have stated theorem [ thmfull ] for discrete variables , it easily generalizes to other classes of random variables . the only difference is the specific choices of sufficient statistics used to define the generalized covariance matrix . this generality becomes apparent in the proof . to provide intuition for theorem [ thmfull ] , we consider its consequences for specific graphs . when the original graph is a tree [ such as the graph in figure [ figgraphs](a ) ] , it is already triangulated , so the set @xmath80 is equal to the edge set @xmath81 , together with singleton nodes . hence , theorem [ thmfull ] implies that the inverse @xmath63 of the matrix of sufficient statistics for vertices and edges is graph - structured , and blocks of nonzeros in @xmath63 correspond to edges in the graph . in particular , we may apply theorem [ thmfull](a ) to the subsets @xmath82 and @xmath72 , where @xmath13 and @xmath83 are distinct vertices with @xmath84 , and conclude that the @xmath73 sub - block @xmath68 is equal to zero . when @xmath14 is not triangulated , however , we may need to invert a larger augmented covariance matrix and include sufficient statistics over pairs @xmath85 as well . for instance , the augmented graph shown in figure [ figgraphs](c ) is a triangulation of the chordless @xmath31-cycle in panel ( b ) . the associated set of maximal cliques is given by @xmath86 ; among other predictions , our theory guarantees that the generalized inverse covariance @xmath63 will have zeros in the sub - block @xmath87 . in fact , it is not necessary to take sufficient statistics over _ all _ maximal cliques , and we may consider a slightly smaller augmented covariance matrix . ( this simpler type of augmented covariance matrix explains the calculations given in section [ secexamples ] . ) by classical graph theory , any triangulation @xmath66 gives rise to a _ junction tree _ representation of @xmath55 . nodes in the junction tree are subsets of @xmath88 corresponding to maximal cliques of @xmath66 , and the intersection of any two adjacent cliques @xmath89 and @xmath90 is referred to as a _ separator set _ @xmath91 . furthermore , any junction tree must satisfy the _ running intersection property _ , meaning that for any two nodes of the junction tree say , corresponding to cliques @xmath92 and @xmath93the intersection @xmath94 must belong to every separator set on the unique path between @xmath92 and @xmath93 . the following result shows that it suffices to construct generalized covariance matrices augmented by separator sets : [ corsep ] let @xmath50 be the set of separator sets in any triangulation of @xmath55 , and let @xmath63 be the inverse of @xmath95 . then @xmath96 whenever @xmath97 . note that @xmath98 , and the set of sufficient statistics considered in corollary [ corsep ] is generally much smaller than the set of sufficient statistics considered in theorem [ thmfull ] . hence , the generalized covariance matrix of corollary [ corsep ] has a smaller dimension than the generalized covariance matrix of theorem [ thmfull ] , which becomes significant when we consider exploiting these population - level results for statistical estimation . the graph in figure [ figgraphs](c ) of example [ exasgraph ] and the associated matrix in equation ( [ eqnaugmented ] ) provide a concrete example of corollary [ corsep ] in action . in this case , the single separator set in the triangulation is @xmath99 , so when @xmath100 , augmenting the usual covariance matrix with the additional sufficient statistic @xmath101 and taking the inverse yields a graph - structured matrix . indeed , since @xmath102 , we observe that @xmath103 in equation ( [ eqnaugmented ] ) , consistent with the result of corollary [ corsep ] . although theorem [ thmfull ] and corollary [ corsep ] are clean population - level results , however , forming an appropriate augmented covariance matrix requires prior knowledge of the graph , namely , which edges are involved in a suitable triangulation . this is infeasible in settings where the goal is to recover the edge structure of the graph . corollary [ corsep ] is most useful for edge recovery when @xmath14 admits a triangulation with only singleton separator sets , since then @xmath104 . in particular , this condition holds when @xmath14 is a tree . the following corollary summarizes our result : [ cortree ] for any graph with singleton separator sets , the inverse @xmath63 of the covariance matrix @xmath105 of vertex statistics is graph - structured . ( this class includes trees as a special case . ) in the special case of binary variables , we have @xmath106 , so corollary [ cortree ] implies that the inverse of the ordinary covariance matrix @xmath107 is graph - structured . for @xmath108-ary variables , @xmath109 is a matrix of dimensions @xmath110 involving indicator functions for each variable . again , we may relate this corollary to example [ exasgraph]the inverse covariance matrices for the tree graph in panel ( a ) and the dino graph in panel ( e ) are exactly graph - structured . although the dino graph is not a tree , it possesses the nice property that the only separator sets in its junction tree are singletons . corollary [ corsep ] also guarantees that inverse covariances may be partially graph - structured , in the sense that @xmath111 for any pair of vertices @xmath112 separable by a singleton separator set , where @xmath113 . this is because for any such pair @xmath112 , we may form a junction tree with two nodes , one containing @xmath13 and one containing @xmath83 , and apply corollary [ corsep ] . indeed , the matrix @xmath63 defined over singleton vertices is agnostic to which triangulation we choose for the graph . in settings where there exists a junction tree representation of the graph with only singleton separator sets , corollary [ cortree ] has a number of useful implications for the consistency of methods that have traditionally only been applied for edge recovery in gaussian graphical models : for tree - structured discrete graphs , it suffices to estimate the support of @xmath114 from the data . we will review methods for gaussian graphical model selection and describe their analogs for discrete tree graphs in sections [ secglobal ] and [ secnodewise ] . theorem [ thmfull ] also has a corollary , that is , relevant for nodewise neighborhood selection approaches to graph selection @xcite , which are applicable to graphs with arbitrary topologies . nodewise methods use the basic observation that recovering the edge structure of @xmath55 is equivalent to recovering the neighborhood set @xmath115 for each vertex @xmath5 . for a given node @xmath15 and positive integer @xmath116 , consider the collection of subsets @xmath117 the following corollary provides an avenue for recovering @xmath118 based on the inverse of a certain generalized covariance matrix : [ corneigh ] for any graph and node @xmath119 with @xmath120 , the inverse @xmath63 of the matrix @xmath121 is @xmath13-block graph - structured , that is , @xmath122 whenever @xmath123 . in particular , @xmath111 for all vertices @xmath124 . note that @xmath125 is the set of subsets of all candidate neighborhoods of @xmath13 of size @xmath116 . this result follows from theorem [ thmfull ] ( and the related corollary [ corsep ] ) by constructing a particular junction tree for the graph , in which @xmath13 is separated from the rest of the graph by @xmath118 . due to the well - known relationship between the rows of an inverse covariance matrix and linear regression coefficients @xcite , corollary [ corneigh ] motivates the following neighborhood - based approach to graph selection : for a fixed vertex @xmath5 , perform a single _ linear regression _ of @xmath126 on the vector @xmath127 . via elementary algebra and an application of corollary [ corneigh ] , the resulting regression vector will expose the neighborhood @xmath118 in an arbitrary discrete graphical model ; that is , the indicators @xmath128 corresponding to @xmath129 will have a nonzero weight only if @xmath130 . we elaborate on this connection in section [ secnodewise ] . we now turn to the proof of theorem [ thmfull ] , which is based on certain fundamental correspondences arising from the theory of exponential families @xcite . recall that our exponential family ( [ eqngeneralmultinomial ] ) has binary - valued indicator functions ( [ eqnsuff ] ) as its sufficient statistics . let @xmath93 denote the cardinality of this set and let @xmath131 denote the multivariate function that maps each configuration @xmath132 to the vector @xmath133 obtained by evaluating the @xmath93 indicator functions on @xmath134 . using this notation , our exponential family may be written in the compact form @xmath135 , where @xmath136 since this exponential family is known to be minimal , we are guaranteed @xcite that @xmath137 \quad\mbox{and}\quad\nabla^2 \phi(\theta ) = { \operatorname{cov}}_\theta \bigl[\mathbb{i}(x ) \bigr],\ ] ] where @xmath138 and @xmath139 denote ( resp . ) the expectation and covariance taken under the density @xmath140 @xcite . the conjugate dual @xcite of the cumulant function is given by @xmath141 the function @xmath79 is always convex and takes values in @xmath142 . from known results @xcite , the dual function @xmath79 is finite only for @xmath143 belonging to the marginal polytope @xmath144 the following lemma , proved in the supplementary material @xcite , provides a connection between the covariance matrix and the hessian of @xmath79 : [ lementropy ] consider a regular , minimal exponential family , and define @xmath145 $ ] for any fixed @xmath146 . then @xmath147 \bigr)^{-1 } & = & \nabla^2 \phi^*(\mu).\end{aligned}\ ] ] note that the minimality and regularity of the family implies that @xmath148 $ ] is strictly positive definite , so the matrix is invertible . for any @xmath149 , let @xmath150 denote the unique natural parameter @xmath77 such that @xmath151 . it is known @xcite that the negative dual function @xmath152 is linked to the shannon entropy via the relation @xmath153 in general , expression ( [ eqndualentropy ] ) does _ not _ provide a straightforward way to compute @xmath154 , since the mapping @xmath155 may be extremely complicated . however , when the exponential family is defined with respect to a triangulated graph , @xmath79 has an explicit closed - form representation in terms of the mean parameters @xmath156 . consider a junction tree triangulation of the graph , and let @xmath157 be the collection of maximal cliques and separator sets , respectively . by the junction tree theorem , we have the factorization @xmath158 where @xmath159 and @xmath160 are the marginal distributions over maximal clique @xmath92 and separator set @xmath161 . consequently , the entropy may be decomposed into the sum @xmath162 where we have introduced the clique- and separator - based entropies @xmath163 and @xmath164 given our choice of sufficient statistics ( [ eqnsuff ] ) , we show that @xmath159 and @xmath160 may be written explicitly as `` local '' functions of mean parameters associated with @xmath92 and @xmath161 . for each subset @xmath165 , let @xmath166 be the associated collection of mean parameters , and let @xmath167 be the set of mean parameters associated with all nonempty subsets of @xmath20 . note that @xmath168 contains a total of @xmath169 parameters , corresponding to the number of degrees of freedom involved in specifying a marginal distribution over the random vector @xmath170 . moreover , @xmath168 uniquely determines the marginal distribution @xmath171 : [ lemmapping ] for any marginal distribution @xmath171 in the @xmath172-dimensional probability simplex , there is a unique mean parameter vector @xmath168 and matrix @xmath173 such that @xmath174 . for the proof , see the supplementary material @xcite . we now combine the dual representation ( [ eqndualentropy ] ) with the decomposition ( [ eqnplainentropy ] ) , along with the matrices @xmath175 from lemma [ lemmapping ] , to conclude that @xmath176 now consider two subsets @xmath75 that are not contained in the same maximal clique . suppose @xmath20 is contained within maximal clique @xmath92 . differentiating expression ( [ eqnentropyexp ] ) with respect to @xmath177 preserves only terms involving @xmath159 and @xmath160 , where @xmath161 is any separator set such that @xmath178 . since @xmath179 , we clearly can not have @xmath180 . consequently , all cross - terms arising from the clique @xmath92 and its associated separator sets vanish when we take a second derivative with respect to @xmath181 . repeating this argument for any other maximal clique @xmath182 containing @xmath20 but not @xmath21 , we have @xmath183 . this proves part ( a ) . turning to part ( b ) , note that if @xmath20 and @xmath21 are in the same maximal clique , the expression obtained by taking second derivatives of the entropy results in an algebraic expression with only finitely many solutions in the parameters @xmath156 ( consequently , also @xmath77 ) . hence , assuming the @xmath77 s are drawn from a continuous distribution , the corresponding values of the block @xmath68 are a.s . nonzero . moving beyond the population level , we now state and prove several results concerning the statistical consistency of different methods both known and some novel for graph selection in discrete graphical models , based on i.i.d . draws from a discrete graph . for sparse gaussian models , existing methods that exploit sparsity of the inverse covariance matrix fall into two main categories : global graph selection methods ( e.g. , @xcite ) and local ( nodewise ) neighborhood selection methods @xcite . we divide our discussion accordingly . we begin by describing how a combination of our population - level results and some concentration inequalities may be leveraged to analyze the statistical behavior of log - determinant methods for discrete graphical models with singleton separator sets , and suggest extensions of these methods when observations are systematically corrupted by noise or missing data . given a @xmath184-dimensional random vector @xmath185 with covariance @xmath186 , consider the estimator @xmath187 where @xmath188 is an estimator for @xmath186 . for multivariate gaussian data , this program is an @xmath0-regularized maximum likelihood estimate known as the _ graphical lasso _ and is a well - studied method for recovering the edge structure in a gaussian graphical model @xcite . although the program ( [ eqnlogdet ] ) has no relation to the mle in the case of a discrete graphical model , it may still be useful for estimating @xmath189 . indeed , as shown in ravikumar et al . @xcite , existing analyses of the estimator ( [ eqnlogdet ] ) require only tail conditions such as sub - gaussianity in order to guarantee that the sample minimizer is close to the population minimizer . the analysis of this paper completes the missing link by guaranteeing that the population - level inverse covariance is in fact graph - structured . consequently , we obtain the interesting result that the program ( [ eqnlogdet])even though it is ostensibly derived from gaussian considerations is a consistent method for recovering the structure of any binary graphical model with singleton separator sets . in order to state our conclusion precisely , we introduce additional notation . consider a general estimate @xmath188 of the covariance matrix @xmath190 such that @xmath191 & \leq & c \exp \bigl(- \psi(n , p ) \bigr)\end{aligned}\ ] ] for functions @xmath192 and @xmath193 , where @xmath194 denotes the elementwise @xmath195-norm . in the case of fully - observed i.i.d . data with sub - gaussian parameter @xmath196 , where @xmath197 is the usual sample covariance , this bound holds with @xmath198 and @xmath199 . as in past analysis of the graphical lasso @xcite , we require a certain _ mutual incoherence _ condition on the true covariance matrix @xmath186 to control the correlation of nonedge variables with edge variables in the graph . let @xmath200 , where @xmath201 denotes the kronecker product . then @xmath202 is a @xmath203 matrix indexed by vertex pairs . the incoherence condition is given by @xmath204,\ ] ] where @xmath205 is the set of vertex pairs corresponding to nonzero entries of the precision matrix @xmath206 , equivalently , the edge set of the graph , by our theory on tree - structured discrete graphs . for more intuition on the mutual incoherence condition , see ravikumar et al . @xcite . with this notation , our global edge recovery algorithm proceeds as follows : [ alggraphlasso ] 1 . form a suitable estimate @xmath188 of the true covariance matrix @xmath190 . 2 . optimize the graphical lasso program ( [ eqnlogdet ] ) with parameter @xmath207 , and denote the solution by @xmath208 . 3 . threshold the entries of @xmath209 at level @xmath210 to obtain an estimate of @xmath206 . it remains to choose the parameters @xmath211 . in the following corollary , we will establish statistical consistency of @xmath209 under the following settings : @xmath212 where @xmath40 is the incoherence parameter in inequality ( [ eqnmatincoherence ] ) and @xmath213 are universal positive constants . the following result applies to algorithm [ alggraphlasso ] when @xmath188 is the sample covariance matrix and @xmath211 are chosen as in equations ( [ eqnlambdatau ] ) : [ corglobising ] consider an ising model ( [ eqnising ] ) defined by an undirected graph with singleton separator sets and with degree at most @xmath116 , and suppose that the mutual incoherence condition ( [ eqnmatincoherence ] ) holds . with @xmath214 samples , there are universal constants @xmath215 such that with probability at least @xmath216 , algorithm [ alggraphlasso ] recovers all edges @xmath112 with @xmath217 . the proof is contained in the supplementary material @xcite ; it is a relatively straightforward consequence of corollary [ corsep ] and known concentration properties of @xmath188 as an estimate of the population covariance matrix . hence , if @xmath217 for all edges @xmath218 , corollary [ corglobising ] guarantees that the log - determinant method plus thresholding recovers the full graph exactly . in the case of the standard sample covariance matrix , a variant of the graphical lasso has been implemented by banerjee , el ghaoui and daspremont @xcite . our analysis establishes consistency of the graphical lasso for ising models on single separator graphs using @xmath219 samples . this lower bound on the sample size is unavoidable , as shown by information - theoretic analysis @xcite , and also appears in other past work on ising models @xcite . our analysis also has a _ cautionary message _ : the proof of corollary [ corglobising ] relies heavily on the population - level result in corollary [ cortree ] , which ensures that @xmath206 is graph - structured when @xmath14 has only singleton separators . for a general graph , we have no guarantees that @xmath206 will be graph - structured [ e.g. , see panel ( b ) in figure [ figgraphs ] ] , so the graphical lasso ( [ eqnlogdet ] ) is _ inconsistent in general_. on the positive side , if we restrict ourselves to tree - structured graphs , the estimator ( [ eqnlogdet ] ) is attractive , since it relies only on an estimate @xmath188 of the population covariance @xmath186 that satisfies the deviation condition ( [ eqnlogdettreecond ] ) . in particular , even when the samples @xmath220 are contaminated by noise or missing data , we may form a good estimate @xmath188 of @xmath186 . furthermore , the program ( [ eqnlogdet ] ) is always convex regardless of whether @xmath188 is positive semidefinite . as a concrete example of how we may correct the program ( [ eqnlogdet ] ) to handle corrupted data , consider the case when each entry of @xmath221 is missing independently with probability @xmath222 , and the corresponding observations @xmath223 are zero - filled for missing entries . a natural estimator is @xmath224 where @xmath225 denotes elementwise division by the matrix @xmath226 with diagonal entries @xmath227 and off - diagonal entries @xmath228 , correcting for the bias in both the mean and second moment terms . the deviation condition ( [ eqnlogdettreecond ] ) may be shown to hold w.h.p . , where @xmath229 scales with @xmath227 ( cf . loh and wainwright @xcite ) . similarly , we may derive an appropriate estimator @xmath188 for other forms of additive or multiplicative corruption . generalizing to the case of @xmath108-ary discrete graphical models with @xmath230 , we may easily modify the program ( [ eqnlogdet ] ) by replacing the elementwise @xmath0-penalty by the corresponding group @xmath0-penalty , where the groups are the indicator variables for a given vertex . precise theoretical guarantees follow from results on the group graphical lasso @xcite . turning to local neighborhood selection methods , recall the neighborhood - based method due to meinshausen and bhlmann @xcite . in a gaussian graphical model , the column corresponding to node @xmath13 in the inverse covariance matrix @xmath231 is a scalar multiple of @xmath232 , the limit of the linear regression vector for @xmath16 upon @xmath233 . based on @xmath234 i.i.d . samples from a @xmath184-dimensional multivariate gaussian distribution , the support of the graph may then be estimated consistently under the usual lasso scaling @xmath235 , where @xmath236 . motivated by our population - level results on the graph structure of the inverse covariance matrix ( corollary [ cortree ] ) , we now propose a method for neighborhood selection in a tree - structured graph . although the method works for arbitrary @xmath108-ary trees , we state explicit results only in the case of the binary ising model to avoid cluttering our presentation . the method is based on the following steps . for each node @xmath11 , we first perform @xmath0-regularized linear regression of @xmath16 against @xmath233 by solving the modified lasso program @xmath237 where @xmath238 is a constant , @xmath239 are suitable estimators for @xmath240 , and @xmath207 is an appropriate parameter . we then combine the neighborhood estimates over all nodes via an and operation [ edge @xmath112 is present if both @xmath13 and @xmath83 are inferred to be neighbors of each other ] or an or operation ( at least one of @xmath13 or @xmath83 is inferred to be a neighbor of the other ) . note that the program ( [ eqnnoisylasso ] ) differs from the standard lasso in the form of the constraint . indeed , the normal setting of the lasso assumes a linear model where the predictor and response variables are linked by independent sub - gaussian noise , but this is not the case for @xmath16 and @xmath233 in a discrete graphical model . furthermore , the generality of the program ( [ eqnnoisylasso ] ) allows it to be easily modified to handle corrupted variables via an appropriate choice of @xmath241 , as in loh and wainwright @xcite . the following algorithm summarizes our nodewise regression procedure for recovering the neighborhood set @xmath118 of a given node @xmath13 : [ algnodewise ] 1 . form a suitable pair of estimators @xmath242 for covariance submatrices ( @xmath243 , @xmath244 ) . 2 . optimize the modified lasso program ( [ eqnnoisylasso ] ) with parameter @xmath207 , and denote the solution by @xmath245 . 3 . threshold the entries of @xmath245 at level @xmath210 , and define the estimated neighborhood set @xmath246 as the support of the thresholded vector . in the case of fully observed i.i.d . observations , we choose @xmath247 to be the recentered estimators @xmath248 and assign the parameters @xmath211 according to the scaling @xmath249 where @xmath250 and @xmath192 is some parameter such that @xmath251 is sub - gaussian with parameter @xmath252 for any @xmath116-sparse vector @xmath253 , and @xmath192 is independent of @xmath253 . the following result applies to algorithm [ algnodewise ] using the pairs @xmath247 and @xmath254 defined as in equations ( [ eqnlassopair ] ) and ( [ eqnlambdatau2 ] ) , respectively . [ propising ] suppose we have i.i.d . observations @xmath220 from an ising model and that @xmath255 . then there are universal constants @xmath256 such that with probability greater than @xmath216 , for any node @xmath257 , algorithm [ algnodewise ] recovers all neighbors @xmath130 for which @xmath258 . we prove this proposition in the supplementary material @xcite , as a corollary of a more general theorem on the @xmath195-consistency of the program ( [ eqnnoisylasso ] ) for estimating @xmath259 , allowing for corrupted observations . the theorem builds upon the analysis of loh and wainwright @xcite , introducing techniques for @xmath195-bounds and departing from the framework of a linear model with independent sub - gaussian noise . regarding the sub - gaussian parameter @xmath192 appearing in proposition [ propising ] , note that we may always take @xmath260 , since @xmath261 when @xmath253 is @xmath116-sparse and @xmath221 is a binary vector . this leads to a sample complexity requirement of @xmath262 . we suspect that a tighter analysis , possibly combined with assumptions about the correlation decay of the graph , would reduce the sample complexity to the scaling @xmath263 , as required by other methods with fully observed data @xcite . see the simulations in section [ secsims ] for further discussion . for corrupted observations , the strength and type of corruption enters into the factors @xmath264 appearing in the deviation bounds ( c.2a ) and ( c.2b ) below , and proposition [ propising ] has natural extensions to the corrupted case . we emphasize that although analogs of proposition [ propising ] exist for other methods of graph selection based on logistic regression and/or mutual information , the theoretical analysis of those methods does not handle corrupted data , whereas our results extend easily with the appropriate scaling . in the case of @xmath108-ary tree - structured graphical models with @xmath265 , we may perform multivariate regression with the multivariate group lasso @xcite for neighborhood selection , where groups are defined ( as in the log - determinant method ) as sets of indicators for each node . the general relationship between the best linear predictor and the block structure of the inverse covariance matrix follows from block matrix inversion , and from a population - level perspective , it suffices to perform multivariate linear regression of all indicators corresponding to a given node against all indicators corresponding to other nodes in the graph . the resulting vector of regression coefficients has nonzero blocks corresponding to edges in the graph . we may also combine these ideas with the group lasso for multivariate regression @xcite to reduce the complexity of the algorithm . moving on from tree - structured graphical models , our method suggests a graph recovery method based on nodewise linear regression for general discrete graphs . note that by corollary [ corneigh ] , the inverse of @xmath266 is @xmath13-block graph - structured , where @xmath116 is such that @xmath267 . it suffices to perform a single multivariate regression of the indicators @xmath268 corresponding to node @xmath13 upon the other indicators in @xmath269 . we again make precise statements for the binary ising model ( @xmath270 ) . in this case , the indicators @xmath271 corresponding to a subset of vertices @xmath7 of size @xmath272 are all @xmath273 distinct products of variables @xmath274 , for @xmath275 . hence , to recover the @xmath116 neighbors of node @xmath13 , we use the following algorithm . note that knowledge of an upper bound @xmath116 is necessary for applying the algorithm . [ algnontreenodecay ] 1 . use the modified lasso program ( [ eqnnoisylasso ] ) with a suitable choice of @xmath247 and regularization parameter @xmath207 to perform a linear regression of @xmath16 upon all products of subsets of variables of @xmath233 of size at most @xmath116 . denote the solution by @xmath245 . threshold the entries of @xmath245 at level @xmath210 , and define the estimated neighborhood set @xmath246 as the support of the thresholded vector . our theory states that at the population level , nonzeros in the regression vector correspond exactly to subsets of @xmath118 . hence , the statistical consistency result of proposition [ propising ] carries over with minor modifications . since algorithm [ algnontreenodecay ] is essentially a version of algorithm [ algnontree ] with the first two steps omitted , we refer the reader to the statement and proof of corollary [ cornontree ] below for precise mathematical statements . note here that since the regression vector has @xmath276 components , of which are nonzero , the sample complexity of lasso regression in step ( 1 ) of algorithm [ algnontreenodecay ] is @xmath277 . for graphs exhibiting correlation decay @xcite , we may reduce the computational complexity of the nodewise selection algorithm by prescreening the nodes of @xmath278 before performing a lasso - based linear regression . we define the nodewise correlation according to @xmath279 and say that the graph exhibits _ correlation decay _ if there exist constants @xmath280 such that @xmath281 for all @xmath282 , where @xmath283 is the length of the shortest path between @xmath13 and @xmath83 . with this notation , we then have the following algorithm for neighborhood recovery of a fixed node @xmath13 in a graph with correlation decay : [ algnontree ] 1 . compute the empirical correlations @xmath284 between @xmath13 and all other nodes @xmath285 , where @xmath286 denotes the empirical distribution . 2 . let @xmath287 be the candidate set of nodes with sufficiently high correlation . ( note that @xmath51 is a function of both @xmath13 and @xmath288 and , by convention , @xmath289 . ) use the modified lasso program ( [ eqnnoisylasso ] ) with parameter @xmath207 to perform a linear regression of @xmath16 against @xmath290 , the set of all products of subsets of variables @xmath291 of size at most @xmath116 , together with singleton variables . denote the solution by @xmath245 . 4 . threshold the entries of @xmath245 at level @xmath210 , and define the estimated neighborhood set @xmath246 as the support of the thresholded vector . note that algorithm [ algnontreenodecay ] is a version of algorithm [ algnontree ] with @xmath292 , indicating the absence of a prescreening step . hence , the statistical consistency result below applies easily to algorithm [ algnontreenodecay ] for graphs with no correlation decay . for fully observed i.i.d . observations , we choose @xmath241 according to @xmath293 and parameters @xmath211 as follows : for a candidate set @xmath51 , let @xmath294 denote the augmented vector corresponding to the observation @xmath221 , and define @xmath295 . let @xmath296 . then set @xmath297 where @xmath192 is some function such that @xmath298 is sub - gaussian with parameter @xmath299 for any @xmath300-sparse vector @xmath253 , and @xmath192 does not depend on @xmath253 . we have the following consistency result , the analog of proposition [ propising ] for the augmented set of vectors . it applies to algorithm [ algnontree ] with the pairs @xmath239 and @xmath211 chosen as in equations ( [ eqngamparam ] ) and ( [ eqnlambdatau3 ] ) . [ cornontree ] consider i.i.d . observations @xmath220 generated from an ising model satisfying the correlation decay condition ( [ eqncorrdecay ] ) , and suppose @xmath301 then there are universal constants @xmath256 such that with probability at least @xmath302 , and for any @xmath257 : the set @xmath51 from step ( 2 ) of algorithm [ algnontree ] satisfies @xmath303 . algorithm [ algnontree ] recovers all neighbors @xmath304 such that @xmath305 the proof of corollary [ cornontree ] is contained in the supplementary material @xcite . due to the exponential factor @xmath306 appearing in the lower bound ( [ eqnham ] ) on the sample size , this method is suitable only for bounded - degree graphs . however , for reasonable sizes of @xmath116 , the dimension of the linear regression problem decreases from @xmath276 to @xmath307 , which has a significant impact on the runtime of the algorithm . we explore two classes of bounded - degree graphs with correlation decay in the simulations of section [ secsims ] , where we generate erds renyi graphs with edge probability @xmath308 and square grid graphs in order to test the behavior of our recovery algorithm on nontrees . when @xmath265 , corresponding to nonbinary states , we may combine these ideas with the overlapping group lasso @xcite to obtain similar algorithms for nodewise recovery of nontree graphs . however , the details are more complicated , and we do not include them here . note that our method for nodewise recovery in nontree graphical models is again easily adapted to handle noisy and missing data , which is a clear advantage over other existing methods . in this section we report the results of various simulations we performed to illustrate the sharpness of our theoretical claims . in all cases , we generated data from binary ising models . we first applied the nodewise linear regression method ( algorithm [ algnodewise ] for trees ; algorithm [ algnontreenodecay ] in the general case ) to the method of @xmath0-regularized logistic regression , analyzed in past work for ising model selection by ravikumar , wainwright and lafferty @xcite . their main result was to establish that , under certain incoherence conditions of the fisher information matrix , performing regularized logistic regression with a sample size @xmath309 is guaranteed to select the correct graph w.h.p . thus , for any bounded - degree graph , the sample size @xmath234 need grow only logarithmically in the number of nodes @xmath184 . under this scaling , our theory also guarantees that nodewise _ linear regression _ with @xmath0-regularization will succeed in recovering the true graph w.h.p . @c@c@ -regularized logistic vs. linear regression methods for graph recovery . each panel plots of the probability of correct graph recovery vs. the rescaled sample size @xmath310 ; solid curves correspond to linear regression ( method in this paper ) , whereas dotted curves correspond to logistic regression @xcite . curves are based on average performance over 500 trials . simulation results for two - dimensional grids with @xmath311 neighbors , and number of nodes @xmath184 varying over @xmath312 . consistent with theory , when plotted vs. the rescaled sample size @xmath313 , all three curves ( red , blue , green ) are well aligned with one another . both linear and logistic regression transition from failure to success at a similar point . analogous results for an erds renyi graph with edge probability @xmath314 . analogous results for a chain - structured graph with maximum degree @xmath315.,title="fig : " ] & -regularized logistic vs. linear regression methods for graph recovery . each panel plots of the probability of correct graph recovery vs. the rescaled sample size @xmath310 ; solid curves correspond to linear regression ( method in this paper ) , whereas dotted curves correspond to logistic regression @xcite . curves are based on average performance over 500 trials . simulation results for two - dimensional grids with @xmath311 neighbors , and number of nodes @xmath184 varying over @xmath312 . consistent with theory , when plotted vs. the rescaled sample size @xmath313 , all three curves ( red , blue , green ) are well aligned with one another . both linear and logistic regression transition from failure to success at a similar point . analogous results for an erds renyi graph with edge probability @xmath314 . analogous results for a chain - structured graph with maximum degree @xmath315.,title="fig : " ] + & + + in figure [ figsimsone ] we present the results of simulations with two goals : ( i ) to test the @xmath316 scaling of the required sample size ; and ( ii ) to compare @xmath0-regularized nodewise linear regression ( algorithms [ algnontreenodecay ] and [ algnontree ] ) to @xmath0-regularized nodewise logistic regression @xcite . we ran simulations for the two methods on both tree - structured and nontree graphs with data generated from a binary ising model , with node weights @xmath317 and edge weights @xmath318 . to save on computation , we employed the neighborhood screening method described in section [ secnontree ] to prune the candidate neighborhood set before performing linear regression . we selected a candidate neighborhood set of size @xmath319 with highest empirical correlations , then performed a single regression against all singleton nodes and products of subsets of the candidate neighborhood set of size at most @xmath116 , via the modified lasso program ( [ eqnnoisylasso ] ) . the size of the candidate neighborhood set was tuned through repeated runs of the algorithm . for both methods , the optimal choice of regularization parameter @xmath207 scales as @xmath320 , and we used the same value of @xmath207 in comparing logistic to linear regression . in each panel we plot the probability of successful graph recovery versus the rescaled sample size @xmath321 , with curves of different colors corresponding to graphs ( from the same family ) of different sizes . solid lines correspond to linear regression , whereas dotted lines correspond to logistic regression ; panels ( a ) , ( b ) and ( c ) correspond to grid graphs , erds renyi random graphs and chain graphs , respectively . for all these graphs , the three solid / dotted curves for different problem sizes are well aligned , showing that the method undergoes a transition from failure to success as a function of the ratio @xmath321 . in addition , both linear and logistic regression are comparable in terms of statistical efficiency ( the number of samples @xmath234 required for correct graph selection to be achieved ) . @cc@ of missing data varying in @xmath322 . panel shows simulation results for nodewise regression applied to chain graphs for varying @xmath184 and @xmath222 . panel shows simulation results for nodewise regression applied to star graphs with maximal node degree @xmath323 and varying @xmath222.,title="fig : " ] & of missing data varying in @xmath322 . panel shows simulation results for nodewise regression applied to chain graphs for varying @xmath184 and @xmath222 . panel shows simulation results for nodewise regression applied to star graphs with maximal node degree @xmath323 and varying @xmath222.,title="fig : " ] + & + + the main advantage of nodewise linear regression and the graphical lasso over nodewise logistic regression is that they are straightforward to correct for corrupted or missing data . figure [ figsimstwo ] shows the results of simulations designed to test the behavior of these corrected estimators in the presence of missing data . panel ( a ) shows the results of applying the graphical lasso method , as described in section [ secglobal ] , to the dino graph of figure [ figgraphs](e ) . we again generated data from an ising model with node weights @xmath324 and edge weights @xmath325 . the curves show the probability of success in recovering the 15 edges of the graph , as a function of the rescaled sample size @xmath321 for @xmath326 . in addition , we performed simulations for different levels of missing data , specified by the parameter @xmath327 , using the corrected estimator ( [ eqnmissdata ] ) . note that all five runs display a transition from success probability 0 to success probability 1 in roughly the same range , as predicted by our theory . indeed , since the dinosaur graph has only singleton , corollary [ cortree ] ensures that the inverse covariance matrix is exactly graph - structured , so our global recovery method is consistent at the population level . further note that the curves shift right as the fraction @xmath222 of missing data increases , since the recovery problem becomes incrementally harder . panels ( b ) and ( c ) of figure [ figsimstwo ] show the results of the nodewise regression method of section [ secnodewise ] applied to chain and star graphs , with increasing numbers of nodes @xmath328 and @xmath329 , respectively . for the chain graphs in panel ( b ) , we set node weights of the ising model equal to @xmath324 and edge weights equal to @xmath325 . for the varying - degree star graph in panel ( c ) , we set node weights equal to @xmath324 and edge weights equal to @xmath330 , where the degree @xmath116 of the central hub grows with the size of the graph as @xmath331 . again , we show curves for different levels of missing data , @xmath332 . the modified lasso program ( [ eqnnoisylasso ] ) was optimized using a form of composite gradient descent due to agarwal , negahban and wainwright @xcite , guaranteed to converge to a small neighborhood of the optimum even when the problem is nonconvex @xcite . in both the chain and star graphs , the three curves corresponding to different problem sizes @xmath184 at each value of the missing data parameter @xmath222 stack up when plotted against the rescaled sample size . note that the curves for the star graph stack up nicely with the scaling @xmath333 , rather than the worst - case scaling @xmath334 , corroborating the remark following proposition [ propising ] . since @xmath335 is fixed for the chain graph , we use the rescaled sample size @xmath321 in our plots , as in the plots in figure [ figsimsone ] . once again , these simulations corroborate our theoretical predictions : the corrected linear regression estimator remains consistent even in the presence of missing data , although the sample size required for consistency grows as the fraction of missing data @xmath222 increases . the correspondence between the inverse covariance matrix and graph structure of a gauss markov random field is a classical fact with numerous consequences for estimation of gaussian graphical models . it has been an open question as to whether similar properties extend to a broader class of graphical models . in this paper , we have provided a partial affirmative answer to this question and developed theoretical results extending such relationships to discrete undirected graphical models . as shown by our results , the inverse of the ordinary covariance matrix is graph - structured for special subclasses of graphs with singleton separator sets . more generally , we have considered inverses of _ generalized covariance matrices _ , formed by introducing indicator functions for larger subsets of variables . when these subsets are chosen to reflect the structure of an underlying junction tree , the edge structure is reflected in the inverse covariance matrix . our population - level results have a number of statistical consequences for graphical model selection . we have shown that our results may be used to establish consistency ( or inconsistency ) of standard methods for discrete graph selection , and have proposed new methods for neighborhood recovery which , unlike existing methods , may be applied even when observations are systematically corrupted by mechanisms such as additive noise and missing data . furthermore , our methods are attractive in their simplicity , in that they only involve simple optimization problems . thanks to the associate editor and anonymous reviewers for helpful feedback .

a comprehensive description of electromagnetism in classical theory is given in terms of the electromagnetic field strength acting locally and directly on charged matter . this is not the case in quantum theory where a state of charged matter is influenced by electromagnetic field even in the situation when the region of the nonvanishing field strength does not overlap with the region accessible to charged matter ; the indispensable condition is that the latter region be non - simply - connected . in particular , a magnetic field of an infinitely long current - carrying solenoid can be shielded from the penetration of charged matter ; such a field configuration may be denoted as an impenetrable magnetic vortex . perhaps , w. ehrenberg and r. e. siday were the first to discover theoretically the purely quantum effect which is due to the influence of the vector potential of the impenetrable magnetic vortex on the motion of a charged particle , and they proposed an interference experiment to detect this @xcite . ten years later the effect was rediscovered by y. aharonov and d. bohm who made an important addition by predicting also an effect which is due to the influence of the electric scalar potential @xcite . immediately , in a year , the first interference experiment to verify the magnetic effect was performed @xcite , and this caused a burst of interest opening a long way of an extensive research both in theory and in experiment , see , e.g. , @xcite and references therein ; the effect conventionally bears the name of the authors of @xcite . although a great progress in understanding the aharonov bohm ( ab ) effect has been achieved since the time of foundational works @xcite , one may agree with the authors of @xcite that `` the investigation and exploitation of the ab effect remain far from finished '' . it should be noted that the concern of paper @xcite was in the elaboration of the consistent theory of refraction of electron rays in the framework of electron wave optics . following thoroughly their flow of reasoning , the authors arrive at the conclusion that the medium in a spatial region outside a magnetic vortex is anisotropic and that there exist wave - optical phenomena which are due to the influence of the vector potential of the vortex . finally , in the very end of the paper , they suggest an interference experiment to verify their findings . an approach of paper @xcite was quite different : the authors heuristically , with the use of few rather general and obvious formulas ( which are even left unnumbered in the text ) , arrive at their suggestion of interference experiments already in the beginning of the paper . a powerful theoretical background behind this intuitive suggestion is hereafter given in section 4 . `` exact solution for scattering problems '' ( which exceeds in its volume two preceding sections ) . the background relies directly on the formalism of quantum mechanics , namely on the schrdinger wave equation , which allows the authors to consider their effect as a scattering event . perhaps , this was a reason , why paper @xcite , unlike paper @xcite , had received an immediate response from the physics community . in this respect it looks rather paradoxical that , whereas interference experiments aiming at the verification of the ab effect were numerous ( see @xcite ) , a scattering experiment aiming at the same was _ never _ performed . one of the motivations of the present paper is to provide an explanation for this paradox . let us commence by recalling the famous formula obtained by the authors of @xcite for the scattering differential cross section per unit length of the magnetic vortex ( see ( 21 ) and ( 22 ) in @xcite ) : @xmath2 } { \sin^2(\varphi/2)},\label{eq1}\ ] ] where @xmath3 is the momentum of a scattered particle , @xmath4 is its charge , @xmath5 is the scattering angle ( with @xmath6 corresponding to the forward direction ) and @xmath7 is the flux of the magnetic vortex directed along the @xmath8-axis . a characteristic feature of ( 1 ) is a periodic dependence in the magnetic flux with the period equal to the london flux quantum , @xmath9 . also , one could suggest that ( 1 ) describes a purely quantum effect , since it vanishes in the formal limit of @xmath10 , or , more physically , in the limit of large enough values of the particle momentum , @xmath11 . however , result ( 1 ) corresponds to the idealized case of the magnetic vortex of the vanishing transverse size . since any real magnetic vortex ( either that produced by a current - carrying solenoid , or that inside a magnet ) is of a nonzero transverse size , the quotient of squared sines in ( 1 ) in the real case becomes a certain function of dimensionless variable @xmath12 , where @xmath13 is a length characterizing the transverse size of the vortex . the transverse size effects were taken into account in @xcite , and results were found to converge with ( 1 ) in the case of @xmath14 . as to the case of @xmath15 , it was only scarcely and insufficiently considered , see @xcite , and , therefore , let us dwell on this case at more length . it should be noted in the first place that the smallest possible values of @xmath13 can be of order 1 @xmath16 ( ferromagnetic whiskers enclosed in screening shells ) @xcite , while , in a scattering event , the largest possible wavelengths , @xmath17 , of the lightest charged particle , electron , are of order 0.1 nm ( slowly moving electrons of energies 10 - 100 ev , e.g. , such as those in famous diffraction experiments by c. j. davisson and l. h. germer in 1927 , nobel prize in 1937 ) . hence , quantity @xmath18 takes values of order @xmath19 at least , and formula ( 1 ) which might be supposed to approximate the case of @xmath20 has no relation to scattering of real particles . albeit this formula looks essentially quantum , the `` quantum '' factor of @xmath21 appears just due to dimensional reasons because of @xmath22 ; it is a somewhat more explicating to write @xmath21 as product @xmath23 , where the second ( dimensionless ) factor is vanishingly small for all charged particles existing in nature . this was certainly understood by the authors of @xcite , who never proposed cross section ( 1 ) for the experimental verification . it may seem quite reasonable to apply the quasiclassical ( called also semiclassical ) method of g. wentzel , h. a. kramers and l. brillouin ( wkb ) @xcite to solving the problem of scattering of real charged particles by an impenetrable magnetic vortex : the size of the interaction region may be estimated as being of order of @xmath13 and larger , then condition @xmath24 allows one to hope that the use of the wkb method will be adequate , see , e.g , @xcite . however , a skepticism immediately arises : how can a purely and essentially quantum effect as is the ab one be deduced with the help of a quasiclassical ( semiclassical ) method which is commonly believed to give nothing else but the yield of classical theory ? in this respect it should be noted that the findings of the authors of @xcite indicate that scattering off an impenetrable magnetic vortex in the case of @xmath15 is reduced to classical scattering of point particles off an impenetrable tube with no flux . nevertheless , in spite of the skepticism substantiated by the results of @xcite , we shall proceed in the present paper by applying the wkb method to the solution of the scattering ab problem . let us recall that interference experiments aiming at the verification of the ab effect were performed with rather energetic electrons . we are considering electrons of the same ( as in the interference experiments ) and even smaller ( down to 10 - 20 ev ) energies and study direct scattering of such particles on an impenetrable magnetic vortex , using the wkb method . whereas the authors of @xcite employed the dirichlet boundary condition to ensure the vortex impenetrability , we go further by employing the most general ( unorthodox in the language of @xcite ) boundary condition that is required by the basic concepts of probability conservation and self - adjointness of the hamilton operator . in the course of our study we shall find , in addition to the classical effect of @xcite , a purely quantum effect that can be denoted as the scattering ab effect with real particles . in the next section we consider a quantum - mechanical charged particle in the background of an impenetrable magnetic vortex and choose a boundary condition for the particle wave function at the vortex edge . the wkb method is used to obtain the partial waves of the solution to the schrdinger equation in section 3 . the scattering amplitude and cross section are obtained in section 4 . the conclusions are drawn and discussed in section 5 . we relegate some details of calculation of the scattering amplitude to appendix a , while the use of the hamilton - jacobi equation of motion to derive the classical effect is outlined in appendix b. defining a scalar product as @xmath25 , we get , using integration by parts , @xmath26\chi\right\}= \nonumber \\ = \!\int\limits_{\omega}\!{\rm d}^3x\left\{\left [ -\frac{1}{2m}\left(\hbar{{\mbox{\boldmath $ \partial$}}}-{\rm i}\frac{e}{c}{\bf a}\right)^2\!+\!v\right]\tilde{\chi}\right\}^*\chi- \frac{\hbar}{2m}\int\limits_{\partial\omega}\!{\rm d}{{\mbox{\boldmath $ \sigma$}}}\cdot \left\{\tilde{\chi}^*\left[\left(\hbar{{\mbox{\boldmath $ \partial$}}}-{\rm i}\frac{e}{c}{\bf{a}}\right)\chi\right]\right.\!-\nonumber \\ \left.-\left[\left(\hbar{{\mbox{\boldmath $ \partial$}}}+{\rm i}\frac{e}{c}{\bf{a}}\right)\tilde{\chi}^*\right]\chi\right\}\equiv \left(h^\dag\tilde{\chi},\chi\right)-{\rm i}\hbar\int\limits_{\partial\omega}{\rm d}{{\mbox{\boldmath $ \sigma$}}}\cdot{\bf u } , \label{eq2}\end{aligned}\ ] ] where @xmath27 is a two - dimensional surface bounding the three - dimensional spatial region @xmath28 , @xmath29 is the formal expression for a general quantum - mechanical hamilton operator in an external electromagnetic field ( electric scalar potential is included in the general potential energy part @xmath30 ) , and @xmath31 -\left[\left(\hbar{\mbox{\boldmath $ \partial$}}+{\rm i}\frac{e}{c}{\bf a}\right)\tilde{\chi}^*\right]\chi\right\}.\label{eq4}\ ] ] operator @xmath32 is hermitian ( symmetric ) , @xmath33 if @xmath34 the latter condition can be satisfied in various ways by imposing different boundary conditions for @xmath35 and @xmath36 . however , among the whole variety , there may exist a possibility that a boundary condition for @xmath36 is the same as that for @xmath35 ; then the domain of definition of @xmath37 ( set of functions @xmath36 ) coincides with that of @xmath32 ( set of functions @xmath35 ) , and operator @xmath32 is called self - adjoint . whether such a possibility exists is in general determined by the weyl von neumann theory of self - adjoint operators , see , e.g. , @xcite . the action of a self - adjoint operator results in functions belonging to its domain of definition only , and , therefore , a multiple action and functions of such an operator ( for instance , the evolution operator ) can be consistently defined . in the present case , the problem of self - adjointness of operator @xmath32 ( 3 ) is resolved by imposing a boundary condition in the form @xmath38\biggr|_{{\bf x}\in\partial\omega}= \nonumber \\ \biggl.=\left[\sin(\rho\pi)\,{\bf n}\cdot\left({\mbox{\boldmath $ \partial$}}-\frac{{\rm i}e}{\hbar c}{\bf a}\right)\tilde{\chi}+ \cos(\rho\pi)\frac{1}{\bf n\cdot{\bf x}}\tilde{\chi}\right]\biggr|_{{\bf x}\in\partial\omega}=0 , \label{eq7}\end{aligned}\ ] ] where @xmath39 is the internal unit normal to boundary @xmath27 and @xmath40 is the real parameter ( the self - adjoint extension parameter ) . defining current @xmath41- \left[\left(\hbar{\mbox{\boldmath $ \partial$}}+{\rm i}\frac{e}{c}{\bf a}\right)\chi^*\right]\chi\right\},\label{eq8}\ ] ] we note that self - adjointness condition ( 7 ) results in the vanishing of the normal component of the current at the boundary : @xmath42 if @xmath35 is a solution to the schrdinger equation , @xmath43 then current ( 8) is the probability current density satisfying the continuity equation @xmath44 thus , self - adjointness of the hamilton operator ensures probability conservation , @xmath45 . in the following we consider a quantum - mechanical charged particle in the background of a static magnetic field , hence @xmath46 . the magnetic field is confined to an infinite tube which is classically impenetrable to the charged particle ; thus , spatial region @xmath28 is an infinite three - dimensional space with the exclusion of the tube containing the magnetic flux lines , and boundary surface @xmath27 is a surface of the tube . assuming the cylindrical symmetry of the flux tube , we use cylindrical coordinates @xmath47 with the @xmath8-axis coinciding with the axis of the tube and choose the components of the vector potential in @xmath28 in the form @xmath48 where @xmath7 is the total magnetic flux confined in the tube . then the hamilton operator takes form @xmath49,\label{eq13}\ ] ] and the solution to schrdinger equation ( 10 ) with @xmath32 ( 13 ) is @xmath50 where @xmath51 is the solution to equation @xmath52\psi(r,\varphi)=0,\label{eq15}\ ] ] and obeys the boundary condition , see ( 7 ) , @xmath53|_{r = r_c}=0;\label{eq16}\ ] ] @xmath13 is the tube radius , @xmath54 is the energy of the stationary scattering state , @xmath55 and @xmath3 are the momenta in the longitudinal and radial transverse directions , respectively . one could recognize that ( 16 ) is known as the robin boundary condition ; the case of @xmath56 corresponds to the dirichlet condition ( perfect reflectivity of the boundary ) and the case of @xmath57 corresponds to the neumann condition ( absolute rigidity of the boundary ) . it should be emphasized that parameter @xmath40 is in general dependent on @xmath8 and @xmath5 . thus , the `` number '' of self - adjoint extension parameters is infinite , moreover , it is not countable but is of power of a continuum . this distinguishes the case of an extended boundary from the case of an excluded point ( contact interaction ) when the number of self - adjoint extension parameters is finite , being equal to @xmath58 for the deficiency index equal to \{@xmath59 } ( see , e.g. , @xcite ) . the motion of the particle in the longitudinal direction is free , and our aim is to determine the particle motion in the orthogonal plane , i.e. to find a solution to ( 15 ) and ( 16 ) . let us start with the decomposition of wave function @xmath51 into partial waves @xmath60 where @xmath61 is the set of integer numbers . the radial component of each partial wave satisfies equation @xmath62r_n(r)=0,\label{eq18}\ ] ] where @xmath63 and @xmath64 @xmath65 . note that the amplitude of @xmath51 ( 17 ) is periodic in the value of flux @xmath7 with the period equal to the london flux quantum . we present the radial component of the partial wave as @xmath66 $ ] , where @xmath67 satisfies the ricatti - type equation @xmath68 expanding @xmath67 into series @xmath69 and collecting terms of the same order in @xmath70 , we obtain the system of equations : @xmath71 ^ 2=p_n^2 , \\ 2\left[\partial_r\sigma_n^{(0)}\right]\left[\partial_r\sigma_n^{(1)}\right]=-pr_c\delta_r\sigma_n^{(0 ) } , \\ 2\left[\partial_r\sigma_n^{(0)}\right]\left[\partial_r\sigma_n^{(2)}\right]=-pr_c\delta_r\sigma_n^{(1)}- \left[\partial_r\sigma_n^{(1)}\right]^2 , \\ \qquad\qquad \vdots \\ 2\left[\partial_r\sigma_n^{(0)}\right]\left[\partial_r\sigma_n^{(2l)}\right]=-pr_c\delta_r\sigma_n^{(2l-1)}- 2\sum\limits_{l'=1}^{l-1}\left[\partial_r\sigma_n^{(l')}\right]\left[\partial_r\sigma_n^{(2l - l')}\right]- \left[\partial_r\sigma_n^{(l)}\right]^2,\\ 2\left[\partial_r\sigma_n^{(0)}\right]\left[\partial_r\sigma_n^{(2l+1)}\right]=-pr_c\delta_r\sigma_n^{(2l)}- 2\sum\limits_{l'=1}^{l}\left[\partial_r\sigma_n^{(l')}\right]\left[\partial_r\sigma_n^{(2l+1-l')}\right ] , \\ \qquad\qquad \vdots \end{array } \label{eq22}\ ] ] consecutively solving equations beginning from the top , one can obtain @xmath67 ( 21 ) up to an arbitrary order in expansion parameter @xmath72 . the terms corresponding to odd @xmath73 contribute to the amplitude of @xmath74 , while the terms corresponding to even @xmath73 contribute to the phase of @xmath74 . however , the power of the wkb method reveals fully itself in those cases when it suffices to take account for the first two terms , @xmath75 and @xmath76 , only . it should be noted that equation ( 18 ) , or ( 20 ) , has its analogue in optics , describing the propagation of waves in media @xcite : quotient @xmath77 corresponds to the index of refraction . terms @xmath78 @xmath79 are negligible , if the variation of the refraction index along the ray of propagation becomes appreciable on the distances which are much larger than the effective wavelength , @xmath80 where @xmath39 points in the direction of the ray which in our case is orthogonal to the boundary . in view of ( 19 ) this yields condition @xmath81 which is satisfied in the quasiclassical region , @xmath82 as well as in the deeply nonclassical region @xmath83 solving the first two equations in ( 22 ) , we get @xmath84.\label{eq27}\ ] ] it should be noted that the second equation in ( 22 ) is nothing more but the continuity equation , @xmath85 , for the partial wave current , @xmath86,\ ] ] in the quasiclassical region . actually , the wkb approximation ( i.e. the neglect of @xmath78 with @xmath87 ) is efficient in a much more extended regions than those given by ( 25 ) and ( 26 ) . the wkb approximation is not valid in the vicinity of @xmath88 which is defined by @xmath89 this point is the turning point of a classical trajectory of the particle . in view of ( 27 ) , the two linearly independent solutions to ( 18 ) in the quasiclassical region are @xmath90 while the two linearly independent solutions to ( 18 ) in the deeply nonclassical region are @xmath91 where @xmath92 the crucial question is how oscillating solutions ( 29 ) in the outer region go over to solutions ( 30 ) in the inner region and vice verse . to answer this question , one has to find a solution in the vicinity of the turning point , where @xmath93 and @xmath94 , and to match it with the solutions in the inner and the outer regions . but a more instructive way , as stated in @xcite , is to consider formally @xmath74 as a function of complex variable @xmath95 and to perform transitions along paths in the complex plane , where condition ( 24 ) is satisfied . namely , a path goes along the real positive semiaxis up to the left vicinity of the turning point , circumvents it along a semicircle in the upper or lower half - plane , and then goes again along the real positive semiaxis . for the transition from the inner to the outer region we have @xmath96 where the ( @xmath97 ) sign corresponds to the path in the upper or lower half - plane , and @xmath98 @xmath99 , in contrast to @xmath100 , is divergent in the inner region . for the transition from the outer to the inner region we have @xmath101 and @xmath102 \rightarrow r_{n,{\rm in}}^{(-)}(r),\label{eq35}\ ] ] where the account is taken for the fact that @xmath103 is real . boundary condition ( 16 ) in terms of partial waves takes form @xmath104|_{r = r_c}=0.\label{eq36}\ ] ] in the case @xmath105 we choose @xmath106e^{{\rm i}\pi/4},\label{eq37}\ ] ] where the factor of @xmath107 is inserted for future convenience . coefficient @xmath108 is determined from condition ( 36 ) : @xmath109|_{r = r_c } } { \cot(\rho\pi)+[r\partial_r\ln r_{n,{\rm out}}^{(+)}(r)]|_{r = r_c}}= \nonumber \\ = \exp\left\{-\frac{2{\rm i}}{\hbar}\int\limits_{r_t}^{r_c}{\rm d}r\,p_n-2{\rm i}\arctan\left [ \frac{r_cp_n(r_c)\hbar^{-1}}{\cot(\rho\pi)-\frac{1}{2}\,\frac{p^2}{p_n^2(r_c)}}\right]\right\ } , \qquad r_c > r_t ; \label{eq38}\end{aligned}\ ] ] note that the explicit form of the turning point ( i.e. the solution to ( 28 ) ) is evidently @xmath110 . in the case @xmath111 we choose @xmath112,\label{eq39}\ ] ] where coefficient @xmath108 is determined from condition ( 36 ) : @xmath113 @xmath114\right|_{r = r_c } } { \cot(\rho\pi)+\left.\left\{r\partial_r\ln \left[r_{n,{\rm in}}^{(-)}(r)+cr_{n,{\rm in}}^{(+)}(r)\right]\right\ } \right|_{r = r_c } } , \,\,r_c < r_t,\ ] ] and constant @xmath115 is to be fixed before long . using correspondence rules ( 33 ) and ( 35 ) , we continue @xmath74 ( 39 ) to the case @xmath116 as @xmath117 @xmath118 + c\left[r_{n,{\rm out}}^{(+)}(r)e^{-{\rm i}\pi/4}+ r_{n,{\rm out}}^{(-)}(r)e^{{\rm i}\pi/4}\right]\right\}.\ ] ] the converging wave , @xmath119 , should disappear from the part with coefficient @xmath108 , cf . this fixes the constant , @xmath120 , yielding in the case @xmath116 @xmath121 with @xmath122^{-1},\,\,r_c < r_t.\label{eq41}\end{aligned}\ ] ] we can rewrite ( 37 ) in the case @xmath105 as @xmath123r_{n,{\rm out}}^{(+)}(r)e^{{\rm i}\pi/4}.\label{eq42}\ ] ] we conclude this section by stating that ( 40 ) and ( 42 ) give the radial components of partial waves of @xmath51 ( 17 ) in the quasiclassical region in the wkb approximation . going over to asymptotics @xmath124 , where @xmath125,\qquad \int\limits_{r_t}^{r}drp_n = pr-\frac{\hbar}{2 } @xmath126 where @xmath127 and @xmath128+\right . \nonumber \\ \left.+\sum\limits_{|n-\mu|>pr_c/\hbar}e^{{\rm i}n\varphi}a_ne^{-\frac{\rm i}{2}|n-\mu|\pi } c_n(r_c,\rho)\right\}.\label{eq46}\end{aligned}\ ] ] to determine coefficient @xmath129 , let us recall the asymptotics of the partial wave decomposition of a plane wave , @xmath130= \nonumber \\ = \sqrt{\frac{2\pi\hbar}{pr}}\left[\delta(\varphi)e^{{\rm i}(pr\hbar^{-1}-\pi/4 ) } + \delta(\varphi-\pi)e^{-({\rm i}pr\hbar^{-1}-\pi/4)}\right],\label{eq47}\end{aligned}\ ] ] where @xmath131 is the angular delta - function , @xmath132 . the asymptotics of the plane wave is naturally interpreted as a superposition of two cylindrical waves : the diverging one , @xmath133 , going in the forward , @xmath6 , direction and the converging one , @xmath134 , coming from the backward , @xmath135 , direction . namely the converging cylindrical wave should be present without distortions in the asymptotics of wave function @xmath51 ( 44 ) , whereas the diverging cylindrical wave is distorted and differs from that in ( 47 ) ; actually , this is the condition that @xmath51 be the scattering state solution . equating terms before @xmath134 in @xmath51 to the terms before @xmath136 in the first line of ( 47 ) , we get @xmath137.\label{eq48}\ ] ] as a result , we obtain the following expression for the @xmath13-independent part of the wave function @xmath138 where @xmath139}=\nonumber \\ = \sqrt{\frac{2\pi\hbar}{pr}}\left[e^{-{\rm i}\mu{\rm sgn}(\varphi)\pi } \delta(\varphi)e^{{\rm i}(pr\hbar^{-1}-\pi/4)}+\delta(\varphi-\pi ) e^{-{\rm i}(pr\hbar^{-1}-\pi/4)}\right],\label{eq50}\end{aligned}\ ] ] is the incident wave and @xmath140\!]+\frac 12\right)\varphi } } { \sin(\varphi/2)}\label{eq51}\ ] ] is the scattering amplitude @xcite ; the sign function is @xmath141 at @xmath142 , @xmath143\!]$ ] denotes the integer part of quantity @xmath144 ( i.e. the integer which is less than or equal to @xmath144 ) and it is implied that @xmath145 . the squared absolute value of ( 51 ) yields differential cross section ( 1 ) . incident wave ( 50 ) differs from the plane wave : the distortions are due to the long - range nature of the vector potential outside the flux tube . the apparent divergence of amplitude ( 51 ) in the forward direction is a phantom , because ( 51 ) is valid under condition @xmath146 , whereas , otherwise , the divergence is absent and the discontinuity in incident wave ( 50 ) at @xmath147 is cancelled , yielding @xmath148 which is continuous and differentiable at @xmath6 , see , e.g. , @xcite and references therein . amplitude @xmath149 ( 51 ) is of order @xmath150 $ ] and is negligible as compared to the contribution from the @xmath13-dependent part of the wave function , which can be presented in the following form : @xmath151 \frac{e^{{\rm i}(pr\hbar^{-1}+\pi/4)}}{\sqrt{r}},\label{eq52}\ ] ] where @xmath152 @xmath153 @xmath154 we show in appendix a ( see ( a.22 ) ) that amplitude @xmath155 ( 55 ) , although exceeding amplitude @xmath149 ( 51 ) , is still negligible , being of order @xmath156 $ ] . amplitude @xmath157 ( 54 ) is of nonnegligible value which is calculated in appendix a : @xmath158-{\rm i}\pi/4\right\}\times\nonumber \\ \times\exp\left\{-2{\rm i}\arctan\left[\frac{2pr_c\hbar^{-1}|\sin^3(\varphi/2)| } { 2\cot(\rho\pi)\sin^2(\varphi/2)-1}\right]\right\}.\label{eq56}\end{aligned}\ ] ] amplitude @xmath159 ( 53 ) is a finite sum of geometric progression , which is straightforwardly calculated : @xmath160\!]+\frac 12)\varphi } \frac{\sin(s_c\varphi/2)}{\sin(\varphi/2)}\cos(\mu\pi+s_c\varphi/2)\label{eq57}\ ] ] in the case @xmath161\!]-[\![\mu]\!]=[\![pr_c\hbar^{-1}-\mu]\!]+[\![\mu]\!]+1=s_c,\label{eq58}\ ] ] or @xmath162\!]+\frac 12\mp\frac 12)\varphi } \left\{\frac{\sin[(s_c+1/2)\varphi/2]}{\sin(\varphi/2)}\cos\left[\mu\pi+(s_c+1/2)\varphi/2\right]-\right . \nonumber \\\left.-\tan(\varphi/4)\pm{\rm i}\right\}\label{eq59}\end{aligned}\ ] ] in the case @xmath161\!]-[\![\mu]\!]-\frac 12\pm\frac 12=[\![pr_c\hbar^{-1}-\mu]\!]+[\![\mu]\!]+ \frac 12\mp\frac 12=s_c.\label{eq60}\ ] ] since @xmath163 , amplitude @xmath159 is strongly peaked in the forward direction ( actually as a smoothed delta - function ) , whereas amplitude @xmath157 is vanishing in the forward direction . therefore , the interference between the amplitudes , @xmath164 , is vanishing in the same manner as @xmath165 . the differential cross section of the scattering process , @xmath166 , is hence a sum of two terms : @xmath167 and @xmath168,\label{eq62}\ ] ] where we have recalled that @xmath65 and introduced function @xmath169 which at @xmath170 can be regarded as a regularized ( smoothed ) delta - function , @xmath171 term ( 61 ) which is independent of the self - adjoint extension parameter , the magnetic flux and even the particle momentum is obtainable in the framework of classical theory , for instance , with the use of the hamilton - jacobi equation of motion , see appendix b. amplitude @xmath157 ( 54 ) can be rewritten as @xmath172\right\},\label{eq64}\ ] ] where @xmath173 is the wkb phase shift , @xmath174 and @xmath175 are given by ( a.2 ) and ( a.3 ) in appendix a. the derivative of the wkb phase shift , @xmath176 , can be related to the classical reflection angle ( equal to the classical incidence angle ) , @xmath177 , see ( b.10 ) in appendix b : @xmath178 thus , the condition of stationarity of the phase in ( 64 ) , @xmath179 ( @xmath180 is negligible , see appendix a ) , coincides in its form with the classical relation between scattering angle @xmath5 and reflection angle @xmath177 , see ( b.12 ) in appendix b , and the classical impact parameter can be introduced as @xmath181 note also that the first equation in ( 22 ) is equivalent to the equation determining the squared @xmath95-derivative of the hamilton - jacobi action , see ( b.5 ) in appendix b. however , the merits of the wkb method are not restricted to determining the wkb phase shift , i.e. to the only description of classical elastic reflection . the use of the method has also allowed us to find term ( 62 ) which is generically independent of the self - adjoint extension parameter but depends on both the magnetic flux and the particle momentum . this term describes the purely quantum effect of diffraction . it should be noted that the quantum effect is well separated in the scattering angle from the classical one : namely , diffraction is in the forward direction where classical reflection is absent . although the range of angles where diffraction is to be observed is quite narrow , the whole ( i.e. integrated over @xmath145 ) contribution of the diffraction effect to the total cross section is the same as that of the reflection effect : @xmath182 and the total cross section is twice the classical one ; the latter is needed for the optical theorem to be maintained ( see @xcite ) . it should be noted in the first place that the ab scattering amplitude ( see ( 51 ) ) and , consequently , cross section ( 1 ) were obtained in @xcite by imposing the condition of regular behavior for the particle wave function at the location of a singular ( i.e. of zero transverse size ) magnetic vortex . however , the regularity condition is not the most general one that is required by probability conservation . the most general condition in the case when a spatial dimension along the vortex is ignored involves four self - adjoint extension parameters ( the deficiency index is \{2,2 } ) @xcite . the dependence on the self - adjoint extension parameters enters into terms which are added to amplitude @xmath149 ( 51 ) . therefore , the cross section depends on four arbitrary parameters or , if a longitudinal dimension is taken into account , on four arbitrary functions of @xmath8 . if an invariance under rotations around the vortex is assumed to be a physical requirement , then the cross section still remains to be dependent on two arbitrary functions . the problem of seeking some , if any , sense for this arbitrariness is of purely academic interest , since the case of a magnetic vortex of zero transverse size has no relation to physics reality . turning now to the physical case of a magnetic vortex of nonzero transverse size , which is addressed in the present paper , we would like to emphasize that the classical initial equation ( see ( b.2 ) in appendix b ) , as well as the quantum - mechanical one ( see ( 10 ) with ( 13 ) ) , involves the vector potential of the vortex ( see ( 12 ) ) . however , in classical theory , this potential has no physical consequences , as is demonstrated , for example , in appendix b. one may agree with the authors of @xcite that `` the ab effect was already implicit in the 1926 schrdinger equation '' , but it takes more than eight decades before a physical consequence of the vector potential from the schrdinger equation is at last pointed out , see ( 62 ) . this explains why a scattering experiment aiming at the verification of the ab effect is still lacking . the main motivation of the present work is to make such an experiment plausible from a theorist s point of view . that is why we have taken care to prevent from any objection of theoretical kind : we have employed the most general boundary condition that is required by probability conservation and have shown how to grasp the ab effect even with the use of the 1926 wkb method . the ab effect , as well as quantization of the flux trapped in superconductors , the van hove singularities in the excitation spectra of crystals etc , explicates an importance of topological concepts for quantum physics @xcite . the ab phase , @xmath183 , which is acquired by the charged - particle wave function after the particle has encircled a magnetic vortex is certainly the same for the particles of different energies . nonetheless , a question always is how to observe this phase in real experiments , and what are physical restrictions on the particle energy in these experiments . for instance , interference experiments to verify the ab effect involve electrons of energies restricted to the range of 10 - 100 kev @xcite . interference is immanently a wave - optical phenomenon which is indirectly described in quantum theory through reference to wave optics ( a quantum - mechanical particle possesses de broglie s wavelength and thus is subject to the laws of wave optics ) . meanwhile , quantum theory provides a direct description of scattering phenomena and , in particular , of the ab effect as a scattering event . a scattering experiment with particles of the wavelength exceeding the vortex radius is hardly realistic , and that is why we are considering a scattering event with particles of the wavelength which is less than the vortex radius . by using the wkb method in the present paper , we obtain the scattering differential cross section , @xmath166 , consisting of two parts : one ( 61 ) describes the classical effect of elastic reflection from the vortex and another one ( 62 ) describes the quantum effect of diffraction on the vortex . the effects are equal to each other , when integrated over the whole range of the scattering angle , see ( 69 ) . it should be noted that , although the amplitude yielding the classical effect depends on the self - adjoint extension parameter ( @xmath40 ) , the flux of the vortex ( @xmath7 ) and the particle momentum ( @xmath3 ) , see ( 56 ) , all the dependence disappears in the differential cross section , see ( 61 ) ; certainly , this cross section can be obtained with the use of purely classical methods ( the hamilton - jacobi equation of motion ) . the amplitude yielding the quantum effect is generically independent of the self - adjoint extension parameter , see ( 57)-(60 ) . this signifies that diffraction ( to be more precise , the fraunhofer diffraction , i.e. the diffraction in almost parallel rays ) is independent of the choice of different boundary conditions which are compatible with the conservation of probability . owing to diffraction the ab effect persists in scattering of real particles , e.g. , electrons of energies in the range from 10 ev to 100 kev . the classical effect opens a window in the forward direction , where the ab effect is to be observed as a fringe shift in the diffraction pattern exhibited by the scattering differential cross section , see ( 62 ) . we present the window for the observation of the ab effect on fig . 1 , where the scattering differential cross section and the scattering angle are normalized in such a way that the plot remains actually the same for @xmath184 ; recall that the present - day realistic restriction is @xmath185 . the classical effect given by ( 61 ) is unobservable at @xmath186 , whereas the diffraction oscillations given by ( 62 ) are unobservable on the classical background at @xmath187^{1/3}<|\varphi|<\pi$ ] ; at @xmath188^{1/3}$ ] , both effects are unobservable being indistinguishable from the @xmath189-background which is of order of @xmath190^{1/3}$ ] , see ( a.22 ) in appendix a. a gate for particles propagating in the strictly forward ( @xmath6 ) direction is opened when the vortex flux equals an integer times the london flux : more than 90% of the diffraction cross section is accumulated in a peak centred at @xmath6 and having width @xmath191 , see the dashed line on fig . the gate for the strictly forward propagation of particles is closed when the vortex flux equals a half - integer times the london flux : a dip at @xmath6 is surrounded by two symmetric peaks accumulating more than 85% of the diffraction cross section ( each peak having width @xmath191 ) , see the solid line on fig . thus , the classically forbidden propagation of particles in the strictly forward direction is controlled by altering the amount of the vortex flux . is along the ordinate axis and @xmath192 is along the abscissa axis ( @xmath193 ) . dashed and solid lines correspond to cases @xmath194 and @xmath195 , respectively ( @xmath196 ) . the area under one central peak by the dashed line ( @xmath197 ) is 0.4514119@xmath970.0000002 , and the area under two central peaks by the solid line ( @xmath198 ) is 0.4278549@xmath970.0000013.,title="fig : " ] + this scattering ab effect was never observed experimentally , since a detailed proposal for its observation was just recently elaborated @xcite . such an experiment is in our opinion of fundamental significance and seems to be quite feasible with the present - day facilities providing for sources of bright and yet coherent electron beams . the work of yu . a. s. was supported by the national academy of sciences of ukraine ( project no.0112u000054 ) , by the the program of fundamental research of the department of physics and astronomy of the national academy of sciences of ukraine ( project no.0112u000056 ) and by the ictp seenet - mtp grant prj-09 `` strings and cosmology '' . the work of n. d. v. was supported by the science and technology center in ukraine and the national academy of sciences of ukraine within the framework of the targeted research & development initiatives ( trdi ) program under grant no.5716 . we acknowledge the support from the state agency for science , innovations and informatization of ukraine under the sffr - brrfr grant f54.1/019 . we present amplitude @xmath157 ( 54 ) in the form @xmath199\right\},\end{aligned}\ ] ] where , see ( 38 ) , @xmath200 @xmath201,\ ] ] and we have introduced dimensionless variable @xmath202 . to calculate the sum in ( a.1 ) in asymptotics @xmath203 , we use the poisson summation formula @xmath204\right\}+\frac 12e^{{\rm i}\eta(s_+,s)}+ \frac 12e^{{\rm i}\eta(-s_-,s)},\ ] ] where @xmath205\!]$ ] . if @xmath206 and @xmath207 is convex upwards , @xmath208 , on interval @xmath209 , then only a finite number of terms in the series on the right - hand side of ( a.4 ) contributes to the leading asymptotics at @xmath203 , and one can use the method of stationary phase for its evaluation . namely , if equation @xmath210\right|_{n = n_j}-2\pi l_j=0\ ] ] determines a stationary point inside the interval , @xmath211 , for some values of @xmath73 denoted by @xmath212 , then @xmath213\right\}\left\{\frac{2\pi e^{-{\rm i}\pi/2 } } { \left.-\left[\frac{\partial^2}{\partial n^2}\eta(n , s)\right]\right|_{n = n_j}}\right\}^{1/2}+o(1).\ ] ] in the present case we have @xmath214.\ ] ] differentiating ( a.2 ) and ( a.3 ) over @xmath215 , we obtain @xmath216 and @xmath217^{1/2 } \left\{\left[1-\left(\frac{n-\mu}{s}\right)^2\right]\cot(\rho\pi)-\frac 32\right\ } } { \left\{\left[1-\left(\frac{n-\mu}{s}\right)^2\right]\cot(\rho\pi)-\frac 12\right\}^2+s^2\left [ 1-\left(\frac{n-\mu}{s}\right)^2\right]^3}.\ ] ] ( a.9 ) is negligible as compared to ( a.8 ) , being of order @xmath218 at @xmath219 , @xmath203 , where the wkb approximation is valid , see ( 25 ) . therefore equation ( a.5 ) takes form @xmath220 choosing the range for @xmath5 as @xmath145 , we find that a solution to ( a.10 ) exists at @xmath221 only . denoting the solution by @xmath222 , we obtain @xmath223 and , hence , @xmath224\right|_{n = n_0}=-\frac { 2}{s|\sin(\varphi/2)|}.\ ] ] we see that the term corresponding to @xmath225 in the right - hand side of ( a.4 ) is of order @xmath226 , whereas all other terms are of order @xmath227 . as a result , we obtain expression ( 56 ) . amplitude @xmath155 ( 55 ) is presented in the form @xmath228}\times \nonumber \\ \times\exp{\left[2\sqrt{(n-\mu)^2-s^2}-2|n-\mu|{\rm arccosh}(|n-\mu|/s)\right]}\times\nonumber \\ \times\left\{\frac{\cot(\rho\pi)+\frac 12\frac{s^2}{(n-\mu)^2-s^2}-\sqrt{(n-\mu)^2-s^2 } } { \cot(\rho\pi)+\frac 12\frac{s^2}{(n-\mu)^2-s^2}+\sqrt{(n-\mu)^2-s^2}}+\right.\nonumber \\\left.+ \frac{\rm i}2\exp\left[2\sqrt{(n-\mu)^2-s^2}-2(n-\mu){\rm arccosh}\left(\frac{|n-\mu|}{s}\right)\right]\right\}^{-1},\end{aligned}\ ] ] where ( 41 ) is taken into account . the sum in ( a.13 ) at @xmath6 is evaluated asymptotically at @xmath229 by converting it into integral : @xmath230 + o(1/\sqrt{s})\right\},\ ] ] where @xmath231 \frac{\cot(\rho\pi)+\frac 12\frac{1}{v^2 - 1}-s\sqrt{v^2 - 1}}{\cot(\rho\pi ) + \frac 12\frac{1}{v^2 - 1}+s\sqrt{v^2 - 1}}\times\nonumber \\ \times\left\{1 + 4\exp\left[4s\left(v{\rm arccosh}v-\sqrt{v^2 - 1}\right)\right ] \left[\frac{\cot(\rho\pi)+\frac 12\frac{1}{v^2 - 1}-s\sqrt{v^2 - 1}}{\cot(\rho\pi)+\frac 12 \frac{1}{v^2 - 1}+s\sqrt{v^2 - 1}}\right]^2\right\}^{-1}\end{aligned}\ ] ] and @xmath232 \left[\frac{\cot(\rho\pi)+\frac 12\frac{1}{v^2 - 1}-s\sqrt{v^2 - 1}}{\cot(\rho\pi)+\frac 12 \frac{1}{v^2 - 1}+s\sqrt{v^2 - 1}}\right]^2\right\}^{-1 } .\end{aligned}\ ] ] in the cases of @xmath233 and @xmath234 , the integrand in ( a.15 ) is maximal at @xmath235 , and one can use the laplace method for the estimation of @xmath236 in this case , see , e.g. , @xcite . namely , let @xmath237 be continuous function with @xmath238 and @xmath239 be continuous differentiable function obeying conditions @xmath240 at @xmath241 and @xmath242 at @xmath243 @xmath244 , then @xmath245 where @xmath246 is the euler gamma - function . in the present case @xmath247 , @xmath248 , @xmath249 , @xmath250 , and we obtain @xmath251 a similar estimate is obtained for @xmath252 ( a.16 ) in the case of @xmath233 : @xmath253 in the case of @xmath254 , the integrand in ( a.16 ) is maximal at @xmath255 where @xmath256 . the integrand in the vicinity of the maximum is approximated as @xmath257(v - v_0)^2\right\},\ ] ] and the integral is estimated as @xmath258 the integrand in ( a.15 ) in the case of @xmath254 changes sign at @xmath255 . similarly to ( a.20 ) , we obtain the estimate for integral @xmath236 in this case : @xmath259 in the case of @xmath234 , the numerical analysis of @xmath252 yields an estimate of order @xmath260 . hence , we obtain the estimate for amplitude @xmath261 ( a.13 ) : @xmath262 where constant @xmath263 is independent of @xmath264 , @xmath265 and @xmath5 . in classical theory , we consider scattering of a point charged particle by an impenetrable tube containing magnetic flux @xmath7 . the hamilton function corresponding to the motion outside of the tube is @xmath266,\ ] ] where @xmath267 , and the hamilton - jacobi equation ( see , e.g. , @xcite ) is @xmath268=0.\ ] ] a general solution to ( b.2 ) is @xmath269 primary conservation laws are @xmath270=e\ ] ] with @xmath271 , @xmath272 and @xmath273 being arbitrary constants ( @xmath274 ) . taking the conservation laws into account , we obtain action @xmath275 ( b.3 ) in the form @xmath276 secondary conservation laws are @xmath277 with @xmath278 , @xmath279 and @xmath280 being arbitrary constants . thus we obtain relations @xmath281 @xmath282 @xmath283 which determine a trajectory of the particle , i.e. the functional dependence of @xmath284 , @xmath8 and @xmath285 on @xmath95 . one can conclude that the classical trajectory is independent of the enclosed magnetic flux , since @xmath286 can be absorbed into constant @xmath271 : @xmath287 . the trajectory is symmetric with respect to the point of reflection , @xmath288 . as a consequence of ( b.7 ) , we obtain the following expression for the incidence angle which is equal to the reflection angle : @xmath289 where @xmath290 @xmath291 . defining the impact parameter as @xmath292 and using elementary tools of plane geometry , we determine the angle of deflection ( called also the scattering angle ) in range @xmath145 as @xmath293 hence , the scattering differential cross section is @xmath294

there is large observational evidence of the presence of supermassive black holes ( smbh ) in the centres of most nearby galaxies @xcite . the dynamical evidence related to the orbital motion of stars in the cluster surrounding sgr @xmath7 indicates the presence of a smbh with mass @xmath8 @xcite . in addition , the observed correlation between the central black hole masses and the stellar velocity dispersion of the bulge of the host galaxies suggests a direct connection between the formation and evolution of galaxies and smbh @xcite . the observation of luminous quasars detected at redshifts higher than 6 in the sloan digital sky survey ( sdss ) implies that smbh with masses @xmath9 , which are believed to be the engines of such powerful quasars , were formed within the first billion years after the big bang ( e.g. fan 2006 for a recent review ) . however , it is still an open question how smbh seeds form and grow to reach such high masses in such a short amount of time @xcite . a number of different routes based on stellar dynamical processes , hydrodynamical processes or a combination of both have been suggested ( e.g. volonteri 2010 for a recent review ) . one of the theoretical scenarios for smbh seed formation is the gravitational collapse of the first generation of stars ( population iii stars ) with masses @xmath10 that are expected to form in halos with virial temperature @xmath11 at @xmath12 where cooling by molecular hydrogen is effective . as a result of the gravitational collapse of such pop iii stars , very massive bhs would form and then grow via merger and accretion @xcite . another possible scenario proposes that if sufficient primordial gas in massive halos , with mass @xmath13 , is unable to cool below @xmath14 , it may lead to the formation of a supermassive object @xcite , which would eventually collapse to form a smbh . this route assumes that fragmentation , which depends on efficient cooling , is suppressed , possibly by the presence of sufficiently strong uv radiation , that prevents the formation of molecular hydrogen in an environment with metallicity smaller than a given critical value @xcite . furthermore , fragmentation may depend on the turbulence present within the inflow of gas , and on the mechanism redistributing its angular momentum @xcite . the `` bars - within - bars '' mechanism @xcite is a self - regulating route to redistribute angular momentum and sustain turbulence such that the inflow of gas can proceed without fragmenting as it collapses even in a metal - enriched environment . depending on the rate and efficiency of the inflowing mass , there may be different outcomes . a low rate of mass accumulation would favor the formation of isentropic supermassive stars ( sms ) , with mass @xmath15 , which then would evolve as equilibrium configurations dominated by radiation pressure @xcite . a different outcome could result if the accumulation of gas is fast enough so that the outer layers of sms are not thermally relaxed during much of their lifetime , thus having an entropy stratification @xcite . a more exotic mechanism that could eventually lead to a sms collapsing into a smbh is the formation and evolution of supermassive dark matter stars ( sdms ) ( spolyar et al 2008 ) . such stars would be composed primarily of hydrogen and helium with only about @xmath16 of their mass in the form of dark matter , however they would shine due to dark matter annihilation . it has recently been pointed out that sdmss could reach masses @xmath17 @xcite . once sdmss run out of their dark matter supply , they experience a contraction phase that increases their baryon density and temperature , leading to an environment where nuclear burning may become important for the subsequent stellar evolution . if isentropic sms form , their quasi - stationary evolution of cooling and contraction will drive the stars to the onset of a general relativistic gravitational instability leading to their gravitational collapse @xcite , and possibly also to the formation of a smbh . the first numerical simulations , within the post - newtonian approximation , of @xcite concluded that for spherical stars with masses greater than @xmath18 thermonuclear reactions have no major effect on the collapse , while less massive stars exploded due to hydrogen burning . later @xcite performed the first relativistic simulations of the collapse of a sms in spherical symmetry . they were able to follow the evolution until the formation of a bh , although their investigations did not include any microphysics . @xcite revisited the work of @xcite and performed simulations of non - rotating sms in the range of @xmath19-@xmath18 with post - newtonian corrections and detailed microphysics that took into account an equation of state ( eos ) including electron - positron pairs , and a reaction network describing hydrogen burning by the cno cycle and its break - out via the rapid proton capture ( rp)-process . they found that sms with zero initial metallicity do not explode , while sms with masses larger than @xmath20 and with metallicity @xmath21 do explode . more recently @xcite carried out general relativistic hydrodynamic simulations of the spherically symmetric gravitational collapse of sms adopting a spacetime foliation with outgoing null hypersurfaces to solve the system of einstein and fluid equations . they performed simulations of spherical sms with masses in the range of @xmath22 using an eos that accounts for contributions from baryonic gases , and in a tabulated form , radiation and electron - positron pairs . they were able to follow the collapse from the onset of the instability until the point of bh formation , and showed that an apparent horizon ( ah ) enclosing about 25% of the stellar material was formed in all cases when simulations stopped . @xcite carried out general relativistic numerical simulations in axisymmetry of the collapse of uniformly rotating sms to bhs . they did not take into account thermonuclear burning , and adopted a @xmath23-law eos , @xmath24 with adiabatic index @xmath25 , where @xmath26 is the pressure , @xmath27 the rest - mass density , and @xmath28 the specific internal energy . although their simulations stopped before the final equilibrium was reached , the bh growth was followed until about 60@xmath29 of the mass had been swallowed by the smbh . they estimated that about 90@xmath29 of the total mass would end up in the final smbh with a spin parameter of @xmath30 . the gravitational collapse of differentially rotating sms in three dimensions was investigated by @xcite , who focused on the post - bh evolution , and also on the gravitational wave ( gw ) signal resulting from the newly formed smbh and the surrounding disk . the gw signal is expected to be emitted in the low frequency lisa band ( @xmath31 hz ) . despite the progress made , the final fate rotating isentropic sms is still unclear . in particular , it is still an open question for which initial metallicities hydrogen burning by the @xmath32-limited hot cno cycle and its break - out via the @xmath33o@xmath34ne reaction ( rp - process ) can halt the gravitational collapse of rotating sms and generate enough thermal energy to lead to an explosion . to address this issue , we perform a series of general relativistic hydrodynamic simulations with a microphysical eos accounting for contributions from radiation , electron - positron pairs , and baryonic matter , and taking into account the net thermonuclear energy released by the nuclear reactions involved in hydrogen burning through the pp - chain , cold and hot cno cycles and their break - out by the rp - process , and helium burning through the 3-@xmath35 reaction . the numerical simulations were carried out with the nada code @xcite , which solves the einstein equations coupled to the general relativistic hydrodynamics equations . greek indices run from 0 to 3 , latin indices from 1 to 3 , and we adopt the standard convention for the summation over repeated indices . unless otherwise stated we use units in which @xmath36 . next we briefly describe how the system of einstein and hydrodynamic equations are implemented in the nada code . we refer to @xcite for a more detailed description of the main equations and thorough testing of the code ( namely single bh evolutions , shock tubes , evolutions of both spherical and rotating relativistic stars , gravitational collapse to a bh of a marginally stable spherical star , and simulations of a system formed by a bh surrounded by a self - gravitating torus in equilibrium ) . we follow the 3 + 1 formulation in which the spacetime is foliated into a set of non - intersecting spacelike hypersurfaces . in this approach , the line element is written in the following form @xmath37 where @xmath35 , @xmath38 and @xmath39 are the lapse function , the shift three - vector , and the three - metric , respectively . the latter is defined by @xmath40 where @xmath41 is a timelike unit - normal vector orthogonal to a spacelike hypersurface . we make use of the bssn formulation @xcite to solve the einstein equations . initially , a conformal factor @xmath42 is introduced , and the conformally related metric is written as @xmath43 such that the determinant of the conformal metric , @xmath44 , is unity and @xmath45 , where @xmath46 . we also define the conformally related traceless part of the extrinsic curvature @xmath47 , @xmath48 where @xmath49 is the trace of the extrinsic curvature . we evolve the conformal factor defined as @xmath50 @xcite , and the auxiliary variables @xmath51 , known as the _ conformal connection functions _ , defined as @xmath52 where @xmath53 are the connection coefficients associated with @xmath44 . during the evolution we also enforce the constraints @xmath54 and @xmath55 at every time step . we use the cartoon method @xcite to impose axisymmetry while using cartesian coordinates . in addition to the bssn spacetime variables ( @xmath56 ) , there are two more quantities left undetermined , the lapse , @xmath35 and the shift vector @xmath57 . we used the so - called `` non - advective 1+log '' slicing @xcite , by dropping the advective term in the `` 1+log '' slicing condition . in this case , the slicing condition takes the form @xmath58 for the shift vector , we choose the `` gamma - freezing condition '' @xcite written as @xmath59 @xmath60 where @xmath61 is a constant that acts as a damping term , originally introduced both to prevent long term drift of the metric functions and to prevent oscillations of the shift vector . the general relativistic hydrodynamics equations , expressed through the conservation equations for the stress - energy tensor @xmath62 and the continuity equation are @xmath63 where @xmath27 is the rest - mass density , @xmath64 is the fluid four - velocity and @xmath65 is the covariant derivative with respect to the spacetime metric . following @xcite , the general relativistic hydrodynamics equations are written in a conservative form in cylindrical coordinates . since the einstein equations are solved only in the @xmath66 plane with cartesian coordinates ( 2d ) , the hydrodynamic equations are rewritten in cartesian coordinates for @xmath66 . the following definitions for the hydrodynamical variables are used @xmath67 @xmath68 @xmath69 @xmath70 @xmath71 where @xmath72 and @xmath73 are the lorentz factor and the specific fluid enthalpy respectively , and @xmath26 is the pressure . the conserved variables are @xmath74 , @xmath75 , @xmath76 . we refer to @xcite for further details . isentropic sms are self - gravitating equilibrium configurations of masses in the range of @xmath77 , which are mainly supported by radiation pressure , while the pressure of electron - positron pairs and of the baryon gas are only minor contributions to the eos . such configurations are well described by newtonian polytropes with polytropic index @xmath78 ( adiabatic index @xmath25 ) . the ratio of gas pressure to the total pressure ( @xmath32 ) for spherical sms can be written as @xcite @xmath79 where @xmath80 is the mean molecular weight . thus @xmath81 for @xmath82 . since nuclear burning timescales are too long for @xmath83 , evolution of sms proceeds on the kelvin - helmholtz timescale and is driven by the loss of energy and entropy by radiation as well as loss of angular momentum via mass shedding in the case of rotating configurations . although corrections due to the nonrelativistic gas of baryons and electrons and general relativistic effects are small , they can not be neglected for the evolution . firstly , gas corrections raise the adiabatic index slightly above @xmath84 @xmath85 secondly , general relativistic corrections lead to the existence of a maximum for the equilibrium mass as a function of the central density . for spherical sms this means that for a given mass the star evolves to a critical density beyond which it is dynamically unstable against radial perturbations @xcite : @xmath86 the onset of the instability also corresponds to a critical value of the adiabatic index @xmath87 , i.e. configurations become unstable when the adiabatic index drops below the critical value @xmath88 this happens when the stabilizing gas contribution to the eos does not raise the adiabatic index above @xmath84 to compensate for the destabilizing effect of general relativity expressed by the second term on the righ - hand - side of eq . ( [ gcri ] ) . rotation can stabilize configurations against the radial instability . the stability of rotating sms with uniform rotation was analyzed by @xcite . they found that stars at the onset of the instability have an equatorial radius @xmath89 , a spin parameter @xmath90 , and a ratio of rotational kinetic energy to the gravitational binding energy of @xmath91 . to close the system of hydrodynamic equations ( eq . [ hydro eqs ] ) we need to define the eos . we follow a treatment which includes separately the baryon contribution on the one hand , and photons and electron - positron pairs contributions , in a tabulated form , on the other hand . the baryon contribution is given by the analytic expressions for the pressure and specific internal energy @xmath92 @xmath93 where @xmath94 the universal gas constant , @xmath95 the temperature , @xmath96 the baryon specific internal energy , and @xmath97 is the mean molecular weight due to ions , which can be expressed as a function of the mass fractions of hydrogen ( @xmath98 ) , helium ( @xmath99 ) and heavier elements ( metals ) ( @xmath100 ) as @xmath101 where @xmath102 is the average atomic mass of the heavy elements . we assume that the composition of sms ( approximately that of primordial gas ) has a mass fraction of hydrogen @xmath103 and helium @xmath104 , where the metallity @xmath105 is an initial parameter , typically of the order of @xmath106 ( see table 1 details ) . thus , for the initial compositions that we consider the mean molecular weight of baryons is @xmath107 ( i.e. corresponding to a molecular weight for both ions and electrons of @xmath108 ) . effects associated with photons and the creation of electron - positron pairs are taken into account employing a tabulated eos . at temperatures above @xmath109 , not all the energy is used to increase the temperature and pressure , but part of the photon energy is used to create the rest - mass of the electron - positron pairs . as a result of pair creation , the adiabatic index of the star decreases , which means that the stability of the star is reduced . given the specific internal energy , @xmath28 and rest - mass density , @xmath27 , as evolved by the hydrodynamic equations , it is possible to compute the temperature @xmath95 by a newton - raphson algorithm that solves the equation @xmath110 for @xmath95 , @xmath111 where @xmath112 is the iteration counter . in order to avoid the small time steps , and cpu - time demands connected with the solution of a nuclear reaction network coupled to the hydrodynamic evolution , we apply an approximate method to take into account the basic effects of nuclear burning on the dynamics of the collapsing sms . we compute the nuclear energy release rates by hydrogen burning ( through the pp - chain , cold and hot cno cycles , and their break - out by the rp - process ) and helium burning ( through the 3-@xmath35 reaction ) as a function of rest - mass density , temperature and mass fractions of hydrogen @xmath98 , helium @xmath99 and cno metallicity @xmath100 . these nuclear energy generation rates are added as a source term on the right - hand - side of the evolution equation for the conserved quantity @xmath113 . the change rates of the energy density due to nuclear reactions , in the fluid frame , expressed in units of [ @xmath114 @xmath115 @xmath116 are given by : pp - chain @xcite : @xmath117 where @xmath118 , and @xmath119 is given by @xmath120 3-@xmath35 @xcite : @xmath121 where @xmath122 . cold - cno cycle @xcite : @xmath123 hot - cno cycle @xcite : @xmath124 rp - process @xcite : @xmath125 since we follow a single fluid approach , in which we solve only the hydrodynamics equations eq . ( [ hydro eqs ] ) ( i.e. we do not solve additional advection equations for the abundances of hydrogen , helium and metals ) , the elemental abundances during the time evolution are fixed . nevertheless , this assumption most possibly does not affect significantly the estimate of the threshold metallicity needed to produce a thermal bounce in collapsing sms . the average energy release through the 3-@xmath35 reaction is about 7.275 mev for each @xmath126c nucleus formed . since the total energy due to helium burning for exploding models is @xmath127 ergs ( e.g. @xmath128 ergs for model s1.c ) ; and even considering that this energy is released mostly in a central region of the sms containing @xmath129 of its rest - mass @xcite , it is easy to show that the change in the metallicity is of the order of @xmath130 . therefore , the increase of the metallicity in models experiencing a thermal bounce is much smaller than the critical metallicities needed to trigger the explosions . similarly , the average change in the mass fraction of hydrogen due to the cold and hot cno cycles is expected to be @xmath131 for exploding models . [ cols="<,^,^,^,^,^,^,^,^,^,^,^ " , ] after each time iteration the conserved variables ( i.e. @xmath132 ) are updated and the _ primitive _ hydrodynamical variables ( i.e. @xmath133 ) have to be recovered . the recovery is done in such a way that it allows for the use of a general eos of the form @xmath134 . we calculate a function @xmath135 , where @xmath136 and @xmath137 depend only on the conserved quantities and the pressure guess @xmath138 . the new pressure is computed then iteratively by a newton - raphson method until the desired convergence is achieved . the eos allows us to compute the neutrino losses due to the following processes , which become most relevant just before bh formation : pair annihilation ( @xmath139 ): most important process above @xmath109 . due to the large mean free path of neutrinos in the stellar medium at the densities of sms the energy loss by neutrinos can be significant . for a @xmath140 sms most of the energy release in the form of neutrinos originates from this process . the rates are computed using the fitting formula given by @xcite . photo - neutrino emission ( @xmath141 ): dominates at low temperatures @xmath142 and densities @xmath143 @xcite . plasmon decay ( @xmath144 ): this is the least relevant process for the conditions encountered by the models we have considered because its importance increases at higher densities than those present in sms . the rates are computed using the fitting formula given by @xcite . the evolution equations are integrated by the method of lines , for which we use an optimal strongly stability - preserving ( ssp ) runge - kutta algorithm of fourth - order with 5 stages @xcite . we use a second - order slope limiter reconstruction scheme ( mc limiter ) to obtain the left and right states of the primitive variables at each cell interface , and a hlle approximate riemann solver @xcite to compute the numerical fluxes in the @xmath145 and @xmath146 directions . derivative terms in the spacetime evolution equations are represented by a fourth - order centered finite - difference approximation on a uniform cartesian grid except for the advection terms ( terms formally like @xmath147 ) , for which an upwind scheme is used . the computational domain is defined as @xmath148 and @xmath149 , where @xmath150 refers to the location of the outer boundaries . we used a cell - centered cartesian grid to avoid that the location of the bh singularity coincides with a grid point . since it is not possible to follow the gravitational collapse of a sms from the early stages to the phase of black hole formation with a uniform cartesian grid ( the necessary fine zoning would be computationally too demanding ) , we adopt a regridding procedure @xcite . during the initial phase of the collapse we rezone the computational domain by moving the outer boundary inward , decreasing the grid spacing while keeping the initial number of grid points fixed . initially we use @xmath151 grid points , and place the outer boundary at @xmath152 where @xmath153 is the equatorial radius of the star . rezoning onto the new grid is done using a polynomial interpolation . we repeat this procedure 3 - 4 times until the collapse timescale in the central region is much shorter than in the outer parts . at this point , we both decrease the grid spacing and also increase the number of grid points @xmath154 in dependence of the lapse function typically as follows : @xmath155 if @xmath156 , @xmath157 if @xmath158 , and @xmath159 if @xmath160 . this procedure ensures the error in the conservation of the total rest - mass to be less than @xmath161 on the finest computational domain . to deal with the spacetime singularity from the newly formed bh we use the method of excising the matter content in a region within the horizon as proposed by @xcite once an ah is found . this excision is done only for the hydrodynamical variables , and the coordinate radius of the excised region is allowed to increase in time . on the other hand , we do neither use excision nor artificial dissipation terms for the spacetime evolution , and solely rely on the gauge conditions . here we define some of the quantities listed in table 1 . we compute the total rest - mass @xmath162 and the adm mass @xmath163 as @xmath164 @xmath165,\end{aligned}\ ] ] where @xmath166 ( @xmath41 being the unit normal to the hypersurface ) and @xmath167 is the scalar curvature associated to the conformal metric @xmath44 . the rotational kinetic energy @xmath168 and the gravitational potential energy @xmath72 are given by @xmath169 where @xmath170 is the angular velocity . @xmath171 where the internal energy is computed as @xmath172 in axisymmetry the ah equation becomes a nonlinear ordinary differential equation for the ah shape function , @xmath173 @xcite . we employ an ah finder that solves this ode by a shooting method using @xmath174 and @xmath175 as boundary conditions . we define the mass of the ah as @xmath176 where @xmath177 is the area of the ah . with zero metallicity ) for three different eos ( @xmath23-law and the microphysical eos with and without the electron - positron pair creation).,width=283 ] and zero metallicity ) with a @xmath23-law and the microphysical eos with electron - positron pair creation . the middle and lower panels display the ah mass and the disk mass as a function of time for the collapse simulation with a @xmath23-law eos.,width=283 ] the initial sms are set up as isentropic objects . all models , except model d2 , are chosen such that they are gravitationally unstable , and therefore their central rest - mass density is slightly larger than the critical central density required for the onset of the collapse of a configuration with given mass and entropy . a list of the different sms we have considered is provided in table 1 . models s1 and s2 represent a spherically symmetric , nonrotating sms with gravitational mass of @xmath178 and @xmath179 , respectively , while models r1 and r2 are uniformly rotating initial models again with masses of @xmath178 and @xmath179 , respectively . the rigidly and maximally rotating initial models r1 and r2 , and the differentially rotating models d1 and d2 are cconstructed with a polytropic eos with the lorene code ( url http://www.lorene.obspm.fr ) . we obtain temperatures for our microphyscial models by inverting the corresponding energy density with our eos of section 3.2 . we also introduce a perturbation to trigger the gravitational collapse by reducing the pressure overall by @xmath180 . in order to determine the threshold metallicity required to halt the collapse and produce an explosion we carry out several numerical simulations for each initial model with different values of the initial metallicity . the initial metallicities along with the fate of the star are given in table 1 . axisymmetric calculations without rotation ( i.e. models s1 and s2 ) retain the spherical symmetry of the initial conditions . there are no physical phenomena like convective or overturn instabilities to produce asphericity , i.e. we can directly compare our 2d non - rotating models with those computed in spherical symmetry ( 1d calculations ) by @xcite and @xcite . the main differences with respect to the results obtained by @xcite are most likely due to two reasons . first , we apply a fully general relativistic treatment while they used a post - newtonian treatment of gravity , and second , there are differences in the treatment of nuclear burning ( fuller et al . 1986 solved the relevant nuclear network without the approximations adopted in our work , see section 3.3 for further details ) . despite these differences , the results agree fairly well . as discussed in detail in section 7.1 the initial metallicities required to produce an explosion are similar , and a thermal bounce can be produced only if sufficient energy is liberated during the phase when the hcno cycle is active . the main difference with respect to the work of @xcite resides in the formulation of einstein s field equations ; in particular , in the foliation of the spacetime ( foliation into a set spacelike hypersurfaces versus a foliation with outgoing null hypersurfaces ) . in order to compare with the results of @xcite , we computed the redshifted total energy output for a model having the same rest - mass , doing the time integration until approximately the same evolutionary stage as in @xcite . we find that the total energies released in neutrinos differ by less than @xmath181 ( for mode details see section 7.3 ) . cm previous simulations of sms collapse to bh in general relativity have been performed with a @xmath23-law eos with @xmath182 ( with the only exception being the work of linke et al . 2001 ) . in order to elucidate the influence of the eos on the dynamics of collapsing sms , we performed three simulations of the same initial model ( model r1.0 , a marginally unstable uniformly rotating sms with zero initial metallicity ) without nuclear burning effects , and with three eos : a @xmath23-law eos with @xmath182 ( i.e. a similar set - up as in shibata & shapiro 2002 ) and the microphysical eos with and without including electrons and the e@xmath183 pairs , i.e. for the last eos case we consider eqs . ( 19 ) and ( 20 ) for the baryons plus @xmath184 and @xmath185 for the photons ( where @xmath186 is the radiation density constant ) . we denote the @xmath23-eos as eos-0 , the full microphysical eos , our canonical one for the studies of this work as eos-1 , and the reduced microphysical case as eos-2 . in figure [ fig1 ] we show with a dotted line the time evolution of the central density of model r1.0 with eos-0 , and with a dashed ( solid ) line the time evolution of the central density with eos-2 ( eos-1 ) . the first thing to note is that the collapse timescale obtained with the @xmath23-law eos is shorter than that obtained with the microphysical eos ( both eos-1 and eos-2 ) , because the ion pressure contribution to the eos raises the adiabatic index above @xmath84 ( see eq . [ ion_gamma ] ) . this increase in the adiabatic index helps to stabilize the star against the gravitational instability , and therefore delays the collapse . on the other hand , the effect of pair creation reduces the adiabatic index below @xmath84 at @xmath187k . this explains the differences between the solid and dashed lines in figure [ fig1 ] at central densities @xmath188 g/@xmath189 , which correspond to central temperatures @xmath190k . once the collapse enters this regime , pair creation becomes relevant enough to reduce the adiabatic index below @xmath84 , which destabilizes the collapsing star . compared to previous works ( shibata & shapiro 2002 ) the use of a microphysical eos instead of a @xmath23-law eos delays the collapse ( mostly due to the baryons while e@xmath183 destabilize ) of an initially gravitationally unstable configuration . we also performed 2d axisymmetric simulations of differentially rotating sms . first , we investigated the influence of the eos on _ gravitationally unstable stars _ using model d1 as a reference . this model corresponds , within the accuracy to which the initial conditions can be reproduced , to a differentially rotating unstable sms discussed by @xcite ( i.e. their model i ) . results are displayed in figure [ fig2 ] . in the upper panel of this figure , we show the time evolution of the central density for model d1 with eos-0 ( dashed line ) , and with the microphysical eos-1 ( solid line ) . opposite to the behavior in the case of the uniformly rotating sms r1.0 , the collapse timescale is longer with the @xmath23-law than with the microphysical eos . the reason for this difference is that the initial central temperature ( @xmath191k ) of model d1 is an order of magnitude higher than the initial central temperature in r1.0 . therefore , electron - positron pair creation reduces the stability of the star ( by reducing @xmath23 ) already during the initial stages of the collapse . this behavior is expected to be present also in 3d , and since the collapse timescale is reduced when using the microphysical eos , nonaxisymmetric instablilities would have even less time to grow before the formation of a bh . it reinforces the conclusions of @xcite , who showed that the three dimensional collapse of rotating stars proceeds in an approximately axisymmetric manner . the lower two panels of figure [ fig2 ] display the growth of the ah mass and the disk mass ( defined as the rest - mass outside the ah of the newly formed bh ) as a function of time ( in units of the gravitational mass for comparison with figures 9 and 10 of saijo & hawke 2009 ) , respectively . the values of both quantities at the end of the simulation agree , within a @xmath192 difference , with those obtained in 3d by @xcite . we note that there is also good agreement ( less than @xmath192 difference ) regarding the time at which an ah is first detected . these observed small differences are likely due to differences in the initial models and numerical techniques rather than to the influence of nonaxisymmtric effects . this suggests that our collapse simulation with the same treatment of physics yields good agreement with the 3d simulations of @xcite we also investigated the influence of electron - positron pair creation on the evolution of _ gravitationally stable differentially rotating _ sms using model d2 which is similar to the stable differentially rotating model iii of @xcite . we performed three simulations of model d2 varying the eos . in figure [ fig3 ] we show the central rest - mass density as a function of time for a @xmath23-law ( dashed line ) , and for the microphysical eos-1 ( solid line ) and eos-2 ( dotted line ) . in agreement with the results obtained by @xcite we find that model d2 represents a stable differentially rotating sms when a @xmath23-law is used . a persistent series of oscillations is triggered by the initial perturbation in the pressure . this is also the case with the microphysical eos-2 without the inclusion of electrons and the e@xmath193 pairs . however , the time evolution of d2 is completely different when e@xmath193 pairs are taken into account . the influence of pairs is large enough to destabilize model d2 against gravitational collapse . we note that unlike all other sms considered in this paper , which are @xmath25 models initially unstable to gravitational collapse , model d2 is an initially stable @xmath25 model which becomes gravitationally unstable only by the creation of electron - positron pairs at high temperatures . hence , using a microphysical eos with electron - positron pairs is crucial to determine the stability of differentially rotating sms . we note that the central temperature of the initial models d1 and d2 is of the order of @xmath194k . at this temperature , the main source of thermonuclear energy is hydrogen burning via the rp - process . it is however expected that such sms would previously experience a phase of hydrogen burning via the cold and hot cno cycles which would significantly affect the evolution of the models such that configurations with high @xmath195 as in models d1 and d2 might never be reached . therefore , models d1 and d2 are not particularly well suited to investigate the existence of a thermal bounce during collapse ( see section 7 ) . exploring in detail the parameter space for the stability of differentially rotating sms with the microphysical eos , and the existence of a thermal bounce during the collapse phase depending on the initial stellar metallicity , is a major task on its own , which is beyond the scope of this paper . for these reasons we do not consider differentially rotating sms in this work . first we consider a gravitationally unstable spherically symmetric sms with a gravitational mass of @xmath178 ( s1.a , s1.b and s1.c ) , which corresponds to a model extensively discussed in @xcite , and therefore allows for a comparison with the results presented here . @xcite found that unstable spherical sms with @xmath178 and an initial metallicity @xmath196 collapse to a bh while models with an initial metallicity @xmath197 explode due to the nuclear energy released by the hot cno burning . they also found that the central density and temperature at thermal bounce ( where the collapse is reversed to an explosion ) are @xmath198 g/@xmath189 and @xmath199 k , respectively . the left panels in figure [ fig4 ] show the time evolution of the central rest - mass density ( upper panel ) and central temperature ( lower panel ) for models s1.a , s1.c , r1.a and r1.d , i.e. , non - rotating and rotating models with a mass of @xmath200 . in particular , the solid lines represent the time evolution of the central density and temperature for model s1.c ( @xmath201 ) and r1.d ( @xmath202 ) . as the collapse proceeds , the central density and temperature rise rapidly , which increases the nuclear energy generation rate by hydrogen burning . since the metallicity is sufficiently high , enough energy can be liberated to increase the pressure and to produce a thermal bounce . this is the case for model s1.c . in figure [ fig4 ] we show that a thermal bounce occurs ( at approximately @xmath203 s ) entirely due to the hot cno cycle , which is the main source of thermonuclear energy at temperatures in the range @xmath204 . the rest - mass density at bounce is @xmath205 g/@xmath189 and the temperature @xmath206 . these values , as well as the threshold metallicity needed to trigger a thermonuclear explosion ( @xmath201 ) , are higher than those found by @xcite ( who found that a spherical nonrotating model with the same rest - mass would explode , if the initial metallicity was @xmath197 ) . on the other hand , dashed lines show the time evolution of the central density and temperature for model s1.a ( @xmath197 ) . in this case , as well as for model s1.b , the collapse is not halted by the energy release and continues until an ah is found , indicating the formation of a bh . we note that the radial velocity profiles change continuously near the time where the collapse is reversed to an explosion due to the nuclear energy released by the hot cno burning , and an expanding shock forms only near the surface of the star at a radius @xmath207 ( i.e. @xmath208 ) where the rest - mass density is @xmath209 . we show in figure [ fig5 ] the profiles of the @xmath145-component of the three - velocity @xmath210 along the @xmath145-axis ( in the equatorial plane ) for the nonrotating spherical stars s1.a ( dashed lines ) and s1.c ( solid lines ) at three different time slices near the time at which model s1.c experiences a thermal bounce . velocity profiles of model s1.c are displayed up to the radius where a shock forms at @xmath211 and begins to expand into the low density outer layers of the sms . the evolutionary tracks for the central density and temperature of the rotating models r1.a and r1.d are also shown in figure [ fig4 ] . a dashed line corresponds to model r1.a , with an initial metallicity @xmath202 , which collapses to a bh . a solid line denotes model r1.d with @xmath196 , which explodes due to the energy released by the hot cno cycle . we find that model r1.c with a lower metallicity of @xmath212 also explodes when the central temperature is the range dominated by the hot cno cycle . as a result of the kinetic energy stored in the rotation of models r1.c and r1.d , the critical metallicity needed to trigger an explosion decreases significantly relative to the non - rotating case . we observe that rotating models with initial metallicities up to @xmath213 do not explode even via the rp - process , which is dominant at temperatures above @xmath214 k and increases the hydrogen burning rate by @xmath215 times relative to the hot cno cycle . we also note that the evolution time scales of the collapse and bounce phases are reduced because rotating models are more compact and have a higher initial central density and temperature than the spherical ones at the onset of the gravitational instability . the right panels in figure [ fig4 ] show the time evolution of the central rest - mass density ( upper panel ) and the central temperature ( lower panel ) for models s2.a , s2.b , r2.a and r2.d , i.e. , of models with a mass of @xmath216 . we find that the critical metallicity for an explosion in the spherical case is @xmath217 ( model s2.b ) , while model s2.a with @xmath218 collapses to a bh . we note that the critical metallicity leading to a thermonuclear explosion is higher than the critical value found by @xcite ( i.e. , @xmath219 ) for a spherical sms with the same mass . the initial metallicity leading to an explosion in the rotating case ( model r2.d ) is more than an order of magnitude smaller than in the spherical case . as for the models with a smaller gravitational mass , the thermal bounce takes place when the physical conditions in the central region of the star allow for the release of energy by hydrogen burning through the hot cno cycle . overall , the dynamics of the more massive models indicates that the critical initial metallicity required to produce an explosion increases with the rest - mass of the star . figure [ fig6 ] shows the total nuclear energy generation rate in erg@xmath220s for the exploding models as a function of time during the late stages of the collapse just before and after bounce . the main contribution to the nuclear energy generation is due to hydrogen burning by the hot cno cycle . the peak values of the energy generation rate at bounce lie between several @xmath221 for the rotating models ( r1.d and r2.d ) , and @xmath222 for the spherical models ( s1.c and s2.b ) . as expected the maximum nuclear energy generation rate needed to produce an explosion is lower in the rotating models . moreover , as the explosions are due to the energy release by hydrogen burning via the hot cno cycle , the ejecta would mostly be composed of @xmath223he . -component of the three - velocity @xmath210 along the @xmath145-axis ( in the equatorial plane ) for the nonrotating spherical stars s1.a ( dashed lines ) and s1.c ( solid lines ) at three different time slices near the time at which model s1.c experiences a thermal bounce . velocity profiles of model s1.c are displayed up to the radius where a shock , that expands into the low density outer layers of the sms , forms.,width=302 ] s for the exploding models ( s1.c , r1.d , r1.c , s2.b and r2.d ) as a function of time near the bounce . the contribution to the nuclear energy generation is mainly due to hydrogen burning by the hot cno cycle . the peak values of the energy generation rate at bounce lie between @xmath224 $ ] for the rotating models ( r1.d and r2.d ) , and @xmath225 $ ] for the spherical models ( s1.c and s2.b).,width=302 ] as a result of the thermal bounce , the kinetic energy rises until most of the energy of the explosion is in the form of kinetic energy . we list in the second but last column of table 1 the radial kinetic energy after thermal bounce , which ranges between @xmath226 for the rotating star r1.c , and @xmath227 for the spherical star s2.b . due to the lack of resolution at the surface of the star , it becomes difficult to compute accurately the photosphere and its effective temperature from the criterion that the optical depth is @xmath228 . therefore , in order to estimate the photon luminosity produced in association with the thermonuclear explosion , we make use of the fact that within the diffusion approximation the radiation flux is given by where @xmath230 is the energy density of the radiation , and @xmath231 is the opacity due to electron thompson scattering , which is the main source of opacity in sms . the photon luminosity in terms of the temperature gradient and for the spherically symmetric case can be written as @xmath232 where @xmath186 is the radiation constant , and @xmath233 the speed of light . as can be seen in the last panel of figure [ fig7 ] the distribution of matter becomes spherically symmetric during the phase of expansion after the thermonuclear explosion . in this figure ( fig . [ fig7 ] ) we show the isodensity contours for the rotating model r1.d . the frames have been taken at the initial time ( left figure ) , at @xmath234s ( central figure ) just after the thermal bounce ( at @xmath235s ) , and at @xmath236 s when the radius of the expanding matter is roughly 4 times the radius of the star at the onset of the collapse . the photon luminosity computed using eq.([photon_lum ] ) for model r1.d is displayed in figure [ fig8 ] , where we also indicate with a dashed vertical line the time at which the thermal bounce takes place . the photon luminosity before the thermal bounce is computed at radii inside the star unaffected by the local dynamics of the low density outer layers which is caused by the initial pressure perturbation and by the interaction between the surface of the sms and the artificial atmosphere . once the expanding shock forms near the surface , the photon luminosity is computed near the surface of the star . the lightcurve shows that , during the initial phase , the luminosity is roughly equal to the eddington luminosity @xmath237 until the thermal bounce . then , the photon luminosity becomes super - eddington when the expanding shock reaches the outer layers of the star and reaches a value of @xmath238 . this value of the photon luminosity after the bounce is within a few percent difference with respect to the photon luminosity @xcite found for a nonrotating sms of same rest - mass . the photon luminosity remains super - eddington during the phase of rapid expansion that follows the thermal bounce . we compute the photon luminosity until the surface of the star reaches the outer boundary of the computational domain @xmath239 after the bounce . beyond that point , the luminosity is expected to decrease , and then rise to a plateau of @xmath240 due to the recombination of hydrogen ( see fuller et al . 1986 for a nonrotating star ) . the outcome of the evolution of models that do not generate enough nuclear energy during the contraction phase to halt the collapse is the formation of a bh . the evolutionary tracks for the central density and temperature of some of these models are also shown in figure [ fig4 ] . the central density typically increases up to @xmath241 and the central temperature up to @xmath242 just before the formation of an ah . three isodensity contours for the rotating model r1.a collapsing to a bh are shown in figure [ fig9 ] , which display the flattening of the star as the collapse proceeds . the frames have been taken at the initial time ( left panel ) , at @xmath243 ( central panel ) approximately when model r1.d with higher metallicity experiences a thermal bounce and , at @xmath244 , where a bh has already formed and its ah has a mass of @xmath245 of the total initial mass . at the temperatures reached during the late stages of the gravitational collapse ( in fact at @xmath246 ) the most efficient process for hydrogen burning is the breakout from the hot cno cycle via the @xmath33o@xmath34ne reaction . nevertheless , we find that models which do not release enough nuclear energy by the hot cno cycle to halt their collapse to a bh , are not able to produce a thermal explosion due to the energy liberated by the @xmath33o@xmath34ne reaction . we note that above @xmath109 , not all the liberated energy is used to increase the temperature and pressure , but is partially used to create the rest - mass of the electron - positron pairs . as a result of pair creation , the adiabatic index of the star decreases , which means the stability of the star is reduced . moreover , due to the presence of e@xmath183 pairs , neutrino energy losses grow dramatically . figure [ fig10 ] shows ( solid lines ) the time evolution of the redshifted neutrino luminosities of four models collapsing to a bh ( s1.a , r1.a , s2.a , and r2.a ) , and ( dashed lines ) of four models experiencing a thermal bounce ( s1.c , r1.d , s2.b , and r2.d ) . the change of the slope of the neutrino luminosities at @xmath247 denotes the transition from photo - neutrino emission to the pair annihilation dominated region . the peak luminosities in all form of neutrino for models collapsing to a bh are @xmath248 . neutrino luminosities can be that important because the densities in the core prior to bh formation are @xmath249 , and therefore neutrinos can escape . the peak neutrino luminosities lie between the luminosities found by @xcite for the collapse of spherical sms , and those found by @xcite ( who only took into account the luminosity in the form of electron antineutrino ) . the maximum luminosity decreases slightly as the rest - mass of the initial model increases , which was already observed by @xcite . in addition , we find that the peak of the redshifted neutrino luminosity does not seem to be very sensitive to the initial rotation rate of the star . we also note that the luminosity of model r1.a reflects the effects of hydrogen burning at @xmath250 . the total energy output in the form of neutrinos is listed in the last column of table 1 for several models . the total radiated energies vary between @xmath5 ergs for models collapsing to a bh , and @xmath6 ergs for exploding models . these results are in reasonable agreement with previous calculations . for instance , @xcite obtained that the total energy output in the form of electron antineutrinos for a spherical sms with a mass @xmath251 and zero initial metallicity was @xmath252 , although their simulations neglected general relativistic effects which are important to compute accurately the relativistic redshifts . on the other hand , @xcite , by means of relativistic one - dimensional simulations , found a total radiated energy in form of neutrinos of about @xmath253 for the same initial model , and about @xmath254 when redshifts were taken into account . in order to compare with the results of @xcite , we computed the redshifted total energy output for model s1.a , having the same rest - mass , until approximately the same evolution stage as @xcite did ( i.e. when the differential neutrino luminosity @xmath255 ) . we find that the total energy released in neutrinos is @xmath256 . the neutrino luminosities for models experiencing a thermonuclear explosion ( dashed lines in fig . [ fig10 ] ) peak at much lower values @xmath258 , and decrease due to the expansion and disruption of the star after the bounce . the axisymmetric gravitational collapse of rotating sms with uniform rotation is expected to emit a burst of gravitational waves @xcite with a frequency within the lisa low frequency band ( @xmath259 ) . although through the simulations presented here we could not investigate the development of nonaxisymmetric features in our axisymmetric models that could also lead to the emission of gws , @xcite have shown that the three dimensional collapse of rotating stars proceeds in an approximately axisymmetric manner . where @xmath264 refers to the second time derivative of the quadrupole moment . the gravitational wave quadrupole amplitude is @xmath265 . following @xcite we compute the second time derivative of the quadrupole moment by finite differencing the numerical results for the first time derivative of @xmath266 obtained by where d is the distance of the source , and @xmath269 the spectral energy density of the gravitational radiation given by @xmath270 with @xmath271 we have calculated the quadrupole gravitational wave emission for the rotating model r1.a collapsing to a bh . we plot in figure [ fig11 ] the characteristic gravitational wave strain ( eq.[h_char ] ) for this model assuming that the source is located at a distance of 50 gpc ( i.e. , @xmath272 ) , together with the design noise spectrum @xmath273 of the lisa detector @xcite . we find that , in agreement with @xcite , @xcite and @xcite , the burst of gravitational waves due to the collapse of a rotating sms could be detected at a distance of 50 gpc and at a frequency which approximately takes the form @xcite furthermore , @xcite have recently investigated , by means of three - dimensional general relativistic numerical simulations of equilibrium tori orbiting bhs , the development of the nonaxisymmetric papaloizou - pringle instability ( ppi ) in such systems @xcite , and have found that a nonaxisymmetric instability associated with the @xmath277 mode grows for a wide range of self - gravitating tori orbiting bhs , leading to the emission of quasiperiodic gws . in particular , @xcite have pointed out that the emission of quasiperiodic gws from the torus resulting after the formation of a smbh via the collapse of a sms could be well above the noise sensitivity curve of lisa for sources located at a distance of 10gpc . such instability appears for tori whose angular velocity in the equatorial plane expressed as @xmath278 has @xmath279 where @xmath280 corresponds to the keplerian limit , i.e. @xmath281 in newtonian gravity . we find that the torus ( defined as the rest - mass outside the ah ) that forms after the collapse to a bh of the uniformly rotating model r1.a ( when the mass of the ah exceeds @xmath245 of the adm mass ) does not fulfill the above condition for the development of the ppi . however , we find that the torus that forms when the differentially rotating model d1 collapses to a bh has a distribution of angular momentum such that @xmath278 with @xmath282 . this suggests that the torus may be prone to the development of the nonaxisymmetric ppi , which would lead to the emission of quasiperiodic gws with peak amplitude @xmath283 and frequency @xmath284 during an accretion timescale @xmath285 . we have presented results of general relativistic simulations of collapsing supermassive stars using the two - dimensional general relativistic numerical code nada , which solves the einstein equations written in the bssn formalism and the general relativistic hydrodynamic equations with high resolution shock capturing schemes . these numerical simulations have used an eos that includes the effects of gas pressure , and tabulated those associated with radiation pressure and electron - positron pairs . we have also taken into account the effects of thermonuclear energy release by hydrogen and helium burning . in particular , we have investigated the effects of hydrogen burning by the @xmath32-limited hot cno cycle and its breakout via the @xmath33o@xmath34ne reaction ( rp - process ) on the gravitational collapse of nonrotating and rotating sms with non - zero metallicity . we have presented a comparison with previous studies , and investigated the influence of the eos on the collapse . we emphasize that axisymmetric calculations without rotation ( i.e. models s1 and s2 ) retain the spherical symmetry of the initial configurations as there are no physical phenomena to produce asphericity and numerical artifacts associated with the use of cartesian coordinates are negligibly small . overall , our collapse simulations yield good agreement with previous works when using the same treatment of physics . we have also found that the collapse timescale depends on the ion contributions to the eos , and electron - positron pair creation affects the stability of sms . interestingly , differentially rotating stars that are gravitationally stable with a @xmath25 eos can become unstable against gravitational collapse when the calculation is performed with the microphysical eos including pair creation . we have found that objects with a mass of @xmath286 and an initial metallicity greater than @xmath1 explode if non - rotating , while the threshold metallicity for an explosion is reduced to @xmath2 for objects which are uniformly rotating . the critical initial metallicity for a thermal explosion increases for stars with a mass of @xmath3 . the most important contribution to the nuclear energy generation is due to the hot cno cycle . the peak values of the nuclear energy generation rate at bounce range from @xmath287 for rotating models ( r1.d and r2.d ) , to @xmath288 for spherical models ( s1.c and s2.b ) . after the thermal bounce , the radial kinetic energy of the explosion rises until most of the energy is kinetic , with values ranging from @xmath289 for rotating stars , to up @xmath290 for the spherical star s2.b . the neutrino luminosities for models experiencing a thermal bounce peak at @xmath291 . the photon luminosity roughly equal to the eddington luminosity during the initial phase of contraction . then , after the thermal bounce , the photon luminosity becomes super - eddington with a value of about @xmath292 during the phase of rapid expansion that follows the thermal bounce . for those stars that do not explode we have followed the evolution beyond the phase of black hole formation and computed the neutrino energy loss . the peak neutrino luminosities are @xmath4 . sms with masses less than @xmath3 could have formed in massive halos with @xmath293 . although the amount of metals that was present in such environments at the time when sms might have formed is unclear , it seems possible that the metallicities could have been smaller than the critical metallicities required to reverse the gravitational collapse of a sms into an explosion . if so , the final fate of the gravitational collapse of rotating sms would be the formation of a smbh and a torus . in a follow - up paper , we aim to investigate in detail the dynamics of such systems ( collapsing of sms to a bh - torus system ) in 3d , focusing on the post - bh evolution and the development of nonaxisymmetric features that could emit detectable gravitational radiation . we thank b. mller and p. cerd - durn for useful discussions . work supported by the deutsche forschungsgesellschaft ( dfg ) through its transregional centers sfb / tr 7 `` gravitational wave astronomy '' , and sfb / tr 27 `` neutrinos and beyond '' , and the cluster of excellence exc153 `` origin and structure of the universe '' .

mri plays a key role in the diagnosis and management of acute ischemic stroke and traumatic brain injury . a rigorous understanding of mri begins with the elementary particles which give rise to the magnetic properties of tissue . elementary particles have two independent properties which manifest magnetic moments : ( i ) orbital angular momentum , and ( ii ) intrinsic spin . these two properties are independent only to first order , as it was their very interaction via spin - orbit coupling which led to detectable energy differences in the hydrogen spectrum , and consequently to the discovery of intrinsic spin . we discuss both properties in this paper in the context of the hydrogen atom . the hydrogen atom @xmath0h is the most frequently targeted nucleus in mri due largely to its biological abundance and high gyromagnetic ratio . the @xmath0h nucleus consists of a single proton . protons are made of quarks , specifically one down and two up quarks , each of which are of spin 1/2 . the charge and spin of the proton are both directly due to its quark composition . the down quark has a charge of -1/3 while the up quarks each have charge + 2/3 , which all sums up to a charge of + 1 . the nuclear spin is 1/2 by virtue of spin cancellation from the antiparallel alignment of two of the three quarks . the remainder of this paper is organized as follows : section ( [ sec : mri ] ) presents an overview of the mri , section ( [ sec : stroke ] ) reviews the mri assessment of acute ischemic stroke , section ( [ sec : h_atom ] ) describes the hydrogen atom and reviews an analytical solution to its schrdinger equation , section ( [ sec : spin ] ) reviews intrinsic spin , section ( [ sec : adding ] ) reviews quantum mechanical addition of orbital angular momentum and spin , section ( [ sec : groups ] ) reviews group theory of su(2 ) and so(3 ) and their relationship , and section ( [ sec : summary ] ) summarizes this review . the mri section , section ( [ sec : mri ] ) , contains a number of subsections which are outlined as follows : subsection ( [ subsec : scales ] ) discusses the hierarchy of scales in mri phenomenology , subsection ( [ subsec : magnetism_e_n ] ) reviews the magnetism of electrons and nucleons , subsection ( [ subsec : mag_potential ] ) reviews the magnetic potential energy and its role in the descriptions of magnetization , torque , force , and larmor precession , subsection ( [ subsec : pulse_mag ] ) reviews pulse magnetization , subsection ( [ subsec : equilibrium_mag ] ) describes equilibrium magnetization , subsection ( [ subsec : t1 ] ) describes spin - lattice or t@xmath1 recovery , subsection ( [ subsec : t2 ] ) describes spin - spin or t@xmath2 decay , subsection ( [ subsec : fid ] ) reviews free inductance decay , subsection ( [ subsec : bloch ] ) reviews the bloch equations , subsection ( [ subsec : contrast ] ) overviews intravenous contrast agents in mri , subsection ( [ subsec : gradient ] ) reviews gradient - based band - selection method for mri space localization , and subsection ( [ subsec : machine ] ) reviews the components of the mri machine . the stroke section , section ( [ sec : stroke ] ) , contains a number of subsections which are outlined as follows : subsection ( [ subsec : dwi ] ) reviews diffusion weighted imaging ( dwi ) , subsection ( [ subsec : pwi ] ) reviews perfusion weighted imaging ( pwi ) , subsection ( [ subsec : dwi_pwi ] ) reviews the combination of dwi and pwi and its utility in imaging the penumbra , subsection ( [ subsec : mrs ] ) reviews magnetic resonance spectroscopy , subsection ( [ subsec : bold ] ) reviews blood oxygen level dependent ( bold ) imaging , and subsection ( [ subsec : mra ] ) reviews mra . appnedix ( a ) reviews the signals processing involved in mri , including the fourier transforms , the canonical sampling theorem , and the convolution theorem . the mri is an optical technology whose core equation is the maxwell - faraday equation . it exploits ensemble phenomena in which the composition of a sample can be probed by sensing its magnetic properties through radio frequency waves . here , we describe the various parts of mri and their respective roles in the whole . the hierarchy of scales in the modeling of magnetic resonance imaging proceeds from the subnuclear scale of quarks and gluons , to the subatomic scale of discrete electrons and nucleons , and finally to the bulk sample scale where ensemble effects and statistical mechanics are the operative physics . in a general sense , the subnuclear scale is coupled to quantum field theory , the subatomic scale is coupled to quantum mechanics , and the bulk sample scale is coupled to classical electromagnetics . certain phenomena in mri are only meaningful over specific regimes . for instance , a magnetic moment experiences a torque when an external magnetic field is applied . however , the concept of torque is a classical concept . it is deterministic and continuous , and is meaningful only at the level of ensemble or bulk effect . on the microscopic level , such as on the scale of single electrons and nucleons , quantum mechanics is the operative physics , and state transitions are quantized . this is encoded in the time evolution of quantum states governed by an appropriate hamiltonian , yielding the specific schrdinger equation that applies in that setting . although pedagogically separable , the scales are intrinsically linked physically . for instance , the bulk magnetization , @xmath3 , is the net sum of the atomic level magnetic moments , @xmath4 . similarly the pulse magnetization frequency used to torque the bulk magnetization vector and change its direction is a radio frequency wave , @xmath5 , oscillating at the larmor frequency of the individual nucleons . furthermore , @xmath5 takes its effect by acting directly on the individual electrons and nucleons , causing them to precess in phase . electrons , protons , and neutrons are magnets . they derive their magnetism from intrinsic spin and orbital angular momentum . in the presence of an externally applied magnetic field , this results in bulk magnetism of sample tissue . the magnetic dipole moment , @xmath6 , of a single unpaired electron or nucleon is given by , @xmath7 where @xmath8 is the gyromagnetic ratio measured in hz / t ( hertz per tesla ) and @xmath9 is the spin . the gyromagnetic ratio for the electron is given by , @xmath10 where @xmath11 is the elementary charge , @xmath12 is the mass of the electron , and @xmath13 is the electron gyromagnetic factor ( g - factor ) , a dimensionless number whose experimental agreement with theory is near unprecedented on this scale , and thereby lends strong credibility to the theory of quantum electrodynamics . table ( [ tab : consts ] ) shows the electron and proton g - factors . the gyromagnetic ratio for nucleons is given by , @xmath14 where @xmath15 is the mass of the proton , and @xmath16 is the g - factor of the nucleon . estimated gyromagnetic ratios of some biologically - relevant nuclei are shown in table ( [ tab : gyromag ] ) @xcite . [ tab : gyromag ] the bohr magneton , @xmath17 , and the nuclear magneton , @xmath18 , are named quantities related to the electron and proton gyromagnetic ratios respectively , and are given by , @xmath19 and @xmath20 the magnetic moment of the electron is much larger than that of the nucleons , and as is discussed below , this corresponds to smaller precession , pulse , and signal frequency for protons than electrons . specifically , nucleons precess , absorb , and emit electromagnetic signals in the radio frequency range . this range is non - ionizing radiation , and makes mri a safer choice than ionizing imaging modalities such as plane and computed tomography x - rays . the magnetic potential energy is , @xmath21 the force on a magnetic moment , the torque on a magnetic moment , and the larmor precession frequency or resonance frequency , can all be described in terms of the magnetic potential energy , @xmath22 . the magnetic moment aligns itself in either parallel or anti - parallel orientation to the external field @xmath23 . the extent of parallelism is constrained by the eigenvalues of the @xmath24 and @xmath25 operators above , such that from the geometry , the parallel and anti - parallel states make angles of @xmath26 degrees with the positive and negative z - axes respectively ( assuming @xmath23 is oriented exclusively along the z - axis ) . this process is called magnetization . in the case of a population sample , upon application of @xmath23 , the spin orientation distribution goes from a somewhat arbitrary array of directions to being composed of only two directions , parallel and antiparallel . the parallel orientation is the lower energy state , and thereby is the more populous state for protons in the sample . the population distribution is temperature dependent , such that the proportion of protons in the higher energy state increases with temperature . the relationship is given by , @xmath27 where @xmath28 is the higher energy state , @xmath29 is the lower energy state , @xmath30 is the energy difference between the two states , @xmath31 is the boltzmann constant , and t is the absolute temperature in kelvin . the bulk magnetization can be modeled as , @xmath32 with @xmath33 , it follows that @xmath34 and @xmath35 . in other words , the equilibrium magnetization is entirely along the @xmath36 ( longitudinal ) direction and there is no component in the @xmath37 ( transverse ) direction . the torque on a magnetic moment in a magnetic field acts to align it with the field , is a function of the magnetic potential energy , and is given by , @xmath38 where @xmath39 is the angle between @xmath40 and @xmath6 , and the minus sign indicates the restoring nature of the torque . the force on a nucleon or electron in a magnetic field is , @xmath41 when an external magnetic field @xmath23 is applied to an electron or neutron particle , the particle s magnetic moment precesses about the direction of @xmath23at a frequency called the larmor frequency given by , @xmath42 in the equilibrium magnetization state with @xmath33 and @xmath35 , there is no precession of the bulk magnetization @xmath3 because @xmath43 . considering only the spin of a spin 1/2 object in a magnetic field , and assuming @xmath44 , we get the following hamiltonian , @xmath45 where the subscript @xmath46 denotes either @xmath11 for electron or @xmath47 for a neutron , and where , @xmath48 it follows that , @xmath49 we see that remarkably , the hamiltonian is proportional to the z - component of spin , therefore the magnetic spin eigenstates are simultaneously energy eigenstates . furthermore the energy eigenvalues are proportional to the magnetic spin quantum numbers . specifically , @xmath50 the above relation led early researchers to the discovery of spin , by its manifestation as the zeeman effect and the splitting of energy levels in the hydrogen optical spectrum . given a bulk magnetization @xmath3 due to a static magnetic field @xmath51 , a pulse of a weaker magnetic field , @xmath5 , oscillating at the larmor frequency and applied perpendicular to @xmath23 , will result in a transition of the individual nucleons from dephased to in - phase oscillation . this manifests as the acquisition of a transverse phase @xmath52 and oscillations of @xmath3 about the z - axis at the larmor frequency @xmath53 . this `` tilting '' of the magnetization vector also implies a decrease in @xmath54 . during the pulse , the net external magnetic field is the sum of @xmath23 and @xmath5 . if the pulse duration was extended indefinitely , the result would be alignment of the bulk magnetization with the new direction , @xmath55 , and there would be no resultant oscillation observed . the pulsed nature of @xmath5 is therefore critical to the technology of magnetic resonance imaging . the net magnetization is given by , @xmath56 while the ratio of @xmath57 to @xmath58 is given in equation ( [ boltz ] ) above as , @xmath59 . at equilibrium , the solution of the above system of two equations yields the equilibrium magnetization , @xmath60 where @xmath61 is the reduced boltzmann constant and @xmath62 is the proton density . the spin - lattice relaxation , also known as @xmath63 relaxation or longitudinal relaxation , is the process of longitudinal magnetization recovery following a @xmath5 perturbation . specifically , @xmath63 is the time constant of longitudinal magnetization recovery , and is given by , @xmath64 where @xmath65 is equilibrium magnetization and @xmath66 is the longitudinal magnetization instantaneously after pulse excitation . in inversion recovery , the initial magnetization is negative , and typically @xmath67 . substituting this value into equation ( [ eqn : t2 ] ) above yields , @xmath68 fluid attenuated inversion recovery ( flair ) is useful in the diagnosis of certain pathologies such as periventricular white - matter lesion in multiple sclerosis . flair can be t1 or t2 weighted . figure ( [ fig : sag_t1_flair ] ) shows a sagittal section of a t1 flair brain image , and figure ( [ fig : t2_flair_b ] ) shows a t2 flair brain image . the spin - spin relaxation , also known as @xmath69 relaxation or transverse relaxation , is the process of transverse magnetization relaxation following a @xmath5 perturbation . specifically , @xmath69 is the time constant of transverse magnetization recovery , and is given by , @xmath70 where @xmath71 is the transverse magnetization instantaneously after pulse excitation . following the @xmath72 induction of in - phase precession , the proton spins again begin to dephase according to @xmath69 time constant . of note , the relaxation of transverse magnetization is faster than the recovery of longitudinal magnetization . in other terms , @xmath73 . figure ( [ fig : t2axfse ] ) shows a t2 fast spin echo image of acute ischemic stroke in the posterior circulation . upon pulse excitation and tilting , the magnetization vector oscillates about the z - axis . this oscillating magnetic moment in turn generates an electromotive force in accordance with faraday - maxwell law of electromagnetic induction , @xmath74 where @xmath75 is the curl operator , @xmath76 is the electric field , and @xmath77 is the magnetic field . the above equation can be rendered in integral form via stokes theorem to yield , @xmath78 where @xmath79 can represent a wire coil bounding a surface @xmath80 , @xmath81 is an infinitesimal segment length along the wire , and @xmath82 is vector normal to an infinitesimal area of @xmath80 . such integral transformation has certain advantages in numerical computation . for instance , green s functions approaches can be invoked and can greatly decrease the number of unknowns , thereby decreasing the computational resource demand of the problem . the radio frequency ( rf ) coils within the mri probe experience an electromotive force due to the rotating magnetization . this results in an electric current signal which is fed as output to a screen . the signal decays with time according to the @xmath69 time constant , hence the name free inductance decay ( fid ) signal @xcite . the bloch equations describe the time evolution of the bulk magnetization vector , and are therefore also referred to as the equations of motion of magnetization . they are given by , @xmath83 @xmath84 and @xmath85 where @xmath8 is the gyromagnetic ratio of the nucleon and @xmath86 there are several contrast agents in current clinical use , most of which are based either on low molecular weight chelates of gadolinium ion ( gd@xmath87 ) @xcite such as gadolinium - dpta , or are based on iron oxide ( feo ) such as fe@xmath88o@xmath89 , fe@xmath2o@xmath88 , and @xmath8-fe@xmath2o@xmath88 ( or gamma phase maghemite ) @xcite . based on their magnetic properties , they confer contrast of vasculature from background , and by proxy of vascularization density , of organs from surround . specifically , the gd@xmath87-based agents predominantly shorten @xmath63 time , while the feo - based agents predominantly shorten @xmath69 time . nanoparticle formulations of contrast agents are in various stages of development and can be based on gadolinium or iron - oxide , or on other agents such as cobalt , nickel , manganese , or copper ions @xcite . optimization of the efficacy of mr contrast agents is an active area of research @xcite . the growing array of mr contrast applications has included efforts to target tissue or tumor - specific receptors @xcite . serious adverse reactions to intravenous mri contrast agents are rare . however in the mid 2000s gadolinium chelates were increasingly hypothesized as playing a role in nephrogenic systemic fibrosis ( nsf ) , a disease entity unknown prior to 1997 . it has since been essentially established that gadodiamide , a gadolinium - containing agent , is associated with an increased risk of developing nsf in susceptible individuals @xcite . superparamagnetic iron oxides have been promoted by some as possible alternatives to gadolinium - based agents in patients at risk for nsf @xcite . for 3d localization , a gradient coil generates a magnetic field gradient in the bore . then in accordance with the larmor frequency relation @xmath90 , it follows that the resonance frequency , @xmath91 acquires space dependence , and thereby serves to effectively label the tissue in space . the process of space - specific excitation of the sample is called _ slice selection_. the excitation pulse frequency is simply chosen as a bandwidth frequency flanking the region - of - interest or the _ slice_. the process of _ scanning _ is an iteration , however simple or complex , through a sequence of slices . and slice thickness is proportional to excitation pulse bandwidth . the slice thickness can be decreased by decreasing the bandwidth , or alternatively by increasing the magnetic field gradient . for instance , given a constant magnetic field gradient , the slice thickness between two points @xmath92 and @xmath93 is given by , @xmath94 } = r_2 - r_1\ ] ] both the bandwidth and magnetic field gradient are graphical user interface ( gui)- adjustable parameters on current day mri machines . this gradient - based localization technique was paul lauterbur s contribution to the development of mri @xcite . the mri is a simple machine in principle . it was was invented by raymond v. damadian by 1971 @xcite . it consists of a magnet , a radio frequency coil , an empty space or bore , and a computer processor . the magnet is for applying the static and pulsed magnetic fields . current day machines typically use superconducting electromagnets . these are wire coils cooled to low temperatures using liquid helium and liquid nitrogen @xcite , allowing for greatly decreased electrical resistance , and consequently high current loops generating strong magnetic fields . the rf coil typically serves both as the generator of the pulsed magnetic field , and as the detector of the fid signal . such duality of function is possible simply because the maxwell - faraday equation is true in both directions . several machines also have a separate gradient coil . the gradient coil has @xmath46 , @xmath95 , and @xmath36 components indicating the spatial direction of the gradient field , and together conferring 3d localization . the computer processor runs the software instructions for pulse sequences and signals processing , including image reconstruction . mri requires strong magnetic fields . though the earth s magnetic field is variable with time and is asymmetric between northern and southern hemispheres @xcite , it is estimated to range between 25,000 and 65,000 nanoteslas ( 0.25 to 0.65 gauss ) at the surface , with a commonly cited average of 46,000 nanotesla ( 0.46 gauss ) @xcite . current clinical use mris commonly range from 1.5 tesla to 7 teslas . therefore mris may require magnetic fields up to 280,000 times the earth s magnetic field . this has been achieved by a shift from permanent magnets to superconducting electromagnets , and presents another interdisciplinary link to both theory and practice . the dominant physical theory of superconductivity is called bcs theory @xcite after its formulators , bardeen , cooper , and schrieffer , and is based on electron pairing via phonon - exchange to form so called cooper pairs . though electrons are fermions and subject to the fermi exclusion principle , precluding any two electrons occupying the same state , cooper pairs act like bosons because their pairing and spin summation results in integer - spin character , a type of boson - like state . upon cooling , these boson - like pairs can condense , analogous to bose - einstein condensation and representing a quantum phase transition . quantum phase transitions have historically been challenging to study in a controlled way in the laboratory , but have recently been demonstrated by h. t. mebrahtu et . al . in tunable quantum tunnelling studies of luttinger liquids @xcite . in superconductivity , upon transition into the condensed state , the coherence of the system imposes a high energy penalty on resistance , effectively resulting in a zero resistance state . this is formally encoded in the theory by an energy gap at low ( subcritical ) temperatures . in superconducting electromagnets , the coil cooling is done through liquid helium and liquid nitrogen and presents a connection to the chemistry of phase transitions , condensed matter and solid state physics , as well as the engineering aspects of attaining and maintaining such low temperatures . in this section , we review the mri modalities used in the assessment of acute ischemic stroke . the diffusion - weighted mri ( dwi ) modality exploits the brownian motion of water molecules within a sample , and the relative phase shift in moving water versus stationary water . the apparent diffusion coefficient ( adc ) is a parameter which can be mapped to provide diagnostic information . in acute ischemic stroke , cytotoxic cellular injury results in axonal edema and a subsequent decrease in brownian motion this is manifested as a hyperintensity on dwi and a hypointensity on the corresponding adc map . figure ( [ fig : dwi_2 ] ) is a dwi showing hyperintensity in the left mca and pca distributions , consistent with acute ischemic stroke . figure ( [ fig : adc_stroke_1 ] ) is an adc map showing hypointensity in the left mca distribution , consistent with acute ischemic stroke . figure ( [ fig : adc_pca_stroke_b ] ) shows an adc map with a left parieto - occipital hypointensity reflecting left posterior circulation acute ischemic stroke . in the subacute setting , the adc may normalize and even increase due likely to ischemia - associated remodeling and loss of structural integrity . the differential for hyperintensity on dwi includes hemorrhagic stroke , traumatic brain injury , multiple sclerosis , and brain abscesses @xcite . dwi has long been shown in animal models to be more efficacious than @xmath69-weighted imaging for the early detection of transient cerebral ischemia @xcite . additionally , dwi has been shown to be highly efficacious in the early detection of acute subcortical infarctions @xcite . dwi in conjunction with echo - planar imaging has been shown to effectively discriminate between high grade ( high cellularity ) and low grade ( low cellularity ) gliomas @xcite . of note , in spite of the relatively high specificity and sensitivity of dwi in the early detection of acute ischemic stroke , there is a small subset of stroke patients who evade dwi detection in spite of clinically evident stroke - like neurological deficits @xcite . dwi tractography or diffusion tensor imaging is a form of principal component analysis in which the dominant eigendirection is used to determine the path of an axonal tract in a given voxel . this method has shown potential for further elucidation of neuronal pathways in the brain . perfusion - weighted imaging ( pwi ) is an ordinary differential equations compartment model of blood perfusion through organs . an arterial input function ( aif ) is determined as input into the particular chosen model . the aif is an impulse , and the model is essentially described by an impulse response or green s function . the computed output are perfusion parameters such as mean transit time , time to peak , cerebral blood flow rate , and cerebral blood volume . there are various protocols for the perfusion conditions . for instance the dynamic susceptibility contrast imaging method uses gadolinium contrast to gather local changes in @xmath96 signal in surrounding tissue . @xmath96 is a type of @xmath69 in which static dephasing effects are not explicitly rf - canceled , therefore dephasing from magnetic field inhomogeneities and susceptibility effects contribute to a tighter fid envelope ( more rapid decay ; @xmath97 ) than is seen in @xmath69 . other pwi protocols include arterial spin labeling and blood oxygen level dependent labeling . not surprisingly , pwi results vary with the choice of perfusion model and computational method with which it is implemented @xcite . both pwi and dwi are efficacious in the early detection of cerebral ischemia and correlate well with various stroke quantitation scales @xcite . furthermore , the combination of diffusion and perfusion - weighted imaging allows for assessment of the ischemic penumbra , that watershed region at the boundary of infarcted and non - infarcted tissue @xcite . it represents tissue which has suffered some degree of ischemia , but remains viable and possibly salvageable by expedient intervention with thrombolytic therapy @xcite , hypothermia , blood pressure elevation , or other experimental methods under study . outside the penumbra , thrombosis or autoregulation of circulation shuts down perfusion to infarcted tissue , resulting in a drastic decrease in perfusion . diffusion is simultaneously decreased due to tissue ischemia , hence no pwi - dwi mismatch occurs . within the penumbra however , the tissue is hypoperfused , yet since still viable , it has yet to sustain sufficient cytotoxic axonal damage to manifest a significant decrease in diffusion . the registration of both imaging modalities therefore demonstrates a diffusion - perfusion mismatch at the penumbra . in set - theoretic terms , the brain territory @xmath98 with compromised diffusion is a proper subset of the territory @xmath99 with compromised perfusion , and the relative complement of @xmath98 in @xmath99 is called the _ ischemic penumbra _ @xmath100 , @xmath101 one caveat to pwi - dwi mismatch assessment is that when done subjectively by the human eye , it can be demonstrably unreliable @xcite . hence development of quantitative metrics and learning algorithms are needed in this area . for instance , generalized linear models have been used and shown promise @xcite . magnetic resonance spectroscopy ( mrs ) is being used in attempts to reliably measure the magnetic resonance of metabolites whose concentrations change in the setting of acute ischemic stroke . lactate ( lac ) levels are well known to increase in ischemia , and is therefore a candidate in development , while n - acetyl aspartate ( naa ) levels have been shown to decrease in acute stroke , and are used in proton spectroscopy studies . most studies confirming the ischemia - associated elevation in lac and decrease in naa have been done in the setting of hypoxic - ischemic encephalopathy in newborns @xcite . in addition to lactate , @xmath39-glx and glycine have also been shown to be increased in the asphyxiated neonate @xcite . phosphorus magnetic resonance spectroscopy detects changes in the resonance spectrum of energy metabolites , and has also shown prognostic significance in hypoxic - ischemic brain injury @xcite . in their current form , mrs methods are less sensitive and practical for the assessment of acute ischemic stroke than the other magnetic resonance modalities discussed here . however , mrs is rapidly finding a place in the routine clinical assessment of hypoxic - ischemic encephalopathy in the asphyxiated neonate . some neonatal intensive care units , for instance , now conduct mrs studies on all neonates below a certain threshold weight . the blood oxygen level - dependent ( bold ) mri is a magnetic resonance imaging method that derives contrast from the difference in magnetic properties of oxygenated versus deoxygenated blood . hemoglobin is the molecule which carries oxygen in the blood , and is located in red blood cells . the oxygen - bound form of hemoglobin is called oxyhemoglobin , while the oxygen - free form is called deoxyhemoglobin . deoxyhemoglobin is paramagnetic and in the presence of an applied magnetic field assumes a relatively higher magnetic dipole moment than oxyhemoglobin which is a diamagnetic molecule . deoxygenated blood has a higher concentration of deoxyhemoglobin than oxyhemoglobin , and this difference is reflected in the mri signal . @xmath96-based bold mri has shown promise in localizing the penumbra in acute ischemic stroke @xcite . it is also used in functional mri ( fmri ) , which is based on the principle that active brain areas have higher resource ( e.g. oxygen , glucose ) demands and higher waste output @xcite . in this context , bold has been used to study the brain s behavior during sensorimotor recovery following acute ischemic stroke . specifically , the coupling between bold and electrical neuroactivity has shed some light on the still poorly understood process of spontaneous motor recovery following a stroke @xcite . bold mri results can be affected by baseline circulatory status , and therefore in the research setting near - infrared spectroscopy ( nirs ) can be used as a control , or as an alternative modality in the clinic @xcite . bold mri has been used in assessing the brain s response to hypercapnia @xcite . hypercapnia can cause changes in multiple variables in the brain such as cerebral blood flow , oxygen consumption rate , cerebral blood volume , arterial oxygen concentration , and red blood cell volume fraction . however , hypercapnia - associated cerebrovascular reactivity has been strongly correlated with arterial spin labeling and other pwi surrogates , suggesting that the brain s reaction to hypercapnia is dominated by changes in cerebral blood flow @xcite . mra is a set of magnetic resonance - based techniques for imaging the circulatory system . figures ( [ fig : mracowax ] ) and ( [ fig : mralat ] ) show mra images of the circle - of - willis . mra techniques can be broadly categorized into flow - dependent and flow - independent groups . the flow - dependent methods derive contrast from the motion of blood in vasculature relative to the static state of surrounding tissue @xcite . two currently well - known and used examples of flow dependent methods are : ( i ) phase contrast mra , ( pc mra ) @xcite and ( ii ) time - of - flight mra ( tof mra ) . the tof mra images can be acquired in either two dimensional ( 2d tof ) or three dimensional ( 3d tof ) formats . figure ( [ fig:2d_tof_fspgr ] ) shows a 2d tof fast spoiled gradient echo sequence ( fsgr ) image of the carotids , and figure ( [ fig:3d_tof ] ) shows a an axial section of a 3d tof image of the carotids . pc mra exploits differences in spin phase of moving blood relative to static surrounding tissue , while tof mra exploits the difference in excitation pulse ( @xmath72 ) exposure of flowing blood relative to static surrounding tissue . this difference occurs because flowing blood spends less time in the field of exposure , and as a result is less spin - saturated then the surrounding tissue . this decreased spin - saturation translates into higher intensity signals on spin - echo sequences . flow - independent methods exploit inherent differences in magnetic properties of blood relative to surrounding tissue . for instance , fresh blood imaging is a method that exploits the longer @xmath69 time constant in blood relative to surround . this method finds specific utility in cardiac and cardio - cerebral imaging by using fast spin echo sequences which can exploit the spin saturation differences between systole and diastole @xcite . other examples include susceptibility weighted imaging ( swi ) and four dimensional dynamic mra ( 4d mra ) . swi derives contrast from magnetic susceptibility differences between blood and surround @xcite , while 4d mra uses time - dependent bit - mask subtraction after injection of gadolinium - dpta or some other pharmacological contrast agent @xcite . figures ( [ fig : swi_1 ] ) and ( [ fig : swi_2 ] ) show swi consistent with acute ischemic infarction in the left pca and mca distributions . of note , in the sense that 4d mra exploits the time interval between injection and initial image acquisition , it is arguably more flow - dependent than the other methods mentioned here in that category . hydrogen is the most commonly imaged nucleus in magnetic resonance imaging , due largely to the great abundance of water in biological tissue , and to the high gyromagnetic ratio of hydrogen . furthermore , hydrogen is the only atom whose schrdinger eigenvalue problem has been exactly solved . other hydrogenic atoms involve screened potentials and require iterative approximation eigenvalue problem solvers such as the hartree - fock scheme and its variants . in what follows we review the electronic orbital configuration of hydrogen . in this section we focus on the electron . specifically that lone electron in the orbit of the hydrogen atom . we make this choice because its simplicity allows us review the attributes of quantum orbital angular momentum and spin , in a context which is not only real , but also highly relevant to magnetic resonance imaging , which as noted targets the hydrogen atom . the electron in @xmath0h has both position and angular momentum . the hamiltonian commutes with the squared orbital angular momentum operator , therefore angular momentum eigenstates are also energy eigenstates . of note , the position and angular momentum observables do not commute . this is equivalent to heisenberg s uncertainty principle . hence definite energy or orbital angular momentum states , are represented in the position basis as probability density functions , which are synonymous with electron clouds or atomic orbitals . orbital angular momentum is the momentum possessed by a body by virtue of circular motion , as along an orbit . in quantum mechanics , the square of the orbital momentum is a quantized quantity , which can take on only certain discrete `` allowed '' values . this reality and discreteness of the orbital momentum eigenvalues is a manifestation of the spectral theorem of normal bounded operators ; given the hermiticity and boundedness of the quantum orbital angular momentum operator over a specified finite volume such as the radius of an atom ; which is itself a manifestation of the negative energy of the electron in orbit yielding a so called _ bound state_. the orbital momentum states can be indirectly represented by the spherical harmonic functions to be derived below . the spherical harmonics are the eigenfunctions of the squared quantum orbital angular momentum operator in spherical coordinates . their product with the radial wave functions yield the wave function , @xmath102 , of the hydrogen s electron . where in accordance with the principal postulate of quantum mechanics , @xmath103 is the probability of finding the electron at a given point @xmath104 . in the time - independent case , the hamiltonian defines the schrdinger equation as follows , @xmath105 the orbital angular momentum states are the angular portion of the solution , @xmath102 , of the schrdinger equation in spherical coordinates . substituting the expression for the hamiltonian of a single non - relativistic particle into equation ( [ eqn : schrodinger ] ) yields , @xmath106 where @xmath107 . in spherical coordinates the above equation becomes , @xmath108\psi(r ) \nonumber \\ + v(r)\psi(r ) = e\psi(r),\end{aligned}\ ] ] where @xmath109 denotes mass . invoking the method of separation of variables , we assume existence of a solution of the form @xmath110 . we illustrate the method on the laplace equation which can be interpreted as a homogeneous form of the schrdinger equation , i.e. one for which @xmath111 , @xmath112 multiplying through by @xmath113 yields , @xmath114 which can be separated as , @xmath115 and @xmath116 we can again assume separability in the form @xmath117 , and substitute into equation ( [ eqn : angular ] ) above to get , @xmath118 from which we extract the following two separated equations , @xmath119 and @xmath120 our homogeneous equation has therefore been separated into three ordinary differential equations : equations ( [ eqn : radial ] ) for the radial part , equation ( [ eqn : azimuth ] ) for the azimuthal part , and equation ( [ eqn : polar ] ) for the polar part . solutions to the radial equation are of the form , @xmath121 where a and b are constant coefficients , @xmath122 is a non - negative integer such that @xmath123 , where @xmath109 is an integer on the right hand side of the azimuthal and polar equations . a regularity constraint at the poles of the sphere yield a sturm - liouville problem which in turn mandates the form @xmath124 . the coefficient @xmath125 is often set to zero to admit only solutions which vanish at infinity . however , this choice is application - specific , and for certain applications it may be appropriate to instead set @xmath126 . solutions to the azimuthal equation are of the form , @xmath127 where @xmath128 and @xmath129 are constant coefficients and @xmath11 is the base of the natural logarithm . to obtain solutions to the polar equation , we proceed in a number of steps . first we substitute @xmath130 , and recast equation ( [ eqn : polar ] ) into , @xmath131 next we compute the derivatives of @xmath132 under the transformation @xmath133 . employing the chain rule yields , @xmath134 and @xmath135 a substitution of the derivatives into the polar equation yields , @xmath136 dividing through by @xmath137 and changing variables via @xmath130 and @xmath138 , yields the associated ( @xmath139 ) legendre differential equation , @xmath140 whose solutions are given by @xmath141 the solution @xmath142 to angular portion of the laplace equation is therefore of the form , @xmath143 where n is a normalization factor given by , @xmath144 and enabling , @xmath145 where @xmath146 is solid angle . @xmath147 are called the spherical harmonic functions and are discussed some more in the following subsection . figures ( [ fig : shf_1_1 ] ) to ( [ fig : shf_5_2 ] ) are plots of a sampling of spherical harmonics with @xmath148 . for the electronic configuration of the hydrogen atom , the electron experiences a potential v(r ) due to the proton . v(r ) is the coulomb potential given by , @xmath149 substituting this into the generic schrdinger equation yields , @xmath150 which in spherical coordinates is , @xmath108\psi(r ) \nonumber \\ -\frac{e^2}{4\pi\epsilon_0r}\psi(r ) = e\psi(r ) , \label{eqn : schrodinger_evp}\end{aligned}\ ] ] where @xmath109 is given by , @xmath151 and is the two - body reduced mass @xmath15 of the proton and @xmath12 of the electron . equation ( [ eqn : schrodinger_evp ] ) above is an eigenvalue problem which after some rearranging , we can solve using the same separation of variables method illustrated above . we recast as , @xmath152r(r)y(\theta,\psi ) \nonumber \\ + \left[\frac{1}{\sin\theta } \frac{\partial}{\partial\theta } \left ( \sin \theta \frac{\partial}{\partial\theta } \right ) + \frac{1}{\sin^2\theta}\frac{\partial^2}{\partial\phi^2}\right]r(r)y(\theta,\psi ) \nonumber\\ = 0\end{aligned}\ ] ] dividing through by @xmath153 and splitting the operator , we get , @xmath154y(\omega ) = -l(l+1 ) \label{eqn : shf2}\ ] ] and @xmath155= l(l+1 ) . \label{eqn : laguerre}\ ] ] equation ( [ eqn : shf2 ] ) is exactly the same as equation ( [ eqn : shf1 ] ) which we solved above , and whose solutions are the spherical harmonic functions , @xmath147 . equation ( [ eqn : laguerre ] ) above is isomorphic to a generalized laguerre differential equation , whose solutions are related to the associated laguerre polynomials . as mentioned in the case of laplace s equation above , regularity conditions at the boundary prescribe the separation constant @xmath156 . to transform equation ( [ eqn : laguerre ] ) into a generalized laguerre equation , we recast into the following form , @xmath157r(r)=0 . \label{eqn : laguerre2}\ ] ] then we proceed with the following sequence of substitutions : @xmath158 executing the first substitution above transforms equation ( [ eqn : laguerre2 ] ) into , @xmath159y(r)=0 . \label{eqn : laguerre3}\ ] ] next , the second substitution transforms the above into , @xmath160y(r)=0 . \label{eqn : laguerre4}\ ] ] and finally , the third substitution transforms the above into , @xmath161y(x)=0 \label{eqn : laguerre5}.\ ] ] dividing through by @xmath162 yields , @xmath163y(x)=0 \label{eqn : laguerre6}.\ ] ] next , we make the substitutions , @xmath164 and @xmath165 which transform equation ( [ eqn : laguerre6 ] ) into the following generalized laguerre equation @xmath166y_j^k(x)=0,\ ] ] whose solutions are readily verified to be the following associated laguerre functions , @xmath167 where @xmath168 are the associated laguerre polynomials . from equation ( [ eqn : lagsubs2 ] ) we see that , @xmath169 which we substitute into equation ( [ eqn : lagsubs3 ] ) to yield , @xmath170 @xmath171 is the principal ( or radial ) quantum number . the corresponding energy eigenvalues , @xmath172 , are obtained by substituting @xmath162 into @xmath173 and solving for @xmath30 , we get , @xmath174 where @xmath175 is the bohr radius and is given by , @xmath176 the value of @xmath172 in the ground state ( @xmath177 ) is called the rydberg constant . its value is shown in table ( [ tab : consts ] ) along with that of other physical constants pertinent to magnetic resonance imaging . the displayed values are from the 2010 committee for data on science and technology ( codata ) recommendations @xcite , and @xmath178 is the relative standard uncertainty . [ tab : consts ] to transform the radial equation solution , equation ( [ eqn : assoc_laguerre ] ) , into a form in terms of radial distance @xmath179 , and principal and azimuthal quantum numbers @xmath171 and @xmath122 , we note the following relations : from equation ( [ eqn : k_l_relation ] ) we get , @xmath180 from equation ( [ eqn : ndef ] ) we get , @xmath181 and from equations ( [ eqn : lagsubs1 ] ) and ( [ eqn : rydberg ] ) we get , @xmath182 therefore , @xmath183 substituting the above derived expressions of @xmath46 , @xmath184 , and @xmath31 into the associated laguerre function , equation ( [ eqn : assoc_laguerre ] ) , we get , @xmath185 next we make the substitution @xmath186 , which yields the radial solutions of the hydrogen atom , @xmath187 where the constant @xmath188 . following multiplication by the spherical harmonics and normalization , we obtain the exact solution of the time - independent schrdinger equation of the hydrogen atom , @xmath189 ^ 3}}e^{-r / na_0}\left(\frac{2r}{na_0}\right)^ll^{2l+1}_{n - l-1}\left(\frac{2r}{na_0}\right)y_{l , m}(\theta,\phi),\end{aligned}\ ] ] where the principal , azimuthal , and magnetic quantum numbers ( @xmath171 , @xmath122 , @xmath109 ) take the following values , @xmath190 in this section , we review the spherical harmonics and the radial wave functions . as noted above , the spherical harmonics , @xmath191 , are characterized by an azimuthal and a magnetic quantum number , @xmath122 and @xmath192 respectively . table ( [ tab : shf ] ) shows the spherical harmonics with @xmath122 = 0,1,2 , and 3 . formalism of state representation by the spherical harmonics is as follows , @xmath193 @xmath194 where , @xmath195 is the polar angle and @xmath196 is the azimuthal angle . the spherical harmonic representing rotation quantum numbers @xmath197 at position state @xmath198 , is given by @xmath199 [ tab : shf ] [ tab : radwavs ] the spherical harmonics are a complete set of orthonormal functions over the unit sphere . in particular , they are eigenfunctions of the square of the orbital angular momentum operator , @xmath200 . and are thereby representations of the allowed discrete states of angular momentum . @xmath200 can be arrived at by representing laplace s equation in spherical coordinates and considering only its angular portion . classically , angular momentum is : @xmath201 where @xmath202 is the position vector and @xmath203 is the momentum vector . by analogy , the quantum orbital angular momentum operator is given by , @xmath204 where @xmath205 is the gradient operator , and we have used the momentum operator of quantum mechanics , @xmath206 in spherical coordinates , @xmath200 is then represented by , @xmath207 and similarly , @xmath208 , @xmath209 , and @xmath210 are represented as follows , @xmath211 @xmath212 @xmath213 given the expressions above for @xmath214 , @xmath200 , and @xmath210 , it is readily shown that , @xmath215 and @xmath216 the square of the radial wave functions is the probability the electron is located a distance @xmath179 from the nucleus . table ( [ tab : radwavs ] ) shows the radial wave functions for @xmath171 = 1,2 , and 3 . @xmath217 = i\hbar \epsilon_{ijk}l_k,\ ] ] where @xmath218 is the levi - civita symbol given by , @xmath219 and i , j , k can have values of x , y , or z. @xmath220 = 0 \qquad \mbox{for } ~j = x , y , z\ ] ] the ladder operators act to increase or decrease the @xmath109 quantum number of a state and are given by , @xmath221 it is readily verified that , @xmath222 @xmath223 = \pm \hbar l_{\pm},\ ] ] and @xmath224 = 2\hbar l_z.\ ] ] spin is an intrinsic quantum property of elementary particles such as the electron and the quark . composite particles such as the neutron , proton , and even atoms and molecules also possess spin by virtue of their composition from their elementary particle constituents . the term spin is itself a misnomer , as a physically spinning object about an axis is not sufficient to account for the observed magnetic moments . the mathematics of spin was worked out by wolfgang pauli , who either astutely or serendipitously neglected to name it . this was wise , as later insight elucidated spin as an intrinsic quantum mechanical property with no classical correlate . spin follows essentially the same mathematics as outlined above for orbital angular momentum . this is by virtue of the isomorphism of the respective lie algebras of so(3 ) and su(2 ) groups . and is elaborated further in section ( [ sec : groups ] ) below . one notable difference is that half - integer eigenvalues are allowed for spin , while orbital angular momentum admits only integer values . the spin algebra is encapsulated in the commutation relations as follows , @xmath225 = i\hbar \epsilon_{ijk}s_k,\ ] ] @xmath226 = 0 \qquad \mbox{for } ~j = x , y , z,\ ] ] @xmath227 @xmath228 @xmath229 = \pm \hbar s_{\pm},\ ] ] and @xmath230 = 2\hbar s_z.\ ] ] the addition of spin and or of orbital angular momentum refers to the process of determining the net spin and or orbital angular momentum of a system of particles . for example , the spin of a hydrogen atom in its ground state , i.e. the @xmath231 state with zero orbital angular momentum , constitutes the composite spin of the electron and the proton . similarly , the nuclear spin , @xmath232 , is the composition of the spins of protons and neutrons in the nucleus of an atom . mathematically , the process is described by a change of basis . a change from the product space basis @xmath233 to the `` net sum '' space basis @xmath234 , where @xmath235 and @xmath236 . the representation of the sum space eigenstates in terms of the product space eigenstates are referred to as the _ clebsch - gordan coefficients_. @xmath237 where @xmath238 are the clebsch - gordan coefficients and are given by , @xmath239 and we have notationally represented the uncoupled product space basis by , @xmath240 here we review a procedural description of the addition of angular momenta . consider a hydrogen atom @xmath0h in the ground state . the spin contributors are the lone electron in the l=0 shell , and a proton which is the sole constituent of the nucleus . the possible configurations of spin are : @xmath241 where @xmath242 indicates spin @xmath243 ( spin up ) and @xmath244 indicates spin @xmath245 ( spin down ) . and where each of the above states are prepared by measuring @xmath246 , @xmath247 , @xmath248 and @xmath249 for each state . we are interested in the combined state of the two particle system , and this requires the quantum mechanical addition of angular momenta , or in this example , spin . the net result is a state @xmath250 whose spin can be determined by measuring @xmath251 and whose @xmath36-component of spin can be determined by measuring @xmath252 , where @xmath253 and @xmath254 . it follows that @xmath235 , and therefore the possible values of @xmath109 are 1 , 0 , 0 , -1 . this suggests a triplet state ( s=1 ) and a singlet state ( s=0 ) . the triplet state corresponds to m=1,0,-1 , and is given by , @xmath255 and the singlet state corresponds to , @xmath256 where upon designation of the top assignment @xmath257 , each of the other states are obtained by application of the ladder operators defined in equation ( [ eqn : spinladder ] ) . for example , @xmath258 to change to a representation in the uncoupled basis , we write , @xmath259 and therefore , @xmath260 which yields , @xmath261 finally to obtain the singlet state , we simply orthogonalize the above , yielding , @xmath262 the above procedure is easily carried out by a computer for any combination of particle spins . for many particle systems , for instance in the computation of nuclear spins or non - ground state @xmath0h configurations , the associativity property is used and the above description applies . in the above example , the factors @xmath263 are the clebsch - gordan coefficients . there are several openly available implementations of clebsch - gordan coefficient calculators , in addition to tabulations in handbooks of physics formulae @xcite . orbital angular momentum and spin are the source of magnetic resonance . their abstract mathematical description is group theoretic , and extends naturally into other fundamental physics such as the strong interaction of quarks in nucleons and nucleons in nuclei . the frenchman henri poincar commented that `` mathematicians do not study objects , but the relationships between objects '' . relationships between symmetry groups have indeed been the essential device for probing the unseen in the realm of high energy particle physics . for spin 1/2 particles such as the electron and the quark , the spin observable can be represented by the su(2 ) group . su(n ) is the group of @xmath264 unitary matrices with unit determinant , in which the group operation is matrix - matrix multiplication . similarly so(n ) is the group of @xmath264 orthogonal matrices of unit determinant , with matrix multiplication as group operator . analogous to the su(2 ) representation of spin , quantum orbital angular momentum can be represented by the so(3 ) group . both groups share a lie algebra , encoded in the commutation relations shown above , because they are isomorphic to each other . in particular , su(2 ) is a double cover of so(3 ) . the su(3 ) group models the interactors of the strong force in quantum chromodynamics ( qcd ) . a counterclockwise rotation by an angle @xmath265 about the x , y , or z - axes , can respectively be represented by @xmath266 the corresponding infinitesimal generators , @xmath267 , @xmath268 , @xmath269 respectively are : @xmath270 the corresponding lie algebra can then be readily shown to be : @xmath271 = i\hbar \epsilon_{ijk } g_k\ ] ] @xmath272 it follows that the generators of su(2 ) are the pauli matrices given by , @xmath273 the su(2 ) lie algebra is then prescribed by the following commutation and anti - commutation relations , @xmath274 = 2i\epsilon_{abc}\sigma_c , \ ] ] and @xmath275 which combine to give : @xmath276 where @xmath277 is the kronecker delta and @xmath278 is the anti - commutator . given @xmath279 , such that , @xmath280 we define an assignment , @xmath281 , such that @xmath282 where @xmath283 . it follows that , @xmath284 where @xmath232 is the @xmath285 identity matrix , @xmath286 are the pauli matrices , and we have used the einstein summation notation over @xmath47 . @xmath287 @xmath281 is a bijective homomorphism between @xmath288 and @xmath289 . and this shows @xmath288 is isomorphic to @xmath289 . the generators of the su(3 ) group are given by the gell - mann matrices : @xmath290 then defining @xmath291 , the commutation relations are , @xmath292 = i\sum_{c=1}^8f_{abc}t_c , \ ] ] and @xmath293 where @xmath294 and @xmath295 are the antisymmetric and symmetric structure constants of su(3 ) respectively . magnetic resonance imaging is an application of quantum mechanics which has revolutionized the practice of medicine . electrons , protons , and neutrons are magnets by virtue of their spin and orbital angular momentum . tissue is made of ensembles of such subatomic particles , and can be imaged by sensing their responses to applied magnetic fields . as discussed in this paper , a wide array of mri modalities exist , each of which derives contrast by specific perturbations of the magnetization vector . the potential of mri in the diagnosis and even the treatment of various diseases is far from fully realized . a quantum level understanding of magnetic resonance technology is essential for further innovation in this field . the author thanks his friends henok t. mebrahtu and peter q. blair for reading this manuscript and providing helpful feedback . and he thanks them for many enjoyable physics conversations over coffee . he thanks daniel ennis for making available his matlab code for plotting spherical harmonics . he thanks his lovely wife lisa m. odaibo for a helpful conversation on the clinical use of magnetic resonance spectroscopy in hypoxic - ischemic injury assessment in the neonate . and he especially thanks her for her love and support . the author thanks his entire family , and celebrates his father , dr . stephen k. odaibo , on his birthday today . he thanks him for being a kind wonderful father and example . he thanks those who contributed and are contributing to his education , including : dr . robert a. copeland jr . , dr . leslie jones , dr . janine smith - marshall , dr . bilal khan , dr . david katz , dr . earl kidwell , dr . william deegan , dr . brian brooks , dr . natalie afshari , dr . isaac o. karikari , dr . xiaobai sun , dr . mark dewhirst , dr . carlo tomasi , dr . john kirkpatrick , dr . nicole larrier , dr . brenda armstrong , dr . phil goodman , dr . joanne wilson and dr . ken wilson , dr . srinivasan mukunduan jr . , david simel , dr . danny o. jacobs , dr . timothy elston , drs . akwari , drs . haynes , dr . john mayer , dr . lex oversteegen , dr . peter v. oneil , dr . anne cusic , dr . henry van den bedem , dr . james r. ward , dr . marius nkashama , dr . michael w. quick , dr . robert j. lefkowitz , dr . arlie petters , and several other excellent educators and role models not mentioned here . signals processing plays an important role in magnetic resonance imaging , and is briefly reviewed in this appendix . the free inductance decay signal is often collected in the frequency domain and fourier transformed into the time domain @xcite . also with a gradual shift towards mri scanners of higher magnetic field , the signal - to - noise ratio requires increasingly more effective noise filtering algorithms . as signal generation and collection methods become increasingly more sophisticated e.g via spirals and other geometric sequences , non - uniform sampling , and exotic @xmath72 pulse sequence configurations , signals processing methods need to advance in tandem to address problems arising . computing speed also takes on new significance with such developments . indeed imaging and graphics applications have significantly driven the interest and funding for fast algorithms and computing . notable advancements include the fast fourier transform , the fast - multipole method of greengard and rokhlin @xcite , and parallel computing hardware such as the compute unified device architecture ( cuda ) multi - core processors by nvidia for graphics processing @xcite . furthermore , a new field is taking form at the interface of high performance software algorithms and multi - core hardware processors @xcite . we take a brief look at some of the basics of signals processing below , the fourier transform converts a time domain signal to its corresponding frequency domain signal , and is given by , @xmath296 = f(\omega ) = \int_{-\infty}^{\infty}f(t ) e^{-i\omega t}dt,\ ] ] while the inverse fourier transform converts a frequency domain signal to its corresponding time domain signal , and is given by , @xmath297=f(t ) = \int_{-\infty}^{\infty}f(\omega ) e^{i\omega t}d\omega.\ ] ] in practice , the discrete versions of the above equations are used instead , and give good results provided that signal data is sampled at an appropriate frequency , the nyquist frequency . given an mri signal of @xmath298 sample points , @xmath299 $ ] , where @xmath300 , the frequency domain representation is given by the discrete fourier transform , @xmath301 = \sum_{n = 0}^{n-1 } f[n]e^{-i2\pi\frac{k}{n } n},\ ] ] and the discrete inverse fourier transform is given by , @xmath302 = \frac{1}{n}\sum_{k = 0}^{n-1 } f[{\omega}_k]e^{i2\pi\frac{k}{n } n}.\ ] ] the above fourier formulas readily allow extension to two and three dimensions as well as various other customizations , tailored to suite the particular mri application and data format . the whittaker- kotelnikov- shannon- raabe- someya- nyquist theorem prescribes a lower bound on the sampling frequency necessary for perfect reconstruction of a band - limited signal . it states that given a band limited signal @xmath303 whose frequency domain representation is such that @xmath304 for all @xmath305 and some @xmath306 , then by sampling at a rate @xmath307 , the image can be perfectly reconstructed . the sampling interval is @xmath308 , and the discrete time signal is expressed as @xmath309 $ ] where @xmath171 is an integer . @xmath310 and is in units of hertz , hence @xmath311 is in seconds . the perfect reconstruction is given by the whittaker - shannon interpolation formula , @xmath312 \cdot \mbox{sinc}\left(\frac{t - nt}{t}\right),\ ] ] where sinc is the _ normalized sampling function _ given by , @xmath313 in practice , low - pass filtering methods can be used to decrease the amplitude of frequency components which exceed the effective bandlimit . the aliasing problem arises when the sampling rate @xmath314 is less than twice the bandlimit @xmath306 . the effect of this is that the reconstructed image is an alias of the actual image . the discrete time fourier transform ( dtft ) of a signal is given by , @xmath315e^{-i\omega n},\ ] ] and in the case of an under - sampled mri signal , the dtft matches that of the alias . therefore the reconstructed image is merely an alias . this phenomenon can be described more formally by invoking the poisson summation formula , @xmath316 where the middle term above is the periodic summation . for a bandlimited under - sampled signal , increasing the sampling rate to satisfy the nyquist criterion will restore perfect reconstruction . the convolution theorem facilitates application - specific image processing before or after reconstruction in either k - space or time domain . the theorem is given by , @xmath317 and states that the fourier transform of the convolution of two functions equals the product of the fourier transforms of both functions . where the convolution is defined as , @xmath318 100 m. abramowitz and i. a. stegun , _ handbook of mathematical functions : with formulas , graphs , and mathematical tables _ 55 , courier dover publications , 1964 . h. ay , f. s. buonanno , g. rordorf , p. w. schaefer , l. h. schwamm , o. wu , r. g. gonzalez , k. yamada , g. a. sorensen , and w. j. koroshetz , _ normal diffusion - weighted mri during stroke - like deficits _ , neurology * 52 * ( 1999 ) , no . 9 , 17841784 . d. azzopardi , j. s. wyatt , e. b. cady , d. t. delpy , j. baudin , a. l. stewart , p. l. hope , p. a. hamilton , and e. o. r. reynolds , _ prognosis of newborn infants with hypoxic - ischemic brain injury assessed by phosphorus magnetic resonance spectroscopy _ , pediatric research * 25 * ( 1989 ) , no . 5 , 445451 . m. babic , d. horak , p. jendelova , v. herynek , v. proks , v. vanecek , p. lesny , and e. sykova , _ the use of dopamine - hyaluronate associate - coated maghemite nanoparticles to label cells _ , international journal of nanomedicine * 7 * ( 2012 ) , 1461 . p. a. bandettini and e. c. wong , _ a hypercapnia - based normalization method for improved spatial localization of human brain activation with fmri _ , nmr in biomedicine * 10 * ( 1997 ) , no . 4 - 5 , 197203 . p. a. barber , d. g. darby , p. m. desmond , q. yang , r. p. gerraty , d. jolley , g. a. donnan , b. m. tress , and s. m. davis , _ prediction of stroke outcome with echoplanar perfusion - and diffusion - weighted mri _ , neurology * 51 * ( 1998 ) , no . 2 , 418426 . j. bardeen , _ theory of the meissner effect in superconductors _ , physical review * 97 * ( 1955 ) , no . 6 , 1724 . j. bardeen , l. n. cooper , and j. r. schrieffer , _ microscopic theory of superconductivity _ , physical review * 106 * ( 1957 ) , no . 1 , 162164 . to3em , _ theory of superconductivity _ , physical review * 108 * ( 1957 ) , no . 5 , 1175 . a. j. barkovich , k. d. westmark , h. s. bedi , j. c. partridge , d. m. ferriero , and d. b. vigneron , _ proton spectroscopy and diffusion imaging on the first day of life after perinatal asphyxia : preliminary report _ , american journal of neuroradiology * 22 * ( 2001 ) , no . 9 , 17861794 . p. barz , a. marmarou , p. fatouros , k. hayasaki , and f. corwin , _ contribution of vasogenic and cellular edema to traumatic brain swelling measured by diffusion - weighted imaging _ , journal of neurosurgery * 87 * ( 1997 ) , no . 6 , 900907 . c. l. bennett , z. p. qureshi , a. o. sartor , l. a. b. norris , a. murday , s. xirasagar , and h.s . thomsen , _ gadolinium - induced nephrogenic systemic fibrosis : the rise and fall of an iatrogenic disease _ , clinical kidney journal * 5 * ( 2012 ) , no . 1 , 8288 . m. a. bernstein , k. f. king , and x. j. zhou , _ handbook of mri pulse sequences _ , academic press , 2004 . m. n. biltcliffe , p. e. hanley , j. b. mckinnon , and p. roubeau , _ the operation of superconducting magnets at temperatures below 4.2 k _ , cryogenics * 12 * ( 1972 ) , no . 1 , 4447 . f. binkofski and r. j. seitz , _ modulation of the bold - response in early recovery from sensorimotor stroke _ , neurology * 63 * ( 2004 ) , no . 7 , 12231229 . e. brcher , _ kinetic stabilities of gadolinium ( iii ) chelates used as mri contrast agents _ , contrast agents i ( 2002 ) , 103122 . j. w. m. bulte and d. l. kraitchman , _ iron oxide mr contrast agents for molecular and cellular imaging _ , nmr in biomedicine * 17 * ( 2004 ) , no . 7 , 484499 . k. butcher , m. parsons , t. baird , a. barber , g. donnan , p. desmond , b. tress , and s. davis , _ perfusion thresholds in acute stroke thrombolysis _ , stroke * 34 * ( 2003 ) , no . 9 , 21592164 . c. calautti and j. c. baron , _ functional neuroimaging studies of motor recovery after stroke in adults _ , stroke * 34 * ( 2003 ) , no . 6 , 15531566 . m. cappellini , g. rapisardi , m. l. cioni , and c. fonda , _ acute hypoxic encephalopathy in the full - term newborn : correlation between magnetic resonance spectroscopy and neurological evaluation at short and long term . _ , la radiologia medica * 104 * ( 2002 ) , no . 4 , 332 . p. caravan , _ strategies for increasing the sensitivity of gadolinium based mri contrast agents _ , chem . * 35 * ( 2006 ) , no . 6 , 512523 . to3em , _ protein - targeted gadolinium - based magnetic resonance imaging ( mri ) contrast agents : design and mechanism of action _ , accounts of chemical research * 42 * ( 2009 ) , no . 7 , 851862 . p. caravan , j. j. ellison , t. j. mcmurry , and r. b. lauffer , _ gadolinium ( iii ) chelates as mri contrast agents : structure , dynamics , and applications _ , chemical reviews * 99 * ( 1999 ) , no . 9 , 22932352 . s. c. chang , p. h. lai , w. l. chen , h. h. weng , j. t. ho , j. s. wang , c. y. chang , h. b. pan , and c. f. yang , _ diffusion - weighted mri features of brain abscess and cystic or necrotic brain tumors : comparison with conventional mri _ , clinical imaging * 26 * ( 2002 ) , no . 4 , 227236 . l. n. cooper , _ bound electron pairs in a degenerate fermi gas _ , physical review * 104 * ( 1956 ) , no . 4 , 1189 . s. b. coutts , j. e. simon , a. i. tomanek , p. a. barber , j. chan , m. e. hudon , j. r. mitchell , r. frayne , m. eliasziw , a. m. buchan , and a. m. demchuk , _ reliability of assessing percentage of diffusion - perfusion mismatch _ , stroke * 34 * ( 2003 ) , no . 7 , 16811683 . s. c. cramer , r. shah , j. juranek , k. r. crafton , and v. le , _ activity in the peri - infarct rim in relation to recovery from stroke _ , stroke * 37 * ( 2006 ) , no . 1 , 111115 . r. a. crisostomo , m. m. garcia , and d. c. tong , _ detection of diffusion - weighted mri abnormalities in patients with transient ischemic attack _ , stroke * 34 * ( 2003 ) , no . 4 , 932937 . e. j. cukauskas , d. a. vincent , and b. s. deaver , _ magnetic susceptibility measurements using a superconducting magnetometer _ , review of scientific instruments * 45 * ( 1974 ) , no . 1 , 16 . r. damadian , _ tumor detection by nuclear magnetic resonance _ , science * 171 * ( 1971 ) , no . 3976 , 11511153 . r. damadian , k. zaner , d. hor , t. dimaio , l. minkoff , and m. goldsmith , _ nuclear magnetic resonance as a new tool in cancer research : human tumors by nmr _ , annals of the new york academy of sciences * 222 * ( 1973 ) , no . 1 , 10481076 . r. v. damadian , _ apparatus and method for detecting cancer in tissue _ , 1974 , us patent 3,789,832 . c. l. dumoulin and h. r. hart jr , _ magnetic resonance angiography . _ , radiology * 161 * ( 1986 ) , no . 3 , 717720 . c. l. dumoulin , s. p. souza , m. f. walker , and w. wagle , _ three - dimensional phase contrast angiography _ , magnetic resonance in medicine * 9 * ( 1989 ) , no . 1 , 139149 . r. r. ernst and w. a. anderson , _ application of fourier transform spectroscopy to magnetic resonance _ , review of scientific instruments * 37 * ( 1966 ) , no . 1 , 93102 . j. faiz kayyem , r. m. kumar , s. e. fraser , and t. j. meade , _ receptor - targeted co - transport of dna and magnetic resonance contrast agents _ , chemistry & biology * 2 * ( 1995 ) , no . 9 , 615620 . d. n. firmin , g. l. nayler , p. j. kilner , and d. b. longmore , _ the application of phase shifts in nmr for flow measurement _ , magnetic resonance in medicine * 14 * ( 1990 ) , no . 2 , 230241 . b. s. geisler , f. brandhoff , j. fiehler , c. saager , o. speck , j. rther , h. zeumer , and t. kucinski , _ blood oxygen level dependent mri allows metabolic description of tissue at risk in acute stroke patients _ , stroke * 37 * ( 2006 ) , no . 7 , 17781784 . c. b. grandin , t. p. duprez , a. m. smith , c. oppenheim , a. peeters , a. r. robert , and g. cosnard , _ which mr - derived perfusion parameters are the best predictors of infarct growth in hyperacute stroke ? comparative study between relative and quantitative measurements _ , radiology * 223 * ( 2002 ) , no . 2 , 361370 . l. greengard and v. rokhlin , _ a fast algorithm for particle simulations _ , journal of computational physics * 73 * ( 1987 ) , no . 2 , 325348 . to3em , _ a new version of the fast multipole method for the laplace equation in three dimensions _ , acta numerica * 6 * ( 1997 ) , no . 1 , 229269 . o. h. j. grhn and r. a. kauppinen , _ assessment of brain tissue viability in acute ischemic stroke by bold mri _ , nmr in biomedicine * 14 * ( 2001 ) , no . 7 - 8 , 432440 . f. hamzei , r. knab , c. weiller , and j. rther , _ the influence of extra - and intracranial artery disease on the bold signal in fmri _ , neuroimage * 20 * ( 2003 ) , no . 2 , 13931399 . p. hanover and b. eilhardt , _ conductor system for superconducting cables _ , january 1 1972 , us patent 3,634,597 . p. harish and p. narayanan , _ accelerating large graph algorithms on the gpu using cuda _ , high performance computing hipc 2007 ( 2007 ) , 197208 . m. hermier and n. nighoghossian , _ contribution of susceptibility - weighted imaging to acute stroke assessment _ , stroke * 35 * ( 2004 ) , no . 8 , 19891994 . m. l. hugh , _ low temperature electric transmission systems _ , 1971 , us patent 3,562,401 . g. e. jackson , s. wynchank , and m. woudenberg , _ gadolinium ( iii ) complex equilibria : the implications for gd ( iii ) mri contrast agents _ , magnetic resonance in medicine * 16 * ( 1990 ) , no . 1 , 5766 . c. w. jung and p. jacobs , _ physical and chemical properties of superparamagnetic iron oxide mr contrast agents : ferumoxides , ferumoxtran , ferumoxsil _ , magnetic resonance imaging * 13 * ( 1995 ) , no . 5 , 661674 . g. w. kabalka , e. buonocore , k. hubner , m. davis , and l. huang , _ gadolinium - labeled liposomes containing paramagnetic amphipathic agents : targeted mri contrast agents for the liver _ , magnetic resonance in medicine * 8 * ( 1988 ) , no . 1 , 8995 . s. kestur , k. irick , s. park , a. al maashri , v. narayanan , and c. chakrabarti , _ an algorithm - architecture co - design framework for gridding reconstruction using fpgas _ , design automation conference ( dac ) , 2011 48th acm / edac / ieee , ieee , 2011 , pp . 585590 . y. r. kim , i. j. huang , s. r. lee , e. tejima , j. b. mandeville , m. p. a. van meer , g. dai , y. w. choi , r. m. dijkhuizen , e. h. lo , and b. r. rosen , _ measurements of bold / cbv ratio show altered fmri hemodynamics during stroke recovery in rats _ , journal of cerebral blood flow & metabolism * 25 * ( 2005 ) , no . 7 , 820829 . h. kobayashi , m. w. brechbiel , et al . , _ dendrimer - based macromolecular mri contrast agents : characteristics and application . _ , molecular imaging * 2 * ( 2003 ) , no . 1 , 1 . s. d. konda , m. aref , s. wang , m. brechbiel , and e. c. wiener , _ specific targeting of folate dendrimer mri contrast agents to the high affinity folate receptor expressed in ovarian tumor xenografts _ , magnetic resonance materials in physics , biology and medicine * 12 * ( 2001 ) , no . 2 , 104113 . p. h. kuo , e. kanal , a. k. abu - alfa , and s. e. cowper , _ gadolinium - based mr contrast agents and nephrogenic systemic fibrosis _ , radiology * 242 * ( 2007 ) , no . 3 , 647649 . s. m. lai , t. y. tsai , c. y. hsu , j. l. tsai , m. y. liao , and p. s. lai , _ bifunctional silica - coated superparamagnetic fept nanoparticles for fluorescence / mr dual imaging _ , journal of nanomaterials * 2012 * ( 2012 ) , 5 . j. a. langlois , w. rutland - brown , and m. m. wald , _ the epidemiology and impact of traumatic brain injury : a brief overview _ , the journal of head trauma rehabilitation * 21 * ( 2006 ) , no . 5 , 375378 . p. c. lauterbur et al . , _ image formation by induced local interactions : examples employing nuclear magnetic resonance _ , nature * 242 * ( 1973 ) , no . 5394 , 190191 . n. lee , y. choi , y. lee , m. park , w. k. moon , s. h. choi , and t. hyeon , _ water - dispersible ferrimagnetic iron oxide nanocubes with extremely high r 2 relaxivity for highly sensitive in vivo mri of tumors _ , nano letters * 12 * ( 2012 ) , no . 6 , 31273131 . d. r. lide , _ crc handbook of chemistry and physics _ , crc press , 2012 . m. j. lipinski , v. amirbekian , j. c. frias , j. g. s. aguinaldo , v. mani , k. c. briley - saebo , v. fuster , j. t. fallon , e. a. fisher , and z. a. fayad , _ mri to detect atherosclerosis with gadolinium - containing immunomicelles targeting the macrophage scavenger receptor _ , magnetic resonance in medicine * 56 * ( 2006 ) , no . 3 , 601610 . a. y. liu , j. a. maldjian , l. j. bagley , g. p. sinson , and r. i. grossman , _ traumatic brain injury : diffusion - weighted mr imaging findings _ , american journal of neuroradiology * 20 * ( 1999 ) , no . 9 , 16361641 . d. h. live and s. i. chan , _ bulk susceptibility corrections in nuclear magnetic resonance experiments using superconducting solenoids _ , analytical chemistry * 42 * ( 1970 ) , no . 7 , 791792 . g. k. malik , m. pandey , r. kumar , s. chawla , b. rathi , and r. k. gupta , _ mr imaging and in vivo proton spectroscopy of the brain in neonates with hypoxic ischemic encephalopathy _ , european journal of radiology * 43 * ( 2002 ) , no . 1 , 613 . s.a . manavski , _ cuda compatible gpu as an efficient hardware accelerator for aes cryptography _ , signal processing and communications , 2007 . icspc 2007 . ieee international conference on , ieee , 2007 , pp . d. m. mandell , j. s. han , j. poublanc , a. p. crawley , j. a. stainsby , j. a. fisher , and d. j. mikulis , _ mapping cerebrovascular reactivity using blood oxygen level - dependent mri in patients with arterial steno - occlusive disease _ , stroke * 39 * ( 2008 ) , no . 7 , 20212028 . c. maneru , c. junque , n. bargallo , m. olondo , f. botet , m. tallada , j. guardia , and j. m. mercader , _ 1h - mr spectroscopy is sensitive to subtle effects of perinatal asphyxia _ , neurology * 57 * ( 2001 ) , no . 6 , 11151118 . p. marckmann , l. skov , k. rossen , a. dupont , m. b. damholt , j. g. heaf , and h. s. thomsen , _ nephrogenic systemic fibrosis : suspected causative role of gadodiamide used for contrast - enhanced magnetic resonance imaging _ , journal of the american society of nephrology * 17 * ( 2006 ) , no . 9 , 23592362 . m. p. marks , d. c. tong , c. beaulieu , g. w. albers , a. de crespigny , and m. e. moseley , _ evaluation of early reperfusion and iv tpa therapy using diffusion - and perfusion - weighted mri _ , neurology * 52 * ( 1999 ) , no . 9 , 17921792 . h. t. mebrahtu , i. v. borzenets , d. e. liu , h. zheng , y. v. bomze , a. i. smirnov , h. u. baranger , and g. finkelstein , _ quantum phase transition in a resonant level coupled to interacting leads _ , nature * 488 * ( 2012 ) , no . 7409 , 6164 . r. w. meyerhoff , _ superconducting power transmission _ , cryogenics * 11 * ( 1971 ) , no . 2 , 91101 . m. mikawa , h. kato , m. okumura , m. narazaki , y. kanazawa , n. miwa , and h. shinohara , _ paramagnetic water - soluble metallofullerenes having the highest relaxivity for mri contrast agents _ , bioconjugate chemistry * 12 * ( 2001 ) , no . 4 , 510514 . j. mintorovitch , m. e. moseley , l. chileuitt , h. shimizu , y. cohen , and p. r. weinstein , _ comparison of diffusion - and t2-weighted mri for the early detection of cerebral ischemia and reperfusion in rats _ , magnetic resonance in medicine * 18 * ( 1991 ) , no . 1 , 3950 . m. miyazaki , s. sugiura , f. tateishi , h. wada , y. kassai , and h. abe , _ non - contrast - enhanced mr angiography using 3d ecg - synchronized half - fourier fast spin echo _ , journal of magnetic resonance imaging * 12 * ( 2000 ) , no . 5 , 776783 . p. j. mohr , b. n. taylor , and d. b. newell , _ codata recommended values of the fundamental physical constants : 2010 . 2012 _ , e - print : arxiv . org / abs/1203.5425 . b. morgan , a. l. thomas , j. drevs , j. hennig , m. buchert , a. jivan , m. a. horsfield , k. mross , h. a. ball , l. lee , et al . , _ dynamic contrast - enhanced magnetic resonance imaging as a biomarker for the pharmacological response of ptk787/zk 222584 , an inhibitor of the vascular endothelial growth factor receptor tyrosine kinases , in patients with advanced colorectal cancer and liver metastases : results from two phase i studies _ , journal of clinical oncology * 21 * ( 2003 ) , no . 21 , 39553964 . m. e. moseley , y. cohen , j. mintorovitch , l. chileuitt , h. shimizu , j. kucharczyk , m. f. wendland , and p. r. weinstein , _ early detection of regional cerebral ischemia in cats : comparison of diffusion - and t2-weighted mri and spectroscopy _ , magnetic resonance in medicine * 14 * ( 1990 ) , no . 2 , 330346 . y. murata , k. sakatani , t. hoshino , n. fujiwara , t. kano , s. nakamura , and y. katayama , _ effects of cerebral ischemia on evoked cerebral blood oxygenation responses and bold contrast functional mri in stroke patients _ , stroke * 37 * ( 2006 ) , no . 10 , 25142520 . c. j. l. murray , a. d. lopez , et al . , _ global mortality , disability , and the contribution of risk factors : global burden of disease study _ , lancet * 349 * ( 1997 ) , no . 9063 , 14361442 . t. neumann - haefelin , h. j. wittsack , f. wenserski , m. siebler , r. j. seitz , u. mdder , and h. j. freund , _ diffusion - and perfusion - weighted mri : the dwi / pwi mismatch region in acute stroke _ , stroke * 30 * ( 1999 ) , no . 8 , 15911597 . e. a. neuwelt , b. e. hamilton , c. g. varallyay , w. r. rooney , r. d. edelman , p. m. jacobs , and s. g. watnick , _ ultrasmall superparamagnetic iron oxides ( uspios ) : a future alternative magnetic resonance ( mr ) contrast agent for patients at risk for nephrogenic systemic fibrosis ( nsf ) & quest _ , kidney international * 75 * ( 2008 ) , no . 5 , 465474 . j. nickolls , i. buck , m. garland , and k. skadron , _ scalable parallel programming with cuda _ , queue * 6 * ( 2008 ) , no . 2 , 4053 . a. nukada , y. ogata , t. endo , and s. matsuoka , _ bandwidth intensive 3-d fft kernel for gpus using cuda _ , high performance computing , networking , storage and analysis , 2008 . sc 2008 . international conference for , ieee , 2008 , pp . 111 . national academies ( us ) . committee on facilitating interdisciplinary research , committee on science , engineering , public policy ( us ) , institute of medicine ( us ) , and national academy of engineering , _ facilitating interdisciplinary research _ , national academy press , 2005 . opdyke and v. mejia , _ earth s magnetic field _ , geophysical monograph series * 145 * ( 2004 ) , 315320 . l. pan , l. gu , and j. xu , _ implementation of medical image segmentation in cuda _ , information technology and applications in biomedicine , 2008 . itab 2008 . international conference on , ieee , 2008 , pp . g. papamakarios , g. rizos , n .p . pitsianis , and x. sun , _ fast computation of local correlation coefficients on graphics processing units _ , proc . of spie vol , vol . 7444 , 2009 , pp . 7444121 . m. w. parsons , p. a. barber , j. chalk , d. g. darby , s. rose , p. m. desmond , r. p. gerraty , b. m. tress , p. m. wright , g. a. donnan , and s. m. davis , _ diffusion - and perfusion - weighted mri response to thrombolysis in stroke _ , annals of neurology * 51 * ( 2002 ) , no . 1 , 2837 . j. e. perthen , f. calamante , d. g. gadian , and a. connelly , _ is quantification of bolus tracking mri reliable without deconvolution ? _ , magnetic resonance in medicine * 47 * ( 2002 ) , no . 1 , 6167 . o. a. c. petroff , j. w. prichard , k. l. behar , j. r. alger , j. a. den hollander , and r. g. shulman , _ cerebral intracellular ph by 31p nuclear magnetic resonance spectroscopy _ , neurology * 35 * ( 1985 ) , no . 6 , 781781 . j. m. provenzale , r. d. tien , g. j. felsberg , and l. hacein - bey , _ spinal dural arteriovenous fistula : demonstration using phase contrast mra . _ , journal of computer assisted tomography * 18 * ( 1994 ) , no . 5 , 811 . k. n. raymond , v. c. pierre , et al . , _ next generation , high relaxivity gadolinium mri agents _ , bioconjugate chemistry * 16 * ( 2005 ) , no . 1 , 38 . g. rordorf , w. j. koroshetz , w. a. copen , s. c. cramer , p. w. schaefer , r. f. budzik , l. h. schwamm , f. buonanno , a. g. sorensen , and g. gonzalez , _ regional ischemia and ischemic injury in patients with acute middle cerebral artery stroke as defined by early diffusion - weighted and perfusion - weighted mri _ , stroke * 29 * ( 1998 ) , no . 5 , 939943 . h. j. rosen , s. e. petersen , m. r. linenweber , a. z. snyder , d. a. white , l. chapman , a. w. dromerick , j. a. fiez , and m. corbetta , _ neural correlates of recovery from aphasia after damage to left inferior frontal cortex _ , neurology * 55 * ( 2000 ) , no . 12 , 18831894 . m. rovaris , a. gass , r. bammer , s. j. hickman , o. ciccarelli , d. h. miller , and m. filippi , _ diffusion mri in multiple sclerosis _ , neurology * 65 * ( 2005 ) , no . 10 , 15261532 . w. rutland - brown , j. a. langlois , k. e. thomas , and y. l. xi , _ incidence of traumatic brain injury in the united states , 2003 _ , the journal of head trauma rehabilitation * 21 * ( 2006 ) , no . 6 , 544 . c. rydahl , h. s. thomsen , and p. marckmann , _ high prevalence of nephrogenic systemic fibrosis in chronic renal failure patients exposed to gadodiamide , a gadolinium - containing magnetic resonance contrast agent _ , investigative radiology * 43 * ( 2008 ) , no . 2 , 141 g. schlaug , a. benfield , a. e. baird , b. siewert , k. o. lvblad , r. a. parker , r. r. edelman , and s. warach , _ the ischemic penumbra _ , neurology * 53 * ( 1999 ) , no . 7 , 15281528 . l. c. shih , j. l. saver , j. r. alger , s. starkman , m. c. leary , f. vinuela , g. duckwiler , y. p. gobin , r. jahan , j. p. villablanca , p. m. vespa , and c. s. kidwell , _ perfusion - weighted magnetic resonance imaging thresholds identifying core , irreversibly infarcted tissue _ , stroke * 34 * ( 2003 ) , no . 6 , 14251430 . m. b. singer , j. chong , d. lu , w. j. schonewille , s. tuhrim , and s. w. atlas , _ diffusion - weighted mri in acute subcortical infarction _ , stroke * 29 * ( 1998 ) , no . 1 , 133136 . m. b. stearns , _ spin - echo and free - induction - decay measurements in pure fe and fe - rich ferromagnetic alloys : domain - wall dynamics _ , physical review * 162 * ( 1967 ) , no . 2 , 496 . g. strangman , j. p. culver , j. h. thompson , and d. a. boas , _ a quantitative comparison of simultaneous bold fmri and nirs recordings during functional brain activation _ , neuroimage * 17 * ( 2002 ) , no . 2 , 719731 . a. sturzu , s. sheikh , u. klose , m. deeg , h. echner , t. nagele , u. ernemann , and s. heckl , _ novel gastrin receptor - directed contrast agents - potential in brain tumor magnetic resonance imaging _ , medicinal chemistry * 8 * ( 2012 ) , no . 2 , 133137 . t. sugahara , y. korogi , m. kochi , i. ikushima , y. shigematu , t. hirai , t. okuda , l. liang , y. ge , y. komohara , et al . , _ usefulness of diffusion - weighted mri with echo - planar technique in the evaluation of cellularity in gliomas _ , journal of magnetic resonance imaging * 9 * ( 1999 ) , no . 1 , 5360 . v. n. thijs , d. m. somford , r. bammer , w. robberecht , m. e. moseley , and g. w. albers , _ influence of arterial input function on hypoperfusion volumes measured with perfusion - weighted imaging _ , stroke * 35 * ( 2004 ) , no . 1 , 9498 . h. s. thomsen , _ nephrogenic systemic fibrosis : a serious late adverse reaction to gadodiamide _ , european radiology * 16 * ( 2006 ) , no . 12 , 26192621 . p. thullen , j. c. dudley , d. l. greene , j. l. smith , and h. h. woodson , _ an experimental alternator with a superconducting rotating field winding _ , power apparatus and systems , ieee transactions on ( 1971 ) , no . 2 , 611619 . d. c. tong , m. a. yenari , g. w. albers , m. obrien , m. p. marks , and m. e. moseley , _ correlation of perfusion - and diffusion - weighted mri with nihss score in acute ( @xmath319 6.5 hour ) ischemic stroke _ , neurology * 50 * ( 1998 ) , no . 4 , 864869 . c. tu and a. y. louie , _ nanoformulations for molecular mri _ , wiley interdisciplinary reviews : nanomedicine and nanobiotechnology ( 2012 ) . f. h. r van der zande , p. a. m hofman , and w. h. backes , _ mapping hypercapnia - induced cerebrovascular reactivity using bold mri _ , neuroradiology * 47 * ( 2005 ) , no . 2 , 114120 . z. wang , j. m. van oort , and m. x. zou , _ development of superconducting magnet for high - field mr systems in china _ , physica c : superconductivity ( 2012 ) . e. c. wiener , s. konda , a. shadron , m. brechbiel , and o. gansow , _ targeting dendrimer - chelates to tumors and tumor cells expressing the high - affinity folate receptor _ , investigative radiology * 32 * ( 1997 ) , no . 12 , 748 . w. wiltschko and r. wiltschko , _ magnetic compass of european robins _ , science * 176 * ( 1972 ) , no . 4030 , 6264 . g. woan , _ the cambridge handbook of physics formulas _ , cambridge univ pr , 2000 . o. wu , w. j. koroshetz , l. stergaard , f. s. buonanno , w. a. copen , r. g. gonzalez , g. rordorf , b. r. rosen , l. h. schwamm , r. m. weisskoff , and a. g. sorensen , _ predicting tissue outcome in acute human cerebral ischemia using combined diffusion - and perfusion - weighted mr imaging _ , stroke * 32 * ( 2001 ) , no . 4 , 933942 . k. yamada , o. wu , r. g. gonzalez , d. bakker , l. stergaard , w. a. copen , r. m. weisskoff , b. r. rosen , k. yagi , t. nishimura , and a. g. sorensen , _ magnetic resonance perfusion - weighted imaging of acute cerebral infarction _ , stroke * 33 * ( 2002 ) , no . 1 , 8794 . h. zhu , d. g. buck , z. zhang , h. zhang , p. wang , v. a. stenger , m. r. prince , and y. wang , _ high temporal and spatial resolution 4d mra using spiral data sampling and sliding window reconstruction _ , magnetic resonance in medicine * 52 * ( 2004 ) , no . 1 , 1418 .

a vehicle moving through the stratosphere ( altitudes 40km-50 km ) at hypersonic velocities ( 8 - 15 mach ) is covered by a plasma sheath . typically , the plasma density @xmath0 can be as high as @xmath1 with corresponding plasma frequency @xmath2 of about @xmath3 . in ( [ plasma_freq ] ) , @xmath4 is the electron charge @xmath5 , @xmath6 and @xmath7 is the electron mass @xmath8 . therefore the plasma is opaque to frequencies lower than @xmath3 . direct communication through such a plasma to and from the vehicle is impossible because frequencies @xmath9 suitable for long distance propagation through the atmosphere are usually much less . for example , the standard frequency used for navigational satellite systems , including the global positioning system ( gps ) , are less than @xmath10 . for the gps , @xmath11 . the challenge is to devise means to maintain continuous contact with the hypersonic vehicle . when such vehicles were principally spacecrafts , a blackout period of up to two minutes was acceptable albeit undesirable . but when the vehicles are of military origin , it is clear that continuous contact is essential for both targeting and rapid abort reasons . it is a challenge which has drawn many responses . they fall into several categories . the first ignores the presence of the plasma by using signals with frequencies well above the plasma frequency . the difficulty with this method is that such signals are heavily attenuated in and scattered by the atmosphere . a second means , which also ignores the plasma , is to use low frequency signals in the @xmath12 range where wavelengths are large compared to the plasma sheath thickness ( typically of the order of a meter ) . but such solutions have high cost and low bit rates and are not well supported by existing infrastructure . a third category of solutions violates the plasma . one approach is to remove , by vehicle reshaping , for example , the plasma from certain points on the vehicle at which one might place an antenna . another is to destroy it by electrophilic injection or by injecting water drops . a third approach is to use powerful magnets to reshape the plasma . such solutions involve a heavy cost in that design features necessary for their implementation must be built into the vehicle a priori . nevertheless some are feasible and worthy of consideration . for example , it is possible to build an antenna into a sharp leading edge which would protrude beyond the plasma and survive for sufficiently long ( it would be eventually destroyed by ablation ) to cover the flight time . the fourth category of solutions , and the one to which we are attracted , uses the properties of the plasma itself to effect transmission in the same way a judo expert uses the strength and motion of an opponent to defeat him . one idea is to create new modes of oscillation and propagation by the introduction of magnetic fields . indeed , for strong enough fields , the larmor frequency @xmath13 is sufficiently large that the window @xmath14 for which the plasma is opaque is small and transmission can be achieved for frequencies below @xmath13 . but the introduction of magnetic fields involves large additional weight and new design features . the second idea is much more simple . its aim is to take advantages of nonlinear properties of plasma to render it effectively transparent to the signal . communications both to and from the vehicle are feasible using basically the same ideas . we shall first describe the `` to the vehicle '' case . consider figure [ plasma_response_fig ] in which we show schematically the response of the plasma to an incoming signal with low frequency @xmath15 from a direction which makes an angle @xmath16 with the normal to the vehicle . there are two principal features to the response . first , there is a reflection from the layer at a point @xmath17 where the plasma frequency at the point @xmath18 is @xmath19 . however , the influence of the signal is felt beyond that point , namely at the resonant layer @xmath20 where @xmath21 . langmuir oscillations are excited there which produce large transversal and longitudinal components of the electric field . the resonant layer acts as an antenna . the task is to find a way to connect the antenna at the resonant layer at @xmath20 to a receiver on board the vehicle at @xmath22 . there are several possibilities which we have outlined before @xcite . the most practical one , however , is also the most simple and first suggested without a detailed numerical simulation in @xcite . we use an onboard source , which we call the pump , to generate electromagnetic signals of sufficiently high frequency @xmath23 ( @xmath24 ) that they can propagate through the plasma . there are several candidates for such a source . for example , available on the open market is a klystron amplifier which can generate @xmath25 of power at frequencies of @xmath26 . these high frequency waves have only to travel distances of a meter or less . they interact nonlinearly with and scatter off the signal wave . not surprisingly , the largest contribution to the scattered wave comes from the nonlinear interaction of the pump wave with the plasma density distortion induced by the incoming signal wave at the resonant layer . we call the scattered wave a stokes wave because the scattering process is a three wave interaction analogous to raman scattering . the stokes wave with frequency @xmath27 carries the information encoded on the signal wave back to the vehicle . we will show that , whereas much of the scattered stokes wave propagates away from the vehicle , a significant fraction is returned to the vehicle . what is remarkable is this . the ratio of the power flux of the stokes wave received at the vehicle to the power flux contained in the signal wave at the plasma edge can be between 0.7 and 2 percent . this means that reception of gps signals may be possible because one simply needs an onboard receiver approximately 100 times more sensitive than commercially available hand held receivers or use sufficiently larger antenna . we shall discuss in the conclusion the sensitivity required for a variety of sources . communications from the vehicle requires two power sources on the vehicle . one , which we term the stokes wave generator , will also carry the signal . the other is the pump wave . both have carrier frequencies above that of the maximum of the plasma frequency . their nonlinear interaction in the plasma produces an oscillations of frequency @xmath28 . consider figure [ from_concept ] . for @xmath29 where @xmath30 is determined by @xmath31 and @xmath16 is calculated from the differences in propagation directions of the pump and stokes waves , the oscillation does not propagate and its strength decays away from the vehicle . nevertheless this oscillation is sufficiently strong to act as a power source for a propagating wave in the region @xmath32 where @xmath33 . in the conclusion we analyze what power is required in order for the signal to be detected by distant receivers . it appears that even if we use usual available on the market generators communication can be put into practice . the plan of the paper is as follows . we begin in _ section 2 _ with a detailed analysis of the two dimensional propagation and interaction of a signal wave of frequency @xmath15 , a pump wave of frequency @xmath23 and a stokes wave of frequency @xmath34 through a plasma with a given density profile @xmath35 where @xmath36 is the direction normal to the vehicle . the key equation is a modification of the well known ginzburg equation @xcite @xmath37,\nonumber\end{aligned}\ ] ] for the magnetic field amplitude @xmath38 of an oscillation of frequency @xmath39 . in ( [ ginzburg_equation ] ) , the effective electric susceptibility is @xmath40 ( @xmath41 is the local plasma frequency and @xmath42 the collision frequency ) . the susceptibility is due to the linear response of the plasma to the electric fields of whichever waves are involved . the nonlinear current @xmath43 will be determined both by the product of the plasma density distortion with the linear current and the nonlinear response of the electric velocity field due principally to dynamic pressure forces . we observe that , for @xmath44 , the electric susceptibility is approximately @xmath45 and the left hand side of the nonlinear ginzburg equation ( [ ginzburg_equation ] ) is the usual wave operator . how do we use ( [ ginzburg_equation ] ) ? for the case of communication to the vehicle , we use it in two ways . first with @xmath46 , we determine for @xmath47 and @xmath48 , the field @xmath49 from which the distortion to the plasma produced by the incoming wave is calculated . in this instance , @xmath49 satisfies @xmath50 a glance at the third term shows that propagation is impossible for @xmath51 or , from ( [ epsilon_z_omega ] ) , for @xmath52 . the importance of the resonance layer where @xmath53 is seen from the denominator in the second term . having solved for @xmath49 from ( [ ginzburg_to_equation ] ) we can then calculate the plasma distortion field @xmath54 . its interaction with the pumping wave then produces a nonlinear current @xmath43 which gives rise to the stokes wave . the stokes wave @xmath55 and its propagation is calculated by solving ( [ ginzburg_equation ] ) with this @xmath43 and appropriate boundary conditions at the plasma edge and at the vehicle . our goal is to determine @xmath56 . we give the results of both the numerical simulation and an analytic estimation . the latter takes advantage of the fact , that , for the stokes wave , @xmath57 and that the principal plasma distortion occurs at the resonance layer . for communicating from the vehicle , we solve ( [ ginzburg_to_equation ] ) with the right hand side given by @xmath58 with @xmath43 calculated from the nonlinear interaction of the pump and stokes waves . here the goal is to calculate the flux of power of the signal wave with frequency @xmath59 as it leaves the plasma edge in the direction of some distant receiver . in _ section 3 _ , we describe the numerical procedure and give detailed results of our calculations . finally , in the _ conclusion _ , we use our results to calculate the powers of both the incoming and outgoing signals at their respective receivers . we discuss in addition several important considerations : * the advantages , particularly in terms of available power , of using pulsed signals . * the possibility of using gps sources for incoming signals . * the challenges involved in making ideas practicable . we shall study a very idealized situation when the plasma sheath is a flat slab . the plasma density is a linear function of the horizontal coordinate @xmath36 @xmath60 in this geometry the vehicle is the vertical wall placed at @xmath22 . the plasma density near the vehicle is @xmath61 . the plasma contacts the vacuum at @xmath62 , where @xmath63 . we shall study two situations : communication to the vehicle and communication from the vehicle . in both cases , three almost monochromatic electromagnetic waves exist in the plasma . two of them have high frequencies @xmath23 ( pumping wave ) , @xmath34 ( stokes wave ) . the third one has low frequency @xmath15 , satisfying the condition @xmath64 in the `` to the vehicle '' case @xmath15 is the circular frequency of the incoming signal . in the `` from the vehicle '' case , @xmath15 is the circular frequency of the outgoing signal . in both these cases , the low - frequency signal plays a key role . because the local plasma frequency at @xmath65 is @xmath15 , @xmath66 let us denote also the langmuir frequency at the vehicle as @xmath67 thus @xmath68 in a realistic situation @xmath69 ( it corresponds to @xmath70 ) , @xmath71 , @xmath72 , and @xmath73 . the wavelength of the incoming signal in the vacuum is @xmath74 , so that @xmath75 . we point out that in the case of low - frequency wave reflection from the ionosphere , the situation is the opposite @xmath76 . we shall assume that the ions positions are fixed and the plasma is cold ( @xmath77 ) . the magnetic field has only one component @xmath78 . the electric field has two components @xmath79 , @xmath80 . neither the electric nor magnetic fields depend on the @xmath81-coordinate . maxwell s equations read @xmath82 @xmath83 @xmath84 @xmath85 + \frac{\mu_0 e}{m}\vec h\right ) - \frac{1}{2}\nabla v^2,\ ] ] @xmath86 the power flux in vacuum is @xmath87 in equation ( [ euler ] ) @xmath42 is the effective friction of the electron fluid with the neutral gas , sometimes called the ion collision frequency . we take @xmath88 . the current @xmath89 . @xmath90 is the linear response of the plasma on the electric field , @xmath43 is the current due to nonlinear effects . for a monochromatic wave of frequency @xmath39 , maxwell s equations can be rewritten in the following form @xmath91 @xmath92 in our geometry , ( [ protoginzburg ] ) is one scalar equation . we should stress that this is an exact equation . the only challenge is the calculation of @xmath43 . finally for the magnetic field , one obtain the ginzburg equation @xmath93 for the high frequency pump and stokes waves @xmath94 . some exact solutions of simplified versions of the homogeneous ginzburg equation for several important cases can be found in appendix [ app : analytics_homogenious ] . what we are going to do is the following : in subsection [ lin_resp ] we shall calculate linear responses of the plasma to an electromagnetic wave , such as the electron velocity , linear current and the electron density profile perturbation ; the calculation of the first nonlinear correction to the linear current is done in subsection [ nonlin_curr ] ; analytic estimations for `` to the vehicle '' and `` from the vehicle '' cases are given in subsections [ analytics_to ] and [ analytics_from ] respectively . in order to calculate the nonlinear current we need to consider the linear responses of the plasma to the presence of an electromagnetic wave . for a field with frequency @xmath39 @xmath95 from ( [ euler ] ) , the linear term in the velocity @xmath96 and @xmath97 from ( [ rot_h ] ) @xmath98 using maxwell equations one can express all responses in terms of magnetic field @xmath99 the expression for a distortion @xmath100 of the electron density in the plasma @xmath101 can be derived from ( [ continuity ] ) and ( [ linear_velocity ] ) , @xmath102 the nonlinear current is due to the first nonlinear correction to the linear response velocities of electrons and a scattering of an electromagnetic wave on the distortion of the charge density profile produced by another wave @xmath103 we introduce the nonlinear velocity @xmath104 which can be found from the following equation @xmath105 + \frac{\mu_0 e}{m}\vec v_l \times \vec h - \frac{1}{2}\nabla v_l^2 = - \frac{1}{2}\nabla v_l^2.\ ] ] here we used a corollary of the maxwell equations and ( [ linear_v ] ) whence to within @xmath106 , @xmath107 = -\frac{\mu_0 e}{m}\vec h.\ ] ] this means that only the dynamic pressure induced by the fields affects the plasma . finally , we have everything for the calculation of the first term in right hand side of ginzburg equation ( [ ginzburg ] ) [ curl_j_nl ] @xmath108 the detailed expression of equation s ( [ ginzburg ] ) right hand side can be found in appendix [ app : current_nl ] . we would like to estimate the ratio @xmath109 of the fluxes of the squared scattered field to the squared incoming signal field and express it as a function of pump power flux @xmath110 measured in watts per square meter . we can make an analytic estimation of the three - wave process efficiency . the main contribution comes from the vicinity of @xmath65 . the reason comes from the fact that the real part of dielectric susceptibility ( [ epsilon_z_omega ] ) for the low frequency signal wave has a zero at this point . it means that the nonlinear current on the right hand side of the ginzburg equation has a very sharp peak near @xmath20 . a typical plot of right hand side is given in figure [ rhs_to ] . this issue is discussed in more detail in appendix [ app : current_nl_to ] . if we consider a high frequency pumping wave we can use the plane wave approximation @xmath111 the low frequency signal wave can be written @xmath112 for the stokes wave , whose frequency is higher than the plasma frequency , one can use the following approximate ginzburg equation @xmath113 where @xmath114 is calculated from the @xmath115 of the nonlinear current given in ( [ curl_j_nl ] ) . to solve , we use the method of variation of constants.we find @xmath116 one can say that @xmath117 is the amplitude of the stokes wave propagating to the vehicle and @xmath118 is the amplitude of the anti - stokes wave propagating from the vehicle . the main contribution to @xmath119 arises from the vicinity of @xmath20 , where @xmath120 is almost singular @xmath121 after some simple but tedious calculations ( see appendix [ app : current_nl_to ] ) one finds @xmath122 where @xmath123 is the pumping incident angle . details of these calculations are given in appendix [ app : current_nl_to ] . the angular dependence of @xmath124 , which we call @xmath125 , can be calculated numerically by solving the homogeneous ginzburg equation . in fig . [ rhosinphi ] , we plot the product @xmath126 against @xmath16 . at the optimal value @xmath127 , @xmath128 , @xmath129 using the expression @xmath130 , one gets @xmath131 for the optimal values of incidence angles ( @xmath132 , @xmath133 ) , the given plasma parameters and @xmath73 , one gets the following maximum value of the efficiency coefficient @xmath134 this is consistent with what we obtain by direct numerical simulation . equation ( [ ginzburg_equation ] ) can be rewritten in the following form @xmath135 it is not too surprising that that the dominant contribution to the rhs of ( [ ginzburg_rewritten_to ] ) is the first term and arises from the neighborhood of @xmath20 . again , just as in the `` to the vehicle '' case , the resonant layer acts as a transmitting antenna which will beam the message contained on the stokes wave to a distant receiver at frequency @xmath28 . in fig . [ rhs_from ] we verify that indeed the dominant contribution comes from the first term on the rhs of ( [ ginzburg_rewritten_to ] ) and from the neighborhood of @xmath20 . hence we can get simple equation for a very good approximation to the approximate particular solution of ( [ ginzburg_from ] ) , namely @xmath136 the general solution is the following @xmath137 where @xmath138 and @xmath139 are solutions of the homogeneous part of equation ( [ ginzburg_from ] ) , and @xmath138 is bounded as @xmath140 , @xmath139 is unbounded ( exponentially ) at the vehicle . thus @xmath141 . see appendix [ app : analytics_homogenious ] for a discussion of solutions to the homogeneous ginzburg equation . using the boundary condition on the edge of the plasma ( @xmath62 ) @xmath142 where @xmath143 is the @xmath36-component of wavevector of the outgoing low frequency signal wave , and @xmath144 , one finds @xmath145 finally , for the magnetic field at @xmath62 we find @xmath146 the function @xmath147 oscillates with @xmath36 with wavenumber @xmath148 . the lower the wavenumber the more will be the contribution in the integral . this gives us a very simple optimal strategy for the choice of pump and stokes wave directions . we should radiate both the stokes and pumping waves in the desired direction of the signal wave propagation . in this case we also have an exact compatibility with the boundary conditions at @xmath62 . if we consider the expression for @xmath147 given in appendix [ app : current_nl ] , we can see that in the case @xmath149 the first term ( [ current_nl_y ] ) is the major one in the vicinity of resonant layer . the resonant layer works like radiating antenna . using the simplified nonlinear current expression and considering the pumping and stokes waves as plane waves one finds @xmath150 where @xmath151 . using the solutions of the approximate homogeneous equations ( [ ginzburgbessel ] ) , we can estimate @xmath152 . thus for @xmath153 , one finds @xmath154 for the power density , we have @xmath155 this result is quite clear from physical point of view . the larger @xmath16 is , the longer is the distance over which the signal wave is generated in the plasma . in our simulations , @xmath156 and in this case we can not use the simplified expression given above . instead we find , @xmath157 here we introduced the coefficient @xmath158 the value of which we obtain from our numerics . finally , we find @xmath159 the subscripts refer to the frequencies of the onboard pump waves . again , we find the magnitude and angular dependence to be consistent with our numerical results . the equation we solve numerically in all cases is the ginzburg equation ( [ ginzburg ] ) including all terms on its right hand side . the boundary conditions are given at @xmath160 , in the vacuum beyond the plasma edge and at @xmath22 , the vehicle . to solve this equation we use a `` sweep''-method described in detail in appendix [ app : numerical_method ] . the method was invented simultaneously in several places for work on classified topics in the middle of the last century . in the soviet union , it was introduced by a group of l.d . landau ( information from i.m . khalatnikov ) ( the first publication @xcite appeared several years later due to obvious reasons ) and was developed to its modern form in @xcite . as the first step in the `` to the vehicle '' case we have to find the profile of the incoming magnetic field in the plasma . we used an incident angle @xmath161 . it will be shown later that this angle is an optimal value but it is good for an initial evaluation of the possibility of communication . we consider the incoming signal as a monochromatic plane wave of a given frequency @xmath162 and amplitude @xmath163 . the current is equal to zero . in this case , the boundary conditions are @xmath164 the resulting profile of the magnetic field is shown in fig . [ h_low_freq_to ] . the profile of @xmath165 is shown in fig . [ ez_low_freq_to ] . at the next stage , we consider an incident low frequency magnetic field profile as a source of distortion of the plasma density profile and take into account currents due to the presence of a pump wave . the pumping wave angle @xmath166 . our goal is to calculate the scattered field @xmath167 with frequency @xmath27 . in this case , the boundary conditions are @xmath168 the profiles of the magnetic fields @xmath167 for two different pumping frequencies are shown in figures [ h_high_freq_to_12ghz ] and [ h_high_freq_to_18ghz ] . we note that the resonant layer @xmath20 acts as if it were a source . in the `` from the vehicle '' case we calculate the magnetic field of the low frequency wave generated by plane pump and stokes waves . following the optimal strategy in this case , described in the analytic part of the paper , we take all angles equal to each other @xmath169 . in this case , the boundary conditions are @xmath170 here @xmath49 is the magnetic field of the signal wave with frequency @xmath59 . the boundary condition at @xmath171 , gives us the worst of all cases by definition . the low frequency magnetic fields for two different pumping frequencies are shown in figs . [ h_low_freq_from_12ghz ] and [ h_low_freq_from_18ghz ] . we tested the robustness of the code by allowing for both finite and zero conductivity of the vehicle surface in the `` to the vehicle case '' . during the simulation in the `` from the vehicle '' case we also redid the simulation with the derivative of the magnetic field at the vehicle equal to zero . in all the cases , the influences of the differing boundary conditions were negligible . in the `` to the vehicle '' case , it is convenient to introduce the function @xmath172 as the ratio @xmath109 of the scattered field flux to the incoming signal flux and express it as a function of pump flux @xmath110 measured in watts per square meter . we found @xmath173 @xmath174 these results are in a good agreement with the analytic estimation ( [ estimate_to_number ] ) . any difference is due to the fact that the pumping frequency is not sufficiently high to neglect the plasma frequency . the reason we used these frequencies and not much higher ones was that they are available on standard microwave equipment and devices . in the `` from the vehicle '' case , we calculate the ratio @xmath175 of the output signal flux to the product of the pump and stokes fluxes and express it as a function of the optimal angle . we found @xmath176 @xmath177 in order to investigate the dependence of the result on the angles @xmath178 , we calculated @xmath179 for various different choices . the results are shown in figs . [ theta_phi_to_12ghz]-[theta_phi_from_18ghz ] . as one can see , in the `` to the vehicle '' case we have a very good agreement between the analytically estimated angular dependence ( [ estimation_to ] ) and the numerical results . namely , we have a maximum at pumping angles close to @xmath132 and the efficiency coefficient @xmath172 goes to zero at the vicinity of @xmath180 in a agreement with the @xmath181 dependence . so we can formulate a simple rule : in order to get the best possible performance , send the pump wave in a direction perpendicular to the plasma edge surface . in the `` from the vehicle '' case , the situation is even simpler . as it was shown in section [ analytics_from ] the power conversion is optimal if we radiate both the pump and stokes waves in the direction of the desired signal wave propagation . the estimated angular dependence ( [ estimation_from ] ) can be fitted with good accuracy to the numerical results using only one tuning coefficient @xmath182 . it is shown that this coefficient weakly depends on the pumping frequency . let us now discuss the practical usage of this approach for receiving at and transmitting from the vehicle . for the `` to the vehicle '' case we consider the problem of receiving even gps signals . let us estimate the resulting attenuation coefficient . given a pump waveguide aperture of @xmath183 and a pump power of @xmath184 , this gives @xmath185 . one can use the pulse regime . in this case , even for pulses @xmath186 long , every pulse still contains more than @xmath187 periods of the low frequency signal and we can get much higher power flux @xmath188 . it gives us the attenuation coefficients @xmath189 @xmath190 the usual level of a gps signal at the earth surface is about @xmath191 ( 1 decibel per milliwatt is equal to @xmath192 ) . indoors , one must use high sensitivity gps receivers . many general purpose chipsets have been available for several years . presently , the market offers sensitivities @xmath193 ( for example @xcite ) . using the definition of @xmath194 one can see , that it is possible to receive a signal with an attenuation about @xmath195 . also it is possible to use a much bigger antenna on the vehicle than in the case of a handheld device . in this case , it is even possible to receive a signal using the continuous rather than pulsed regime for a klystron pump . so even at the angles far from optimal , one can receive gps signals . further , we used maximum value of the plasma thickness . if the plasma sheath is thinner , the angular dependence is broader . some characteristics of klystron amplifiers available on the open market are given in table [ table_1 ] @xcite . in the `` from the vehicle '' case , because of sensitive land based receivers , all we need is to have a reasonable signal . let us estimate an incoming power on some land based antenna . first of all , for any real antenna we have to take into account the decrease of a signal due to diffraction broadening . if the diameter of the land - based antenna ( figure [ antenna ] ) is @xmath196 , the diameter of the signal flux after some long distance @xmath197 will be @xmath198 it means that if we have power flux at an antenna @xmath199 , the power flux at the edge of the plasma after a distance @xmath197 will be @xmath200 for example , for an antenna of the diameter equal to 5 m , after 100 km @xmath201 now one can calculate the sensitivity of the receiver needed . let us suppose that the signal beam outgoing from the vehicle has diameter @xmath202 , signal frequency @xmath203 and corresponding wave length @xmath204 , the land based antenna has a diameter @xmath205 and is situated at a distance @xmath206 . using the previous results for diffraction , the pumping klystrons powers from the table above and the expression @xmath207 , one can get for the power on the land based receiver @xmath208 we now list for two different frequencies , the corresponding powers in watts at the receiving antenna . @xmath209 the gps receiver mentioned above has a sensitivity about @xmath210 . even with such a modest size of the antenna and ordinary klystrons one can receive the signal at almost any angle . as a final remark one can conclude that proposed method for communication with and from the supersonic vehicle is realistic even using standard devices available on the open market . in a future work , there are several additional points we would like to consider further . first , it might be worthwhile to discuss the effect of the plasma density profile at @xmath211 on the wave interaction . one may expect that the transition will be much narrower than @xmath212 , because of the shock formation . but shock waves take place not in the plasma but in air . the air and plasma densities are not connected directly . the plasma density is defined by the level of ionization which is governed by a density distribution in accordance with the saha equation . in a real plasma sheath the characteristic temperature is much less then the ionization potential , and a typical level of ionization is low ( @xmath213-@xmath214 ) . under this conditions the plasma density depends on the temperature dramatically . the temperature jumps inside the shock then grows gradually toward the vehicle . still , just after the shock it is too low to provide a strong ionization , and the most essential increase in ionization takes place far behind the shock . for this reason we can neglect the jump of plasma density inside the shock wave and treat it as smooth . in this case an approximation by a linear function seems to be reasonable . let us note that there is no blackout if the plasma sheath is as thick as @xmath212 and the plasma density is as low as @xmath215 . in this case , the incident wave can reach the vehicle due to the skin - effect . anyway , our numerical code can be used for an arbitrary density profile . another question is the following : will the shock and the flow behind it suffer from hydrodynamic instabilities ? actually , hydrodynamic - type instabilities as well as hydrodynamic turbulence are slow processes in our time - scale . we can treat the plasma sheath as `` frozen '' . the distortion of the density profile , although frozen , can slightly change the results . it could look quite surprising that collision frequency @xmath42 drops out from the final results after integration over the spatial variable . this is a common mathematical trick , working perfectly as long as @xmath216 . in our case we have @xmath217 . in fact , the shape of the resonant layer can be distorted by nonlinear effects . they are essential if the ongoing signal is powerful enough . at very small @xmath42 , the resonant peak would become very narrow . could then the electron thermal motion start affecting the structure of the resonance ? this question is very important . a resonant layer can not be thinner than the debye length . in our case @xmath218 while the thickness of the resonant layer is @xmath219 . thus the influence of the temperature can be neglected . in the case of much thinner resonant layer this phenomenon should be taken into account and considered separately . anyway , such influence could lead only to a radiation of langmuir waves from the resonant area . but this effect is nothing but additional dissipation , dropping out from the final formulae . when the signals come from the external source , what would be the role of the @xmath220-polarized component ? one can expect that , for some orientations of the vehicle , the s - component will be dominant , and the resonance will disappear.this could somewhat reduce the efficiency . in our idealized model ( the vehicle is an infinite wall ) there is a difference between @xmath220 and @xmath221 polarizations . for a real vehicle such a difference could be important . this is a subject for future study . as one can see there are still a lot of open questions . the physics of the problem is very rich . our present work may be best described as a `` proof of concept '' research from which we can make some orders of magnitude estimates and gain some insights . the results can be made more accurate by including some of the physical effects discussed above . this work was supported by afosr contract number fh95500410090 . korotkevich was supported by rfbr grant 06 - 01 - 00665-a , the programme `` nonlinear dynamics and solitons '' from the ras presidium and `` leading scientific schools of russia '' grant . by neglecting @xmath43 , we obtain the linear ginzburg equation . it takes an especially simple form if @xmath47 , @xmath222 and @xmath223 . in this case @xmath224 and equation ( [ ginzburg ] ) is @xmath225 here @xmath226 and @xmath227 is another length . in our case @xmath228 , @xmath229 and @xmath230 . one can introduce the dimensionless variable @xmath231 . then equation ( [ ginzburglinear ] ) simplifies to , @xmath232 here @xmath233 is a dimensionless constant . equation ( [ ginzburglinear ] ) has two linearly independent solutions @xmath234 , @xmath235 . we assume @xmath236 the wronskian of these solutions is proportional to @xmath237 . we can put @xmath238 it means that @xmath239 equation ( [ ginzburglineardimensionless ] ) can not be solved in terms of any known special functions . in the `` outer '' area @xmath240 it reduces to the form @xmath241 one can check that equation ( [ ginzburgairy ] ) can be solved in terms of the airy functions @xmath242 and @xmath243 . namely , @xmath244 from ( [ wronskian ] ) one gets @xmath245 in the `` inner '' area @xmath246 , the equation ( [ ginzburglineardimensionless ] ) is reduced to the form @xmath247 equation ( [ ginzburgbessel ] ) can be solved in terms of bessel functions @xcite . two linearly independant solutions of equation ( [ ginzburgbessel ] ) @xmath248 , @xmath249 behave in neighborhood of @xmath250 as follows @xmath251 both solutions , which are some linear combinations of @xmath248 , @xmath249 are bounded . thus the magnetic field has no singularity at @xmath20 . consider the ginzburg equation for a wave @xmath252 and calculate the right hand side of ( [ ginzburg ] ) in terms if the fields @xmath253 . in the `` to the vehicle '' case , @xmath254 will represent the pump wave , @xmath255 the signal wave and @xmath256 the stokes wave . in the `` from the vehicle '' case , @xmath256 will be the signal and @xmath254 and @xmath255 the pump and signal carrying stokes waves respectively . in all cases @xmath257 . we find @xmath258_x = \nonumber\\ = -\frac{e^3 n_0'(z ) k_3}{2m^2 \omega_3 \left(1+{\mathrm i}\nu/\omega_3\right ) } \left ( \frac{1}{\varepsilon_1^ * \left(1 - { \mathrm i}\nu/\omega_1\right)\omega_1 ^ 2 \varepsilon_2 \left(1 + { \mathrm i}\nu/\omega_2\right)\omega_2 ^ 2 } \frac{\partial h_1^*}{\partial z}\frac{\partial h_2}{\partial z } + \right.\\ \left.+ \frac{k_1 k_2}{\varepsilon_1^ * \left(1 - { \mathrm i}\nu/\omega_1\right)\omega_1 ^ 2 \varepsilon_2 \left(1 + { \mathrm i}\nu/\omega_2\right)\omega_2 ^ 2}h_1^ * h_2\right ) + \nonumber\\ + \frac{e^3}{m^2}\left(\frac{k_2k_1 ^ 2 } { \left(1 - { \mathrm i}\nu/\omega_1\right)\omega_1 ^ 3 \varepsilon_2 \left(1 + { \mathrm i}\nu/\omega_2\right)\omega_2 ^ 2}\frac{\partial}{\partial z}\left(\frac{n_0(z)}{\varepsilon_1^*}\right)\right.+\nonumber\\ + \left.\frac{k_2 ^ 2k_1 } { \left(1 - { \mathrm i}\nu/\omega_1\right)\omega_2 ^ 3 \varepsilon_1^ * \left(1 + { \mathrm i}\nu/\omega_2\right)\omega_1 ^ 2}\frac{\partial}{\partial z}\left(\frac{n_0(z)}{\varepsilon_2}\right)\right)h_1^*h_2\\ + \frac{e^3}{m^2}\left(\frac{1}{\omega_2 ^ 2 \varepsilon_2\left(1+{\mathrm i}\nu/\omega_2\right)}\frac{\partial h_2}{\partial z } \frac{k_1}{\omega_1 ^ 3\left(1-{\mathrm i}\nu/\omega_1\right)}\frac{\partial } { \partial z}\left(h_1^*\frac{\partial } { \partial z}\left(\frac{n_0(z)}{\varepsilon_1^*}\right)\right)+\right.\nonumber\\ + \left.\frac{1}{\omega_1 ^ 2\varepsilon_1^*\left(1-{\mathrm i}\nu/\omega_1\right)}\frac{\partial h_1^*}{\partial z } \frac{k_2}{\omega_2 ^ 3\left(1+{\mathrm i}\nu/\omega_2\right)}\frac{\partial } { \partial z}\left(h_2\frac{\partial } { \partial z}\left(\frac{n_0(z)}{\varepsilon_2}\right)\right)\right)-\\ -\frac{\mu_0 e^3}{m^2}\left(\frac{k_2}{\omega_2 ^ 3\left(1+{\mathrm i}\nu/\omega_2 \right)}\frac{\partial}{\partial z}\left(\frac{n_0(z)}{\varepsilon_2}\right ) + \frac{k_1}{\omega_1 ^ 3\left(1-{\mathrm i}\nu/\omega_1\right)}\frac{\partial}{\partial z}\left(\frac{n_0(z)}{\varepsilon_1^*}\right)\right)h_1^ * h_2.\end{aligned}\ ] ] using formulae ( [ current_nl_general ] ) , @xmath259 in the `` to the vehicle '' case , the main contribution comes from the terms containing poles @xmath260_x \simeq\\ \simeq \frac{e^3}{m^2}\frac{k_p k^2}{\left(1 - { \mathrm i}\nu/\omega_p\right)\omega^3 \varepsilon_p \left(1 + { \mathrm i}\nu/\omega_p\right)\omega_p^2}\frac{\partial}{\partial z}\left(\frac{n_0(z)}{\varepsilon^*}\right)h^ * h_p+\nonumber\\ + \frac{e^3}{m^2}\frac{1}{\omega_p^2 \varepsilon_p\left(1+{\mathrm i}\nu/\omega_p\right)}\frac{\partial h_p}{\partial z } \frac{k}{\omega^3\left(1-{\mathrm i}\nu/\omega\right)}\frac{\partial } { \partial z}\left(h^*\frac{\partial } { \partial z}\left(\frac{n_0(z)}{\varepsilon^*}\right)\right)-\nonumber\\ -\frac{\mu_0 e^3}{m^2}\frac{k}{\omega^3\left(1-{\mathrm i}\nu/\omega\right)}\frac{\partial}{\partial z}\left(\frac{n_0(z)}{\varepsilon^*}\right)h^ * h_p.\nonumber\\\end{aligned}\ ] ] assume that the high frequency pumping wave remains undisturbed . then @xmath261 we find @xmath262_x { \mathrm e}^{-{\mathrm i}\kappa_p z}\ ] ] and @xmath263 after several integrations by parts in the second term of @xmath264_x$ ] , taking into account @xmath265 , one finds @xmath266 for most pumping angles , and using the fact that @xmath267 , one can substitute @xmath268 and consider incidence angles of pumping and stokes wave to be close in absolute value . following fig . [ plasma_response_fig ] the pumping incidence angle is @xmath123 and low frequency signal incidence angle is @xmath16 . @xmath269 using integration by parts once more one can get @xmath270 calculating this integral by residues and taking into account @xmath271 we finally get @xmath272 here we briefly present formulae for the `` sweep '' method in a very general way following the approach given in @xcite . consider the ode @xmath273 in the region @xmath274 with boundary conditions @xmath275 we use for ( [ num_equation ] ) a second order finite difference scheme on a grid of @xmath276-nodes ( @xmath277 ) with constant step @xmath278 @xmath279 this equation is only valid for inner nodes of the grid . the boundary conditions take the form @xmath280 we can rewrite equation ( [ num_equation_diff ] ) as @xmath281 in the same way for ( [ num_boundary_diff ] ) , one finds @xmath282 the result is a tridiagonal matrix @xmath283 equation for @xmath284 . the solution of the linear system of equations with tridiagonal matrix is well described in numerous sources ( for instance @xcite ) . it can be shown that one can find a solution in the following form @xmath285 from the left boundary , we have from ( [ num_matrix_left ] ) , that @xmath286 in this case @xmath287 next we derive a recurrence relation for @xmath288 and @xmath289 . after substituting ( [ sweep_relation ] ) in ( [ num_matrix_inner ] ) , we find @xmath290 then , comparing with ( [ sweep_relation ] ) , we see that @xmath291 using the initial values ( [ initial_for_pq ] ) and the recurring relations ( [ direct_sweep_recurring ] ) , one can get all @xmath292 coefficients up to @xmath293 ( `` direct sweep '' from left to right ) . than we use second ( right ) boundary condition ( n - th equation ) @xmath294 immediately one finds @xmath295 finally , performing a recurrent `` backward sweep '' ( from right to left ) , using the already known @xmath292 , `` sweep''-relations ( [ sweep_relation ] ) and the initial condition ( [ y_initial ] ) , we get values for all @xmath296 . l.d . landau , n.n . meiman and i.m . khalatnikov , proceedings of the iiird ussr mathematics meeting , moscow , * 3 * , 92 ( 1956 ) . also in a book `` l.d . collection of papers '' , vol . 2 , p.393 , `` science '' , moscow ( 1969 ) . w.h . press , b.p . flannery , s.a . teukolsky and w.t . vetterling , `` numerical recipes : the art of scientific computing '' , cambridge university press + ( isbn-13 : 9780 - 5214 - 3108 - 8 | isbn-10 : 0 - 5214 - 3108 - 5 ) http://nr.com

theoretical models based on direct numerical simulations have unveiled a new way toward a better comprehension of the rich and complex phenomenology associated with astrophysical plasmas . finite difference codes such as zeus @xcite or nirvana+ @xcite inaugurated this novel era and have been used by an increasingly large fraction of researchers nowadays . however , as reported in @xcite , the lack of upwinding techniques and conservation properties have progressively moved scientist s attention toward more accurate and robust methods . in this respect , the successful employment of the so - called high resolution shock - capturing ( hrsc ) schemes have revealed a mighty tool to investigate fluid dynamics in non - linear regimes . some of the motivations behind their growing popularity is the ability to model strongly supersonic flows while retaining robustness and stability . the unfamiliar reader is refereed to the books of @xcite , @xcite and references therein for a more comprehensive overview . implementation of hrsc algorithms is based on a conservative formulation of the fluid equations and proper upwinding requires an exact or approximate solution @xcite to the riemann problem , i.e. , the decay of a discontinuity separating two constant states . this approach dates back to the pioneering work of @xcite , and it has now become the lead line in developing high resolution codes examples of which include flash ( * ? ? ? * for reactive hydrodynamics ) , the special relativistic hydro code genesis @xcite , the versatile advection code ( vac , * ? ? ? * ) or the new nirvana @xcite . most hrsc algorithms are based on the so - called reconstruct - solve - average ( rsa ) strategy . in this approach volume averages are first reconstructed using piecewise monotonic interpolants inside each computational cell . a riemann problem is then solved at each interface with discontinuous left and right states and the solution is finally evolved in time . it turns out that this sequence of steps is quite general for many systems of conservation laws and , therefore , it provides a general framework under which we have developed a multi - physics , multi - algorithm high resolution code , pluto . the code is particularly suitable for time - dependent , explicit computations of highly supersonic flows in presence of strong discontinuities and it can be employed under different regimes , i.e. , classical , relativistic unmagnetized and magnetized flows . the code is structured in a modular way allowing a new module to be easily incorporated . this flexibility turns out to be quite important , since many aspects of computational fluid dynamics are still in rapid development . besides , the advantage offered by a multi - physics , multi - solver code is also to supply the user with the most appropriate algorithms and , at the same time , provide inter - scheme comparison for a better verification of the simulation results . pluto is entirely written in the c programming language and can run on either single processor or parallel machines , the latter functionality being implemented through the message passing interface ( mpi ) library . the code has already been successfully employed in the context of stellar and extragalactic jets @xcite , radiative shocks @xcite , accretion disks @xcite , magneto - rotational instability , relativistic kelvin - helmholtz instability and so forth . the paper is structured as follows : in [ sec : code_design ] we give a description of the code design ; in [ sec : phys ] we introduce the physics modules available in the code ; in [ sec : source ] we give a short overview on source terms and non - hyperbolicity and in [ sec : verification ] the code is validated against several standard benchmarks . [ fig : flow ] pluto is designed to integrate a general system of conservation laws which we write as @xmath0 here @xmath1 denotes a state vector of conservative quantities , @xmath2 is a rank-2 tensor , the rows of which are the fluxes of each component of @xmath1 and @xmath3 defines the source terms . additional source terms may implicitly arise when taking the divergence of @xmath2 in a curvilinear system of coordinates . an arbitrary number of advection equations may be added to the original conservation law ( [ eq : cl ] ) . although the components of @xmath1 are the primary variables being updated , fluxes are more conveniently computed using a different set of physical quantities which we take as the primitive vector @xmath4 . this choice is supported , moreover , by the fact that interpolation on primitive variables enforces physical constraints such as pressure positivity and sub - luminal speeds in the case of relativistic flows . numerical integration of the conservation law ( [ eq : cl ] ) is achieved through shock - capturing schemes using the finite volume ( fv ) formalism where _ volume averages _ evolve in time . generally speaking , these methods are comprised of three steps : an interpolation routine followed by the solution of riemann problems at zone edges and a final evolution stage . in pluto , this sequence of steps provides the necessary infra - structure of the code , see the schematic diagram in fig . [ fig : flow ] . at the higher level , the original system of equations is integrated by following this general sequence of steps , regardless any knowledge of the physics involved . the explicit form of @xmath1 , @xmath4 , @xmath2 and @xmath3 , on the other hand , depends on the particular physical module selected . thus , at the lower level , a physical module collects the set of algorithms required to compute the terms involved in the discretization of the right hand side of eq . ( [ eq : cl ] ) . this set should provide one or more riemann solver(s ) , mapper routines for the conversion between primitive and conservative variables , a flux routine giving the components of @xmath2 in each direction , a source term function ( if any ) and a routine to compute the maximum and minimum characteristic speeds of the jacobian matrix . of course , additional features may be easily added by exploiting the independent modularity . from the user s perspective , a particular configuration can be defined through a friendly interface entirely written in the python scripting language . the interface allows the user to specify all problem - dependent attributes and algorithms , such as number of dimensions , geometry , physics module , reconstruction method , time stepping integration and so forth . lccc @xmath5 & @xmath6 & @xmath7 & @xmath7 + @xmath8 & @xmath9 & @xmath10 & @xmath11 + @xmath12 & @xmath13 & @xmath13 & @xmath10 + @xmath14 & @xmath15 & @xmath16 & @xmath17 + @xmath18 & @xmath19 & @xmath20 & @xmath21 + @xmath22 & @xmath23 & @xmath23 & @xmath24 + @xmath25 & @xmath26 & @xmath27 & @xmath28 + @xmath29 & @xmath26 & @xmath26 & @xmath30 + @xmath31 & @xmath26 & @xmath26 & @xmath32 pluto employs logically rectangular grids in a generic system of orthogonal curvilinear coordinates @xmath33 ( see table [ tab : geom ] ) . let @xmath34 , @xmath35 and @xmath36 be the number of points in the three directions . the lower and upper coordinate bounds of zone @xmath37 are @xmath38 and @xmath39 , respectively , where @xmath40 , @xmath41 , @xmath42 . the zone widths are then simply given by @xmath43 , @xmath44 , @xmath45 . the mesh spacing can be either uniform or ( geometrical or logarithmic ) stretched . parallelization is achieved through domain decomposition : the global domain is divided in sub - domains with an equal number of points and each of the sub - domains is assigned to a processor . by default the sub - domains are created as maximally cubic , however the user , at run time can specify a different strategy for the creation of the sub - domains , imposing that a given direction can not be subdivided . in the code , parallelization is handled through the mpi library . grid adaptivity techniques are also being provided . a one dimensional adaptive mesh refinement ( amr ) module based on the @xcite method is distributed with the code and extension to multiple spatial dimensions is currently being developed using the chombo library available at http://seesar.lbl.gov / anag / chombo/. however , since the main goal of this paper is to focus on the code modularity and its performance , amr implementation will be described in future works . in order to avoid formulas with cluttered notations we omit integer - valued subscripts when referring to three - dimensional quantities . thus @xmath46 becomes simply @xmath4 . we introduce the standard two - point difference one - dimensional operator @xmath47 where @xmath48 is a given direction , @xmath49 and @xmath50 are , respectively , the cell s right ( @xmath51 ) and left ( @xmath52 ) interface areas and cell volume in that direction ( see table [ tab : geom ] ) . here @xmath53 when @xmath48 , respectively . scalar components of the vectors appearing in eq . ( [ eq : rhs ] ) will be denoted with a pair of square brackets ; e.g. @xmath54}$ ] is the component of @xmath55 contributing to the @xmath56 equation . furthermore , since several of the building block algorithms are one - dimensional , we will also omit the superscript @xmath57 when unnecessary . the numerical flux functions @xmath58 in eq . ( [ eq : rhs ] ) follow the solution of 1-d riemann problems at cell interfaces . since more than one riemann solver may be available in each physics module , we set , without loss of generality , @xmath59 where @xmath60 is the riemann solver being used and @xmath61 , @xmath62 are suitable left and right states at the zone edges perpendicular to the @xmath6 direction . interpolation routines are designed to reconstruct a piecewise polynomial approximations @xmath63 to @xmath4 inside each cell starting from its cell averages : @xmath64 where @xmath65 ( @xmath66 ) at @xmath67 ( @xmath68 ) . here @xmath69 is an interpolation routine designated to provide left and right edge interpolated values inside each cell , that is , @xmath70 and @xmath71 , where the l " and r " subscripts refer to the sides of the interface . reconstruction methods have to satisfy monotonicity constraints in order to avoid spurious oscillations in proximity of discontinuities and steep gradients , see the books from @xcite , @xcite and references therein . for second - order linear interpolants , for instance , one has @xmath72 with the slopes @xmath73 computed following a limiting procedure applied to primitive or characteristic variables . in the latter case one has @xmath74 where @xmath75 and @xmath76 are , respectively , the right and left eigenvectors of the primitive form of the equations , @xmath77 are forward ( @xmath51 ) and backward ( @xmath52 ) derivatives and @xmath78 labels the @xmath78-th characteristic field . different slope limiters ( @xmath79 ) are characterized by distinct steepening properties and can be independently assigned to each primitive variable or characteristic field . at the time of this writing , pluto allows to choose between either flat ( @xmath80 order in space ) , linear ( @xmath81 order ) , third - order convex eno @xcite , parabolic reconstructions ( as in * ? ? ? * ) , or the 5th order finite difference weno scheme of @xcite ( only for the hydrodynamics equations ) . as already anticipated , computation of the numerical flux function @xmath82 at a zone edge ( @xmath83 ) requires the solution @xmath84 , for @xmath85 , to the initial value problem @xmath86 complemented with the one - dimensional evolutionary equation for @xmath1 . since this section deals with interface quantities , we shall omit , in this section only , the @xmath51 subscripts . the exact solution to the riemann problem ( [ eq : riemann ] ) involves the decay of a set of non - linear waves and can be a rather cumbersome task to achieve . with the exception of few simple cases , existing riemann solvers routinely involved in upwind schemes are based on different levels of approximation . the lax - friedrichs rusanov flux is robust , but also the most diffusive solver and is available for all modules . it computes the fluxes according to : @xmath87 \,,\ ] ] where @xmath88 is the projection of the tensor flux on the @xmath89 unit vector , @xmath90 is kronecker - delta symbol and @xmath91 is the largest local signal velocity . a somewhat different approximation comes from the hll - family solvers : @xmath92 where the solution to the riemann problem is approximated by @xmath93 waves with @xmath94 and @xmath95 and separated by @xmath96 states . the @xmath97 are computed from a suitable parameterization " of the rankine - hugoniot jump conditions across each wave : @xmath98 with @xmath99 . the hll and hllc solvers can be consistently derived following this recipe with @xmath100 and @xmath101 , respectively , see @xcite , @xcite for the hydro equations , @xcite for the mhd equations , @xcite for the relativistic equations . similarly , we have also implemented the multi - state ( @xmath102 ) hlld solver of @xcite in the mhd module . the hll approach has some attractive features in that it guarantees positivity of pressure and , in the case of relativistic flows , it preserves the condition @xmath103 . linearized riemann solvers are more accurate since the averaging process inherent to eq . ( [ eq : tvdlf ] ) or ( [ eq : hll ] ) is removed and all jumps are computed by linearization around some average state . the roe solver computes the numerical fluxes according to : @xmath104 \,,\ ] ] where the rows ( columns ) of @xmath105 ( @xmath106 ) are the left ( right ) eigenvectors of the jacobian @xmath107 . the advection upstream splitting method ( ausm , originally proposed * ? ? ? * ) provides an alternative and efficient upwinding strategy by splitting the flux into convective and pressure terms , respectively associated with linear and nonlinear fields : @xmath108 where @xmath109 is a suitably defined convective velocity and the @xmath110 flux is governed by the acoustic speed . the original ausm scheme has been substantially improved in the work of @xcite and @xcite . the most accurate approach consists in directly solving the whole set of rankine - hugoniot jump conditions to find @xmath111 , from which the flux can be computed , @xmath112 however , this approach is computationally expensive since it generally involves the solution of highly nonlinear equations . different physics modules come with different sets of riemann solvers , and additional methods of solution may be easily accommodated . we warn that more accurate riemann solvers ( such linearized or exact schemes ) may introduce insufficient numerical dissipation for certain flow configurations . sporadically , this could lead to a number of numerical pathologies such as the carbuncle phenomena , odd - even coupling or lack of positivity - preserving properties . see the work by @xcite for a comprehensive review . time marching algorithms provide a general ( i.e. physics - independent ) framework for the discretization of the left hand side of eq ( [ eq : cl ] ) . for example , in the simplest case of forward euler discretization one has @xmath113 where @xmath114 is the right hand side operator ( [ eq : rhs ] ) or a sum of them for dimensionally split or unsplit schemes , respectively . the time step @xmath115 is limited by the courant - friedrichs - lewy ( cfl , * ? ? ? * ) condition : @xmath116 with @xmath117 and @xmath118 being , respectively , the smallest cell length and largest signal velocity in the @xmath57 direction . @xmath119 identifies the courant number , restricted by the conditions given in the last column of table [ tab : nriem ] . pluto provides a number of time - marching schemes for the explicit numerical integration of the conservation law ( [ eq : cl ] ) . we discriminate between 1 ) fully discrete , zone - edge extrapolated and 2 ) semi - discrete methods . evolution in more than one dimension may be achieved by either operator splitting @xcite or fully multidimensional integration . lcccc mh ( s ) & 1 & 2 & 3 & 1 + mh ( u ) & - & 4 & 12 & 1 + chtr ( s ) & 1 & 2 & 3 & 1 + chtr ( u ) & - & 4 & 12 & 1 + rk2 ( s ) & 2 & 4 & 6 & @xmath26 + rk2 ( u ) & 3 & 6 & 9 & @xmath120 + rk3 ( s ) & 2 & 4 & 6 & @xmath26 + rk3 ( u ) & 3 & 6 & 9 & @xmath120 zone - edge extrapolated methods achieve second order temporal accuracy by midpoint in time quadrature : @xmath121 and thus are based on a single - step . for directional splitting methods one has @xmath122 , whereas @xmath123 in the case of a dimensionally unsplit method . the input states for the riemann solver are computed by taylor expansion of the primitive variables at half time step , @xmath124 where @xmath125 is the matrix of the quasi - linear form of the equations and @xmath126 is the identity matrix . an alternative form which does not require the primitive form of the equations can be written in terms of conservative variables . both variants yield the well - known muscl - hancock scheme ( mh henceforth , * ? ? ? * ; * ? ? ? when eigenvalues and eigenvectors are available , upwind limiting may be used to select only those characteristics which contribute to the effective left and right states . this approach is employed , for instance , in the original ppm advection scheme of @xcite and we will refer to as characteristic tracing ( chtr henceforth ) . these methods require boundary conditions to be assigned at the beginning of the time step only . the advantage offered by dimensionally split single - step algorithms is the lower computational cost ( only one riemann solver per cell per direction is required , see tab [ tab : nriem ] ) and storage requirement . an alternative , dimensionally unsplit version of these schemes , however , adopts the corner transport upwind ( ctu ) method of @xcite , @xcite and is more expensive . in this case , an extra correction term is needed in eq . ( [ eq : states ] ) ; in two dimensions , for example , the input states for the riemann problem are modified to @xmath127 with @xmath128 being the right - hand side operator corresponding to the transverse direction . as in the dimensionally split case , the cfl stability condition is not affected by the dimensionality of the problem ( i.e. @xmath129 , see also tab [ tab : nriem ] ) . contrary to its dimensionally split version , on the other hand , this method is computationally more expensive since 4 instead of 2 ( in 2d ) and 12 instead of 3 ( in 3d ) riemann problems must be solved at each interface , see table [ tab : nriem ] . semi - discrete methods are based on the classical method of lines , where the spatial discretization is considered separately from the temporal evolution which is left continuous in time . in this framework ( [ eq : cl ] ) is discretized as a regular ode . pluto implements the @xmath130 and @xmath131 order total variation diminishing ( tvd ) runge - kutta schemes of @xcite . the second - order method ( rk2 ) advances the system of conservation law ( [ eq : cl ] ) as @xmath132 @xmath133 \,,\ ] ] with @xmath134 or @xmath135 in the case of a dimensionally split or unsplit method , respectively . the third - order runge - kutta method ( rk3 ) may also be used , at the cost of an additional step : @xmath136 @xmath137 \,,\ ] ] @xmath138\,.\ ] ] for this class of methods , input states for the riemann solver are given by the output of the interpolation routine , see [ sec : recon ] . besides , boundary conditions must be assigned before each step . a total of two and three riemann problems per cell per direction must be solved by the rk2 and rk3 marching scheme , respectively . furthermore , fully unsplit runge - kutta integrators require a stronger time step limitation , see table [ tab : nriem ] . pluto is distributed with four independent physics modules for the explicit numerical integration of the fluid equations under different regimes and conditions . the hydrodynamics ( hd ) , magnetohydrodynamics ( mhd ) , relativistic ( rhd ) , and relativistic mhd ( rmhd ) modules solve , respectively , the euler equations of gas dynamics , the ideal / resistive mhd equations the energy - momentum conservation laws of a special relativistic perfect gas , and the equations for a ( special ) relativistic magnetized ideal plasma . in what follows , @xmath139 , @xmath140 , and @xmath141 will denote , respectively , the proper density , thermal pressure and total energy density . vector fields such as @xmath142 , @xmath143 , @xmath144 , define the momentum density , velocity and magnetic field . finally , @xmath145 will be used to define the ratio of specific heats for the ideal equation of state . this module implements the equations of classical fluid dynamics with an ideal equation of state . the conservative variables @xmath1 and the flux tensor are : @xmath146 where @xmath147 is the momentum density and @xmath126 is the unit , @xmath148 tensor . the total energy density @xmath141 is related to the gas pressure @xmath140 by the ideal gas closure : @xmath149 the set of primitive variables @xmath150 is given by density , velocity @xmath151 and thermal pressure @xmath140 . this module comes with a set of several riemann solvers , including the nonlinear riemann solver based on the two - shock approximations @xcite , the roe solver @xcite , the ausm+ scheme @xcite , the hll @xcite , hllc @xcite solvers and the lax - friedrichs solver @xcite . the hd module contains an implementation of the fast eulerian transport algorithm for differentially rotating disk ( fargo , * ? ? ? * ) on polar grids . the fargo scheme allows much larger time steps than the standard integration where the courant condition is traditionally limited by the fast orbital motion at the inner boundary . the mhd module deals with the equations of classical ideal or resistive magnetohydrodynamics ( mhd ) . in the ideal case , @xmath1 and @xmath152 may be written as @xmath153 with @xmath147 and @xmath154 being the total ( thermal + magnetic ) pressure , respectively . the additional constraint @xmath155 complements the magnetic field evolution ( see [ sec : divb ] ) . resistivity is introduced by adding appropriate parabolic terms to the induction and energy equations , see [ sec : diffusion ] . available equations of state implemented are the ideal gas law , @xmath156 and the isothermal equation of state @xmath157 where @xmath158 is the ( constant ) isothermal speed of sound . the set of primitive variables is the same one used for the hd module , with the addition of magnetic fields . the user can choose among the following available riemann solvers : the roe solver of @xcite , the hll @xcite , hllc @xcite , hlld @xcite and the lax - friedrichs solvers . the solution to the mhd equations must fulfill the solenoidal constraint , @xmath155 , at all times . unfortunately it is well known that numerical scheme do not naturally preserve this condition unless special discretization techniques are used . among the variety of monopole control strategies proposed in literature ( for a review see * ? ? ? * ) , we have implemented 1 ) the eight wave formulation @xcite and 2 ) the constrained transport ( ct henceforth ) of @xcite , and @xcite . the ct framework has been incorporated into the unsplit ctu integrator following the recent work by @xcite . a similar approach is used by @xcite . in the first strategy , the magnetic field has a cell centered representation and an additional source term is added to the mhd equation . the discretization of the source term is different depending on the riemann solver ( following * ? ? ? * ) , a feature which we found to greatly improve robustness . in the ct formulation , on the other hand , the induction equation is integrated directly using stoke s theorem and the magnetic field has a staggered collocation . this module deals with the motion of an ideal fluid in special relativity . the equations are given by particle number conservation and energy - momentum conservation @xcite . conservative variables and tensor flux are : @xmath159 where @xmath160 is the laboratory density and @xmath161 is the lorentz factor . for convenience , velocities are normalized to the the speed of light . the relation between conserved and primitive variables @xmath162 is more involved than its classical counterpart . specifically we have @xmath163 where the specific enthalpy @xmath164 depends on the equation of state . the inverse map requires the solution of a nonlinear equation . @xcite address this topic for different choices of @xmath165 . two equations of state are available for this module : the constant-@xmath145 ideal gas law @xmath166 with @xmath167 and the equation of state ( tm henceforth ) already introduced in @xcite : @xmath168 which satisfies taub s inequality @xcite and approximates ( within @xmath169 ) the single - specie relativistic perfect gas ( for a thorough discussion , see * ? ? ? * ; * ? ? ? * and references therein ) . the two shock nonlinear riemann solver described in @xcite has been incorporated into the set of available riemann solvers , together with the recently developed hllc algorithm by @xcite . the hll and lax - friedrichs riemann solvers have also been coded . the motion of an ideal relativistic magnetized fluid is described by conservation of mass , energy - momentum , and by maxwell s equations , see , for example , @xcite . pluto implements the equation of special relativistic magnetohydrodynamics where the vector of conservative variables and respective fluxes can be cast as @xmath170 where @xmath171 , @xmath172 , @xmath173 and @xmath174 is the total ( thermal + magnetic ) pressure . the components of @xmath1 are related to the primitive variables @xmath175 through @xmath176 the inverse map is highly nonlinear and different methods of solution have been proposed when a constant @xmath145 law is adopted ( for a recent review , see * ? ? ? * ) . in pluto , we follow the approach described in @xcite which was recently generalized to include the tm equation of state described in [ sec : rhd ] . the interested reader should also refer to the work by @xcite . the rmhd module may be used with the lax - friedrichs , hll or the recently developed hllc riemann solver , see @xcite . since the induction equation takes the same form as in the classical case , the divergence of the magnetic field can be kept under control by any of the methods already introduced in the mhd module , see [ sec : divb ] . in addition to the homogeneous terms previously described , a number of additional features may be added in the code . local source terms are functions of the variables themselves but not of their derivatives and are included either during the advection step or via operator splitting . examples include the centripetal and coriolis terms implicitly arising when working in polar or spherical coordinates , external forces ( e.g. gravity ) or optically thin radiative losses . non - ideal effects such as viscosity , resistivity and thermal conduction , on the other hand , introduce parabolic corrections to the equations and involve the solution of diffusion equations . the divergence of a rank-2 tensor in curvilinear coordinates breaks down the homogeneous properties of eq . ( [ eq : cl ] ) . this loss of conservation comes from the additional source terms inevitably introduced by those unit vectors that do not have fixed orientation in space ( see , for instance , the appendix in * ? ? ? * ) . in the case of symmetric or anti - symmetric tensors , however , some of the source terms can be eliminated by properly re - arranging the derivatives . this ensures conservation of the angular momenta rather than the linear ones . for all physics modules previously introduced , for example , the components of the flux tensor in the momentum equation form a @xmath148 rank 2 symmetric tensor @xmath177 , where @xmath140 is the isotropic pressure term . however , since @xmath178 , the differencing terms involving pressure are separately treated as gradient components and never contribute to the source terms . the symmetric character of @xmath179 leads to further simplifications in polar coordinates , since the @xmath10-component of the divergence of @xmath179 may be written as @xmath180 this leads to the so - called angular momentum - conserving form " where the radial contribution to @xmath56 is computed by the new operator @xmath181 } = -\frac{1}{r\delta{{\mathcal v}}^r}\left(r^2_+ f^r_{+,[m_\phi ] } - r^2_-f^r_{-,[m_\phi]}\right)\,,\ ] ] thus leaving the radial component of momentum as the only non - homogeneous equation , with @xmath182 } = m_{\phi\phi}/r$ ] . similar considerations lead to the angular momentum conserving form in spherical geometry . geometrical source terms are integrated explicitly during the advection step by adding their contribution to the right hand side operator ( [ eq : rhs ] ) . in the semi - discrete approach , this is straightforward . for edge - extrapolated methods , this requires 1 ) augmenting ( [ eq : states ] ) with the cell - center source term for a half step and then 2 ) averaging the resulting time centered left and right edge values to compute the @xmath183 for the final conservative step . for many astrophysical applications , radiative processes may become important during the evolution . this is particularly true when the cooling timescale becomes comparable to , or smaller than the typical timescale for the dynamical evolution . optically thin radiative losses are treated as local source terms depending on the temperature , density and an arbitrary number of ions of different elements ( e.g. h , he , c , n , o , and so on ) . the ionization fractions are advected with the fluid and are subject to collisional ionization , recombination and charge exchange processes . currently , we have implemented a cooling module @xcite similar to the one coded by @xcite , extending the temperature range of applicability ( @xmath184 k ) by using an increased number of ion species . this module employs @xmath185 ions and is valid for densities up to @xmath186 @xmath187 . operator splitting is employed to evolve the chemical network in time . a @xmath188 order fully implicit scheme is adopted in regions of rapid variations , whereas a second order explicit integrator is used otherwise . a comprehensive description is outside the goal of this work and will be available in future works . parabolic terms introduce second - order spatial derivatives and their treatment requires the solution of diffusion equations . typical examples include electric resistivity , @xmath189 & = & 0 \,,\end{aligned}\ ] ] where @xmath190 is the current and @xmath191 is the resistivity ; or thermal conduction , @xmath192 where @xmath193 is the energy advection flux and @xmath194 is the thermal conductivity coefficient . these terms may be included into the original conservation law with the further time step limitation @xmath195 , where @xmath196 is the advective time step ( eq . [ eq : cfl ] ) and @xmath197\,.\ ] ] is the diffusion step . here @xmath198 characterizes the physical process , e.g. , @xmath191 or @xmath194 . for advection - dominated problems ( i.e. @xmath199 ) , eq . ( [ eq : diff_timestep ] ) does not pose severe restrictions and diffusion terms can be treated explicitly . we achieve this by adding centered in space , second - order finite difference approximations to the right hand side operator @xmath200 , see eq . ( [ eq : rhs ] ) . however , for diffusion - dominated problems and/or increasingly high resolution simulations , @xmath201 will eventually drop below the typical advection scale . in such situations the explicit integration can be considerably accelerated using the super time stepping ( sts ) method @xcite . in sts , diffusion terms are included via operator splitting , and the solution vector is evolved over a super time step @xmath202 consisting of @xmath93 smaller sub - steps , the stability of which is closely related to the properties of chebyshev polynomials . it can be proved @xcite that @xmath203 which makes sts almost @xmath93 times faster than the standard explicit scheme . thus , if @xmath202 is taken to be the advective time step , sts will require ( approximately ) @xmath204 iterations rather than @xmath205 , typical of a normal sub - cycling explicit time stepping . this approach offers dramatic ease of implementation over implicit schemes . resistivity has been extensively tested through the direct comparison with analytical solutions of linear diffusion problems of magnetic field . our results confirm the efficiency and accuracy of the sts approach already highlighted by previous investigators . this has encouraged further experiments towards nonlinear problems ( e.g. spitzer - like conductivity ) , where preliminary results suggest that the sts methodology may be succesfully applied . this issue will be presented in future works . lcccccccc double mach reflection([sec : hd_dmr ] ) & hd & ( a ) & cartesian & 2 & parabolic & hllc & rk3 ( s ) & 0.8 + & & ( b ) & & & ceno3 & hllc & rk3 ( s ) & 0.8 + & & ( c ) & & & weno5 & roe f - s & rk3 ( s ) & 0.8 + & & ( d ) & & & parabolic & two - shock & chtr ( s ) & 0.8 + & & ( e ) & & & linear & roe & mh ( u ) & 0.8 + & & ( f ) & & & linear & hll & mh ( u ) & 0.8 + under - expanded jet ( [ sec : hd_jet ] ) & hd & ( a ) & cylindrical & 2 & linear & ausm+ & mh ( u ) & 0.4 + & & ( b ) & & & ceno3 & hll & rk3 ( s ) & 0.4 + rotated shock tube ( [ sec : mhd_st2.5 ] ) & mhd & ( a ) & cartesian & 2.5 & linear & roe & rk2 ( u ) & 0.4 + & & ( b ) & & & linear & hlld & rk2 ( u ) & 0.4 + & & ( c ) & & & linear & hll & rk2 ( u ) & 0.4 + fast rotor ( [ sec : mhd_rotor ] ) & mhd & ( a ) & cartesian & 2 & parabolic & hllc & chtr ( u ) & 0.6 + & & ( b ) & polar & & linear & roe & rk3 ( u ) & 0.4 + magnetized torus ( [ sec : mhd_torus ] ) & mhd & ( a ) & spherical & 2.5 & linear & hlld & rk2 ( u ) & 0.4 + & & ( b ) & spherical & & parabolic & hll & rk3 ( u ) & 0.4 + & & ( c ) & cylindrical & & linear & hlld & rk2 ( u ) & 0.4 + & & ( d ) & cylindrical & & parabolic & hll & rk3 ( u ) & 0.4 + spherical shock tube ( [ sec : mhd_torus ] ) & rhd & ( a ) & spherical & 1 & parabolic & two - shock & chtr(s ) & 0.8 + & & ( b ) & spherical & 1 & linear & hll & mh ( s ) & 0.8 + & & ( c ) & cartesian & 3 & linear & hllc & chtr(s ) & 0.8 + magnetized blast wave ( [ sec : rmhd_blast3d ] ) & rmhd & ( a ) & cartesian & 3 & linear & hll & rk2 ( u ) & 0.2 + & & ( b ) & cylindrical & 2 & linear & hll & mh ( u ) & 0.4 pluto has been validated against several benchmarks typically adopted for other numerical schemes . several algorithms for relativistic flows presented in @xcite and @xcite are now part of the code and the related verification problems will not be repeated here . in what follows , we propose additional tests aimed to check the code performances on problems of different nature , geometry and dimension . computational details for each test are given in table [ tab : tests ] . the initial condition for this test problem @xcite consists in a planar shock front making an angle @xmath206 with a reflecting wall , taken to be the @xmath6 axis : @xmath207 where @xmath208 is the shock coordinate . the ideal equation of state with @xmath209 is used throughout the calculation . the computational domain is the rectangle @xmath210\times[0,1]$ ] covered with a uniform grid with @xmath211 . outflow boundary conditions apply at @xmath212 and reflective conditions are imposed at the bottom boundary @xmath213 for @xmath214 . fluid variables are kept constant to their initial values at @xmath213 for @xmath215 and at the leftmost boundary @xmath216 . at the top boundary @xmath217 the exact motion of the shock is prescribed . we carry out integration until @xmath218 using the six different combinations of algorithms described in table [ tab : tests ] . panels in the left column of fig . [ fig : hd_dmr ] show , from top to bottom , the results obtained with high order ( @xmath219 ) interpolants : piecewise parabolic reconstruction ( a ) , @xmath220 order central eno ( b ) and the weno scheme of @xcite ( c ) . panels on the right show the original ppm scheme of @xcite ( case d , top ) , the @xmath188 order unsplit muscl - hancock with a roe solver ( case e , middle ) and with the hll solver ( case f , bottom ) . after the reflection , a complicated flow structure develops with two curved reflected shocks propagating at directions almost orthogonal to each other and a tangential discontinuity separating them . at the wall , a pressure gradient sets up a denser fluid jet propagating along the wall . kelvin - helmholtz instability patterns may be identified with the rolls " developing at the slip line . this feature is visible only for some of the selected schemes . indeed , the use of high - order interpolants , such as ppm ( a , d ) or weno(c ) , and the ability of the riemann solvers to resolve contact and shear waves result in smaller numerical viscosity when compared to a second - order slope - limited reconstruction and/or a more approximate riemann solver . in this respect , case a ) shows the smallest amount of dissipation , whereas the hll solver ( f , bottom right ) has the largest numerical diffusion . in terms of cpu time , we found that case a ) to f ) followed the ratios @xmath221 , that is , with the @xmath222 order weno schemes being by far the most expensive ( more than a factor of @xmath223 compared to the simple hll ) and the original ppm ( split ) scheme being the fastest . in doing the comparison , however , one has to bear in mind that the amount of computing time depends , in the first place , by the number of riemann problems solved in each zone at each time step , as shown in table [ tab : nriem ] . this number is 6 for cases a ) , b ) and c ) , but @xmath224 for case e ) and f ) ( which are dimensionally unsplit ) and only @xmath225 for the original dimensionally split ppm scheme , case d ) . thus one can safely conclude , comparing case e ) and f ) , that the roe solver is by a factor @xmath226 slower than hll or , by comparing a ) , b ) and c ) , that the ceno and ppm are considerably faster than weno , although the combinations of algorithms in a ) better resolves small scale structures . finally , we note that computations carried with more accurate schemes ( with the exception of the @xmath222 order weno ) exhibit a slight tendency to form a spurious kinked " mach stem on the @xmath6 axis . this feature is engendered by a numerical flaw @xcite caused by insufficient numerical dissipation and it becomes particularly dramatic when the resolution is further increased . code validation , in addition to the typical tests described above , can be carried out against laboratory experiments of under - expanded free jets as well . this kind of experiments consists of injecting , through a converging nozzle , a gas at stagnation pressure @xmath227 into a chamber kept at lower pressure @xmath228 . the shock structure that forms is typically axially symmetric and consists of a quasi - stationary mach - disk , located at a distance @xmath229 from the nozzle , and of a barrel and reflected shocks @xcite . after series of experiments with different gases , an empirical expression was found @xcite that relates the mach - disk location @xmath229 with the pressure ratio @xcite : @xmath230 with @xmath231 the effective sonic nozzle radius . furthermore , @xcite obtained for the mach number @xmath232 on the jet axis the following empirical expression : @xmath233^{-\frac{\gamma-1}{4 } } \left ( \frac{\gamma+1}{\gamma-1 } \right)^{\frac{\gamma+1}{4 } } \left ( \frac{z}{2r^ * } \right)^{\gamma-1 } \,.\ ] ] we have carried out 2d numerical simulations in cylindrical coordinates on a uniform grid @xmath234 $ ] , @xmath235 $ ] with @xmath236 zones per beam radius . the grid has been further extended in both directions ( up to @xmath237 and @xmath238 ) by adding a second patch of geometrically stretched zones ( @xmath239 and @xmath240 zones in @xmath7 and @xmath13 , respectively ) in order to avoid spurious reflections at the boundaries . free outflow is set at the outer boundaries and reflective conditions are imposed on the axis @xmath241 and for @xmath242 at @xmath243 . considering the actual nozzle as the injection zone at @xmath243 , @xmath244 , we have obtained the values of the pressure @xmath245 and density @xmath246 by employing the isentropic laws for a perfect gas @xcite for a converging nozzle with stagnation pressure @xmath227 and density @xmath247 : @xmath248 we choose @xmath249 and @xmath250 ( typical of an argon jet in helium at ambient temperature ) and accordingly choose a sonic injection velocity at the the nozzle ( @xmath251 ) . densities , velocities and lengths are normalized to the ambient @xmath139 , sound speed and beam radius , respectively . fig . [ fig : hdjet_a ] shows the density maps for the two selected numerical schemes ( see table [ tab : tests ] ) at @xmath252 . the results obtained with the ausm+ riemann solver ( scheme a ) disclose a somewhat greater amount of small scale structure than the hll integration ( scheme b ) . this is not surprising because of the former s ability to properly capture contact and shear waves . on the contrary , the latter greatly simplifies the wave pattern of the riemann fan at the cost of introducing extra numerical dissipation . this is compensated , to some level , by the choice of the @xmath220 order interpolant ( ceno3 ) which , nevertheless , is better combined with an equally accurate time marching scheme ( rk3 ) . the overall balance , therefore , favors the second order scheme since only @xmath224 riemann problems must be solved instead of the @xmath253 required by rk3 , see table [ tab : nriem ] . this conclusion is supported not only by the decreased numerical viscosity inherent to scheme a ) , but also by the reduced computational cost , which sees scheme b ) being @xmath254 times slower than scheme a ) . in both cases we see that the mach number profiles plotted in fig [ fig : hdjet_b ] exhibit an excellent agreement with the empirical relation ( [ eq : knuth ] ) . it should be emphasized that the saturated position of the mach disk oscillates around the experimental value , eq . ( [ eq : young ] ) , by a few percents . the oscillations are caused by disturbances originated at the intersection between the mach disk and the barrel shock surrounding the jet . this test sets up a riemann problem between left and right states given , respectively , by @xmath255 where parallel ( @xmath256 ) and perpendicular ( @xmath257 ) components refer to the direction of structure propagation . the ratio of specific heats is @xmath258 . the line @xmath259 ( corresponding to a rotation angle of @xmath260 with respect to the y - axis ) is chosen as the surface of discontinuity separating the two constant states and the computational domain @xmath261\times[0,1/n_{\mathrm{x}}]$ ] is discretized with @xmath262 zones , as in @xcite . this setup yields a mesh spacing ratio @xmath263 thus ensuring that the initial magnetic field is divergence - free ( ct is used to advect magnetic fields ) . at the boundaries in the @xmath9 direction we impose translational invariance by applying shifted boundary condition , i.e. , for any pair of indexes @xmath264 spanning the ghost zones , we set @xmath265 at the lower @xmath9-boundary and @xmath266 at the upper @xmath9-boundary . the structure involves three - dimensional vector fields and the outcoming wave pattern is bounded by two fast shocks ( located @xmath267 and @xmath268 ) enclosing two rotational discontinuities ( visible only in the magnetic field ) , two slow shocks and one contact wave in at @xmath269 . [ fig : mhd_st2.5 ] shows a cut along the @xmath6 axis at @xmath270 together with the high - resolution reference solution ( solid line ) computed from a one - dimensional integration at @xmath271 . the second order unsplit runge - kutta method with a cfl number of @xmath272 together with @xmath188 order linear interpolation on primitive variabels and the riemann solver of roe were used for integration . computations were repeated at different resolutions starting from @xmath273 ( lowest ) up to @xmath274 ( highest ) , by doubling the number of zones each time . the performances and accuracies of the roe , hlld and hll riemann solvers were compared in terms of errors ( in @xmath275 norm ) and cpu time , as shown in fig [ fig : mhd_st2.5_error ] . as indicated , hlld and roe yields comparable errors ( @xmath276 and @xmath277 at the highest resolution ) , whereas the hll solver results in errors which are @xmath278 larger . in terms of cpu time , the hlld solver is considerably faster than the roe scheme and this is reflected in the right panel of fig [ fig : mhd_st2.5_error ] . indeed , for a given accuracy , the computing time offered by hlld significantly improves over the roe solver and it is still slightly better than the cheaper hll scheme . in this sense , hlld offers the best trade off between accuracy and efficiency . a caveat : the previous considerations hold , strictly speaking , for the particular combinations of algorithms applied to the problem in question and should not be trusted as general statements . one may find , for example , that if interpolation is carried on characteristic variables ( instead of primitive ) the overall trend which favours hll over roe ( in terms of cpu time ) is reversed . still , our results illustrate how the choice of an algorithm or another may considerably bear upon important issues of accuracy and efficiency . a rapidly spinning ( @xmath279 ) cylinder with higher density is embedded in a static background medium with uniform pressure ( @xmath280 ) , threaded by a constant magnetic field ( @xmath281 ) . hydrodynamical variables smoothly change their values between the disk and the external environment using a taper function : @xmath282 the disk has radius @xmath283 and the taper function @xmath284 terminates at @xmath285 . the initial radial velocity is zero everywhere and the adiabatic index is @xmath209 . this problem has been considered by several authors , see for example @xcite , and references therein . because of the geometrical setting , we have setup the problem in both cartesian and polar coordinates . on the cartesian grid , the domain is the square @xmath286 ^ 2 $ ] with outflow boundary conditions applied everywhere . on the polar grid , we choose @xmath287 , @xmath288 with periodic boundary conditions in @xmath10 and zero - gradient at the outer radial boundary . at the inner boundary we set @xmath289 . for each geometry , we adopt the combination of algorithms listed in table [ tab : tests ] and compare two different strategies to control the solenoidal constraint @xmath155 , i.e. , the ct and powell s methods . [ fig : mhd_rotor ] shows density and magnetic pressure contours computed at @xmath290 on the cartesian and polar grids at a resolution of @xmath291 zones . as the disk rotates , strong torsional alfvn waves form and propagate outward carrying angular momentum from the disk to the ambient . our results fully agree with those previously recovered by the above - mentioned authors and computations obtained with the ct and powell s scheme shows excellent agreement . despite the higher complexity of the ct scheme where both staggered and zone - centered magnetic field are evolved in time , the computational cost turned out to be comparable ( ct / powell @xmath292 ) for each system of coordinates . moreover , the solutions computed in the two different geometries show nearly identical patterns , although polar coordinates ( even with their intrinsically higher numerical viscosity at higher radii , due to the diverging nature of the grid ) better fit the geometrical configuration of the problem . however , computations carried out on the polar grid were @xmath293 times slower than the cartesian one , because of the more severe time step limitation at the inner boundary , where the grid narrows . these considerations can considerably affect the choice of geometry , specially in long - lived simulations . we now discuss an application of the code to a problem of astrophysical interest , along the lines of @xcite . the problem involves a magnetized , constant angular momentum ( @xmath294 ) torus in a ( properly normalized ) pseudo - newtonian gravitational potential , @xmath295 . the torus has an equilibrium configuration described by the integral relation @xmath296 here @xmath297 is determined by the radial location of the inner edge of the torus ( @xmath298 in our case ) and @xmath299 is the keplerian specific angular momentum at the pressure maximum ( @xmath300 ) where the orbital period is @xmath301 . density and pressure are tied by the polytropic relation @xmath302 , with @xmath258 . a poloidal magnetic field is initialized inside the torus using the @xmath10 component of the potential vector @xmath303 , normalized by the condition @xmath304 . we solve the problem in spherical as well as cylindrical coordinates . the spherical grid used for the problem is @xmath305 and @xmath306 , with @xmath307 uniform zones covering the region @xmath308 and @xmath309 . the remaining regions were covered by a geometrically stretched grid with @xmath310 zones . outflow boundaries are imposed at the innermost and outermost radii , whereas axisymmetry holds at @xmath311 and @xmath312 . in cylindrical coordinates , the computational box is @xmath313 and @xmath314 , with a uniform grid @xmath315 in the region @xmath308 and @xmath316 . axisymmetry boundary conditions are imposed at the axis , while all the rest are set to outflow . in addition , each case is further investigated by adopting 1 ) the hlld solver with second order slope limited reconstruction -case a and c in table [ tab : tests]- and 2 ) the hll solver combined with piecewise parabolic reconstruction ( case b and d in table [ tab : tests ] ) . second and third order runge kutta schemes are used for integration , respectively . figure [ fig : mhd_torus ] shows the density logarithm at @xmath317 for the four cases respectively . initially , the strong shear generates a toroidal magnetic field component leading to a pressure - driven expansion of the torus and accretion takes place at @xmath318 . the growth of the poloidal field develops into a highly turbulent regime accounted for by the magneto rotational instability ( mri , * ? ? ? * ) at @xmath319 . the effect is a net transport and redistribution of angular momentum . even though linear reconstruction presumably introduces more diffusion than parabolic interpolation , the hlld solver , being more accurate , results in computations with a higher level of fine structure ( compared to hll ) around the tori during the turbulent phase . besides , by normalizing the cpu time to case c ) , we found the ratio a : b and c : d to be @xmath320 ( in cylindrical coordinates ) and @xmath321 ( in spherical coordinates ) , respectively . indeed , since higher than @xmath188 order interpolants are computationally more expensive and are used in conjunction with @xmath220 order time accurate schemes , the hll integration comes at an extra computational cost . this suggests that the use of a more accurate riemann solver can balance the lower order of reconstruction in space and result , at the same time , in more efficient computations . moreover , note that the equatorial symmetry breaks down more rapidly when parabolic reconstruction is employed , but it is retained to a higher level for the @xmath188 order case this is likely due to the larger number of operations introduced by the ppm reconstruction step , inevitably leading to faster - growing round off errors . finally we note that computations in spherical geometry suffer from the major drawbacks of introducing excessive numerical viscosity at the outer radii and to severely limiting the time step at the inner boundary . both aspects are indeed confirmed in our numerical tests . on the other hand , a cylindrical system of coordinate does not experience a similar behavior and thus a higher turbulence level is still observed at larger radii . an initial spherical discontinuity separates a high pressure region ( @xmath322 for @xmath323 ) from a uniform medium with @xmath324 ( for @xmath325 ) . the density is set to unity everywhere and the gas is initially static . computations proceed by investigating the influence of two different equations of state ( eos ) , the constant @xmath145 law with @xmath326 ( eq . [ eq : rhd_ideal ] ) and the tm equation ( [ eq : rhd_tm ] ) . moreover , we compare the performances of the ppm scheme ( with the two - shock riemann solver ) with the simpler mh ( with the hll solver ) which avoids characteristic decomposition of the equations ( schemes a and b in table [ tab : tests ] ) . spherical symmetry is assumed in the one - dimensional calculation . the ensuing wave pattern ( solid line in fig [ fig : rhd_sph ] ) is comprised of a left rarefaction wave , a contact discontinuity in the middle and a right going shock . the overall morphology is significantly affected by the choice of the eos . indeed , when the more realistic eos is adopted , waves propagate at smaller velocities and this is particularly evident at the head and the tail points of the left - going rarefaction wave . this results in a stronger adiabatic expansion for the constant @xmath145 gas and in a denser shell between the shock and the trailing contact wave for the tm eos . [ fig : rhd_sph_error ] reports the @xmath275 norm errors ( for density ) and the normalized cpu time as functions of the resolution @xmath327 ( @xmath328 ) for both schemes and for the selected equations of state . errors are computed against a reference solution of @xmath329 zones . the cheaper scheme b ) is certainly faster , although almost twice the resolution should be employed to achieve comparable accuracy with scheme a ) . the opposite trend is found for the ppm scheme which is slower by almost a factor of @xmath225 when compared to mh / hll . employment of the tm eos results in an additional computational cost of about @xmath330 independently of the riemann solver . however , the simple results obtained here suggest that a more realistic eos can significantly impact the solution and thus justify the direct use of eq . ( [ eq : rhd_tm ] ) instead of ( [ eq : rhd_ideal ] ) . further reading on this topic , together with its extension to rmhd , may be found in @xcite . spherically symmetric problems can serve as useful verification benchmarks to tests the code performance on three dimensional cartesian grids . for this purpose , we use the second - order chtr scheme with the recently developed hllc riemann solver @xcite ( scheme c ) and adopt a resolution of @xmath331 computational zones . by exploiting the symmetry , computations are carried out on one octant ( @xmath261 ^ 3 $ ] ) only by adopting reflecting boundary conditions . results are over - plotted using symbols in fig . [ fig : rhd_sph ] . small propagation structures such as the thin density shell pose serious computational challenges . indeed we see that the maximum density peaks achieve @xmath332 and @xmath333 of the reference values for @xmath326 and the tm equation of state , respectively . the smooth rarefaction wave is correctly captured at constant entropy with very small overshoots at the back . results at higher resolution ( not shown here ) show increasingly better agreement . scaling performances have been tested for this problem on different parallel platforms , all yielding satisfactory scaling properties . [ fig : scaling ] shows , for instance , the normalized execution time on the ibm sp cluster 1600 p5 - 575 with @xmath334 processors ( @xmath335 ) for the same problem at a resolution of @xmath331 , indicating an almost ideal scaling ( @xmath336 ) . a spherical region with density @xmath337 and thermal pressure @xmath280 is embedded in a static uniform medium with @xmath338 and @xmath339 . the sphere is centered around the origin and has radius @xmath340 . a linear smoothing function is applied for @xmath341 . the whole region is threaded by a constant vertical magnetic field in the @xmath13 direction , @xmath342 . the ideal equation of state with specific heat ratio @xmath343 is used . a similar setup in two dimensional planar geometry was earlier considered by @xcite . our computational grid employs @xmath344 zones on the box @xmath345 ^ 3 $ ] . by exploiting the symmetric nature of the problem we reduce the computations to one octant ( @xmath346 ) at half the resolution this is easily achieved by symmetrizing all quantities with respect to the coordinate planes and flipping the signs of @xmath347 at @xmath213 , @xmath348 at @xmath216 , @xmath349 at @xmath243 . figs . [ fig : rmhd_blast3d ] shows the magnetic field distribution , thermal pressure and lorentz factor for the two configurations at @xmath350 . as a reference solution , computations on a cylindrical axisymmetric grid with @xmath351 $ ] zones are also shown . the expanding region is delimited by a fast forward shock propagating radially at almost the speed of light . because of the strong magnetic confinement a jet - like structure develops in the @xmath13 direction with a maximum lorentz factor of @xmath352 . this problem is particularly challenging because of the very small plasma @xmath353 and instructive in checking the robustness of the code and the algorithm response to different kinds of degeneracies . scaling tests have been conducted in a way similar to [ sec : rhd_sph ] and corresponding performance are shown in fig . [ fig : scaling ] . in the context of astrophysical problems , where high mach number compressible flow dynamics plays a crucial role , a sizable number of numerical codes is now available to the community of scientists . most of them are based on the reconstruct - solve - average strategy typical of the high - resolution shock - capturing godunov - type codes . the reasons for choosing one code or another can be the most diverse , however , one can attempt to list a number of characteristics that a code should hold to be appealing for the skilled user , but not necessarily expert in numerics : 1 . _ a modular structure : the possibility to easily code and combine different algorithms to treat different physics ; _ + pluto offers a number of features which can be combined together . besides four physical modules ( e.g. newtonian / relativistic hydrodynamics with or without magnetic fields ) , gravity , radiative cooling , resistivity , multidimensional geometries , equations of state may be included where required ; 2 . _ provide the user with a number of numerical schemes tested against the most severe benchmarks ; _ + several algorithms ( e.g. riemann solvers , interpolations , choice of boundary conditions and so on ) have been coded in pluto and their choice is dictated by the problem at hand and/or by efficiency and robustness issues . 3 . _ portability ; _ + we have successfully ported pluto to a number of different unix - based systems among which ibm sp5/sp4 , sgi irix , linux beowulf clusters , power macintosh . in addition the code can run in either serial , single - processor or parallel machines . _ grid adaptivity : the ability to resolve flow features with different spatial and temporal scales on the same computational domain ; _ + the code provides a one - dimensional amr integrator and multidimensional extensions are being developed by taking advantage of the chombo library ( http://seesar.lbl.gov/anag/chombo/ ) . this feature will be distributed with future versions of the code . _ last but not least , user friendliness ; _ + a simple user - interface based on the python scripting language is available to setup a physical problem in a quick and self - explanatory way . the interface is conceived to minimize the coding efforts left to the user . we would like to thank s. ritta for his useful contributions to the under - expanded jet test and e. beltritti for extensively testing the code performances on several platforms . the present work was partially supported by the european community marie curie actions - human resource and mobility within the jetset network under contract mrtn - ct-2004 005592 . numerical calculations were partly performed in cineca ( bologna , italy ) thanks to the support by inaf . a. m. would like to thank r. rosner , t. linde and t. plewa and all the people at the flash center at the university of chicago for their helpful suggestions and discussions during the early development of the code .

the x - ray sky as seen by x - ray satellites over the past 40 years , shows a substantially different picture than for example the optical band . while the visual night sky is dominated by main sequence stars , galactic binary systems and super nova remnants form the brightest objects in the x - rays . common to both regimes is the dominance of active galactic nuclei ( agn ) toward lower fluxes . in the x - ray range itself , one observes a slightly different population of agn at soft and at hard x - rays . below 5 kev the x - ray sky is dominated by agn of the seyfert 1 type , above the absorbed seyfert 2 objects appear to become more numerous . these type 2 agn are also the main contributors to the cosmic x - ray background above 5 kev @xcite , although only @xmath5 of the xrb above @xmath6 can be resolved @xcite . beside the rather persistent seyfert galaxies , also blazars are detectable in the hard x - rays . because of the fact that we look into a highly relativistic jet in the case of those sources , the blazars exhibit strong variability on all time - scales and are especially variable in the x - ray and high - energy gamma - ray domain . for an overview over the various extragalactic x - ray surveys see brandt & hasinger ( 2005 ) , and for early _ integral _ results see krivonos et al . ( 2005 ) . the hard x - ray energy range is not currently accessible to x - ray telescopes using grazing incidence mirror systems . instead detectors without spatial resolution like the pds @xcite on _ bepposax _ ( boella et al . 1997 ) and osse @xcite on _ cgro _ ( gehrels et al . 1993 ) have been applied . a synopsis of these previous results is as follows : the 2 10 kev seyfert 1 continua are approximated by a @xmath7 powerlaw form @xcite . a flattening above @xmath8 has been noted , and is commonly attributed to compton reflection @xcite . there is a great deal of additional detail in this spectral domain - `` warm '' absorption , multiple - velocity component outflows , and relativistic line broadening - which are beyond the scope of this paper . the seyfert 2 objects are more poorly categorized here , but the general belief is that they are intrinsically equivalent to the seyfert 1s , but viewed through much larger absorption columns [ see e.g. vron - cetty & vron ( 2000 ) for a recent review ] . above 20 kev the empirical picture is less clear . the @xmath9 continuum shape of both seyfert types is consistent with a thermal comptonization spectral form , although in all but a few cases the data are not sufficiently constraining to rule out a pure powerlaw form . nonetheless , the non - thermal scenarios with pure powerlaw continua extending to @xmath10 energies reported in the pre-_cgro _ era are no longer widely believed , and are likely a result of background systematics . however , a detailed picture of the comptonizing plasma - its spatial , dynamical , and thermo - dynamic structure - is not known . among the critical determinations which _ integral _ or future hard x - ray instruments will hopefully provide are the plasma temperature and optical depth ( or compton `` y '' parameter ) for a large sample of objects . the other major class of gamma - ray emitting agn - the blazars ( fsrqs and bl lac objects ) are even more poorly constrained in the _ integral _ spectral domain . the detection of @xmath11 of theses objects with the egret instrument on _ cgro _ in the 1990s has established the presence of @xmath12 relativistic plasma at small angles to the observers line of sight . the so called `` tev '' blazars are similar objects but with the peak power shifted to even greater energies , probably due to an even smaller viewing angle ( see for example , fossati et al . 1999 , and for earlier _ integral _ results pian et al . 2005 ) . the most common interpretation of the broad - band continuum emission is the synchrotron self compton ( ssc ) model , although the external compton models ( ec ) are still also plausible . broad - band temporal coverage can in principle discriminate between these possibilities , although each of these components can separately contribute to the hard - x - ray band under extreme circumstances . the _ integral _ bandpass covers a critical region which can in principle establish the amplitude and shape of the high - end synchrotron component or or low - end ssc component . critical to each of these issues is the need to obtain improved continuum measurements over the hard x - ray to soft gamma - ray range for as large a sample of objects as possible . _ integral _ @xcite , since its launch in october 2002 , offers unprecedented @xmath13 collecting area and state of the art detector electronics and background rejection capabilities . thus it offers hope of substantial gains in our knowledge of the agn phenomenon . we were thus motivated at this juncture some three years into the mission , to formulated and present the collective look at the data so far obtained and assess the potential for progress in the _ integral _ era . as the satellite spent most of its observing time up to now concentrating on the galactic plane , some low - latitude agn have been detected . observations at higher galactic latitude have also led to studies of individual sources : 3c 273 @xcite , ngc 4388 @xcite , ( sazonov et al . 2004 ) , the s5 0716 + 714 field @xcite , @xcite , ngc 4151 ( beckmann et al . 2005 ) , six agn near the galactic plane ( soldi et al . 2005 ) , and an early study of all extragalactic sources seen through the galactic plane @xcite . two of the new sources discovered by _ integral _ , and igr j21247 + 5058 , have also been shown to be agn @xcite . with the on - going observation of the sky by _ integral _ , a sufficient amount of data is now accessible to compile the first _ integral _ catalog of agn . in this paper we present analysis of recent observations performed by the _ integral _ satellite instruments ibis / isgri , spi , and jem - x , and compare the results with previous studies . observations in the x - ray to soft gamma - ray domain have been performed by the instruments on - board the _ integral _ satellite @xcite . this mission offers the unique possibility to perform simultaneous observations over the @xmath14 energy region . this is achieved by the x - ray monitor ( 230 kev ) jem - x @xcite , the soft gamma - ray imager ( 201000 kev ) isgri @xcite , and the spectrograph spi @xcite , which operates in the 20 8000 kev region . each of these instruments employs the coded - aperture technique @xcite . in addition to these data an optical monitor ( omc , mas - hesse et al . 2003 ) provides photometric measurements in the @xmath15 band . data used for the analysis presented here were all in the public domain by the end of march 2005 . this includes data from orbit revolutions 19 - 137 and revolutions 142 - 149 . data before revolution 19 have been excluded as the instruments settings changed frequently and therefore the data from this period are not suitable to be included into a homogeneous survey . most of the _ integral _ observing time has been spent on the galactic plane so far , with some deep exposures at high galactic latitude , for example in the virgo and the centaurus region . our sample objects have been selected in three ways . the first criterion was a detection in the isgri data in a single pointed observation . the duration of one pointing of the spacecraft is usually 2 - 3 ks long . the standard analysis of the data is performed at the _ integral _ science data centre ( isdc ; courvoisier et al . these results are publicly available . within one pointing the detection threshold for a source in the center of the field of view of isgri is about @xmath16 . a second level search was done using combined data sets of individual observations and contained within separate revolutions which are about 3 days long . in cases where an observation lasted longer than a revolution , the data set was split by revolution . the results from this analysis are publicly available through the heasarc archive . clearly the sensitivity of this second method depends strongly on the duration of the analyzed observation . the sensitivity also depends on the dither pattern of the observation and on the position of the source within the field of view . in addition to those two ways of combining the list of agn seen by _ integral _ , we added all agn which have been reported as _ integral _ detections so far in the literature . this combined list of objects is presented in table [ catalog ] . in order to get meaningful results for each of the sources in this agn sample , it was necessary to perform an individual analysis for each one . we used the data for which the spacecraft was pointed within 10 degrees radius about each agn for isgri and spi , and within 5 degrees for jem - x . the exposure times listed in tab . [ catalog ] refer to the effective isgri values . these values are approximately the same for the spectrograph spi , but for the jem - x monitor it has to be taken into account that they cover a much smaller sky area . thus in the case of dithering observation , the source is not always in the field of view of the monitors , and the exposure times are correspondingly shorter . the analysis was performed using the offline science analysis ( osa ) software version 5.0 distributed by the isdc ( courvoisier et al . this version shows substantial improvement compared to the previous osa 4.2 version , except for the jem - x spectral extraction , for which we used the march 15 , 2005 version of the osa . the significances listed in tab . [ catalog ] have been derived by using the osa software for isgri ( 20 40 kev ) , spi ( 20 40 kev ) , and jem - x ( 2 20 kev ) . in cases where there were no publicly available data covering the source position , we put `` n / a '' in the significance column . in those cases where no jem - x or spi detection was achieved , `` '' is written . for comparison we list the photon flux measured by _ cgro_/osse in the @xmath17 energy band , as reported by johnson et al . ( 1997 ) for seyfert galaxies and by mcnaron - brown et al . ( 1995 ) for blazars . apart from gamma - ray bursts , no extragalactic source detections by ibis / picsit have been reported . the analysis of the _ integral_/ibis data is based on a cross - correlation procedure between the recorded image on the detector plane and a decoding array derived from the mask pattern @xcite . the isgri spectra have been extracted from the count rate and variance mosaic images at the position of the source , which in all cases corresponds to the brightest pixel in the @xmath18 band . the spi analysis was done using the specific analysis software @xcite including version 9.2 of the reconstruction software spiros @xcite which is based on the `` iterative removal of sources '' technique ( iros ; hammersley et al . 1992 ) : a simple image of the field of view is made using a mapping technique which is optimized for finding a source assuming that the data can be explained by only that source plus background . the mapping gives the approximate location and intensity of the source , which are then improved by maximizing a measure of the goodness of fit . the residuals of the fit are used as the input for a further image reconstruction and source search @xcite . the fitting process is easier when an input catalog is provided . in this work we used the source positions found with isgri as an input catalog for the spi analysis , as the sensitivity of isgri is higher than that of spi below 200 kev . with respect to the calibration uncertainty , the ibis instrument team stated that the systematic error is of the order of 1.5% ( 2005 , private communication ) . nevertheless , this value corresponds to on - axis observations within a short period of time with no disturbing influence , such as enhanced background activity . a combined fit of crab spectra taken in revolution 43 , 44 , 120 , 170 , and 239 , i.e. over a 1.5 year span of the mission , shows a larger uncertainty in the flux . the scale of diversity , assuming that crab is a source with constant flux , gives some hint what is the scale of discrepancies in count rates observed in various conditions , i.e. , with different dithering patterns or instrumental settings . clearly the absolute calibration can not be more precise than the observed variations . hence , it seems that for the 22 - 120 kev band we may conclude that the joint @xmath19 uncertainty of the spectral extraction method and the calibration files is about @xmath20 . it may happen that for other sources and in some special circumstances we may encounter a larger discrepancy especially for faint sources , but a systematic error of 5% should be a valid approximation . in most cases the fit to the data from the three instruments used in this work did not show any improvement when adding variable cross - calibration . this is mainly due to the fact that most spectra have low signal - to - noise and adding a further parameter to the fit procedure worsens the fit result , according to an f - test . the spectral shape calibration was also tested on _ integral _ crab observations . with the so - called canonical model for the crab showing a single power law with photon index @xmath21 @xcite , the values retrieved from _ integral _ are @xmath21 ( jem - x ) , and @xmath22 for spi and isgri @xcite . it should be kept in mind though that the crab is significantly brighter than the sources discussed here , and systematic effects might depend on source brightness . all the extracted images and source results are available in electronic form . _ cgro_/osse @xcite covered the energy range @xmath23 . it therefore preferentially detected agn with hard and bright x - ray spectra , which we also expect to be detectable by _ integral_. several of the agn detected by osse have not been seen by _ integral _ so far . we list those sources and the _ integral_/isgri exposure time on the particular agn in table [ ossenonintegral]@xmath24 . the overall spectra extracted from the _ integral _ jem - x , isgri and spi data for 10 agns are shown in fig . [ fig : ngc1275 ] to fig . [ fig:3c279 ] ( see sect . 4 for details ) . a single powerlaw fit with absorption from material in the line of sight has been applied to the _ integral _ spectra of all sources and the results are presented in table [ fitresults ] and in figure [ fig : photonindices ] . the errors on the photon indices are 90% confidence errors . the intrinsic absorption was determined by subtracting the galactic hydrogen column density from the total absorption as measured in the soft x - rays . in the cases where the _ integral_/jem - x data were not available or did not restrict the absorption column density sufficiently , values from the literature have been used to fix this component in the spectral fit ( tab . [ fitresults ] ) . no systematic error has been added to the data . the spectral fitting was done using version 11.3.2 of xspec @xcite . in most cases a single powerlaw gave an acceptable fit to the data , which also is based on the fact that the signal - to - noise of many of the spectral data is rather low . in six cases a cut - off power law fit led to better fits results . we report those results in table [ cutoff ] . in order to investigate the spectra of agn subtypes , we have derived averaged spectra of the seyfert 1 and 2 types , as well as for the intermediate seyferts and the blazars , and according to the intrinsic absorption . the average seyfert 1 spectrum was constructed using the weighted mean of 10 isgri spectra , the seyfert 2 composite spectrum includes 15 sources , and 8 objects form the intermediate seyfert 1.5 group . the two brightest sources , cen a and ngc 4151 , have been excluded from the analysis as their high signal - to - noise ratio would dominate the averaged spectra . the average spectra have been constructed by computing the weighted mean of all fit results on the individual sources as shown in table [ fitresults ] . in order to do so , all spectra had been fit by an absorbed single powerlaw model . when computing the weighted average of the various sub classes , the seyfert 2 objects show flatter hard x - ray spectra ( @xmath2 ) than the seyfert 1.5 ( @xmath25 ) , and seyfert 1 appear to have the steepest spectra ( @xmath26 ) together with the blazars ( @xmath27 ) . the classification according to the seyfert type of the objects is based on optical observations . an approach to classify sources due to their properties in the x - rays can be done by separating the sources with high intrinsic absorption ( @xmath28 ) from those objects which do not show significant absorption in the soft x - rays . it has to be pointed out that not all objects which show high intrinsic absorption in the x - rays are classified as seyfert 2 galaxies in the optical , and the same applies for the other agn sub - types . nevertheless a similar trend in the spectral slopes can be seen : the 21 absorbed agn show a flatter hard x - ray spectrum ( @xmath29 ) than the 13 unabsorbed sources ( @xmath30 ) . the blazars have again been excluded from these samples . spectra are presented in this paper for _ integral _ sources which are as yet not published and for which at least two instruments yield a detection of @xmath31 . spectral fits did not require any cross - calibration correction between the instruments , except for grs 1734 - 292 and for the two brightest sources , cen a and ngc 4151 . in cases where the spectra of different instruments overlap , the spectral plots have been performed in count space , otherwise in photon space . * ngc 1275 * , a seyfert 2 galaxy at @xmath32 , was detected in the isgri and jem - x data ( fig . [ fig : ngc1275 ] ) . the spectrum represents the blend of the perseus cluster with the spectrum of the agn . the measured spectrum in the @xmath33 energy range is fitted by a slightly absorbed ( @xmath34 ) bremsstrahlung model ( @xmath35 ) plus a powerlaw with a photon index fixed to the measurement from the isgri data ( @xmath36 ) . in addition a gaussian line with @xmath37 and an equivalent width of @xmath38 has been applied to achieve a reasonable fit result ( @xmath39 ) . we see no evidence of flattening of the continuum above @xmath8 , which is often attributed to compton reflection , nor can we quantitatively identify a spectral break . * ngc 4507 * ( seyfert 2 ; @xmath40 ) was detected by isgri and spi ( fig . [ fig : ngc4507 ] ) . the combined data are best represented by a power - law ( @xmath41 ) with a high energy cut - off ( @xmath42 ) . this is consistent with results from _ ginga _ and osse , which showed @xmath43 and @xmath44 @xcite . * ngc 4593 * , a seyfert 1 galaxy in the virgo region at @xmath45 , shows a cut - off powerlaw with @xmath46 and @xmath47 and is detected up to 150 kev ( fig . [ fig : ngc4593 ] ) with a flux of @xmath48 ( @xmath49 ) . this is significantly different from the spectrum measured by _ bepposax _ : @xmath50 and @xmath51 @xcite with a similar flux ( @xmath52 ) . * cen a * , a seyfert 2 ( @xmath53 ) is one of the brightest objects in the sample . the statistics allow in this case to determine the intercalibration factors between jem - x , isgri , and spi ( @xmath54 ) . an absorbed power law plus a gaussian line gives the best fit to the data with @xmath55 , @xmath56 , and an equivalent width of the iron k@xmath57 line of @xmath58 . adding an exponential cut - off does not improve the fit ( @xmath59 ) . the results are consistent with results published on a subset of _ integral _ data and with _ bepposax _ data @xcite . * mcg 0630015 * : the seyfert 1 galaxy ( @xmath60 ) was detected by all three instruments ( fig . [ fig : mcg6 ] ) . the spectrum is well represented ( @xmath61 ) by a power law with @xmath62 and weak evidence for absorption of @xmath63 ( @xmath64 ) . * 4u 134460 * is a bright ( @xmath65 ) x - ray source , which has been unidentified until michel et al . ( 2004 ) reported the identification of this source as a low redshift agn but no further information was provided . in order to clarify the nature of this object , we analyzed 31 ks of xmm - newton data of the centaurus b field , taken in august 2001 . although 4u 134460 appears to be near to the rim of the epic camera , a spectrum can be extracted from epic / pn data which shows an iron line at @xmath66 with a flux of @xmath67 and an equivalent width of @xmath68 . if we consider this line to be the iron k@xmath57 fluorescence line at 6.4 kev , the redshift of 4u 134460 is @xmath69 . the simultaneous fit with the _ integral _ isgri and spi data shows that the spectrum is well represented by an absorbed power law plus gaussian line with @xmath70 and @xmath71 ( @xmath72 for 711 dof ; fig . [ fig:4u1344 ] ) . * ic 4329a * ( seyfert 1.2 ; @xmath73 ) is well represented by a powerlaw ( @xmath74 ) with exponential cut - off ( @xmath75 ) , see fig . [ fig : ic4329a ] . the results are consistent on a @xmath76 level with the spectrum measured by _ bepposax_(@xmath77 , @xmath78 ; perola et al . 2002 ) . : the jem - x and isgri spectrum of this seyfert 1.9 ( @xmath79 ) shows an absorbed ( @xmath80 ) high - energy cut - off spectrum with @xmath81 and @xmath82 ( @xmath83 ; fig . [ fig : ngc5506 ] ) . no spi data are available for this source , because spi was undergoing an annealing cycle during the observations of ngc 5506 . the results are consistent with _ bepposax _ measurements @xcite , even though the iron k@xmath57 line and the reflection component are not detectable in the _ integral _ data . * ic 4518 * ( vv 780 ) : this seyfert 2 galaxy ( @xmath73 ) was reported to be detectable in a 700 ks observation in january 2004 on sn 1006 @xcite . the source was detected with @xmath84 in the @xmath18 band with a photon index of @xmath85 and a flux of @xmath86 . note that in tab . [ catalog ] and tab . [ fitresults ] we refer to our measurements , which are consistent with the values derived by kalemci et al . * grs 1734292 * ( ax j1737.42907 ) : this seyfert 1 galaxy ( @xmath87 ) is located near to the galactic center ( @xmath88 , @xmath89 ) . because this region is monitored regularly by _ integral _ , 3.3 msec of observations are available in the public data . even though an isgri spectrum of this source has been published already @xcite , we discuss this object in more detail , because sazonov et al . included neither the jem - x , nor the spi data in their study . the fact that the source is located in a crowded region , makes the analysis more difficult , especially for spi data with its @xmath90 spatial resolution . figure [ fig : grs1734 ] shows the combined jem - x2 , isgri , and spi spectrum . the data are represented by a simple powerlaw with @xmath91 and in this case intercalibration factors had to be applied to achieve an acceptable fit result ( @xmath92 , 161 d.o.f . ) . adding a cut - off or an absorption component to the model does not improve the fit . the average flux is @xmath93 . this source is of additional interest , as it lies within the error circle for the egret source 3eg j1736 - 2908 @xcite . if this is a real association it would be the only known case of a `` gamma - ray loud '' seyfert 1 . * mrk 509 * ( seyfert 1 ; @xmath94 ) was not detected by spi and yielded only low signal - to - noise spectra with jem - x and isgri ( fig . [ fig : mrk509 ] ) . no absorption component was necessary to fit the combined spectrum . a pure single powerlaw ( @xmath95 ) resulted in @xmath96 , consistent with measurements by _ xmm - newton _ ( @xmath97 ; pounds et al . 2001 ) and with _ bepposax_(@xmath98 ) , although the latter data showed in addition a cut - off at @xmath99 @xcite . * s5 0716 + 714 * : results on _ integral _ observations of this bl lac object have been published by pian et al . using the osa reduction version 4.0 they achieved a marginal detection of @xmath100 in the @xmath101 band . using more data and the current version of the software the source is not detectable , but the strong variability of the source might have resulted in a low average flux . * 3c 279 * : this blazar ( @xmath102 ) led to a weak detection with jem - x and to none with spi . the spectrum , model shown in fig . [ fig:3c279 ] , is dominated by the isgri statistics and is well represented by a single powerlaw with photon index @xmath103 ( @xmath39 ) , consistent with _ ( @xmath104 , ballo et al . 2002 ) and _ cgro_/osse observations ( @xmath105 ; mcnaron - brown et al . the jem - x data show an excess in the 2 7 kev energy band . this excess might be caused by the transition from the synchrotron to the inverse compton branch in this energy range , which would lead to a convex spectrum . several sources may have been marginally suspected in the _ integral _ data . they were either targets of dedicated observations or were detected in some single science windows . we performed the same analysis as for the other sources presented here , but found only spurious source candidates which result from image reconstruction problems like ghosts , mask patterns , etc . among those sources are * mrk 231 * , , , * mcg 052316 * , , , * ngc 4418 * , , and @xmath24 . in order to explore the space distribution of hard x - ray selected objects , one generally needs to consider a statistically complete flux - limited sample . however , this is difficult to achieve with a data set comprised of various lines of sight and widely differing exposures . in addition , the telescope effective area can be highly position dependent within the coded aperture field of view . thus , we opt to leave the compilation of a log - n log - s distribution , and perhaps a hard x - ray agn luminosity distribution to a future paper . alternatively , the @xmath106-test is a simple method developed by avni & bahcall ( 1980 ) based on the @xmath107 test of schmidt ( 1968 ) . @xmath108 stands for the volume , which is enclosed by the object , and @xmath109 is the accessible volume , in which the object could have been found ( e.g. due to a flux limit of a survey ) . avni & bahcall showed that different survey areas with different flux limits in various energy bands can be combined by the @xmath106-test . in the case of no evolution @xmath110 is expected and the error @xmath111 for a given mean value @xmath112 based on @xmath113 objects is @xmath114 this evolutionary test is applicable only to samples with a well - defined flux limit down to which all objects have been found . it can therefore also be used to test the completeness of a sample . we performed a series of @xmath106-tests to the _ integral _ agn sample , assuming completeness limits in the range of @xmath115 up to @xmath116 isgri 20 40 kev significance , adding 5% systematic error to the flux uncertainty as described in section [ data ] . for a significance limit below the true completeness limit of the sample one expects the @xmath106-tests to derive a value @xmath117 , where @xmath118 is the true test result for a complete sample . above the completeness limit the @xmath119 values should be distributed around @xmath118 within the statistical uncertainties . the results of the tests are shown in figure [ fig : veva ] . it appears that the sample becomes complete at a significance level around @xmath120 , including 31 agn . the average value above this significance is @xmath121 . this value is consistent with the expected value of 0.5 which reflects no evolution and an even distribution in the local universe . if we do not apply the 5% systematic error to the data , the average value is lower ( @xmath122 ) , indicating that a systematic error is indeed apparent in the isgri data . the typical _ integral _ spectrum can be described by a simple powerlaw model with average photon indices ranging from @xmath123 for obscured agn to @xmath21 for unabsorbed sources ( tab . [ averagespectra ] ) . the simple model does not give reasonable results in cases of high signal - to - noise , where an appropriate fit requires additional features such as a cut - off and a reflection component @xcite . the results presented here show slightly steeper spectra than previous investigations of agn in comparable energy ranges . the same trend is seen in the comparison of crab observations , where the _ integral_/isgri spectra also appear to be slightly steeper than in previous observations @xcite , and in comparison of _ rxte _ and _ integral_/isgri spectra of cen a @xcite . zdziarski et al . ( 1995 ) did a study of seyfert galaxies observed by _ ginga _ and _ cgro_/osse , with a similar redshift distribution ( @xmath124 ) as the seyferts detected by _ integral _ ( @xmath125 ) . the same trend in spectral slope from @xmath126 for radio - quiet seyfert 1 down to @xmath127 for seyfert 2 was detectable in the sample , and a potential cut - off appeared at energies @xmath128 . the average of all seyfert spectra ( @xmath129 ) appears to be similar to the one in our sample ( @xmath130 ) . gondek et al . ( 1996 ) confirmed that seyfert 1 galaxies show a continuum with photon index @xmath131 with high energy cut - off @xmath132 and moderate reflection component ( @xmath133 ) , based on _ exosat _ , _ ginga _ , _ heao-1 _ , and _ cgro_/osse data . more recent studies by perola et al . ( 2002 ) on _ bepposax_data of 9 seyfert 1 galaxies confirmed this picture . malizia et al . ( 2003 ) used the _ bepposax _ data ( 3 200 kev ) of 14 seyfert 1 and 17 seyfert 2 to study average spectral properties . also this work finds seyfert 1 galaxies to be softer ( @xmath134 ) than seyfert 2 ( @xmath135 ) , which also show stronger absorption . both types show iron lines with an equivalent width of 120 ev and a compton reflection component of @xmath136 , but the energy cut - off of the type 1 objects appears to be at higher energies ( 200 kev compared to @xmath137 for the seyfert 2 ) . deluit & courvoisier ( 2003 ) used for their analysis only the _ bepposax_/pds spectra of 14 seyfert 1 , 9 seyfert 1.5 , and 22 seyfert 2 . this study lead to a somewhat different result . contrary to other studies seyfert 2 objects show a steeper spectrum ( @xmath138 ) than seyfert 1 ( @xmath139 ) , but with the large error on the results , the difference is not significant . a cut - off does not seem to be required by the _ bepposax_data . a marginally significant cut - off is detected in the seyfert 1 spectra with @xmath140 . the average spectra show moderate reflection for seyfert 1/2 objects ( @xmath141 and @xmath142 , respectively ) , while the seyfert 1.5 appear to have @xmath143 @xcite . it has to be kept in mind that the results in this study are based on _ bepposax_/pds spectra alone , without taking into account the valuable information from the lower energy lecs / mecs data , which could have significantly constrained the shape of the underlying continuum . the _ integral _ agn sample confirms the previously found trend with harder spectra for the obscured objects . this can be roughly understood by the argument that in order to overcome a substantial absorption , obscured objects have to have hard x - ray spectra to be detectable . comparing the ratio @xmath144 of obscured ( @xmath28 ) to unobscured agn we find in the _ integral _ data that @xmath145 . excluding the targets of observations , leads to @xmath146 . the ratios change slightly when taking into account only those objects which belong to the complete sample with an isgri significance of @xmath0 or higher ( tab . [ catalog ] ) . this sub - sample includes 32 agn , with 18 obscured and 10 unobscured objects ( absorption information is missing for the remaining four objects ) . using only the complete sample changes the ratio to @xmath147 and @xmath148 ( targets excluded ) , respectively . splitting this result up into objects near the galactic plane ( @xmath149 ) and off the plane shows for all objects a ratio of @xmath150 and @xmath151 , respectively . this trend shows that the harder spectra of those objects , where the absorption in the line of sight through the galaxy is low compared to the intrinsic absorption , are more likely to shine through the galactic plane . one would expect a higher ratio in order to explain the unresolved x - ray background in this energy range ( e.g. worsley et al . but it has to be taken into account that our sample only represents the high flux and low luminosity sources ( @xmath152 ) in the local universe ( @xmath153 ) . gilli et al . ( 1999 ) showed that the ratio between type 2 and all qsos should be at most 2 in the local universe . the _ bepposax _ hellas survey showed that the fraction of unobscured agn increases with the flux of the objects @xcite . evidence that the fraction of obscured agn depends on the redshift was found by ueda et al . ( 2004 ) when studying the hard x - ray luminosity function . they find that the density evolution of agn depends on the luminosity which will lead to an increase of the type 2 fraction with redshift . in a recent study la franca et al . ( 2005 ) find that both effects combined ( the fraction of absorbed agn decreases with the intrinsic x - ray luminosity , and increases with the redshift ) can be explained by a luminosity - dependent density evolution model . they further show that the luminosity function of agn , like those presented here , peaks at @xmath154 while high luminosity agn peak at @xmath155 . consistent with our study , la franca et al . also find a ratio of @xmath156 at @xmath157 in the @xmath158 energy band . unified models predict , depending on the applied model , a ratio of @xmath159 for high flux low redshift agn @xcite , while a recent study of _ chandra _ data showed that @xmath160 for low redshift , low luminosity agn @xcite , indicating a larger dominance of obscured objects . these investigations have been focused so far on the x - rays below 10 kev , and _ integral _ can add substantial information to the nature of bright agns in the local universe . considering the expected composition of the hard x - ray background , it is remarkable that our sample includes only 4 compton thick agn . if compton thick agn are dominant in the hard x - rays , they would have to be fainter than the objects detected by _ integral _ so far . the energy range in the 15 200 kev is now also accessible through the bat instrument aboard _ swift _ @xcite . a study by markwardt et al . ( 2005 ) of observations of the extragalactic sky indicates a ratio of @xmath161 between obscured and unobscured agn , fully consistent with the results from the _ integral _ agn sample . although the _ swift_/bat survey covers different areas of the sky , the energy band is similar to the _ integral_/isgri range and the type of agns detectable should be the same . the sample presented here is still too small to constrain the ratio of obscured to unobscured sources , but it might indicate that the unified model predicts this value of @xmath162 correctly , differing from earlier claims @xcite . the agn sample derived from the _ integral _ public data archive comprises 42 low luminosity and low redshift ( @xmath152 , @xmath163 , excluding the blazars ) objects including 5 blazars in the hard x - ray domain . only one galaxy cluster is also detected , but no star - burst galaxy has been as yet . the results from the jem - x2 , isgri , and spi data show that within the statistical limitations of the data the high - energy continua of agn can be described by a simple powerlaw with photon index @xmath4 ( seyfert 1 ) , @xmath3 ( seyfert 1.5 ) , and @xmath2 ( seyfert 2 ) . a similar trend is seen when dividing the seyfert galaxies according to their intrinsic absorption into unabsorbed ( @xmath164 ; @xmath165 ) and absorbed sources ( @xmath166 ) . blazars naturally show strongly variable spectra , but the average spectral slope is of the same order ( @xmath27 ) . it has to be taken into account that the simple model does not represent the data as soon as a sufficient signal - to - noise is apparent , like in the circinus galaxy and cen a @xcite or ngc 4151 @xcite . but the average spectra indicate that a cut - off in most cases , if apparent , appears at energies @xmath167 . the agn sample presented here is complete down to a significance limit of @xmath168 . this results in a complete sample of 32 agn . within this complete sample the ratio between obscured and unobscured agn is @xmath147 , consistent with the unified model for agn and with recent results from the _ swift_/bat survey . only 4 compton thick agn are found in the whole and only 3 in the complete sample . sazonov et al . ( 2005 ) report five additional agn in recent observations , not detectable in the data presented here . additionally one agn seen by osse , the blazar ( mcnaron - brown et al . 1995 ) , has recently been detected by _ integral _ during an outburst @xcite . with the ongoing observations of the _ integral _ and _ swift _ mission , the number of detectable agns will increase further , especially with a number of already performed and scheduled observations far off the galactic plane . this will enable us to solve the mystery of the hard x - ray background above 10 kev in the near future . we like to thank the anonymous referee for the comments which helped to improve the paper . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , of data obtained from the high energy astrophysics science archive research center ( heasarc ) , provided by nasa s goddard space flight center , and of the simbad astronomical database which is operated by the centre de donnes astronomiques de strasbourg . this research has made use of the tartarus ( version 3.1 ) database , created by paul oneill and kirpal nandra at imperial college london , and jane turner at nasa / gsfc . tartarus is supported by funding from pparc , and nasa grants nag5 - 7385 and nag5 - 7067 . arnaud , k.a . 1996 , in : astronomical data analysis software and systems v , eds . jacoby g. and barnes j. , p17 , asp conf . series 101 avni , y. & bahcall , j. n. 1980 , apj , 235 , 694 ballo , l. , et al . 2002 , apj , 567 , 50 barger , a. j. , cowie , l. l. , mushotzky , r. f. , yang , y. , wang , w .- h . , steffen , a. t. , & capa , p. 2005 , aj , 129 , 578 bassani , l. , malagutti , g. , jourdain , e. , roques , j. p. , johnson , w. n. 1995 , apj , 444 , l73 bassani , l. , et al . 2004 , in : proc . 5th integral workshop , esa sp-552 , 139 , astro - ph/0404442 beckmann , v. , gehrels , n. , favre , p. , walter , r. , courvoisier , t. j .- l . , petrucci , p .- o . , malzac , j. 2004 , apj , 614 , 641 beckmann , v. et al . 2005 , apj accepted , bianchi , s. , balestra , i. , matt , g. , guainazzi , m. , perola , g. c. 2003 , a&a , 402 , 141 boella , g. , butler , r. c. , perola , g.c . , piro , l. , scarsi , l. , bleeker , j. a. m. 1997 , a&as , 122 , 299 brandt , w. n. , & hasinger , g. 2005 , ara&a , 43 caroli , e. , stephen , j. b. , di cocco , g. , natalucci , l. , spizzichino , a. 1987 , space sci . , 45 , 349 comastri , a. , setti , g. , zamorani , g. , hasinger , g. 1995 , a&a , 296 , 1 comastri , a. , fiore , f. , vignali , c. , matt , g. , perola , g. c. , & la franca , f. 2001 , mnras , 327 , 781 courvoisier , t.j .- l . , beckmann , v. , bourban , g. , et al . 2003a , a&a , 411 , l343 courvoisier , t.j .- l . , walter , r. , beckmann , v. , et al . 2003b , a&a , 411 , l53 deluit , s. , & courvoisier , t. j .- 2003 , a&a , 399 , 77 de rosa , a. , et al . 2005 , a&a , 438 , 121 diehl , r. , baby , n. , beckmann , v. , et al . 2003 , a&a , 411 , l117 foschini , l. , et al . 2005 , atel , 497 fossati , g. , maraschi , l. , celotti , a. , comastri , a. , & ghisellini , g. 1999 , mnras , 299 , 433 frontera , f. , costa , e. , & dal fiume , d. 1997 , a&as , 122 , 357 gehrels , n. , et al . 2005 , apj , 611 , 1005 gehrels , n. , chipman , e. , & kniffen , d. a. 1993 , a&as , 97 , 5 george , i. m. , & fabian , a. c. 1991 , mnras , 249 , 352 gilli , r. , salvati , m. , hasinger , g. , 2001 , a&a , 366 , 407 gilli , r. , risaliti , g. , & salvati , m. 1999 , a&a , 347 , 424 goldwurm , a. , et al . 2003 , a&a , 411 , l223 gondek , d. , et al . 1996 , mnras , 282 , 646 guainazzi , m. , et al . 1999 , a&a , 346 , 407 hammersley , a. p. , ponman , t. j. , & skinner , g. k. 1992 , nims , a311 , 585 johnson , w. n. , et al . 1993 , apjs , 86 , 693 johnson , w. n. , zdziarski , a. a. , madejski , g. m. , paciesas , w. s. , steinle , h. , & lin , y .- c . 1997 , in the fourth compton symposium , c. d. dermer et al . new york : aip , p. 283 kalemci , e. , boggs , s. e. , lund , n. 2005 , atel , 410 kirsch , m. g. , et al . 2005 , in proc . spie annual meeting , san diego , usa , august 2005 , 5898 - 03 , astro - ph/0508235 krivonos , r. , vikhlinin , a. , churazov , e. , lutovinov , a. , molkov , s. , & sunyaev , r. 2005 , apj , 625 , 89 la franca , f. , et al . 2005 , apj in press , astro - ph/0509081 lebrun , f. , et al . 2003 , a&a , 411 , l141 levenson , n. a. , weaver , k. a. , & heckman , t. m. 2001 , apjs , 133 , 269 lund , n. , et al . 2003 , a&a , 411 , l231 lutz , d. , maiolino , r. , spoon , h. w. w. , moorwood , a. f. m. 2004 , a&a , 418 , 465 malizia , a. , bassani , l. , stephen , j. b. , di cocco , g. , fiore , f. , & dean , a. j. , 2003 , apj , 589 , l17 markwardt , c. b. , tueller , j. , skinner , g. k. , gehrels , n. , barthelmy , s. d. , mushotzky , r. f. 2005 , apjl accepted , astro - ph/0509860 mas - hesse , j. m. , et al . 2003 , a&a , 411 , l261 masetti , n. , palazzi , e. , bassani , l. , malizia , a. , stephen , j. b. 2004 , a&a , 426 , l41 matsumoto , c. , nava , a. , maddox , l. a. , leighly , k. m. , grupe , d. , awaki , h. , ueno , s. 2004 , apj , 617 , 930 matt , g. , et al . 1997 , a&a , 325 , l13 mcnaron - brown , k. , et al . 1995 , apj , 451 , 575 michel , l. , herent , o. , motch , c. , pye , j. , & watson , m.g . 2004 , asp conf . series , 314 , 570 paul , j. , et al . 1991 , adv.sp.res . 11 , 8 , 289 perola , g. c. , et al . 2002 , a&a , 389 , 802 pian , e. , et al . 2005 , a&a , 429 , 427 pounds , k. , reeves , j. , obrien , p. , page , k. , turner , m. , & nayakshin , s. 2001 , apj , 559 , 181 rothschild , r. e. , et al . 2006 , apj submitted sazonov , s. y. , revnivtsev , m. g. , lutovinov , a. a. , sunyaev , r. a. , & grebenev , s. a. 2004 , a&a , 421 , l21 sazonov , s. y. & revnivtsev , m. g. 2004 , a&a , 423 , 469 sazonov , s. y. , churazov , e. , revnivtsev , m. g. , vikhlinin , a. , sunyaev , r. , 2005 , a&a accepted , astro - ph/0508593 schmidt , m. , 1968 , apj , 151 , 393 setti , g. , woltjer , l. 1989 , a&a , 224 , l21 skinner , g. , & connell , p. 2003 , a&a , 411 , l123 soldi , s. , beckmann , v. , bassani , l. , et al . 2005 , a&a accepted , astro - ph/0509123 toor , a. , & seward , f. d. 1974 , aj , 79 , 995 treister , e. , & urry , c. m. 2005 , apj , 630 , 115 ubertini , p. 2003 , a&a , 411 , l131 vedrenne , g. , et al . 2003 , a&a , 411 , l63 vron - cetty , m. p. & vron , p. 2000 , a&a rev . , 10 , 81 winkler , c. , et al . 2003 , a&a , 411 , l1 worsley , m. a. , et al . 2005 , mnras , 357 , 1281 young , a. j. , wilson , a. s. , terashima , y. , arnaud , k. a. , smith , d. a. 2002 , apj , 564 , 176 zdziarski , a. a. , johnson , w. n. , done , c. , smith , d. , & mcnaron - brown , k. 1995 , apj , 438 , l63 lclrrrcrcccc ngc 788 & sy 1/2 & 0.0136 & 02 01 06 & 06 46 56 & 311 & & 10.1 & & & + ngc 1068 & sy 2 & 0.003793 & 02 42 41 & 00 00 48 & 311 & & 4.4 & & & + ngc 1275 & sy 2 & 0.017559 & 03 19 48 & + 41 30 42 & 264 & & 19.1 & & 78 & @xmath169 + 3c 111 & sy 1 & 0.048500 & 04 18 21 & + 38 01 36 & 67 & & 9.5 & 0.1 & & @xmath170 + mcg + 81111 & sy 1.5 & 0.020484 & 05 54 54&+46 26 22 & 21 & & 5.5 & & & @xmath171 + mrk 3 & sy 2 & 0.013509 & 06 15 36 & + 74 02 15 & 472 & & 13.3 & 2.9 & & @xmath172 + mrk 6 & sy 1.5 & 0.018813 & 06 52 12 & + 74 25 37 & 482 & & 8.8 & 1.1 & & + ngc 4051 & sy 1.5 & 0.002336 & 12 03 10 & + 44 31 53 & 443 & & 8.8 & 3.0 & & + ngc 4151 & sy 1.5 & 0.003320 & 12 10 33 & + 39 24 21 & 483 & x & 163.3 & 27.2 & 2089 & @xmath173 + ngc 4253 & sy 1.5 & 0.012929 & 12 18 27 & + 29 48 46 & 715 & & 4.3 & 3.1 & & + ngc 4388 & sy 2 & 0.008419 & 12 25 47 & + 12 39 44 & 215 & x & 37.9 & 4.0 & 19 & @xmath174 + ngc 4395 & sy 1.8 & 0.001064 & 12 25 49 & + 33 32 48 & 739 & & 10.0 & & & + ngc 4507 & sy 2 & 0.011801 & 12 35 37 & 39 54 33 & 152 & & 15.8 & 3.6 & n / a & @xmath175 + ngc 4593 & sy 1 & 0.009000 & 12 39 39 & 05 20 39 & 723 & & 20.9 & 3.8 & 3 & @xmath176 + coma cluster & gclstr & 0.023100 & 12 59 48 & + 27 58 48 & 516 & x & 6.6 & 1.4 & 8.9 & + ngc 4945 & sy 2 & 0.001878 & 13 05 27 & 49 28 06 & 276 & x & 33.4 & 6.7 & & @xmath177 + eso 323g077 & sy 2 & 0.015014 & 13 06 26 & 40 24 53 & 761 & & 9.9 & 1.3 & & + ngc 5033 & sy 1.9 & 0.002919 & 13 13 28 & + 36 35 38 & 377 & & 2.9 & 1.2 & & + cen a & sy 2 & 0.001830 & 13 25 28 & 43 01 09 & 532 & x & 166.1 & 24.8 & 1479 & @xmath178 + mcg0630015 & sy 1 & 0.007749 & 13 35 54 & 34 17 43 & 567 & x & 2.9 & 2.2 & 9 & @xmath179 + 4u 134460 & ? & 0.043 & 13 47 25 & 60 38 36 & 603 & & 40.9 & 5.9 & & + ic 4329a & sy 1.2 & 0.016054 & 13 49 19 & 30 18 36 & 440 & x & 43.5 & 6.2 & 1 & @xmath180 + circinus gal . & sy 2 & 0.001448 & 14 13 10 & 65 20 21 & 589 & & 79.5 & 8.6 & 2 & + ngc 5506 & sy 1.9 & 0.006181 & 14 13 15 & 03 12 27 & 101 & x & 19.6 & n / a & 39 & @xmath181 + pg 1416 - 129 & sy 1 & 0.129280 & 14 19 04 & 13 10 44 & 117 & & 12.0 & 1.1 & n / a & + ic 4518 & sy 2 & 0.015728 & 14 57 43 & 43 07 54 & 338 & & 4.8 & 0.5 & & + ngc 6221 & sy 1/2 & 0.004977 & 16 52 46 & 59 13 07 & 523 & & 7.6 & 7.0 & & + ngc 6300 & sy 2 & 0.003699 & 17 17 00 & 62 49 14 & 173 & & 9.8 & & 3 & + grs 1734 - 292 & sy 1 & 0.021400 & 17 37 24 & 29 10 48 & 3332 & & 43.8 & 16.9 & 15.2 & + igr j18027 - 1455&sy 1 & 0.035000 & 18 02 47 & 14 54 55 & 942 & & 12.1 & 5.8 & 14 & + eso 103-g35 & sy 2 & 0.013286 & 18 38 20 & 65 25 39 & 36 & & 12.2 & 1.5 & & + 1h 1934 - 063 & sy 1 & 0.010587 & 19 37 33 & 06 13 05 & 684 & & 6.6 & 4.4 & & + ngc 6814 & sy 1.5 & 0.005214 & 19 42 41 & 10 19 25 & 488 & & 13.9 & 1.7 & & @xmath182 + cygnus a & sy 2 & 0.056075 & 19 59 28 & + 40 44 02 & 1376 & & 7.3 & 2.5 & & + mrk 509 & sy 1 & 0.034397 & 20 44 10 & 10 43 25 & 73 & x & 10.1 & 2.6 & 6 & @xmath183 + igr j21247 + 5058&radio gal . ? & 0.3 & 21 24 39&+50 58 26 & 213 & & 30.8 & 5.2 & & + mr 2251 - 178 & sy 1 & 0.063980 & 22 54 06 & 17 34 55 & 489 & x & 35.9 & 1.9 & & + mcg -02 - 58 - 022 & sy 1.5 & 0.046860 & 23 04 44 & 08 41 09 & 489 & & 5.9 & 1.6 & & @xmath184 + s5 0716 + 714 & bl lac & 0.3 & 07 21 53 & + 71 20 36 & 482 & x & 0.7 & & & + s5 0836 + 710 & fsrq & 2.172 & 08 41 24 & + 70 53 42 & 391 & & 8.5 & 0.6 & & + 3c 273 & blazar & 0.15834 & 12 29 07 & + 02 03 09 & 655 & x & 35.6 & 6.0 & 32 & @xmath185 + 3c 279 & blazar & 0.53620 & 12 56 11 & + 05 47 22 & 497 & x & 6.8 & & 3 & @xmath186 + pks 1830211&blazar & 2.507 & 18 33 40 & 21 03 40 & 1069 & & 16.1 & 1.4 & & + lcr name & type & exposure@xmath187 + & & [ ks ] + 3c 120 & sy1 & 3 + 3c 390.3 & sy1 & + 3c 454.3 & blazar & 2 + cta 102 & blazar & 6 + eso 141-g55 & sy 1 & 23 + h 1517 + 656 & bl lac & 2 + iii zw2 & sy1/2 & 79 + mrk 279 & sy1.5 & 13 + mrk 841 & sy1.5 & + m82 & starburst & 84 + ngc 253 & starburst & + ngc 2110 & sy2 & + ngc 3227 & sy1.5 & + ngc 3783 & sy1 & + ngc 526a & sy1.5 & + ngc 5548 & sy1.5 & + ngc 7172 & sy2 & + ngc 7213 & sy1.5 & + ngc 7314 & sy1.9 & 489 + ngc 7469 & sy1.2 & 2 + ngc 7582 & sy2 & + pks 0528 + 134 & blazar & 1040 + pks 2155 - 304 & bl lac & + qso 1028 + 313 & sy1.5 & + lrrrcrc ngc 788 & @xmath189 & @xmath190 & & 43.52 & @xmath191 & @xmath192 + ngc 1068 & @xmath193 & @xmath194 & & 41.92 & @xmath195 & @xmath196 + ngc 1275 & @xmath197 & @xmath198 & @xmath199 & 43.47 & 3.75 & @xmath200 + 3c 111 & @xmath201 & @xmath202 & & 44.90 & 0.634 & @xmath203 + mcg + 8 - 11 - 11 & @xmath204 & @xmath205 & & 44.06 & @xmath206 & @xmath207 + mrk 3 & @xmath208 & @xmath209 & & 43.53 & 110 & @xmath210 + mrk 6 & @xmath211 & @xmath212 & & 43.57 & 10 & @xmath213 + ngc 4051 & @xmath214 & @xmath215 & & 43.60 & @xmath216 & @xmath217 + ngc 4151 & @xmath218 & @xmath219 & @xmath220 & 43.15 & 4.9 & @xmath221 + ngc 4253 & @xmath222 & @xmath223 & & 42.93 & 0.8 & @xmath224 + ngc 4388 & @xmath225 & @xmath226 & @xmath227 & 43.55 & 27 & @xmath228 + ngc 4395 & @xmath229 & @xmath230 & & 40.64 & 0.15 & @xmath231 + ngc 4507 & @xmath232 & @xmath233 & & 43.68 & 29 & @xmath234 + ngc 4593 & @xmath235 & @xmath236 & @xmath237 & 43.08 & 0.02 & @xmath238 + coma cluster & @xmath239 & @xmath240 & @xmath241 & 43.40 & @xmath242 & @xmath243 + ngc 4945 & @xmath244 & @xmath245 & & 42.30 & 400 & @xmath246 + eso 323-g077 & @xmath247 & @xmath248 & & 43.21 & 55 & @xmath249 + ngc 5033 & @xmath250 & @xmath251 & & 41.55 & 2.9 & @xmath252 + cen a & @xmath253 & @xmath254 & @xmath255 & 42.75 & 12.5 & @xmath256 + mcg-06 - 30 - 015 & @xmath257 & @xmath258 & @xmath259 & 42.93 & 17.7 & @xmath260 + 4u 1344 - 60 & @xmath261 & @xmath262 & & 44.36 & 2.19 & @xmath263 + ic 4329a & @xmath264 & @xmath265 & @xmath266 & 44.04 & 0.42 & @xmath267 + circinus gal . & @xmath268 & @xmath269 & @xmath270 & 41.97 & 360 & @xmath271 + ngc 5506 & @xmath272 & @xmath273 & @xmath274 & 42.83 & 3.4 & @xmath275 + pg 1416 - 129 & @xmath276 & @xmath277 & & 45.78 & 0.09 & @xmath278 + ic 4518 & @xmath279 & @xmath280 & & 42.92 & ? & @xmath281 + ngc 6221 & @xmath282 & @xmath283 & & 42.39 & 1 & @xmath284 + ngc 6300 & @xmath285 & @xmath286 & @xmath287 & 42.36 & 22 & @xmath288 + grs 1734 - 292 & @xmath289 & @xmath290 & @xmath291 & 43.88 & 3.7 & @xmath292 + igr j18027 - 1455 & @xmath293 & @xmath294 & @xmath295 & 44.03 & 19.0 & @xmath296 + eso 103-g35 & @xmath297 & @xmath298 & & 43.51 & 13 16 & @xmath299 + 1h 1934 - 063 & @xmath300 & @xmath301 & & 42.51 & ? & @xmath302 + ngc 6814 & @xmath303 & @xmath304 & & 42.52 & @xmath305 & @xmath306 + cygnus a & @xmath307 & @xmath308 & & 44.71 & 20 & @xmath309 + mrk 509 & @xmath310 & @xmath311 & @xmath312 & 44.42 & @xmath313 & @xmath314 + igr j21247 + 5058 & @xmath315 & @xmath316 & & 44.00@xmath317 & ? & @xmath318 + mr 2251 - 178 & @xmath319 & @xmath320 & & 44.40 & 0.02 0.19 & @xmath321 + mcg -02 - 58 - 022 & @xmath322 & @xmath323 & & 44.18 & @xmath324 & @xmath325 + s5 0716 + 714 & @xmath326 & @xmath327 & & 45.21@xmath317 & @xmath216 & + s5 0836 + 710 & @xmath328 & @xmath329 & & 47.87 & @xmath330 & @xmath331 + 3c 273 & @xmath332 & @xmath333 & @xmath334 & 45.92 & @xmath335 & @xmath336 + 3c 279 & @xmath337 & @xmath338 & @xmath339 & 46.37 & @xmath340 & @xmath341 + pks 1830211 & @xmath342 & @xmath343 & & 48.09 & @xmath344 & @xmath345 lcc source & @xmath346 & @xmath347 + & powerlaw & [ kev ] + ngc 4507 & @xmath348 & @xmath349 + ngc 4593 & @xmath238 & @xmath350 + eso 323-g077 & @xmath351 & @xmath352 + ic 4329a & @xmath353 & @xmath354 + circinus galaxy & @xmath355 & @xmath356 + ngc 5506 & @xmath357 & @xmath358 +

the sky - average , or global , component of the redshifted @xmath0-cm signal represents a direct tracer of the bulk characteristics of the intergalactic medium ( igm ) during cosmic dawn and the epoch of reionization ( eor ) at redshidfts @xmath7 @xcite . the observable quantity corresponds to a differential brightness temperature that depends on the fraction of neutral hydrogen and the spin temperature of the gas . this temperature encodes the effects on the igm of uv and x - ray radiation from the first generations of stars and stellar remnants @xcite . for @xmath7 , the large - scale evolution of the igm is captured as wideband features in the frequency spectrum of the brightness temperature below @xmath8 mhz , with expected absolute amplitudes lower than @xmath8 mk @xcite . the most direct instrumental approach to attempt the global measurement is the single - antenna , wideband - spectrometer design implemented by edges @xcite , bighorns @xcite , saras @xcite , and sci - hi @xcite . these ground - based experiments cover differing portions of the redshift domain , targeting specific features of the global @xmath0-cm signal . using a similar approach , the dare experiment aims to conduct this measurement from the far side of the moon to mitigate the impact of radio - frequency interference and the earth s ionosphere @xcite . the leda experiment also pursues the global signal through single - antenna total - power measurements , with the addition of interferometric measurements to help estimate instrumental and foreground parameters @xcite . edges has conducted sky measurements from western australia since 2006 . until 2012 the experiment relied on relative calibration provided by three - position switching at the input of the receiver . this approach did not fully account for the impedance mismatch between the antenna and the receiver , since reflections from the receiver input were assumed to be zero , and the system gain and antenna reflections were not referenced to the same calibration plane . in addition , the calibration did not consider a correction for beam chromaticity . despite this simplified approach , edges was able to place initial constraints on the duration of the eor and obtain a first - order estimate for the spectral index of diffuse emission in low - foreground regions @xcite . in 2013 , edges deployed the first instrument that implemented the current end - to - end absolute calibration . this approach consists of 1 ) converting the noise power measured by the antenna at the receiver input to an absolute antenna temperature scale , properly accounting for impedance mismatches between the receiver and the antenna , 2 ) removing the effect of antenna losses , and 3 ) compensating for the effect of beam chromaticity . although in its first iteration it did not achieve the sensitivity for a cosmological detection , its high performance enabled quantification of perturbations in the ionosphere @xcite . starting in 2015 , edges has operated a low - band ( @xmath9 mhz ) and a high - band ( @xmath2 mhz ) instrument . both share the same design and absolute calibration approach , and target redshifts that nominally correspond to the cosmic dawn ( @xmath10 ) and eor ( @xmath1 ) periods , respectively . they represent a significant upgrade with respect to previous iterations , with the objective of reducing systematic effects below the cosmological signal . in this revision , the fourpoint antenna @xcite was replaced with a blade model due to its lower beam chromaticity @xcite . the antenna reflection is periodically measured in situ to high accuracy . the receiver is calibrated in the laboratory to higher precision and accuracy than in 2013 , and in the field it operates underground with active temperature control . these and other upgrades have recently enabled improved long - term ( @xmath11 days ) stable measurements of the spectral index of diffuse foregrounds @xcite . ) and duration ( @xmath12 ) of reionization.,scaledwidth=48.0% ] this paper describes the laboratory calibration of the @xmath3 edges high - band receiver , which nominally targets the eor signal . for reference , figure ( [ figure_models_eor ] ) shows three phenomenological models for this signal , parameterized in terms of the redshift ( @xmath13 ) and duration ( @xmath12 ) of reionization @xcite . our receiver calibration follows the method introduced in @xcite , which involves measuring the spectra , reflection coefficients , and physical temperatures of four absolute calibrators connected externally to the receiver input , in place of the antenna . these calibration measurements are used to determine the function that converts the noise power measured by the instrument to antenna temperature referenced to the receiver input . we use this function to calibrate the sky measurements obtained in the field with the receiver operating at the same temperature as in the lab . as part of the description of our calibration , we expand on the methodology presented in @xcite by providing a model for the ambient and hot calibrator used as the absolute temperature reference . we also introduce two new frequency - dependent parameters a scale and an offset that relate the relative calibration obtained through the internal switching to the absolute calibration at the receiver input . after determining the fiducial receiver calibration , we model the uncertainties encountered in the laboratory measurements and propagate them to the calibrated antenna temperature using a monte carlo approach . we use simulated sky measurements as inputs in this process , and quantify the impact of potential calibration errors via the number of polynomial terms required to fit the corrupted spectrum and reduce the residuals below a nominal threshold of @xmath4 mk . robust estimates are obtained for the impact of each source of uncertainty , as well as for their combined effect . in order to focus on receiver uncertainties , these simulations assume perfect removal of antenna losses and beam chromaticity . the paper is organized as follows : section [ section_edges ] describes the edges instrument and summarizes the receiver calibration strategy and nomenclature , section [ section_calibration_measurements ] describes the calibration measurements and results , section [ section_uncertainty_propagation ] describes the propagation of receiver uncertainty to the antenna temperature using monte carlo simulations , section [ section_results ] presents the results of the uncertainty propagation , section [ section_discussion ] discusses the limiting factors in the calibration performance and some alternatives for improvement , and section [ section_conclusion ] summarizes the findings of this work . figure [ figure_block_diagram ] presents a conceptual block diagram of the edges instrument . the receiver that operates in the field corresponds to the actual unit calibrated in the lab and it is part of an identical setup , except for the antenna and ground plane which only exist in the field . physically , the receiver consists of a metal enclosure that houses the first - stage low - noise amplifier ( lna ) , noise references for relative calibration , additional stages of amplification , filtering , and conditioning , and electronics for remote measurement of the antenna reflection coefficient . the receiver operates at @xmath14c at all times in order to keep the noise and reflection characteristics of the lna stable during calibration and sky observations . temperature stability to better than @xmath15c is achieved using a thermal controller that reads out a thermistor mounted inside the receiver . based on the reading , the controller acts on a hot / cold plate attached to the bottom of the receiver to compensate for temperature drifts . from the output of the receiver the rf signal is taken along a coaxial cable into an equipment rack where it is amplified and filtered by back - end electronics . the signal is then sent to a pc - based @xmath16-ms / s analog - to - digital converter ( adc ) that samples with @xmath17-bit resolution . finally , blocks of @xmath18 time samples are fourier transformed to the frequency domain to obtain spectra with @xmath19 points and @xmath20-khz resolution using a four - term blackman - harris window function . the lna input switches continuously between the antenna and an internal calibration position using switch sw2 in figure [ figure_block_diagram ] . when in the calibration position , the lna receives sequentially two noise levels : 1 ) load and 2 ) load + noise source . in the load state , the noise comes from the output of a @xmath21-db attenuator , and in the load + noise source state the active noise source connected to the input of the attenuator is turned on for additional noise . the noise from the antenna only represents a few percent of the total power at the receiver output . most of the output power is due to out - of - band noise injected below @xmath21 mhz as part of the conditioning stage . this injected noise remains constant during the input switching and provides a stable signal level to the back - end electronics and digitizer , improving linearity and dynamic range in the measurements . each of the three input levels are measured for @xmath22 seconds , which corresponds to an antenna duty cycle of @xmath23% . a total of @xmath24 spectra are accumulated for each position . an implicit assumption of this switching scheme is that the amplification chain remains stable within each @xmath25-second cycle . the circuitry for measurement of antenna reflections , shown in figure [ figure_block_diagram ] , functions as a remote calibration unit for a vector network analyzer ( vna ) . it is centered around a four - position mechanical rf switch where three of the ports are connected to open , short , and matched calibration standards , and the fourth port serves as a pass - through to the antenna . the efficiency of the instrument is highest below @xmath26 mhz . therefore , the calibration described in the rest of this paper was conducted in the range @xmath2 mhz . here we summarize the calibration strategy used by edges , which is based on the method introduced in @xcite . for each three - position cycle of the receiver , the power spectral density ( psd ) from the antenna ( @xmath27 ) , load ( @xmath28 ) , and load + noise source ( @xmath29 ) , are used to compute an initial _ uncalibrated _ antenna temperature , @xmath30 where @xmath31 and @xmath32 represent realistic assumptions for the noise temperatures of the load and noise source , respectively . this computation serves to calibrate out the time - dependent system gain , which includes the complex bandpass of the filters , amplifiers , cables , and adc . all the parameters in equation ( [ equation_uncalibrated_temperature ] ) , and in what follows , are frequency - dependent . we do not explicitly show this dependence for simplicity of notation . to derive the expression for calibration of @xmath33 it is necessary to write the psds in equation ( [ equation_uncalibrated_temperature ] ) in terms of the specific instrument response contributions . the psd for the antenna is given by : @xmath34 , \label{equation_power_antenna}\end{aligned}\ ] ] with @xmath35 here , @xmath36 corresponds to the calibrated antenna temperature . the quantities @xmath37 and @xmath38 represent the system gain referenced to the receiver input , and the receiver noise offset , respectively . @xmath39 is the reflection coefficient of the antenna and @xmath40 is the reflection coefficient of the receiver , both referenced to a @xmath41-@xmath42 system impedance . the temperatures @xmath43 , @xmath44 , and @xmath45 are called noise wave parameters following the formalism introduced by @xcite . they are associated with the noise emitted by the lna input toward the antenna . this noise is reflected back due to imperfect impedance match and re - enters the receiver with phase @xmath46 . the @xmath43 temperature represents the portion of input noise that is uncorrelated with the noise at the lna output , while @xmath44 and @xmath45 are components of the correlated portion . the psds for the internal load and load + noise source follow the same form as equation ( [ equation_power_antenna ] ) . however , at this stage we assume that the reflection coefficients of the load and noise source are zero . in reality , they are not zero but very low ( @xmath47 db ) . thus , these psds are modeled as : @xmath48,\label{equation_power_internal_calibrators1}\\ p_{\text{l+ns } } & = g^*\left[\left(t_{\text{l } } + t_{\text{ns}}\right)\left(1-|\gamma_{\text{rec}}|^2\right ) + t_0^*\right].\label{equation_power_internal_calibrators2}\end{aligned}\ ] ] in these equations , the system gain ( @xmath49 ) and noise offset ( @xmath50 ) are not exactly the same as in equation ( [ equation_power_antenna ] ) because the noise from the internal references is injected at sw2 instead of at the receiver input itself . this aspect , in addition to the assumption for the reflection coefficients , is accounted for below with the introduction of two new parameters fitted during calibration that are able to absorb these small effects . plugging the three psd definitions into equation ( [ equation_uncalibrated_temperature ] ) results in the following identity : @xmath51 \nonumber\\ + \ ; & t_{\text{unc}}\left[\frac{|\gamma_{\text{ant}}|^2|f|^2}{\left(1-|\gamma_{\text{rec}}|^2\right)}\right ] \nonumber\\ + \ ; & t_{\cos}\left[\frac{|\gamma_{\text{ant}}||f|}{\left(1-|\gamma_{\text{rec}}|^2\right)}\cos\alpha\right ] \nonumber\\ + \ ; & t_{\sin}\left[\frac{|\gamma_{\text{ant}}||f|}{\left(1-|\gamma_{\text{rec}}|^2\right)}\sin\alpha\right ] . \label{equation_main_identity}\end{aligned}\ ] ] this equation establishes the relationship between the uncalibrated antenna temperature @xmath33 , computed with equation ( [ equation_uncalibrated_temperature ] ) , and the calibrated antenna temperature @xmath36 at the receiver input . more generically , @xmath36 represents the calibrated noise temperature of any device under measurement at the receiver input . any losses in the device under measurement have to be corrected for externally . the quantities @xmath52 and @xmath53 introduced on the left - hand side of equation ( [ equation_main_identity ] ) represent a scale and an offset that correct the first - order assumptions used for @xmath54 and @xmath32 in equation ( [ equation_uncalibrated_temperature ] ) . they also account for the small path difference between the internal calibration position of sw2 and the receiver input . finally , they also account to first order for the non - zero reflection coefficient of the internal load and noise source . unaccounted higher - order effects are expected to be negligible due to the low reflection coefficients of these devices . the reflection coefficient of the receiver input and the antenna are measured directly with a vna . @xmath55 and @xmath46 ( equations ( [ equation_f ] ) and ( [ equation_alpha ] ) ) are computed from these coefficients . therefore , the remaining calibration task consists of estimating the scale @xmath52 , offset @xmath53 , and the noise wave parameters @xmath43 , @xmath44 , and @xmath45 , in order to apply equation ( [ equation_main_identity ] ) . to solve for these five frequency - dependent quantities we conduct laboratory measurements of four absolute calibration standards connected to the input of the receiver in place of the antenna . following @xcite , the four calibrators are : 1 ) an ambient load , 2 ) a hot load , 3 ) a long ( @xmath56 @xmath57-m ) open - ended coaxial cable , and 4 ) the same long cable but short - circuited at its far end . the ambient and hot loads provide the main temperature references , while the open and shorted cable produces ripples in its spectra that manifest the noise properties of the receiver and enable the estimation of the noise wave parameters . the specific quantities that need to be measured for receiver calibration are : 1 . the uncalibrated temperature spectra for each calibrator @xmath58@xmath59 , @xmath60 , @xmath61 , @xmath62@xmath63 via equation ( [ equation_uncalibrated_temperature ] ) , 2 . the physical or noise temperature of the calibrators @xmath58@xmath64 , @xmath65 , @xmath66 , @xmath67@xmath63 , and 3 . the reflection coefficient of the calibrators @xmath58@xmath68 , @xmath69 , @xmath70 , @xmath71@xmath63 . the subscripts @xmath72 , @xmath73 , @xmath74 , and @xmath75 correspond to the ambient load , hot load , open cable , and shorted cable , respectively . and [ equation_available_power_gain ] ) . the nomenclature given in the figure corresponds to the device operating as a hot load.,scaledwidth=48.0% ] in this section we report details of the laboratory measurements . we begin by describing the calibrator used as ambient and hot load . we then discuss the measurements of spectra , physical temperature , and reflection coefficient of the calibrators . we also introduce the main sources of uncertainty for each calibration measurement , but defer detailed discussion to section [ section_uncertainty_propagation ] . we conclude the section with the calculation and modeling of the five frequency - dependent receiver calibration quantities . the ambient and hot loads used for receiver calibration are implemented as a single device , depicted in figure [ figure_hot_load_device ] . it consists of an rf termination of @xmath41-@xmath42 nominal impedance connected to an @xmath57-cm semi - rigid cable inside a thermally insulated metal enclosure . when acting as an ambient load the device operates at room temperature ( @xmath56 @xmath76 k ) and when used as a hot load the termination is heated up to @xmath56 @xmath16 k by powering a resistor that is thermally connected to the termination . the effective noise temperature of the device when operating as a hot load ( @xmath77 ) is related to the physical temperature of the termination ( @xmath78 ) and the physical temperature of the cable ( @xmath79 ) by @xmath80 where @xmath81 is the available power gain of the assembly , defined as @xcite : @xmath82 in this equation , @xmath83 and @xmath84 are two of the s - parameters of the cable , with port 1 attached to the rf termination and port 2 corresponding to the output connector of the device . @xmath85 is the reflection coefficient of the termination alone and @xmath69 is the reflection coefficient of the device as a whole . the uncalibrated temperature spectra of the absolute calibrators are measured by connecting each calibrator to the receiver input in place of the antenna . these measurements involve the same internal three - position switching as with the antenna , where in each @xmath25-second cycle the setup measures the psd of the calibrator and the two internal noise references . after the measurements , equation ( [ equation_uncalibrated_temperature ] ) is applied offline to produce the uncalibrated temperature spectra . the noise temperatures assumed for the internal load and noise source at this step are @xmath86 k and @xmath87 k , constant across frequency . the final uncalibrated spectra , @xmath59 , @xmath60 , @xmath61 , and @xmath62 , are obtained by averaging @xmath88 hours of data from each calibrator . this is done in order to reduce thermal noise . residual noise is the main source of uncertainty in these quantities . the physical temperatures , @xmath64 , @xmath78 , @xmath66 , and @xmath67 , are measured using thermistors attached to the termination of the ambient / hot load and to the long cable . these measurements are conducted in parallel to the spectra measurement of each calibrator and with a similar time resolution . they are averaged in time to have a direct correspondence with the spectra averages . uncertainties in these measurements are dominated by potential inaccuracies in the resistance - to - temperature model used for the thermistors and unaccounted thermal gradients in the calibrators . the reflection coefficients and s - parameters required in the calibration are measured with a vna . in each measurement , hundreds of traces are averaged to reduce the noise to levels so that they are an insignificant source of uncertainty . the calibrators are passive devices and , thus , they are measured using a typical vna power of @xmath89 dbm ( @xmath4 mw ) . in contrast , the receiver input has to be measured at @xmath90 dbm to avoid saturating the active electronics designed for low - level signals . lower vna power results in higher measurement noise , but we compensated through averaging . we calibrate the vna immediately before every measurement and , thus , do not assume long - term vna stability . the main uncertainty associated with these measurements arises from imperfect vna calibration . the isolation between the inputs of each of the mechanical switches ( sw1 and sw2 in figure [ figure_block_diagram ] ) is about @xmath91 db . however , we assume that the isolation is perfect , which has a negligible impact on the calibration . the switching repeatability is assessed through high - precision s - parameter measurements and the scatter is found to be within the measurement uncertainties . we model and fit the uncalibrated temperature spectra , the reflection coefficients , and the s - parameters , to avoid propagating measurement noise to the fiducial calibration quantities and to bring all the measurements to the same frequency resolution . due to the smooth frequency behavior of the ambient and hot load spectra , and of the s - parameters of the semi - rigid cable , we model these measurements as polynomials in frequency . on the other hand , we use fourier series in frequency to model the spectra of the open and shorted cable , as well as all the reflection coefficients . fourier series are more efficient than polynomials at capturing the higher frequency structure encountered in these measurements . the model parameters are computed through least squares using qr decomposition for better numerical stability . the fit residuals for the spectra are noise - like . they are accounted for as a source of uncertainty in the mc analysis of section [ section_uncertainty_propagation ] . for the reflection coefficients and s - parameters , the rms residuals are @xmath92 db in magnitude and @xmath93 in phase their effects are negligible compared to that from the uncertainty in vna calibration , addressed in section [ section_uncertainty_propagation ] . : scale , @xmath53 : offset , @xmath43 : uncorrelated noise wave , and @xmath94 , @xmath95 : components of the correlated noise wave . they satisfy equation ( [ equation_main_identity ] ) when it is evaluated using the lab measurements of the absolute calibrators . all the quantities in this figure are used as the fiducial quantities in the uncertainty propagation analysis discussed in section [ section_uncertainty_propagation].,scaledwidth=48.0% ] the frequency - dependent calibration quantities ( @xmath52 , @xmath53 , @xmath43 , @xmath44 , and @xmath45 ) are computed from the modeled laboratory measurements through an iterative process . specifically , in each iteration the scale and offset are computed as @xmath96 here , @xmath77 is the noise temperature of the hot load from equation ( [ equation_cable_loss ] ) , @xmath64 is the physical temperature of the ambient load , and @xmath97 are @xmath98 are the calibrated temperature spectra of the ambient and hot loads from equation ( [ equation_main_identity ] ) evaluated at the @xmath99 iteration . the initial values are @xmath100 and @xmath101 . the noise wave parameters ( @xmath43 , @xmath44 , and @xmath45 ) also take an initial value of zero , and in subsequent iterations they are modeled as polynomials in frequency . the polynomial coefficients are computed through a least squares fit to equation ( [ equation_main_identity ] ) evaluated using the measurements of the open and shorted cable and the current values of @xmath52 and @xmath53 . this process converges in three iterations , after which @xmath52 and @xmath53 are also modeled as polynomials in frequency . to find the optimum number of terms in the polynomials , the calibration quantities are modeled with increasing number of terms until the rms difference between the physical temperature and the calibrated temperature spectra of the calibrators themselves reaches a minimum . it is found that seven terms are needed to model each of the five calibration quantities . figure [ figure_calibration_quantities ] presents the derived calibration quantities , computed as just described , along with the reflection coefficient of the receiver input . section [ section_calibration_measurements ] presented the computation of the receiver calibration quantities from laboratory measurements and introduced potential sources of uncertainty . in this section , we discuss these uncertainties in more detail and assess their impact on the final calibrated antenna temperature spectrum through a set of monte carlo ( mc ) simulations . we begin by outlining our mc uncertainty propagation pipeline and then motivate our choices for the measurement uncertainties used in it . figure [ figure_mc_diagram ] presents a diagram of our uncertainty propagation simulation scheme . from left to right , first , an ideal simulated input antenna temperature @xmath102 is computed by convolving a sky model with an antenna beam model . then , this antenna temperature is _ uncalibrated _ by solving equation ( [ equation_main_identity ] ) for @xmath33 using the fiducial derived calibration quantities and receiver reflection coefficient from section [ section_calibration_measurements ] , as well as a model for the reflection coefficient of the antenna . next , inside the block labeled mc , the antenna temperature is recalibrated with equation ( [ equation_main_identity ] ) but using calibration quantities and reflection coefficients that potentially differ from their fiducial values due to simulated errors or noise estimated from our measurement uncertainties . this step is conducted for thousands of mc realizations of the perturbed calibration measurements . the output antenna temperatures are labeled @xmath103 . finally , for each simulated output spectrum , we fit and subtract a polynomial to match the foreground subtraction procedure applied to actual measurements . we use the resulting residuals to quantify the magnitude of the propagated calibration errors . our present interest is in understanding the role of the receiver in the experimental error budget . hence , we restrict our analysis here to only the effects of uncertainties from calibration of the receiver . our uncertainty propagation simulations assume perfect removal of antenna losses and beam chromaticity . in addition , the simulations do not address potential changes in receiver performance during field operations due to , for instance , unaccounted temperature gradients or component aging . field performance will be discussed in forthcoming papers that present the analysis of sky measurements . the rest of this section provides details of the simulations . the input antenna temperature is modeled by the convolution of a sky model with an antenna beam model : @xmath104 where @xmath105 is the sky model , @xmath106 is the beam model , @xmath107 and @xmath108 are the zenith and azimuth angles respectively , and @xmath42 represents coordinates above the horizon . below @xmath109 mhz , the sky brightness temperature is dominated by foregrounds that are more than four orders of magnitude stronger than the expected cosmological signal @xcite . since the dominant effects of receiver mis - calibration and uncertainty operate on the total sky signal , it is acceptable to neglect the cosmological @xmath0-cm signal in our sky model and only include the foreground contribution . we model the foregrounds using the global sky model ( gsm ) described in @xcite . for the antenna beam model , we use the simple frequency - independent azimuthally - symmetric expression used by @xcite , @xmath110 real beams are known to be more complex than this expression , introducing spectral structure into the antenna temperature that could limit the detection of the cosmological signal @xcite . however , in these simulations it is assumed that the chromatic effects of the beam are perfectly known and removed . the sensitivity of the instrument to errors in receiver calibration is a function of the antenna temperature . ground - based instruments such as edges observe the sky , and thus foregrounds , continuously drifting over the antenna instead of conducting deep integrations on a single sky region . therefore , the strength of the antenna temperature used for science analysis will vary as different parts of the sky drift through the beam . to account for the different foreground levels , we explore two scenarios in the simulations : with the 1 ) lowest and 2 ) highest foreground contamination available at the edges observation latitude . these cases are labeled _ quiet _ and _ loud _ sky , respectively , and their convolution with the beam is shown in figure [ figure_antenna_temperature ] . ) ) with the foreground sky model ( gsm , from @xcite ) . the quiet and loud skies correspond to high galactic latitudes and the galactic plane transit , respectively.,scaledwidth=48.0% ] since we are mainly interested in understanding the effect of errors in receiver calibration , we assume a simple model for the antenna reflection coefficient . our model for the reflection magnitude is flat in frequency with a level of @xmath111 db . we model the phase as a linear decrease by @xmath112 between @xmath113 and @xmath114 mhz . this is realistic and corresponds to a delay of @xmath115 ns . this model reproduces , to first order , the measured edges antenna properties and is generally consistent with the performance achievable from a dipole - based antenna operating over an octave bandwidth . clcc + & & + + @xmath59 & ambient load & 66 mk & + @xmath60 & hot load & 66 mk & ( i ) + @xmath61 & open cable & 95 mk & + @xmath62 & shorted cable & 95 mk & + + & & + + @xmath64 & ambient load & & + @xmath78 & hot load termination & @xmath116 mk & ( ii ) + @xmath66 & open cable & & + @xmath67 & shorted cable & & + + & & + + @xmath117 & ambient load & & + @xmath118 & hot load & & + @xmath119 & open cable & 0.0001 & ( iii ) + @xmath120 & shorted cable & & + @xmath121 & receiver & & + @xmath122 & antenna & & + + & & + + @xmath123 & ambient load & & + @xmath124 & hot load & & + @xmath125 & open cable & 0.015@xmath126 & ( iv ) + @xmath127 & shorted cable & & + @xmath128 & receiver & & + @xmath129 & antenna & & + + & & + + @xmath130 & semi - rigid cable & 0.015 & ( v ) + here , we describe the uncertainties assigned to each of the calibration measurements from section [ section_calibration_measurements ] , as well as the antenna reflection coefficient . these uncertainties are applied inside the mc block of the uncertainty propagation , to create slightly perturbed realizations of the calibration parameters . a summary of the nomenclature and values is presented in table [ table_nomenclature ] . in the top block of table [ table_nomenclature ] , we show the @xmath131 uncertainties associated with each of the uncalibrated spectra of the absolute calibration loads ( @xmath59 , @xmath60 , @xmath61 , @xmath62 ) . the uncertainties are set equal to the rms of the residuals from polynomial or fourier series fits to the actual measured spectra . in all cases , the residuals are noise - like . the residual rms for the ambient and hot load measurements is @xmath132 mk , while for the open / shorted cable measurements it is @xmath133 mk . for our mc realizations of the perturbed spectra , we begin with the models for the fiducial spectra and add to each frequency channel realizations of gaussian noise , uncorrelated from channel to channel , drawn from the assigned uncertainties . the second block of table [ table_nomenclature ] shows the uncertainties associated with the physical temperatures ( @xmath64 , @xmath78 , @xmath66 , @xmath67 ) of the calibrator sources . we measure all the physical temperatures with thermistors and a resistance - to - temperature model . the accuracy of the setup is estimated to be @xmath116 mk , hence we use that value for the uncertainty of all physical temperature measurements . for the mc realizations of the calibrator temperatures , we draw from gaussian distributions centered at the fiducial values with a @xmath131 width of @xmath116 mk . the third and fourth blocks of table [ table_nomenclature ] summarize the uncertainties applied to the amplitudes and phases of reflection coefficients , respectively . our mc modeling propagates the reflection coefficient uncertainties for : 1 ) the absolute calibration loads ( @xmath68 , @xmath69 , @xmath70 , @xmath71 ) , 2 ) the receiver input ( @xmath40 ) , and 3 ) the antenna ( @xmath39 ) . although the antenna reflection coefficient is not measured as part of the receiver calibration , it is tightly coupled to final receiver performance and we account for its uncertainty here . we also consider the uncertainty ( listed in the fifth block in table [ table_nomenclature ] ) on the transmission coefficient , @xmath130 , of the semi - rigid cable inside the hot load because it is relevant in the computation of the effective noise temperature from the device ( equation ( [ equation_available_power_gain ] ) ) . the main uncertainties in reflection coefficient measurements arise from uncertainty in vna calibration . vna calibration involves measuring three reflection standards at the vna measurement plane : an open standard , a short standard , and a @xmath41-@xmath42 standard . as part of the edges efforts to increase the accuracy in calibration measurements , we improved upon the manufacturer - specified vna tolerances through more accurate modeling of the @xmath41-@xmath42 standard @xcite . with the improved model for the standard , the additive uncertainty in the magnitude of measured reflection coefficients was reduced to a @xmath131 linear voltage ratio of @xmath134 , which is equivalent to @xmath135 db for a reference reflection of @xmath111 db . in addition to residual modeling errors , sub - dominant effects that contribute to this uncertainty include vna drifts and imperfect connection repeatability . thus , the uncertainties assumed in the mc simulation for the reflection magnitudes are modeled as an additive contribution , fully correlated across the band , with an amplitude drawn from a gaussian with a @xmath131 width of @xmath134 . figure [ figure_reflection_accuracy ] presents a verification of the magnitude accuracy of our vna measurements using open - ended rf attenuators . in the figure , we show reflection coefficient measurements of an attenuator compared to forecasts based on a direct dc measurement of the attenuator resistance . in particular , for @xmath136-db reflection , the direct agreement is better than @xmath137 db and remains better than @xmath138 db even if we assume a pessimistic error in the resistance measurement . we model reflection phase uncertainty using the form @xmath139 found in typical vna specifications . here , @xmath140 is a constant that represents the uncertainty value for the special case of a total reflection ( @xmath141 ) . this model captures the fact that it is more difficult to determine the phase for smaller reflections than for larger reflections . in our simulations , we use @xmath142 and model phase uncertainty as an additive contribution with a @xmath131 width given by @xmath139 . for reference , this corresponds to a @xmath143 range of @xmath144 at the @xmath111-db magnitude of our antenna reflection coefficient model . as was the case for the reflection magnitude , the additive contribution to the phase is modeled as fully correlated across the band . this is a realistic first - order approximation for our vna measurements of @xmath116-mhz bandwidth . the uncertainty of transmission coefficient ( @xmath130 ) measurents also benefits from the improvements in the measurement of reflection coefficient . we assign @xmath131 uncertainty of @xmath145 to the transmission coefficient of the semi - rigid cable in the hot load calibrator and model it in the same way as the reflection magnitudes . - and @xmath146-db respectively ) after a high - accuracy vna calibration ( black line ) , and comparing these measurements with the expectations for the attenuators from their dc resistance ( solid red line ) . the vna calibration requires the dc resistance of the @xmath41-@xmath42 calibration load . this resistance , as well as that of the attenuators , can be measured to better than @xmath147 @xmath42 . for reference , in the figure , the gray lines represent a pessimistic @xmath138-@xmath42 error in the calibration load resistance and the dashed red lines represents a @xmath138-@xmath42 error in the attenuator resistance . even when assuming this pessimistic resistance error , the accuracy is better than @xmath148 db at a reflection level of @xmath149 db , and better than @xmath150 db at @xmath151 db.,scaledwidth=48.0% ] crrrrrrrr + @xmath59 & 5 & 4 & 4 & 4 & 3 & 3 & 2 & 2 + & 25 & 22 & 20 & 18 & 15 & 12 & 6 & 1 + + @xmath60 & 5 & 5 & 4 & 4 & 3 & 2 & 1 & 1 + & 23 & 22 & 20 & 19 & 17 & 14 & 9 & 2 + + @xmath64 & 381 & 357 & 56 & 1 & & & & + & 1852 & 340 & 63 & & & & & + + @xmath78 & 398 & 258 & 40 & 1 & & & & + & 1958 & 240 & 47 & 1 & & & & + + @xmath66 & 1 & 1 & 1 & 1 & 1 & & & + & 2 & 1 & 1 & 1 & 1 & & & + + @xmath67 & 1 & 1 & 1 & 1 & & & & + & 1 & 1 & 1 & 1 & 1 & & & + + @xmath117 & 19 & 19 & 7 & 5 & 1 & & & + & 94 & 27 & 26 & 4 & 2 & & & + + @xmath123 & 60 & 43 & 41 & 9 & 3 & 1 & & + & 228 & 136 & 27 & 22 & 2 & 1 & & + + @xmath118 & 22 & 16 & 9 & 5 & 1 & & & + & 136 & 42 & 36 & 3 & 2 & & & + + @xmath124 & 74 & 38 & 35 & 10 & 2 & 1 & & + & 320 & 198 & 27 & 26 & 2 & 1 & & + + @xmath119 & 5 & 4 & 4 & 2 & 1 & 1 & 1 & 1 + & 5 & 3 & 3 & 2 & 1 & 1 & 1 & 1 + + @xmath125 & 2 & 2 & 2 & 2 & 1 & 1 & 1 & + & 2 & 2 & 2 & 2 & 1 & 1 & 1 & + + @xmath120 & 4 & 4 & 3 & 2 & 2 & 1 & 1 & 1 + & 4 & 4 & 3 & 2 & 2 & 1 & 1 & 1 + + @xmath127 & 1 & 1 & 1 & 1 & 1 & 1 & 1 & + & 1 & 1 & 1 & 1 & 1 & 1 & 1 & + + @xmath121 & 15 & 15 & 15 & 14 & 10 & 5 & 5 & 1 + & 58 & 57 & 57 & 56 & 42 & 23 & 19 & 5 + + @xmath128 & 47 & 41 & 35 & 27 & 25 & 20 & 6 & 6 + & 176 & 160 & 138 & 118 & 105 & 81 & 27 & 24 + + @xmath130 & 104 & 68 & 11 & & & & & + & 520 & 63 & 13 & & & & & + + @xmath122 & 32 & 12 & 12 & 12 & 9 & 6 & 3 & 1 + & 112 & 33 & 33 & 32 & 26 & 15 & 10 & 3 + + @xmath129 & 36 & 33 & 29 & 28 & 23 & 16 & 7 & 4 + & 100 & 92 & 79 & 73 & 63 & 46 & 19 & 12 + + all & 549 & 443 & 94 & 42 & 36 & 26 & 12 & 8 + & 2803 & 504 & 178 & 146 & 128 & 96 & 37 & 27 final cosmological parameter estimation for edges and similar experiments is performed by simultaneously fitting a signal model with a parametrized foreground model . in this step , the foreground model consists of a low - order polynomial or set of basis functions . these functions are able to fit the foreground and beam chromaticity structure in the measured spectra , but also tend to absorb some of the expected @xmath0-cm signal . calibration errors that are similar to foreground structures i.e . those that exhibit large spectral coherence will be associated with the foreground model terms during the final parameter estimation and will have little additional impact on the @xmath0-cm signal estimation . calibration errors that vary relatively rapidly in frequency , on the other hand , will not be fit by the foreground model and will yield more interference with the signal estimation . in order to take this effect into account in our analysis here , we characterize the output antenna temperatures from our uncertainty propagation after fitting and removing a foreground model . we use the polynomial model given by : @xmath152 this model was introduced in @xcite as an efficient expression for removing foreground and beam effects . in the context of this paper , it shares the flexibility of a generic polynomial but is more efficient for residuals with a predominant @xmath153 power - law behavior . it is chosen as a model that could help remove the foreground and residuals from calibration simultaneously with few terms . in this section , we report the results of our uncertainty propagation . the effect of each calibration uncertainty source is first analyzed individually by holding all other sources of uncertainties to zero . then , all the uncertainties are propagated simultaneously to examine their combined impact . we characterize the significance of the calibration uncertainties by calculating the rms of residuals to the foreground model fit for differing numbers of polynomial terms in the foreground model , exploring between zero and seven terms . distributions of rms values are produced from thousands of mc repetitions . we use @xmath154 repetitions when a single effect is studied in isolation and @xmath155 when all the effects are considered simultaneously . these rms distributions are characterized in terms of their @xmath5 upper bounds , rms@xmath156 . the results are presented in table [ table_residuals ] . the table lists the uncertainty sources and the corresponding @xmath157 levels after removing the foreground model , with increasing numbers of polynomial terms shown in separate columns from left to right . the two rows next to each source represent the residuals for the quiet ( top ) and loud ( bottom ) skies respectively . for residuals below @xmath4 mk , the cells are left empty . in the case of the the open and shorted cable spectra ( @xmath61 , @xmath62 ) the residuals are always below @xmath4 mk and hence not shown . figure [ figure_residuals ] illustrates typical residuals for a quiet sky . each row corresponds to a different source of uncertainty . the left column shows two representative cases ( in blue and red ) drawn arbitrarily from our mc simulations before removing any term . the right column shows the residuals for the same two cases after fitting and removing equation ( [ equation_model2 ] ) with five terms . several trends can be identified from table [ table_residuals ] . first and as expected , for most error sources and polynomial orders the residuals are larger when observing a loud sky . this motivates the preference of low - foreground observations for estimation of the cosmological signal , or the downweighting of strong - foreground regions as suggested , for instance , in @xcite . the table also makes evident the low sensitivity of the calibration to errors in measurements of the open and shorted cable . as stated before , the residuals for their spectra ( @xmath61 , @xmath62 ) remain consistently below @xmath4 mk and are not shown in the table . for their physical temperatures ( @xmath66 , @xmath67 ) the initial residuals are @xmath158 mk and for the reflections ( @xmath70 , @xmath71 ) they are @xmath159 mk . at five terms they all reach @xmath4 mk or less . the largest initial residuals occur for errors in the physical temperatures of the ambient and hot loads ( @xmath64 , @xmath78 ) which , for the loud sky , reach up to @xmath160 k. however , as shown in figure [ figure_residuals ] , they strongly follow a power law and can be removed with few terms . specifically , three terms are needed to reduce the residuals to @xmath161 mk for our foreground model . the same applies to the transmission coefficient magnitude of the cable inside the ambient / hot load ( @xmath130 ) since this quantity is involved in the computation of the hot load noise temperature ( section [ section_absolute_hot_load ] ) . the initial impact of thermal noise in the ambient and hot spectra ( @xmath59 , @xmath60 ) is at the @xmath162-mk and @xmath163-mk level for the quiet and loud skies respectively . these residuals decrease slowly as terms are added . more than five terms would be needed to reach @xmath4 mk . for the reflections of these loads ( @xmath68 , @xmath69 ) the initial residuals are significantly higher ( up to @xmath164 mk ) . however , five terms are sufficient to reduce them to @xmath161 mk due to their smooth spectral shape . for the receiver and antenna reflection coefficients ( @xmath40 , @xmath39 ) the initial residuals are comparable to those for the ambient and hot loads . however , it takes more than seven terms to reduce them to @xmath161 mk . with three or more terms removed these residuals are the largest among all the effects . the similarity in behavior between the two sources is not surprising considering their strong interaction in the calibration equations ( section [ section_calibration_formalism ] ) . finally , the last rows of the table present the residuals when all the uncertainties are combined in a comprehensive monte carlo analysis . as expected , initially they are larger than those from individual sources , amounting to about @xmath165 and @xmath166 k for the quiet and loud skies respectively . for the quiet sky they drop to @xmath167 mk with five terms . in the full uncertainty propagation , errors in reflection coefficients are assumed uncorrelated . however , a test was also conducted with correlated errors where , in any given mc repetition , the same error was applied to all the magnitudes / phases . this exercise simulates a case where all these measurements are affected by the same vna bias . the results are qualitatively and quantitatively very similar to those shown in the table , thus not shown for brevity . as shown in @xcite and @xcite through simulations , over a frequency range similar to edges high - band and with sub mk channel noise , it is possible to probe a significant range of eor models when the foreground and instrument spectral response are simultaneously modeled with less than six polynomial terms . some recent cosmological models yield large or late absorption features , or rapid reionization histories within the edges high - band range @xcite . these models should also be accessible with a similar instrument performance . the edges beam chromaticity has been modeled by @xcite to be less than @xmath4 mk at quiet sky locations after a fit to equation ( [ equation_model2 ] ) with five terms . ideally , this number of terms and residual level should not increase due to systematics from receiver calibration . therefore , in order to avoid being limited by receiver systematics , the natural goal for the receiver calibration performance corresponds to keeping the residuals below @xmath4 mk after a fit to equation ( [ equation_model2 ] ) with five terms . this goal is already being met for most of the measurements . the three exceptions correspond to 1 ) the spectra of the ambient and hot loads , 2 ) the reflection coefficient of the receiver , and 3 ) the reflection coefficient of the antenna . simulations indicate that reducing the impact of noise in the ambient and hot load spectra to @xmath161 mk with five terms requires lowering the noise level to @xmath168% of its nominal value . with the current digitization efficiency , this is equivalent to integrating for more than @xmath169 days , since the nominal value is the result of one day of integration . meeting this requirement is realistic and planned for future calibration revisions . the situation is more challenging for the reflection coefficients . it would be necessary to reduce the magnitude and phase uncertainties to @xmath170% and @xmath171% of the nominal values , respectively . achieving such tight tolerances is an area of active research and we are identifying promising techniques . for instance , the current uncertainty levels could be reduced by incorporating measurements of custom reflection standards , such as well characterized air - dielectric coaxial lines , into the vna calibration process @xcite . cccccccccc + @xmath122 & ( a ) & 32 & 12 & 12 & 12 & 9 & 6 & 3 & 1 + & ( b ) & 29 & 7 & 6 & 6 & 5 & 3 & 1 & 1 + & ( c ) & 21 & 12 & 12 & 11 & 9 & 6 & 3 & 1 + & ( d ) & 34 & 12 & 11 & 6 & 3 & 1 & & + & ( e ) & 18 & 6 & 6 & 6 & 5 & 3 & 1 & 1 + & ( f ) & 19 & 7 & 6 & 3 & 2 & & & + + @xmath129 & ( a ) & 36 & 33 & 29 & 28 & 23 & 16 & 7 & 4 + & ( b ) & 18 & 17 & 16 & 15 & 13 & 8 & 4 & 1 + & ( c ) & 35 & 32 & 29 & 27 & 23 & 15 & 7 & 4 + & ( d ) & 36 & 33 & 25 & 16 & 6 & 3 & & + & ( e ) & 18 & 17 & 15 & 15 & 12 & 7 & 4 & 1 + & ( f ) & 18 & 17 & 12 & 9 & 2 & 1 & & an alternative approach to reduce the sensitivity of edges to errors in reflection measurements consists of improving the impedance match between the antenna and receiver . as stated in section [ section_calibration_formalism ] , our reflection coefficients are referenced to the @xmath41-@xmath42 system impedance . therefore , improvement involves bringing both impedances closer to @xmath41-@xmath42 or , equivalently , lowering both reflection coefficients . table [ table_antenna_residuals ] presents five simulation examples with potential for reducing sensitivity to errors in the antenna reflection coefficient ( magnitude , @xmath122 , and phase , @xmath129 ) for different levels of reflection from the antenna and receiver . the computations assume a quiet sky and the nominal uncertainties of table [ table_nomenclature ] . case ( a ) corresponds to the same residuals shown in table [ table_residuals ] , presented again for reference . in case ( b ) , the magnitude of the receiver reflection coefficient is lowered from @xmath172 to @xmath90 db . this change produces a reduction in residuals from the antenna reflection magnitude and phase by a factor of about two for almost any number of terms . in case ( c ) , the nominal settings have been modified by lowering the magnitude of the antenna reflection coefficient from @xmath111 to @xmath136 db . relative to ( a ) , this scenario produces an improvement in the initial residuals of the magnitude , which decreases from @xmath173 to @xmath0 mk . however , when introducing terms in the model , the residuals become almost identical to those in the nominal case . in case ( d ) the delay of the antenna has been reduced from @xmath174 ns to @xmath175 ns , i.e. , the linear change in phase between @xmath113 and @xmath114 mhz has been reduced from @xmath112 to @xmath176 . this change has no impact on the initial residuals , but as more terms are introduced the improvement becomes significant . with five terms , the magnitude residuals are reduced from @xmath177 to @xmath4 mk , and in phase they are reduced from @xmath178 to @xmath179 mk . case ( e ) corresponds to the combination of cases ( b ) and ( c ) , i.e. , both reflection magnitudes have been reduced . case ( f ) combines all the suggested improvements and reduces the residuals to the lowest figures . they start at 19 and 18 mk for the magnitude and phase with no terms removed , and converge to @xmath161 mk with five terms . the improvements described above have the potential to reduce sensitivity to errors . at the same time , their implementation has to consider tradeoffs typically encountered in wideband instrument design . for instance , reducing the antenna reflection coefficient can have an impact on the spectral smoothness of the beam . similarly for the receiver , it is necessary to balance its input reflection coefficient with its noise performance . in addition , the phase of the antenna reflection depends on the physical dimensions of the antenna and on the length of any transmission line between the antenna and the reflection measurement plane . we are currently investigating these alternatives to converge to an optimal instrumental solution . the results presented thus far correspond to @xmath133% residuals in the range @xmath2 mhz . on top of refinements in the instrumentation , a decrease in residuals could be achieved by reducing the bandwidth used in the science analysis motivated , for instance , by the interest in probing late ( @xmath180 ) reionization transitions @xcite . here we show the improvement in performance when reducing the bandwidth to @xmath91 mhz by computing residuals in the range @xmath181 mhz ( @xmath182 ) from the mc simulations of section [ section_uncertainty_propagation ] . discarding a @xmath6-mhz section at the low - frequency end results in less spectral structure , and in structure with lower amplitude due to the lower sky temperature in the remaining band . as depicted in figure [ figure_residuals ] , this especially occurs for key sources of uncertainty such as the antenna and receiver reflection coefficients . with a quiet sky and after removing five terms in the range @xmath181 mhz , the residuals for the ambient and hot load spectra are reduced to @xmath4 mk . for the magnitude and phase of the receiver reflection coefficient they go down to @xmath179 and @xmath162 mk , respectively . for the antenna reflection they are reduced to @xmath183 and @xmath57 mk . finally , when all the uncertainty sources are considered the result is @xmath184 mk . this result for the combined case is about three times better than the @xmath167 mk obtained for the @xmath2 mhz range . this suggests that for a given signal model it may be possible to optimize the bandwidth processed to maximize the constraints on the model . this paper describes the laboratory calibration of the edges high - band receiver used in the @xmath3 observational campaign . the calibration was performed prior to instrument deployment and involved measuring the spectra , reflection coefficients , and physical temperatures of four absolute calibrators connected at the receiver input . these measurements are used to determine the function that converts the antenna noise power to noise temperature . the calibration was done in the range @xmath2 mhz , corresponding to @xmath1 . to evaluate the impact on science analysis of realistic calibration uncertainties , we developed a monte carlo uncertainty propagation pipeline . the results are presented as @xmath133% confidence levels for rms residuals after modeling and fitting the systematics from simulated calibration errors with low - order polynomials , using between zero and seven terms . we focus on the results from the five - term model because this is the number of terms required to account for beam chromaticity from the edges blade antenna . after polynomial subtraction , the systematics from a combination of all the error sources considered in this work amount to @xmath167 mk for observations of a quiet , low - foreground sky . to a large extent , this value is the result of uncertainties in the reflection coefficient of the antenna and the receiver input , which contribute individually with @xmath170 mk . for almost all other error sources the residuals are @xmath161 mk . the receiver calibration is determined in the lab and applied to data taken in the field . while every effort is made to equalize the instrument characteristics in the two circumstances , the validity of the assumptions that go into the mc simulations can only be tested with real sky measurements . this will be addressed in forthcoming papers . assuming that the receiver performance determined in the lab adequately reflects the instrument performance in the field , it should be possible for edges high - band to probe some @xmath0-cm models described in the literature , particularly those with astrophysical parameter combinations that yield large or late absorption features , or with rapid reionization histories . astrophysical results will also be reported in future papers . there are promising near - term opportunities for reduction of systematics from reflection measurements : 1 ) the incorporation of custom reflection standards into the calibration of the vector network analyzer used for measurements , 2 ) the improvement of the impedance match between the antenna and the receiver input , 3 ) the reduction of the changes in antenna phase with frequency . current efforts are focused on investigating and implementing these improvements . we thank the referee for suggestions that helped improve the quality of this manuscript . this work was supported by the nsf through research awards for the experiment to detect the global eor signature ( ast-0905990 and ast-1207761 ) and by nasa through cooperative agreements for the lunar university network for astrophysics ( nna09db30a ) and the nancy grace roman technology fellowship ( nnx12ai17 g ) . raul monsalve acknowledges support from the nasa ames research center ( nnx16af59 g ) .

there has arisen considerable interest in the use of weak value " techniques to improve the accuracy of precision measurement . while it has been recognized for some time that these techniques , in and of themselves , do not overcome fundamental limits for coherent light sources ( standard quantum limit " ) ( see _ e.g. _ @xcite ) , there are technical advantages in that these methods make the experimental approach to these limits relatively easy with common experimental equipment . indeed , these techniques have already been successfully applied in the lab to measure with high precision the optical spin hall effect and other polarization dependent beam deflections @xcite , interferometric deflections of optical beams @xcite , phase shifts @xcite , frequency shifts @xcite , temperature shifts @xcite , and velocity measurements @xcite . in most of these experiments , the weak value amplification ( wva ) technique met and even surpassed the sensitivity of standard techniques in the field . for a recent review of this and related weak value research , see ref . @xcite . although these experimental findings have been employed in a number of different research groups and applied to metrological questions of a number of different physical parameters , there are still some open questions and even controversy @xcite about this technique : precisely how and to what extent can wva techniques help against technical noise , or give some kind of technical advantage in comparison to the standard measurement techniques ? starling _ considered a particular parameter estimator , showing that wva could give an advantage @xcite . an important step was made in this question when feizpour , xingxing , and steinberg @xcite were able to consider a more general kind of technical noise , and showed that so long as it had a long correlation time , wva also help suppress it in the signal to noise ratio ( snr ) . in other closely related work , kedem @xcite , brunner and simon @xcite and nishizawa @xcite also showed an increased performance of the snr in the presence of technical noise . in contrast to these results , recent papers have claimed that wva gives no technical advantage @xcite . the argument is justified by using a fisher information analysis of technical noise applied to the signal carrier ( e.g. , such as beam displacement jitter ) . ( we note the fisher information analysis has been recently applied to wva by other authors as well @xcite . ) however , this particular form of technical noise does not represent the complete picture . there are many forms of technical noise that are not incorporated in this model . for example , in optical beam deflection , noise sources include : electronics noise , transverse displacement and angular jitter , analog - to - digital discretization noise , turbulence , vibration noise of the other optical elements , spectral jitter , etc . in light of this criticism , it is our aim in this work to analyze some of these models and examples using fisher information and maximum likelihood methods in order to understand in precisely what sense they give or fail to give a technical advantage , as well as describe other technical advantages in beam deflection ( and derivative ) experiments where the imaginary wva technique does lead to the optimal fisher information even in the presence of some types of noise sources mentioned above . the paper is organized as follows . in sec . [ fisher ] , we introduce the concepts of fisher information and maximum likelihood techniques , and illustrate how to apply them to gaussian random measurements . we introduce weak value amplification and postselection in sec . uncorrelated , displacement technical noise is discussed in sec . [ kg - sec ] . time correlated technical noise is analyzed in sec . air turbulence is discussed briefly in sec . the combination of displacement jitter and turbulence is discussed in sec . [ imwv ] , showing the weak value technique can suppress both . the weak value technique is shown to better suppress angular jitter in deflection measurements in sec . [ angular ] . we examine a recent photon recycling proposal in sec . [ recycle ] and show the fisher information is boosted by the inverse postselection probability . our conclusions are summarized in sec . fisher information describes the available information about an unknown parameter in a given probability distribution . consider an unknown parameter @xmath0 , upon which some random variable @xmath1 depends . let the probability distribution of @xmath1 , given @xmath0 be @xmath2 . the score of the distribution is defined as @xmath3 . assuming @xmath4 is a smooth function , the average of @xmath5 over @xmath4 is 0 , and its variance ( second moment ) is defined as the fisher information , @xmath6 fisher information is additive over independent trials , so for @xmath7 statistically independent measurements ( such as the collection of @xmath7 photons from a coherent source ) , @xmath8 . consider an unbiased estimator of @xmath0 , called @xmath9 . this is any statistical estimator whose expectation value is @xmath0 . the variance of @xmath9 is bounded from below by the cramr - rao bound ( crb ) , or @xmath10 \ge { \cal i}(d)^{-1}$ ] . thus , the fisher information sets the minimal possible estimate on the uncertainty of @xmath0 , for any unbiased estimator . to illustrate how this works , let us consider a gaussian distribution for @xmath11 , which will describe @xmath7 independent measurements @xmath12 of an unknown mean with known variance @xmath13 for each measurement , p_g(\ { x_i } |d ) = _ i=1^n ( - ) . in this example , the score is given by @xmath14 , so indeed @xmath15 , and the fisher information is given by _ g(d ) = n ^-2 . thus , the crb on the variance is simply @xmath16 . consequently , the minimum resolvable signal @xmath17 will be of order of @xmath18 . in order to achieve this minimum bound on the variance , the optimal unbiased estimator @xmath19 ( often called the efficient estimator ) must be used . we can find it with maximum likelihood methods , by setting the score to zero , and replacing @xmath0 by @xmath19 . in the case of @xmath20 , we have s(d _ opt ) = _ j=1^n ( x_j - d_opt)/^2 = 0 , [ max0 ] so we find the efficient estimator , _ opt = ( 1/n ) _ j=1^n x_j , [ maxl ] which is simply the average of the data in this case . we can check that the variance gives the inverse crb : @xmath21 . in an optical context , this is the `` standard quantum limit '' scaling with @xmath7 , which we can interpret as the photon number . here the parameter @xmath0 can be interpreted as the displacement of a beam with transverse width @xmath22 . one can immediately see that the maximum fisher information occurs for the smallest allowable beam waist . this is intuitively obvious . if a beam of very small waist experiences a small shift in its mean value , a position sensitive detector ( e.g. , a split detector ) would see a large change in intensity as a function of its position compared to a beam with a large waist . in the rest of this paper , we will consider coherent gaussian distributions for simplicity . while this is somewhat restrictive , it is also quite reasonable since most of the experiments have been performed using coherent gaussian probability densities . this is also quite nice theoretically , owing to the fact that the log likelihood function is twice differentiable and we can use variations of the simple form of the fisher information derived above . to apply the above results to recent optical experiments , we briefly recall a few facts about weak values @xcite . if a quantum system is prepared in an initial state @xmath23 , has a system operator @xmath24 that is measured by weakly interacting with a meter prepared in a state of spatial variance @xmath13 , and then postselected in a final state @xmath25 with probability @xmath26 , the meter degree of freedom will be shifted by a multiplicative factor a_w = f | a|i / f |i , [ wv ] where @xmath27 is the `` weak value '' of the operator @xmath28 , while leaving the width @xmath22 unchanged ( for the moment we consider the real weak value case for simplicity ) . such behavior is in contrast to the non - postselected case , where if the initial state is an eigenstate of @xmath24 , so that @xmath29 , the average meter shift is @xmath30 , which can be much smaller in size than the weak value shift , @xmath31 . this process gives rise to a ( normalized ) gaussian meter probability distribution consisting of @xmath32 measurement events @xmath33 , p_g(\ { x_i } |d ) = _ i=1^n ( - ) . computing the crb on the variance as before gives the score to be @xmath34 , so the fisher information is _ g(d ) = ( a_w)^2 n ^-2 = f | a|i^2 n ^-2 . note that the post - selection probability @xmath35 canceled out . therefore , the fisher information is the same as before , except for a factor @xmath36 , a number that can be arranged to approach 1 for a two - level system with a judicious choice of operator , pre- and post - selection ( see appendix ) . this is consistent with the snr analysis of ref . similar points were made by hofmann _ we can then extract all the fisher information from the weak value , showing that ideally , the wv technique can put all of the fisher information into the post - selected events , which matches the fisher information in the standard methods using all of the events . we note that it is not surprising that considering any sub - ensemble gives less information than the whole ensemble . what _ is surprising _ is that by using this particular small sub - ensemble gives you all the fisher information . this fact alone gives us technical advantages , as we shall see . following the maximum likelihood method presented in eqs . ( [ max0],[maxl ] ) , the weak value efficient estimator is given by @xmath37 . as a next step , we include the effect of one kind of technical noise . we first consider the technical noise model of knee and gauger , where the @xmath7 random variables @xmath38 are independently distributed with mean @xmath0 and known variance @xmath13 @xcite . gaussian white technical noise @xmath39 is added to @xmath38 with mean 0 and covariance @xmath40 . we note that similar models have also been previously discussed in refs . @xcite . knee and gauger comment that this kind of noise might represent transverse beam displacement jitter ( in a collimated beam ) , for example . we call this _ type 1 _ technical noise . the measured signal will be the sum of @xmath41 , so the distribution function for @xmath42 is @xmath43 , where @xmath44 integrates over all variables . integrating over @xmath45 gives the distribution of the measured results as a convolution of the two distributions , @xmath46 where @xmath47 is a normalization constant . calculating the fisher information as before , for @xmath7 independent measurements ( photons ) gives = . [ ikg ] the technical noise simply broadens the width , decreasing the fisher information . in the weak value case , we follow the same procedure , except that @xmath48 , and @xmath49 , where @xmath35 is the post - selection probability . we add the same technical noise to the resulting distribution of post - selected events @xmath50 , and get @xmath51 calculating the fisher information for the postselected case ( which must be reduced by @xmath35 , the post - selection probability ) gives _ kg = = , [ ikg ] where we used the weak value formula ( [ wv ] ) . consequently , as before we find the fisher information is modified by a factor of order 1 ( which will decrease or keep the fisher information the same , as discussed before ) . this is the same conclusion reached by knee and gauger @xcite and ferrie and combes @xcite , that the weak value amplification offers no increase of fisher information for technical noise . this result was previously found by feizpour , xingxing , and steinberg @xcite , who noted that this type 1 technical noise could not be suppressed by the real weak values technique . however , as we will show in sec . [ imwv ] , this is not the case for techniques implementing an imaginary weak value . we go on to find the efficient estimator in the post - selected case , following the maximum likelihood method presented in eq . ( [ max0],[maxl ] ) . the efficient estimator is given by @xmath52 . in other words , one can efficiently estimate the parameter from the snr . we note that in either the standard or wva scheme , the fisher information can be further improved by reducing the width of the meter state , but only until @xmath53 , after which the technical noise dominates the variance . we will return to this point in sec . in order to achieve the crb , the fisher information must not only be calculated , but the associated estimator must also be practical to implement . the point of practicality of the efficient estimator can be strongly made by considering the analysis of feizpour , xingxing , and steinberg @xcite . consider an experiment with single photons , such that the measurement is only triggered by the photon detection at the detector , and consists of @xmath7 measured variables @xmath54 described by an average @xmath55 plus a noise term @xmath56 that is correlated in general , @xmath57 ( we define this as _ type 2 _ technical noise ) . the noise term contains both quantum and technical noise . it is then straightforward to check that the average is given by @xmath58 , and the variance is given by @xmath59 . the two limits considered in ref . @xcite are ( i ) the white noise limit , @xmath60 , and ( ii ) the fully correlated limit @xmath61 . in case ( i ) we have @xmath62 , while in case ( ii ) , we have @xmath63 . the snr is given by @xmath64 , thus averaging helps the snr for white noise , but not for fully correlated noise . now consider the postselected case , where @xmath65 , and @xmath66 . in case ( i ) , the variance is now @xmath67 , while the signal is @xmath68 , so the snr scales like @xmath69 , so the small post - selection probability drops out , indicating there is no advantage to using post - selection in this case ( exactly as we calculated in the previous sections for type 1 technical noise ) . however , in case ( ii ) , the signal is boosted the same amount , while the variance remains @xmath70 in the fully correlated case . consequently , the snr shows an advantage over the non - postselected case . feizpour , xingxing , and steinberg go on to consider a particular noise model consisting of a combination of white and time - correlated noise , showing this advantage remains so long as the correlation time remains long compared to the photon production rate ( as it is in many optical implementations ) . we can now revisit this model in the context of the fisher information metric to see how it compares to the snr metric . the joint probability distribution of all of the variables @xmath71 can be written , assuming gaussian statistics , as p_s(*x * ) = . [ multivariable ] here , @xmath72 is a vector of elements @xmath71 , and @xmath73 is a vector of the means ( in this case , @xmath74 is a vector with the same mean in every element ) . the matrix @xmath75 is the covariance matrix and has elements @xmath76 . @xmath77 is the inverse of the covariance matrix , and @xmath78 is its determinant . in this case , we may calculate the fisher information ( [ fi ] ) about @xmath55 contained in this distribution , and obtain , _ s = _ | * * ^t^-1 _ | * * = _ i , j=1^n c^-1_ij . [ fi ] notice this fisher information contains a double sum of the elements of the inverse covariance matrix . the `` diagonal '' terms describe the self - correlation terms , and scale proportionally to @xmath7 , recovering the independent trials for the case when the covariance matrix is diagonal . however , for strong correlations , such as the type feizpour , xingxing , and steinberg consider , each element of the inverse covariance matrix can be of comparable value , so the fisher information can scale at most like @xmath79 , giving a much larger fisher information than for uncorrelated noise . the reason is because of the correlations between the different measurements that the snr metric misses . when we go to the postselected case , the dimension of the matrix shrinks from @xmath7 to @xmath80 , while the mean is boosted by @xmath81 . the fisher information is boosted by @xmath82 as before , but the double sum in ( [ fi ] ) now only goes to @xmath80 in the upper limit , @xmath83 . the covariance matrix is different in general , since it now describes the correlations between the postselected photons only . consequently , for white noise the fisher information is the same , up to a factor of order 1 , while for highly correlated noise , the fisher information scales at most as @xmath84 , so it is actually decreased by a factor of @xmath35 by the weak value technique , compared to the non - postselected case . it is easy to see why : the correlations of any single postselected photon with any rejected photon are lost in the detection scheme ( unless further processing of the correlated missing photons is done ) , and consequently can not be harnessed to further suppress the noise . in contrast to this , if the photon is correlated only with itself , then taking a random postselection will not hurt , and the fisher information can stay the same . however , this is not the whole story . we must ask what estimator should be used that saturates the crb , and whether it is practical to implement it . we can find this estimator using maximum likelihood methods described earlier ( [ max0],[maxl ] ) to find the estimator @xmath85 . we assume that the covariance matrix is positive definite , symmetric , and invertible . it then follows that its inverse is also symmetric . we find the estimator in the non post - selected case to be _ s= . [ estimator ] we can check the variance of this estimator matches the crb , ( _ s - |)^2 = , because the correlation of the two random variables gives the elements of the covariance matrix , @xmath86 , which cancels one of the inverse matrices in the sums , giving one factor of the denominator , resulting in the crb we found above , the inverse of ( [ fi ] ) . however there is a difficulty in this result : the experiment implementing the estimator ( [ estimator ] ) must multiply every data point @xmath54 by a different weighting factor , @xmath87 , which knows about the rest of the data points . the experimenter must know exactly what the correlations are , how many data points are being collected , and must be able to weight each data point by a different factor in order to extract the maximal information in the data average . this is generally a very challenging experimental task . therefore , the snr , which treats every type of noise on equal footing ( so the weighting assignment is @xmath88 for all @xmath89 ) is usually the most practical option . consequently , the snr , although suboptimal as a means for estimating in this case , is still superior since the optimal estimator is impractical to implement for typical experiments . this is why the snr is used by experimentalists : because a complete categorization of the noise correlations is a formidable task requiring detailed knowledge of noise correlations and extensive post - processing . if we go on to consider non - gaussian correlated noise , such as @xmath90 noise , the problem of implementing the estimator becomes even worse . another type of technical noise that is very important in open air experiments is turbulence , which we refer to here as type 3 technical noise . while we will not give a quantitative analysis of this effect here , it gives additional beam width broadening beyond the diffraction limit because of beam breathing ( on a short time scale ) , and beam wander ( on a longer time scale ) because of the propagation through the random medium @xcite . this becomes an important problem when there is a large optical path length from the position where the deflection occurs to the detector where it is measured . beam jitter from turbulence can not be underestimated when dealing with extremely small deflections . the beam wander effects become important on time scales longer than the ratio of the beam width to the typical air velocity . typical experiments can run between seconds and hours , so this effect must be accounted for . the weak values schemes have a distinct advantage over the standard beam deflection measurement in having short optical path lengths . we will see this in detail in the next section . we proceed to consider situations where imaginary wva has technical advantages for combined technical noise types . kedem points out that in the case of imaginary weak values , noise in the average position does not appear in making measurements in the momentum basis , and that noise on the average momentum helps the snr @xcite . we focus on a different effect unique to imaginary weak values . for definiteness , consider beam deflection measurements for a coherent gaussian beam using standard deflection techniques versus a sagnac interferometer weak value experiment . most experiments used to date that show technical advantages use imaginary weak values . to start , assume that the system has no technical noise . the unknown parameter of interest is the deflection @xmath91 of the beam , which can be interpreted as the transverse momentum kick given by a mirror . _ showed that one can interfere a local oscillator of a first order tem mode to achieve the optimal fisher information @xcite . however , for simplicity and to understand the role of beam diameters we consider a more standard approach . a tilt or deflection on a beam requires propagation to observe a displacement . the standard method of measuring an unknown mirror tilt @xmath91 with a beam of width @xmath22 is to propagate the light and focus it with a lens , thereby taking the fourier transform of the beam by measuring the beam with a split detector in the back focal plane of the lens ( see fig . 1(a ) assuming @xmath92 ) . the lens transforms the tilt @xmath91 into a displacement @xmath93 on the detector , where @xmath94 is the focal length and @xmath95 is the wavenumber . the lens takes the beam width @xmath22 at the mirror to a beam width at the focus given by @xmath96 . given angular jitter of momentum @xmath97 . ( a ) standard method : after the momentum kick , the beam propagates a distance @xmath98 before getting a momentum kick @xmath91 from the mirror , that is the parameter we wish to estimate . the beam then passes through a lens with a focus @xmath99 , chosen to equal the distance traveled until it comes to the split detector , where the beam reaches its focus at a position sensitive detector ( psd ) . ( b ) weak value type method : after the momentum jitter kick @xmath97 , the light enters a sagnac interferometer , comprised of a 50/50 beam splitter ( bs ) , soleil - babinet compensator ( sbc ) , which introduce a relative phase shift @xmath100 between the paths , other mirrors , including the one that has the momentum kick @xmath91 we wish to measure . the beam ends at a position sensitive detector ( psd).,width=321 ] the single photon probability distribution function is then p_h(x| k ) = n , where @xmath47 is a normalization . if we send @xmath7 independent photons to measure @xmath91 , this yields a fisher information of _ h = = 4 n ^2 , which is an intuitive result . this means that in order to get the smallest beam waist in the back focal plane , one wants the largest beam waist possible before the lens . we now compare this with the sagnac interferometer weak values result . the weak value technique for measuring the beam deflection ( fig . 1(b ) with @xmath92 ) gives the post - selected distribution , that when properly normalized , is a gaussian distribution with mean @xmath101 and width @xmath13 . here , @xmath100 is the phase difference of clockwise and counter - clockwise photons in the interferometer applied by the soleil - babinet compensator ( sbc ) . this applies only to the post - selected fraction @xmath32 of the photons that satisfy the postselection criterion , where @xmath102 . the postselected single photon probability distribution function is then p_h(x| k ) = n . the fisher information for the postselected measurements , reduced by the postselection probability @xmath35 ( @xmath103 ) is then found to be _ 4 ^4/(^2)= 4 ^2 n. [ fisherimaginary ] thus , the weak values and standard method result in _ the same _ amount of classical fisher information , as also discussed in the appendix @xcite . it is important to realize that this calculation is only valid in the regime of small angles @xmath104 . in other words , in the small angle or weak value regime , all of the fisher information is in the photons leaving the dark port of the interferometer . so how does noise affect the fisher information for the two methods ? as a first step , we will assume that there is no transverse beam deflection jitter , but only transverse detector jitter . in other words , there is a small transverse deflection @xmath91 with no technical noise on the beam , but the detector used to measure the beam has a transverse jitter @xmath105 that we only sample at the photon arrival times , giving type 1 technical noise . we first consider the standard method . as before , the new likelihood function is the convolution of the technical noise - free likelihood function with a gaussian of width @xmath106 . this simply increases the average beam waist at the detector by @xmath107 , resulting in a fisher information _ this is a rather interesting result . it means that this approach can achieve the maximum fisher information ( as if there were no noise at all ) , but only in the limit of large focal length ( @xmath108 ) . a large focal length is equivalent to opting for a larger displacement ( @xmath109 ) while allowing for a large focal spot ( @xmath110 ) . this is once again an intuitive result : if a very small focal spot lands on a detector that has gaussian random shifts , large differential intensity fluctuations will occur , due to detector jitter , compared with a large focal spot . for the imaginary weak value approach , the results are quite different . the beam waist for this case is given by @xmath22 , not the focused beam waist @xmath111 . this gives a fisher information of _ therefore , since @xmath112 and there is freedom to choose @xmath22 as large as one wishes , we can make the beam waist much larger than @xmath106 , which both suppresses @xmath106 and increases the fisher information ( [ fisherimaginary ] ) . this shows that fisher information for the imaginary weak values approach remains unchanged in the presence of transverse detector jitter , while the focused beam approach requires long focal lengths . hence , the measurement geometry plays a very important role in several ways . ideally one would like a detector with a continuous detection distribution , but practical considerations mean there will always be dead space between finite width detectors . a simple method for obtaining nearly optimal beam detection is split detection . however , even in this case there is always a gap between the detectors , which sets a minimum beam diameter and thus a finite propagation through horizontal turbulence . this is not a problem for weak values since the detector can be placed immediately after the last beam splitter which mitigates the type 3 technical noise while also suppressing type 1 technical noise . the standard method operating in the regime where type 1 noise is completely ameliorated will suffer from the type 3 technical noise . we now consider angular beam jitter by modeling it as a random momentum kick @xmath97 given to the beam before it enters the interferometer , or before it approaches the signal mirror in the standard method . this kick could be from air turbulence or mirror jitter . we first consider the standard method of measuring the beam displacement illustrated in fig . the fourier optics of this geometry is described by a series of unitary operators that act on the transverse degree of freedom in the paraxial approximation . starting from an initial transverse state @xmath113 , that we take to be a gaussian in transverse position with zero mean , and variance @xmath13 , the unitary @xmath114 gives the first random momentum kick @xmath97 to the beam . this is followed by a propagation of distance @xmath98 , given by the unitary @xmath115 , where @xmath95 is the wavenumber of the light . this is followed by a momentum kick @xmath91 described by @xmath116 , and then the lens which gives a quadratic phase front , @xmath117 . the final propagation @xmath118 puts the beam at the measurement device , which is a position - sensitive detector ( psd ) . taken together , we can describe the final state as _ d(x ) = x | u_l_2 u_f u_k u_l_1 u_q | . to obtain an explicit form for the state , and the expectation of the position and its variance in this state , a series of complete sets of states are inserted between the unitaries . as an intermediate step , let us define state @xmath119 as the state after the @xmath91 momentum kick , but before the lens . the action of the lens is diagonal in the position basis , but the subsequent propagation is diagonal in the momentum basis . defining the coordinate at the detector as @xmath1 , the state at the detector @xmath120 is given by _ d(x ) = dy _ i(y ) e^-i k_0 y^2/2f - i py e^-i p^2 l_2/2k_0 + i px . we can reverse the order of integration and perform the @xmath4 integration first . this gives _ d(x ) = dy _ i(y ) e^-i k_0 y^2/2f + i ( y - x)^2k_0/2l_2 . let us choose @xmath121 , so the quadratic terms in the exponential cancel out , which will lead to focusing the beam . this leaves the state as _ d(x ) = e^ i k_0 x^2/2f dy _ i(y ) e^-i y k_0 x / f , which simply gives the scaled fourier transform of @xmath119 , but with a particular value of the momentum , @xmath122 . the remainder of the solution involves finding the intermediate state . this is straightforward in the momentum basis , because the first three operators are diagonal in this basis . therefore , the momentum - space expression for @xmath123 is given by _ i(p ) = ( 8 ^2)^1/4 ( - ^2 ( p + q + k)^2 - i l_1 ( p+k)^2/2k_0 ) . putting these results together , we find that the final state at the detector gives a gaussian distribution in position @xmath1 with average and variance of x = - ( k+q ) f / k_0 , ( x - x)^2= f^2/(2 k_0)^2 . [ y ] the net result is that the random momentum kick @xmath97 simply adds to the signal @xmath91 . averaging this distribution of gaussian random jitter of momentum @xmath97 with zero mean and variance @xmath124 gives another gaussian distribution for @xmath1 , of mean @xmath125 and wider variance @xmath126 $ ] . thus , the fisher information , for @xmath7 independent measurements , about @xmath91 in this distribution is given by _ [ idirect ] a similar analysis can be carried out for the weak value interferometer case . in addition to the momentum kick and the propagation steps , there are two unitaries that depend on the which - way operator @xmath127 , where the states @xmath128 are clockwise and counterclockwise moving photon states inside the sagnac interferometer . superpositions of these states are created by the 50 - 50 beam splitter operating on the incoming beam . the unitaries are the relative phase shift operator , @xmath129 , induced by the sbc , and the which - path momentum kick operator , @xmath130 delivered by the signal mirror . as shown in fig . 1(b ) , starting in the state @xmath131 , and ending in state @xmath132 , we seek the transition amplitude _ f | u_l_2 u_kw u_u_l_1 u_q | _ i , which describes photons entering in one interferometer port and exiting the other interferometer port . fortunately , the which - path states pass through all but two of the operators , so this amplitude may be simplified to eliminate the which - path states and operators , giving the state of the transverse beam at the detector , @xmath133 , to be _ wv(x)= x| e^-i p^2l_2/2k_0 ( k x + /2 ) e^-i p^2l_1/2k_0 e^i q x | . the detailed calculation of this state involves inserting complete sets of states , resulting in a complicated expression . the result is simplified by expanding to linear order in @xmath100 and @xmath91 since we assume the weak value ordering of parameters , @xmath134 . we must renormalize the post - selected distribution by the probability @xmath35 of a photon exiting the dark port , given by = ( /2)^2 + k^2 ^2 + o(k^2 q^2 ) + o(k q ) + . we drop the other terms compared with @xmath135 , so the phase shift @xmath100 controls the post - selection . this gives an average displacement of x= + + + where we suppress higher order terms in @xmath136 . the first term of order @xmath91 is the usual weak value term , amplified by @xmath137 @xcite . the second term comes from the diffraction effects which gives a small correction to the first term because we take @xmath138 . the remaining term , @xmath139 is just the free propagation from the momentum kick , and has a geometric optics interpretation of the imparted deflection angle @xmath140 times the total length of propagation . we note that the way @xmath97 enters into the average displacement has a very different form than eq . ( [ y ] ) . we will return to this shortly . the displacement variance at the detector may be similarly calculated to find , ( x - x)^2 = ^2 + ( ) ^2 , plus further corrections of order @xmath141 . thus , we see that the diffraction effects broaden the width of the beam , proportional to the total path length . if we approximate the distribution as a gaussian with the mean and variance discussed above , the fisher information about @xmath91 can be found by averaging over @xmath97 as before to find _ wv = . consequently , even if @xmath142 is large compared to 1 , so as to degrade the fisher information in the standard method ( [ idirect ] ) , in the weak values technique , there is an additional suppression factor of the amount of diffraction , @xmath143 , indicating that the weak value technique outperforms the standard method when dealing with angular jitter . this result can be understood intuitively because the angular jitter directly adds to the detected deflection in the standard measurement case , whereas it is only a small correction to the deflection that is controlled by diffraction in the weak value technique . the rather remarkable properties associated with imaginary weak value experiments may not be entirely attributed to weak values , but to geometric terms associated with the experiments . these terms , such as beam waist diameter , arise even when wva is small . however , it is important to note only in the limit of large wva that all of the information in the measurement can be placed in the measured photons and that optimal snr estimation of the parameter can be made . therefore , these geometric terms and wva work hand - in - hand to achieve the technical noise suppression . we can also consider the possible benefit to the fisher information of the recent photon recycling proposal of dressel _ et al . _ @xcite . considering photons as a resource , one can ask what the maximum amount of information can be extracted from a given set of photons . in much the same way as a high finesse cavity can increase phase sensitivity in an interferometer , recycling photons from the bright port of a weak value experiment can increase the information about a parameter . the central idea is to postselect all of the photons while keeping the large weak value amplification . the scheme is to recycle the rejected photons by closing off the interferometer , so the remaining @xmath144 photons are re - injected ( @xmath145 and once again sample the unknown parameter @xmath91 . the authors show that it is critical that the rejected light be reshaped to once again be in its original profile , otherwise all amplification of the split detector snr is erased over many cycles . below , we consider the simplest version where no propagation effects are included . in that case , the second round of postselected light has exactly the same distribution on the detector , with the same post - selection probability @xmath35 , so the fisher information for this cycle is @xmath146 . this process is now repeated many times , where @xmath147 is the number of rejected photons on the @xmath148 round , and we define @xmath149 . the uniform postselection probability indicates that @xmath150 , implying that @xmath151 . since each measurement is independent from the last , the fisher information simply adds , giving a total of _ tot = 4 ^2 _ i=0^n_i = 4 ^2 n _ i=0^(1-)^i = . therefore , the fisher information has been boosted by a factor of @xmath82 compared to the standard method , or the single pass weak value method . here , we are considering the photon number @xmath7 as the resource . of course , as is pointed out in ref . @xcite , one could have been sending more light onto the detector using the standard method in the time taken for the light recycling or employed other standard schemes , but there could be many technical reasons why this may be impossible or inadvisable given a laboratory set - up . there may be a minimum quiet time between laser pulses , for example , or the detector may have a low light power threshold . we note that the profile reshaping process actually removes information about the parameter @xmath91 , but does so in a way that the estimator can be simply and easily implemented as the split detection snr . together with the technical advantages already discussed , this is an important improvement over previous techniques . when further combined with quantum light techniques , this method gives a powerful advantage for estimating a parameter . the fact that all photons are now collected permits one to further mine interphoton correlations for noise suppression . we have shown how weak value based measurement techniques can give certain technical advantages to precision metrology . first and foremost , the fisher information in a weak value measurement ( which uses a small fraction of the available light ) can be as large as the fisher information of a standard measurement ( which uses all of the available light ) . this is remarkable because the remaining light can be sent to another experiment @xcite , or recycled @xcite to give even higher amounts of fisher information . we have also explored technical advantages the weak value experiments can have over standard measurement techniques . obvious advantages , such as when the detectors saturate at a certain light intensity , have been pointed out in previous works @xcite . the possible advantages for different types of technical noise should be investigated on a case by case basis . there are cases where the weak value techniques gives advantages , and other cases where it is at a disadvantage , and yet other cases where there is no difference . it is clear , for example , that dephasing noise will reduce the size of the weak value , which will be detrimental to this technique @xcite . we have shown how detector noise , together with air turbulence can both be eliminated by a weak value deflection measurement , whereas the conventional standard method must suffer from turbulence jitter if detector noise is suppressed in open air experiments . we also demonstrated that for angular jitter , the weak value technique for beam deflection measurements can have much higher fisher information than the standard technique . considering the wide range of experiments that have now successfully employed weak value techniques to make high precision measurements @xcite , this should not be too surprising . another conclusion we reach is that it is not sufficient to show an estimator does not reach the crb to decide it should be rejected . rather , the optimal estimator should be found , and must be practically implementable . if it is not , the inefficient - but practical - estimator is superior . we have argued that time - correlated technical noise is one example where the difficulty of implementing the optimal estimator is outweighed by the option of implementing the postselection with amplification . we acknowledge support from the us army research office grants no . 62270phii : stir , no . w911nf-09 - 0 - 01417 , no . w911n-12 - 1 - 0263 , as well as the national science foundation grant no . dmr-0844899 . we thank y. aharonov , j. tollaksen , g. c. knee , e. m. gauger , j. dressel , and h. wiseman for helpful comments on the manuscript . our calculations are based on an interaction of the type @xmath152 , where @xmath0 is the unknown small parameter . tracking / measurements of the meter degree of freedom @xmath1 after the postselection gives a snr of @xmath153 . if instead , measurements of the conjugate momentum @xmath4 are performed , the result is @xmath154 @xcite . we will show that under the right choices for pre- and post - selection , @xmath155 or @xmath156 . we chose to consider the operator @xmath157 such that @xmath158 , where @xmath159 and @xmath160 form an orthonormal basis for the hilbert space of a two - level system . the initial state of the system can be written as @xmath161 where @xmath162 and @xmath163 represent the qubit state on the surface of the bloch sphere . the postselection state for the system is defined as @xmath164 where @xmath165 and @xmath100 are angles representing the deviation ( in both angular directions on the bloch sphere , see fig . [ sphere ] ) from the state orthogonal to @xmath166 . representation on the bloch sphere of the preselection and postselection states for @xmath167 . the state @xmath166 corresponds to a unit vector anywhere on the equatorial plane , and the state @xmath168 is off to be antiparallel to @xmath166 in the two angular directions . the angle @xmath169 ( @xmath170 ) gives origin to the real ( imaginary ) part of the weak value . ] note that the angles @xmath165 and @xmath100 are the respective generators of the real and the imaginary part of the weak value . for example , a pure imaginary weak value can be obtained making @xmath172 , so @xmath173 and @xmath174 . in order to maximize @xmath35 we choose @xmath167 , so the results are @xmath175/\sqrt{2}$ ] and @xmath176\sqrt{2}$ ] . finally , in the small angle approximation , @xmath177 and @xmath178 , making @xmath179 . on the other side , a pure real weak value can be obtained making @xmath180 , so @xmath181 , and @xmath182 . in order to make @xmath27 large for this case , we choose again @xmath167 . the preselection is identical to the former case @xmath175/\sqrt{2}$ ] , and the postselection takes the form @xmath183/\sqrt{2}+\cos\theta\left[|+\rangle - e^{i\phi}|-\rangle\right]/\sqrt{2}$ ] . taking the small angle approximation for this case we obtain a similar result to the pure imaginary weak value case , @xmath184 , @xmath185 , and @xmath186 . setting @xmath167 turns out to be the best choice for practical purposes . we therefore calculate the products @xmath187 and @xmath188 for such a choice , and plot them in fig . [ plots ] . it is shown in fig . [ plots](a ) that the largest value @xmath189 is closest reached if @xmath180 and @xmath190 , which corresponds to a pure real weak value and an almost orthogonal postselection . similarly , fig . [ plots](b ) shows that @xmath191 only if @xmath172 and @xmath192 , defining a pure imaginary weak value with almost orthogonal postselection . in this case the interaction takes the form @xmath193 , where tracking the beam s transverse position corresponds to an imaginary weak value . we also note that while knee and gauger treat imaginary weak values in their appendix b , they do not consider the addition of technical noise in this case , and further compare the fisher information of @xmath91 in the imaginary weak value case against a standard strategy measurement of @xmath1 ( equivalent to not having a lens in fig . 1(a ) ) , which are different parameters .

this article is on the one hand on models of weak arithmetics and on the other on proof complexity , i.e. the question of how long formal proofs of tautologies have to be in given proof systems . therefore the introduction will consist of two parts , one for each subject . models of weak arithmetics , like @xmath3 , have been extensively studied for several reasons . they are possibly the simplest objects whose theories bear enough strength to do a good part of mathematics in , yet they are weak enough to allow for a certain kind of constructiveness . the latter has been demonstrated over and over again by various results connecting weak arithmetic theories with complexity classes and computability . we are interested in the strength of the theory obtained by restricting our objects of reasoning to a small initial part of a given model . since a two - sorted theory , such as , is much stronger on its number part than on its set part , it is likely that such a cut is a model of a supposedly much stronger theory . indeed we will see in section [ sec : polylog cuts ] that certain cuts of models of are models of the provably stronger theory @xmath1 . this strengthens a result by paris and wilkie @xcite@xcite , who show that such cuts are models of @xmath4 . in fact they work in a more general setting and , following our argumentation , their result readily implies the sub exponential simulation of @xmath5-frege by @xmath2-frege from bonet , domingo , gavald , maciel , and pitassi @xcite . proof complexity , on the other hand , more or less began when cook and reckhow @xcite discovered the close connection between the lengths of formal proofs in propositional proof systems and standard complexity classes . this connection yields a possibility of dealing with the @xmath6 question by asking , whether there exists a propositional proof system that is polynomially bounded . we will not directly address this question here , but rather explore the relative strengths of two major proof systems , frege and bounded depth frege . these proof systems have been extensively studied , due to their natural appearance as classical calculi , such as gentzen s pk , and it is well known that frege systems are stronger than bounded depth frege systems , as the former system has polynomial size proofs for the pigeonhole principle ( see @xcite ) , while the latter does not ( see @xcite and @xcite ) . lately , filmus , pitassi and santhanam @xcite have proved a sub exponential simulation of frege by bounded depth frege using a combinatoric argument . in section [ sec : impl proof cmpl ] we will obtain the same result by an application of our result about cuts to the provability of the reflection principle for frege in bounded depth frege . currently cook , ghasemloo and nguyen @xcite are working on a purely syntactical proof that gives a slightly better result with respect to the strength of the simulated proof system . the paper is built - up as follows . in section [ sec : preliminaries ] we briefly recapture some basics about complexity theory , bounded arithmetic , proof complexity and the various connections between them . as this is only expository it might be helpful to consult some of the references for a more detailed introduction ( see @xcite , @xcite and @xcite ) . after that , in section [ sec : polylog cuts ] we prove a formalization of nepomnjascij s theorem in the polylogarithmic cut of a model of @xmath0 . using a standard algorithm for evaluating circuits and then applying the formalized version of nepomnjascij s theorem we can conclude that this cut is indeed a model of @xmath1 . finally , in section [ sec : impl proof cmpl ] , we apply this result to prove that a version of the bounded reflection principle of frege is provable in @xmath0 . this , together with a standard argument linking the provability of reflection principles with simulation results , yields the sub exponential simulation of frege by bounded depth frege . we assume familiarity with turing machines , circuits and standard complexity classes such as @xmath7 , @xmath8 , @xmath9 , @xmath10 , @xmath11 and so on . see for example @xcite for an introduction . we will not work a lot within these classes , but rather apply known relations between such classes and weak arithmetic theories . we will work in a two - sorted arithmetic setting , having one sort of variables representing numbers and the second sort representing bounded sets of numbers . we identify such bounded subsets with strings . see @xcite for a thorough introduction . the underlying language , denoted , consists of the following relation , function and constant symbols:@xmath12 an -structure @xmath13 consists of a first - sort universe @xmath14 of numbers and a second - sort universe @xmath15 of bounded subsets of numbers . if @xmath13 is a model of the two - sorted theory @xmath0 ( see [ sec vz ] ) , then the functions @xmath16 and @xmath17 are the addition and multiplication on the universe of numbers . @xmath18 and @xmath19 are interpreted as the appropriate elements zero and one with respect to addition and multiplication . the relation @xmath20 is an ordering relation on the first - sort universe . the function @xmath21 maps an element of the set sort to its largest element plus one ( i.e. to an element of the number sort ) . the relation @xmath22 is interpreted as equality between numbers , @xmath23 is interpreted as equality between bounded sets of numbers . the relation @xmath24 holds for a number @xmath25 and a set of numbers @xmath26 if and only if @xmath25 is an element of @xmath26 . the standard model of two - sorted peano arithmetic will be denoted as @xmath27 . it consists of a first - sort universe @xmath28 and a second - sort universe @xmath29 of all finite subsets of @xmath30 . the symbols are interpreted in the usual way . we denote the first - sort ( number ) variables by lower - case letters @xmath31 , and the second - sort ( set ) variables by capital letters @xmath32 . in case it helps to describe the meaning of a variable we will use lower case words for first - sort and words starting with a capital letter for second - sort variables . we can build formulas in the usual way , using two sorts of quantifiers , number quantifiers and string quantifiers . a number quantifier @xmath33 ( @xmath34 ) is bounded if it is of the form @xmath35 ( @xmath36 ) for some number term @xmath37 . a string quantifier @xmath38 ( @xmath39 ) is bounded if it is of the form @xmath40 ( @xmath41 ) for some number term @xmath37 . a formula is bounded iff all its quantifiers are . all formulas in this paper will be bounded . a formula @xmath42 is in @xmath43 ( or @xmath44 ) if it uses no string quantifiers and all number quantifiers are bounded . a formula @xmath42 is a @xmath45 ( or @xmath46 ) if it is of the form @xmath47 ( or @xmath48 ) , where @xmath49 ( or @xmath50 , respectively ) . if a relation or predicate can be defined by both a @xmath51 and a @xmath52 formula , then we call it @xmath53 definable . the _ depth _ of a formula is the maximal number of alternations of its logical connectives and quantifiers . as mentioned before we will represent a bounded set of numbers @xmath26 by a finite string @xmath54 such that @xmath55 if and only if @xmath56 . we will abuse notation and identify bounded sets and strings , i.e. @xmath26 and @xmath57 . further , we will encode monotone propositional formulas inductively as binary trees in the standard way , giving a node the value @xmath19 if it corresponds to a conjunction and the value @xmath18 , if it corresponds to a disjunction . binary trees are encoded as strings as follows . if position @xmath58 contains the value of a node @xmath59 , then the value of its left successor is contained in position @xmath60 , while the value of its right successor is in @xmath61 . we restate some basic definitions introduced in @xcite . [ def pps ] a _ propositional proof system ( pps ) _ is a surjective polynomial - time function @xmath62 , where @xmath63 is the set of propositional tautologies ( in some natural encoding ) . a string @xmath64 with @xmath65 is called a _ @xmath66-proof _ of @xmath67 . we can define a quasi ordering on the class of all pps as follows . [ def : simulation ] let @xmath68 be propositional proof systems . @xmath69 @xmath66 simulates @xmath70 ( in symbols @xmath71 ) , iff there is a polynomial @xmath72 , such that for all @xmath73 there is a @xmath74 with @xmath75 , such that for all @xmath76 with @xmath77 , @xmath78 . if there is a polynomial time machine that takes @xmath70-proofs and produces @xmath66-proofs for the same formula we say that @xmath66 p - simulates @xmath70 ( in symbols @xmath79 ) . if @xmath66 and @xmath70 mutually ( p-)simulate each other , we say that they are ( p-)equivalent ( in symbols @xmath80 and @xmath81 , respectively ) . in this article we will be mainly interested in bounded depth frege systems and some of their extensions . a frege system is a typical textbook proof system , such as gentzen s propositional calculus pk . we will only sketch a single rule of such a system as an example and refer the interested reader to standard logic textbooks . here , @xmath82 and @xmath83 are sets of formulas while @xmath84 is a formula . @xmath85 is read as `` the conjunction of all formulas in @xmath83 implies the disjunction of all formulas in @xmath82 '' . the cut rule therefore says that , if @xmath83 implies @xmath84 or @xmath82 , and @xmath83 and @xmath84 imply @xmath82 , then @xmath83 already implies @xmath82 . the formula @xmath84 is called the _ cut formula_. in a bounded depth frege system the depths of all formulas in a derivation are bounded by some global constant . this is equivalent to being representable by an @xmath2 circuit . thus we also call bounded depth frege @xmath2-frege . if the formulas are unbounded , we speak of @xmath86-frege or simply of frege . we readily get @xmath87 . a pps @xmath66 is _ polynomially bounded _ iff there is a polynomial @xmath72 such that every tautology @xmath67 has a @xmath66-proof @xmath64 with @xmath88 . we are interested in the existence of polynomially bounded pps . this is , at least in part , due to the following theorem . @xmath89 an easier task than searching for a polynomially bounded pps might be to find some pps with sub exponential bounds to the lengths of proofs . this corresponds to the question , whether sub exponential time nondeterministic turing machines can compute @xmath90-complete languages . to explore the existence of such systems we generalize definition [ def : simulation ] . let @xmath68 be propositional proof systems and @xmath91 a family of increasing functions on @xmath30 . @xmath69 @xmath66 @xmath91-simulates @xmath70 ( in symbols @xmath92 ) , iff there is a function @xmath93 , such that for all @xmath73 there is a @xmath74 with @xmath75 , such that for all @xmath76 with @xmath77 , @xmath94 . if there is an @xmath95-machine that that takes @xmath70-proofs and produces @xmath66-proofs for the same formula we say that @xmath66 @xmath91-computably simulates @xmath70 ( in symbols @xmath96 ) . if @xmath66 and @xmath70 mutually @xmath91-(computably ) simulate each other , we say that they are @xmath91-(computably ) equivalent ( in symbols @xmath97 and @xmath98 , respectively ) . we say a pps @xmath66 _ sub exponentially simulates _ a pps @xmath70 iff the above @xmath91 can be chosen as a class of @xmath99 functions . the base theory we will be working with is @xmath0 . it consists of the following axioms : here , the axioms * basic 1 * through * basic 12 * are the usual axioms used to define peano arithmetic without induction ( @xmath100 ) , which settle the basic properties of addition , multiplication , ordering , and of the constants 0 and 1 . the axiom * l1 * says that the length of a string coding a finite set is an upper bound to the size of its elements . * l2 * says that @xmath101 gives the largest element of @xmath102 plus @xmath103 . * se * is the extensionality axiom for strings which states that two strings are equal if they code the same sets . finally , * @xmath43-comp * is the comprehension axiom schema for @xmath43-formulas ( it is an axiom for each such formula ) and implies the existence of all sets , which contain exactly the elements that fulfill any given @xmath43 property . the theory proves the induction axiom schema for @xmath43 formulas @xmath104 : @xmath105 when speaking about theories we will always assume that the theories are two - sorted theories as in @xcite . the following is a basic notion : [ def : two - sorted definabilty ] let @xmath106 be a theory over the language @xmath107 and let @xmath104 be a set of formulas in the language @xmath108 . a number function @xmath37 is @xmath109-definable in a theory @xmath110 iff there is a formula @xmath111 in @xmath104 such that @xmath110 proves @xmath112 and it holds that @xmath113 a string function @xmath91 is @xmath109-definable in a theory @xmath110 iff there is a formula @xmath114 in @xmath104 such that @xmath110 proves @xmath115 and it holds that @xmath116 finally , a relation @xmath117 is @xmath109-definable iff there is a formula @xmath118 in @xmath119 such that it holds that @xmath120 moreover we wish to talk about sequences coded by strings or numbers . for a string @xmath102 we let @xmath121 $ ] be the @xmath122th bit of @xmath102 . assuming a tupling function @xmath123 we can also talk of @xmath124-ary relations for any constant @xmath124 . we refer to @xmath125 $ ] , to say that the objects @xmath126 are in the relation @xmath102 ( which is equivalent to saying that the predicate @xmath102 holds for the number @xmath127 , i.e. that the @xmath102 contains that number as an element ) . for the sake of simplicity we also refer to @xmath128 $ ] by @xmath129 $ ] . using @xmath124-ary relations we can also encode sequences of bounded numbers @xmath130 by @xmath131x[\langle i,1\rangle]\dots x[\langle i , k\rangle]$ ] in binary . matrices and so on can obviously be formalized in the same way . given a string @xmath132 $ ] representing a @xmath124-ary relation , we denote the @xmath133-ary substring with parameters @xmath134 by @xmath135 $ ] . for example we refer to the element @xmath136 of a given matrix @xmath137 $ ] as @xmath138 $ ] , a string representing @xmath136 in binary . observe that this substring can be @xmath43 defined in @xmath0 . given a number @xmath58 we denote by @xmath139 the @xmath140th number in the sequence encoded by @xmath58 . to do this we assume a fixed @xmath43 definable encoding of numbers that is injective . the sequence itself will be addressed as @xmath141 . as above , we can also talk about matrices , etc . in this way , i.e. read such a sequence as a sequence of @xmath124-tuples . we want to identify strings of short length with sequences of numbers . thus , given a string @xmath102 of length @xmath142 we can @xmath43-define ( in @xmath0 ) a number @xmath143 that codes a sequence @xmath141 , such that @xmath121=\langle x\rangle_i$ ] for all @xmath144 and vice versa . we will use @xmath145 and @xmath146 $ ] to denote the above identification . observe that @xmath25 has to be very small in order to be able to do the above in @xmath0 . given a polynomially bounded turing machine @xmath84 in a binary encoding , we can @xmath147 define a predicate @xmath148 , that states that @xmath102 is accepted by @xmath84 . this can readily be observed , since , provided some machine @xmath84 , there is a constant number of states @xmath149 and the whole computation can be written into a matrix @xmath150 of polynomial size . that @xmath150 is indeed a correct computation can then be easily checked , because the computations are only local . more precisely let @xmath151 be given , where the @xmath152 are different states , with @xmath153 being the initial state and @xmath154 being the accepting state and @xmath155 is the transition function with domain @xmath156 and range @xmath157 , which describes what the machine does . i.e. if @xmath158 , then if the machine is in state @xmath159 and reads @xmath160 , it replaces @xmath160 by @xmath161 , goes into state @xmath162 and moves one position on the tape in the direction @xmath163 . for our formalization we will assume a function @xmath164 and interpret it in the following way , @xmath165 , where we identify @xmath166 . let the polynomial @xmath72 bound the running time of @xmath84 , then we can formalize @xmath148 as follows @xmath167\rangle_1 = x[i]\land i > 0\rightarrow \langle w[\langle 0,i,\cdot \rangle]\rangle_2 = 0 \land \langle w[\langle 0,0,\cdot \rangle]\rangle_2=1)\land\\ & i\geq{\left\vertx\right\vert}\rightarrow(\langle w[\langle 0,i,\cdot \rangle]\rangle_1 = 0 \land \langle w[\langle 0,i,\cdot\rangle]\rangle_2=0)\land\\ & \langle w[\langle j , i,\cdot \rangle]\rangle_2 = 0\rightarrow ( \langle w[\langle j+1,i,\cdot \rangle]\rangle_1= \langle w[\langle j , i,\cdot \rangle]\rangle_1)\land\\ & \langle w[\langle j , i,\cdot \rangle]\rangle_2 = \alpha\rightarrow ( \langle w[\langle j+1,i,\cdot \rangle]\rangle_1= \langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_2\land\\ & \hspace{0.2cm}(\langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_3=0 \rightarrow \langle w[\langle j+1,i,\cdot \rangle]\rangle_2= \langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_1)\land\\ & \hspace{0.2cm}(\langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_3=1 \rightarrow \langle w[\langle j+1,i{\mathbin{\setbox0\hbox{$-$}\setbox2\hbox to\wd0{\hss$^{\mkern1mu\cdot}$\hss}\wd2=0pt\box2\box0}}1,\cdot \rangle]\rangle_2= \langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_1)\land\\ & \hspace{0.2cm}(\langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_3=2 \rightarrow \langle w[\langle j+1,i+1,\cdot \rangle]\rangle_2= \langle\delta(\alpha,\langle w[\langle j , i,\cdot \rangle]\rangle_1)\rangle_1))\land\\ & ( i\neq i'\rightarrow ( \langle w[\langle j , i,\cdot \rangle]\rangle_2>0\rightarrow \langle w[\langle j , i',\cdot \rangle]\rangle_2=0))\land w[\langle p({\left\vertx\right\vert}),i,\cdot \rangle]\rangle_2 = k ) . \end{split}\ ] ] thus , in plain english , @xmath148 says that there exists a matrix @xmath150 of pairs of numbers that witnesses an accepting computation of @xmath84 . here , @xmath168\rangle$ ] is supposed to code the @xmath140th cell on the turing machine s tape after @xmath122 steps of computations on input @xmath102 . as noted above , @xmath168\rangle_1 $ ] is a binary number , which is the value of the cell , @xmath169\rangle_2 $ ] is a number coding the state the machine is in iff the pointer is on that cell . the second and third line of the definition say that the tape in the initial step contains @xmath102 padded with zeroes in the end to get the proper length @xmath170 and that the read / write head is in its starting state and position . the fourth line says that if the read / write head is not on cell @xmath122 , then nothing happens to the content of cell @xmath122 . the fifth line says that the content of the cell , where the read / write head is in step @xmath140 , is changed according to @xmath155 . the next three lines tell us where the read / write head moves . the last line says that there is at most one position on the tape where the read / write head may be at any step and that the state after the last step is accepting . we also define a @xmath147-predicate @xmath171 that says that @xmath84 reaches configuration @xmath172 from configuration @xmath173 in at most @xmath174 many steps . this is essentially the same predicate as @xmath175 , with the constraints on the initial and accepting state lifted and instead a constraint added that the first line of computation is @xmath173 and the last is @xmath172 . we omit the details as it does not severely differ from the above definition of @xmath175 . the theory serves as our base theory to describe complexity classes by arithmetical means . the problem , whether a given monotone formula @xmath176 of size @xmath177 and depth @xmath178 is satisfiable under a given assignment @xmath179 is @xmath2-complete for @xmath86 . therefore cook and nguyen ( @xcite ) define the class @xmath1 as @xmath0 augmented by the axiom @xmath180 , where @xmath181 so , @xmath182 states that there is an evaluation @xmath173 of the monotone formula represented by @xmath183 under the assignment given by @xmath179 of length at most @xmath184 . more specifically , @xmath183 is a tree - encoding of the formula , where @xmath185 is true , if node @xmath58 is @xmath186 and false , if @xmath58 is @xmath187 . the evaluation @xmath173 takes the value of the variables given by @xmath179 and then evaluates the formula in a bottom - up fashion using a standard tree encoding . thus , the value of the formula can be read at @xmath188 . it is interesting to observe that @xmath182 does not hold in @xmath0 . this is , since an application of the witnessing theorem for @xmath0 to a proof of @xmath182 would yield an @xmath2-definition of satisfaction for monotone @xmath86 circuits . this implies that monotone @xmath189 , which is known to be false . in this section we will remind the reader of a connection between the theory @xmath0 and some of its extensions and certain propositional proof systems ( see also @xcite@xcite ) . the following predicates will be subsequently used . they are definable with respect to @xmath0 ( see @xcite ) . @xmath69 @xmath190 is a @xmath43 formula that says that the the string @xmath102 codes a formula . @xmath191 is the @xmath192 definable property that the truth assignment @xmath193 satisfies the formula @xmath102 . @xmath194 is the @xmath195 formula @xmath196 , where @xmath197 is an upper bound to the number of variables in formulas coded by strings of length @xmath101 . @xmath198 is a @xmath43 definable predicate meaning @xmath199 is a depth @xmath161 frege proof for @xmath84 . @xmath200 is a @xmath43 definable predicate meaning @xmath199 is a frege proof for @xmath84 . the following holds [ fact : vz and bd frege ] the theory @xmath0 proves that @xmath2-frege is sound , i.e. for every @xmath161 @xmath201 [ fact : vnc1 and nc1 frege ] the theory @xmath1 proves that frege is sound , i.e. @xmath202 on the other hand , provability of the universal closure of @xmath43 formulas in @xmath0 and @xmath1 implies the existence of polynomial size proofs of their propositional translations in @xmath2-frege and frege , respectively . the propositional translation @xmath203 of a @xmath43 formula @xmath204 is a family of propositional formulas built up inductively ( on the logical depth ) as follows . if @xmath176 is atomic and does not contain second sort variables , we evaluate @xmath176 in @xmath27 , if it contains second sort variables , we have to introduce propositional variables . if @xmath176 is a boolean combination of formulas @xmath205 of lower depth , the translation is simply the same boolean combination of the translations of the @xmath205 . if @xmath176 is @xmath206 or @xmath207 we translate it to the disjunction or conjunction of the translations , respectively . for a proper definition see @xcite . [ fact : polynomial simulation ] there exists a polynomial @xmath72 such that for all @xmath43 formulas @xmath204 the following holds @xmath69 if @xmath208 , then there exist bounded depth frege proofs of all @xmath209 of length at most @xmath210 , for any @xmath211 . if @xmath212 , then there exist frege proofs of all @xmath209 of length at most @xmath210 , for any @xmath211 . these proofs are effective in the sense that for any such @xmath176 there exists a polynomial - time computable function @xmath213 that maps any tuple @xmath214 to the above proofs of @xmath215 . facts [ fact : vz and bd frege ] and [ fact : vnc1 and nc1 frege ] are examples of general principles , the so called _ reflection principles _ , which are defined as follows . [ reflection principle ] let @xmath66 be a pps . then the _ reflection principle _ for @xmath66 , @xmath216 , is the @xmath217-formula ( w.r.t . @xmath0 ) @xmath218 where @xmath219 is a @xmath192-predicate formalizing @xmath66-proofs . reflection principles condense the strength of propositional proof systems . in what follows we will summarize some such results for the proof systems and theories used here . a detailed exposition can be found in @xcite , chapter x , or in @xcite , chapter 9.3 . [ thm simulation by reflection ] if @xmath220 then bounded depth frege p - simulates frege . we will only give a brief sketch of the proof here and leave out the technical details . let @xmath176 be a formula and @xmath221 a frege proof of @xmath176 which is witnessed by a turing machine @xmath222 ( cf def [ def pps ] ) . since @xmath0 proves @xmath223 , by facts [ fact : vz and bd frege ] and [ fact : polynomial simulation ] we have polynomial size proofs of its translations @xmath224 in bounded depth frege . bounded depth frege itself , however , is strong enough to verify that a proper encoding of the computation of @xmath222 on input @xmath225 is correct . thus it can verify that @xmath221 is a frege - proof and , using the translation of the reflection principle and the cut rule , conclude @xmath226 . from this @xmath176 follows , cf . @xcite lemma 9.3.7 . given a term @xmath197 and a variable @xmath58 , we can also introduce the @xmath197-bounded version of the reflection principle for some given pps @xmath66 , @xmath227 that claims soundness only for @xmath197-bounded proofs . [ bounded reflection ] let @xmath197 be a @xmath228-term , @xmath58 a first - sort variable and @xmath66 a pps . then the _ bounded reflection principle _ @xmath227 is the formula @xmath229 we can now generalize theorem [ thm simulation by reflection ] in the following way . [ thm simulation by bounded reflection ] let @xmath197 be a @xmath228-term and @xmath58 a number variable . if @xmath230 for @xmath58 large enough and if @xmath231 then for every propositional formula @xmath176 with a frege proof of length @xmath232 there is a bounded depth frege proof of @xmath176 of length @xmath233 . this proof can be efficiently constructed . the proof is the same as that of theorem [ thm simulation by reflection ] . using the bounded reflection principle we can encode frege proofs of length @xmath232 as bounded depth frege proofs of length @xmath233 . as a corollary we get if @xmath234 for all @xmath235 , then bounded depth frege sub exponentially simulates frege : for all @xmath236 exists @xmath237 , such that the existence of a frege proof of length @xmath238 of a depth @xmath239 formula implies the existence of a depth @xmath161 frege proof of length at most @xmath240 . we will first introduce the notion of a cut @xmath242 of a given two - sorted arithmetic model @xmath243 . this model theoretic approach provides a very good insight on what actually happens semantically with the small elements of arithmetical models . [ cut ] let @xmath244 be a two - sorted arithmetic theory and @xmath245 a model of @xmath244 . @xmath246 of @xmath26 is any substructure such that @xmath69 @xmath247 , @xmath248 , @xmath249 , @xmath250 , @xmath251 is closed under @xmath252 and downwards with respect to @xmath253 , @xmath254 , and @xmath255 is the restriction of @xmath256 to @xmath251 and @xmath257 for all relation and function symbols @xmath258 . we call this cut the _ polylogarithmic cut _ iff @xmath259 to examine the strength of the theory of such cuts of models of @xmath0 , we will show that a formal connection between efficient computability and @xmath43-definability holds . this stands in contrast to general bounded subsets , where the connection is presumably only with respect to @xmath147-definability via the predicate @xmath175 ( see on page ) . the intended theorem is a formalization of nepomnjascij s theorem @xcite ( see also @xcite pg.20 ) . we will sketch the original proof before starting the formalization . [ thm nepomnjascij ] let @xmath260 and @xmath261 be constants . then if the language @xmath262 , the relation @xmath263 is definable by a @xmath43-formula over @xmath30 . we will prove the theorem by induction on @xmath124 for @xmath264 . let @xmath265 and @xmath266 . for any @xmath267 the whole computation can be coded by a number @xmath268 of size @xmath269 . the existence of such a computation gives the desired @xmath43-definability . for @xmath270 we write a sequence @xmath271 of intermediate results coding the computation , where @xmath272 codes the starting configuration on input @xmath58 , such that we can verify that @xmath273 is computable from @xmath274 in @xmath275 . by assumption there exists a @xmath43-formula @xmath276 such that @xmath277 holds iff @xmath273 is computed from @xmath274 . additionally , the whole sequence has length @xmath142 and so we can write the sequence of intermediate results @xmath274 as a number @xmath268 of length @xmath142 . now , the @xmath43-definition of @xmath263 is simply @xmath278 we will now formalize this result in @xmath0 as follows [ thm nepomnjascij_formalized ] let @xmath279 . let @xmath280 for some @xmath281 and let @xmath282 and @xmath283 . if @xmath284 ( for strings of length @xmath238 ) is computed by turing machine @xmath84 , then there exists a @xmath43 definition in @xmath26 of the @xmath147-predicate @xmath285 on the interval @xmath286 $ ] . i.e. any @xmath287 , bounded by @xmath288 is @xmath43-definable in @xmath26 and therefore exists in the polylogarithmic cut of @xmath26 . the following version of the proof stems from a discussion with stephen cook and neil thapen during the sas programme in cambridge . it is more explicit than the original one and clarifies the argument . we will inductively on @xmath161 define a @xmath43 relation @xmath289 that states that the @xmath290th cell of the work tape of @xmath84 , starting on configuration @xmath179 and computing for @xmath291 steps via the computation @xmath292 is @xmath293 . we will bound the quantifiers in such a way that we can conclude that both variables can be of the number sort . as @xmath161 depends only on @xmath84 and @xmath124 we will be doing this induction outside of the theory to construct @xmath161 many formulas . we will then prove the above mentioned properties of @xmath294 by @xmath43 induction on @xmath295 . keep in mind that a cell is given as a pair @xmath296 , where @xmath297 is the actual value of the cell and @xmath298 is a number @xmath299 coding the state the turing machine is in iff the pointer is on that cell and @xmath18 otherwise . as before the transition function is denoted by @xmath155 . we let @xmath179 be a string coding the input at the start of the computation . that is , @xmath179 is a sequence of length @xmath300 , such that @xmath301=1 $ ] , @xmath302=0 $ ] for all @xmath303 and @xmath304 $ ] is the @xmath140th input bit . we let @xmath305 be a sequence encoding the computation of @xmath84 , such that @xmath306 is the state , the machine is in after @xmath140 steps ( 0 denotes that the read / write head is not on cell @xmath307 , while a greater number gives the state and witnesses that the read / write head is on cell @xmath307 ) . @xmath308 is the value of cell @xmath307 after @xmath140 steps of the computation . observe that this also implies that the computation can be encoded as a number , that is , it has to be very short . this is straight forward from the quantifier bounds . we can now define @xmath309\land \langle\mathsf{comp}\rangle_{\langle 1,j',1\rangle } \approx i[j',1]{\large ) } \land\\ & \phantom{{}\equiv { } } [ \forall j<\lceil len(i)^{\frac{1-k\epsilon}{k}}\rceil , j'<\lceil len(i)^{\epsilon}\rceil,\alpha<{\left\verta\right\vert}\\ & \phantom{{}\equiv{}}\hspace{16pt } ( \langle\mathsf{comp}\rangle_{\langle j , j',1\rangle } = 0\rightarrow ( \langle\mathsf{comp } \rangle_{\langle j , j',0\rangle } = \langle\mathsf{comp}\rangle_{\langle j+1,j',0\rangle}))\land\\ & \phantom{{}\equiv{}}\hspace{16pt } ( \langle\mathsf{comp}\rangle_{\langle j , j',1\rangle } = \alpha\rightarrow(\\ & \phantom{{}\equiv{}}\hspace{24pt } ( \langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_3=0\rightarrow ( \langle\mathsf{comp}\rangle_{\langle j+1,j',1\rangle}=\langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_1 \land\\ & \hspace{183 pt } \langle\mathsf{comp}\rangle_{\langle j+1,j',0\rangle}=\langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_2))\land\\ & \phantom{{}\equiv{}}\hspace{24pt } ( \langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_3=1\rightarrow \langle\mathsf{comp}\rangle_{\langle j+1,j'{\mathbin{\setbox0\hbox{$-$}\setbox2\hbox to\wd0{\hss$^{\mkern1mu\cdot}$\hss}\wd2=0pt\box2\box0}}1,1\rangle}=\langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_1 \land\\ & \hspace{179 pt } \langle\mathsf{comp}\rangle_{\langle j+1,j',0\rangle}=\langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_2))\land\\ & \phantom{{}\equiv{}}\hspace{24pt } ( \langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_3=2\rightarrow \langle\mathsf{comp}\rangle_{\langle j+1,j'+1,1\rangle}=\langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_1 \land\\ & \hspace{179 pt } \langle\mathsf{comp}\rangle_{\langle j+1,j',0\rangle}=\langle \delta(\alpha,\langle\mathsf{comp}\rangle_{\langle j , j',0\rangle})\rangle_2))))\land\\ & \phantom{{}\equiv{}}\hspace{16pt } \forall\ell\neq\ell'<\lceil len(i)^{\epsilon}\rceil ( \langle\mathsf{comp}\rangle_{\langle j,\ell,1\rangle}>0\rightarrow \langle\mathsf{comp}\rangle_{\langle j,\ell',1\rangle}=0)]\land\\ & \phantom{{}\equiv { } } ( \langle \mathsf{cell}\rangle_1 = \langle\mathsf{comp}\rangle_{\langle p_1,p_2,0\rangle})\land ( \langle \mathsf{cell}\rangle_2 = \langle\mathsf{comp}\rangle_{\langle p_1,p_2,1\rangle } ) . \end{split}\ ] ] it is straightforward to prove by induction on the number of lines in @xmath292 that @xmath292 is uniquely defined by @xmath310 . we let @xmath311 where @xmath312 is some polynomial depending on the encoding . here it is vital that @xmath312 can be defined such that @xmath313 is a number in @xmath26 . this is possible due to the quantifier bounds we used when defining @xmath310 . thus , @xmath314 is defined by a @xmath43 formula . informally @xmath314 formalizes that there is a computation @xmath315 that is correct in the sense that we can verify that we get from line to line via the transition function of @xmath84 and gives the appropriate values of the cell at @xmath316 . observe that the size of the whole computation as presented above is linear in @xmath238 , i.e. that it can be coded as a number in @xmath26 . we will now proceed by inductively defining @xmath294 and @xmath317 . assume that @xmath318 has already been defined by a @xmath43 formula over @xmath26 . we then let @xmath319\land \langle\mathsf{comp}\rangle_{\langle 1,j',1\rangle } \approx i[j',1])\land\\ & \phantom{{}\equiv { } } ( \forall j<\lceil len(i)^{\frac{1-k\epsilon}{k}}\rceil\exists\mathsf{comp'}<q(len(i))\forall j'<\lceil len(i)^{\epsilon}\rceil\exists \mathsf{cell}'<{\left\verta\right\vert}\forall j''<{\left\verta\right\vert}\\ & \phantom{{}\equiv { } } ( \langle\mathsf{comp}\rangle_{\langle j+1,j',j''\rangle}\leftrightarrow \langle \mathsf{cell}'\rangle_{j''})\land\\ & \phantom{{}\equiv { } } \mathsf{reach}^{d{\mathbin{\setbox0\hbox{$-$}\setbox2\hbox to\wd0{\hss$^{\mkern1mu\cdot}$\hss}\wd2=0pt\box2\box0}}1}_a(\langle \mathsf{comp}\rangle_{\langle j,\cdot,\cdot\rangle},m^{\frac{1-k\epsilon}{k } } , j',\mathsf{cell}',\mathsf{comp'}))\land\\ & \phantom{{}\equiv { } } ( \langle \mathsf{cell}\rangle_1\leftrightarrow\langle\mathsf{comp}\rangle_{\langle p_1,p_2,0\rangle})\land ( \langle \mathsf{cell}\rangle_2=\langle\mathsf{comp}\rangle_{\langle p_1,p_2,1\rangle } ) . \end{split}\ ] ] again , we can prove uniqueness of the computation by induction on the number of its lines and let @xmath320 that this is a @xmath43 definition follows by induction and the same argument as for @xmath314 . here , the predicate @xmath321 takes the role of the transition function in witnessing that each line follows from the preceding one . the total size again is linear in @xmath238 . we now can give a @xmath43 definition of the predicate @xmath322 $ ] coding the computation as in @xmath285 on input @xmath102 of length @xmath288 . we let @xmath322=\mathsf{cell}\equiv$ ] @xmath323 where @xmath72 is a polynomial depending on the encoding and @xmath324 is the starting configuration of @xmath84 on input @xmath102 . informally the above formula says that we compute the configurations of @xmath84 by using the predicates @xmath317 through @xmath314 . that is , after the application of @xmath317 ( i.e. after making the biggest steps ) we have reached configuration @xmath325 , which we plug into @xmath326 to get configuration @xmath327 and so on . it remains to show that this definition of @xmath328 $ ] coincides with the real one , i.e. that @xmath329 $ ] follows from an application of the transition function of @xmath84 from @xmath330 $ ] . we will prove this inductively , depending on @xmath122 . again let @xmath331 be such that @xmath332 . if @xmath333 the assumption follows straightforwardly from the definition of @xmath314 . now for bigger @xmath122 . if the @xmath334 , given as above , is bigger then @xmath18 then again the assumption follows from the definition of @xmath310 . now let @xmath335 be the first index with @xmath336 . we the have to argue that @xmath337 has the desired property . this , however , follows straightforward if we can verify this assertion for @xmath337 . observe that @xmath161 is a constant depending only on @xmath84 and @xmath124 , so we need to make this argument only a constant number of steps to reach @xmath310 , where we know that the assertion holds . since we can code the whole computation as a @xmath43-formula ( in @xmath26 ) , we can easily deduce a @xmath43-definition of the related set by simply stating that the computation accepts ( i.e. that in the last line of the computation the state is accepting ) . this concludes the proof . we can now prove our main result . [ thm : conclusion nepo ] let @xmath338 and @xmath339 be the polylogarithmic cut . then @xmath340 . we have to prove that for all strings @xmath341 , representing a formula @xmath176 as a tree and assignments @xmath342 to its variables ( i.e. leafs in the tree representation ) a string @xmath173 exists in @xmath257 that contains all values of @xmath176 s subformulas as in the definition of @xmath182 in section [ sec : ext of vz ] and satisfies the inductive conditions of @xmath182 . however , by @xmath43-comprehension and the formalized nepomnjascij s theorem it suffices to describe an algorithm that computes , for given @xmath343 and @xmath179 , whether @xmath122 belongs to @xmath173 in @xmath344 for some @xmath235 and @xmath345 . the following is a recursive algorithm computing the value of @xmath346 $ ] , given @xmath347 and @xmath122 . ` nodevalue(g , i , i ) ` * ` boolean left ; boolean right ; ` * ` if i>2`@xmath348`|g| ` * * ` output ( 0 ) ; end ; ` * ` else if i>|g| ` * * ` output ( i[i-|g| ] ) ; end ; ` * ` else if g[i]=1 ` * * ` left : = nodeval(g , i,2i ) ; ` * * ` right : = nodeval(g , i,2i+1 ) ; ` * * ` output ( left and right ) ; end ; ` * ` else if g[i]=0 ` * * ` left : = nodeval(g , i,2i ) ; ` * * ` right : = nodeval(g , i,2i+1 ) ; ` * * ` output ( left or right ) ; end ; ` * ` else ` * * ` output ( 0 ) ; end ; ` observe that the algorithm at any given point only stores a constant amount of data per level of the tree @xmath183 and therefore uses only @xmath349 space . the number steps the algorithm makes is clearly polynomial in the size of @xmath183 . therefore by theorem [ thm nepomnjascij_formalized ] , for every monotone formula @xmath176 , representable as a tree in @xmath13 , we get a @xmath43 formula @xmath350 , such that @xmath351 $ ] . observe that @xmath350 depends on the size of @xmath176 and on its logical depth , as the first is essentially the size of the input for the machine , that @xmath352 codes , while the latter determines the longest iterations in the recursive algorithm . applying the comprehension schema in @xmath0 , i.e. in @xmath26 , this verifies the existence of a @xmath173 as in @xmath182 for all formulas represented by trees in @xmath13.therefore @xmath182 holds in @xmath13 and so @xmath353 . we now wish to apply the above results to propositional proof systems . more precisely we wish to show that theories of small cuts of a model of a given theory @xmath110 correspond to stronger proof systems than @xmath110 does . an elegant way of showing such a statement is via the _ reflection principles _ of the given proof systems , i.e. the statement that the proof system is correct , as explained in section [ sec : rel ba pps ] . with their help we can conclude the following recent result of filmus , pitassi and santhanam @xcite . [ thm filpitsat ] every frege system is sub exponentially simulated by @xmath2-frege systems . by theorem [ thm simulation by bounded reflection ] we have to prove the polylogarithmically bounded reflection principle for frege in @xmath0 . this , by theorem [ thm : conclusion nepo ] however , corresponds to proving the reflection principle for frege in @xmath1 , which holds by fact [ fact : vnc1 and nc1 frege ] . it also follows from theorem [ thm simulation by bounded reflection ] that this proof can be efficiently computed from the frege proof . another , related , application is in the separation of propositional proof systems . in @xcite we proved the following . [ prop iddo me ] for almost every random 3cnf @xmath84 with @xmath25 variables and @xmath354 clauses , where @xmath162 is a large constant , @xmath355 has polynomially bounded @xmath5-frege proofs . on the other hand it is well known ( see for example @xcite ) that such formulas have no subexponential refutation in resolution . thus , this yields an average case separation between resolution and @xmath5-frege . we can now extend this result to an average case separation between resolution and @xmath2-frege as follows . for almost every random 3cnf @xmath84 with @xmath25 variables and @xmath354 clauses , where @xmath162 is a large constant , @xmath355 has subexponentially bounded @xmath2-frege proofs . by theorem [ thm : conclusion nepo ] the polylogarithmic cut of any @xmath0-model is a model of @xmath1 , therefore also of @xmath4 . this yields , as in our proof of theorem [ thm filpitsat ] , that @xmath2-frege subexponentially simulates @xmath5-frege . the result now follows from proposition [ prop iddo me ] . as we have seen cuts of models of weak arithmetics constitute an appropriate way for reasoning about super - polynomial simulations between proof systems . an advantage in comparison to syntactic arguments is the possible applicability of results in model theory and a more uniform treatment . this can readily be observed as with our argument , e.g. the work of paris and wilkie @xcite@xcite immediately imply the simulation results from bonet et al . @xcite . this leads to interesting possibilities for further research , especially towards the weak automatizability of weak propositional proof systems such as resolution . the underlying theory , which was @xmath0 in our argument , must be significantly weakened , however . if we could take @xmath356 as our base theory , we could reason about whether @xmath357 has the feasible interpolation property in the same way as kraj ' iek and pudl ' ak @xcite , bonet , pitassi and raz @xcite or bonet , domingo , gavald , maciel , and pitassi @xcite . now , if @xmath357 does not have quasi - polynomial feasible interpolation we know by a result from atserias and bonet @xcite that resolution is not weakly automatizable , so we would be finished . whether we can actually do it depends on the strength of the theory the polylogarithmic cut of @xmath356 models and if we can formalize some sort of iterated multiplication ( such as in @xcite ) in that theory . also , the security of diffie - hellman seems to be a more appropriate assumption than that of rsa , as the computational power needed to verify the correctness of diffie - hellman seems to be lower . i want to thank steve cook , jan kraj ' iek and neil thapen for helpful suggestions and discussion , emil je ' abek for his comments and for answering my questions and the participants of the maloa special semester in proof complexity in prague 2011 for enduring a sloppy and sometimes faulty exposition of this proof and still coming up with helpful comments . i also want to thank the anonymous referees for pointing out various mistakes and for giving interesting suggestions . a similar construction can be extracted from @xcite and leads to similar results , if perceived in the way we did it here . i want to thank leszek kolodziejczyk for pointing this out .

the seminal paper @xcite introduced an important class of sequential optimization problems , otherwise known as multi - armed bandits . these models have since been used extensively in such fields as statistics , operations research , engineering , computer science and economics . the traditional multi - armed bandit problem can be described as follows . consider @xmath0 statistical populations ( arms ) , where at each point in time it is possible to sample from ( pull ) only one of them and receive a random reward dictated by the properties of the sampled population . the objective is to devise a sampling policy that maximizes expected cumulative rewards over a finite time horizon . the difference between the performance of a given sampling policy and that of an oracle , that repeatedly samples from the population with the highest mean reward , is called the _ regret_. thus , one can re - phrase the objective as minimizing the regret . when the populations being sampled are homogeneous , that is , when the sequential rewards are independent and identically distributed ( i.i.d . ) in each arm , the family of upper - confidence - bound ( ucb ) policies , introduced in @xcite , incur a regret of order @xmath1 , where @xmath2 is the length of the time horizon , and no other `` good '' policy can ( asymptotically ) achieve a smaller regret ; see also @xcite . the elegance of the theory and sharp results developed in @xcite hinge to a large extent on the assumption of homogenous populations and hence identically distributed rewards . this , however , is clearly too restrictive for many applications of interest . often , the decision maker observes further information and based on that , a more _ customized _ allocation can be made . in such settings , rewards may still be assumed to be independent , but no longer identically distributed in each arm . a particular way to encode this is to allow for an exogenous variable ( a covariate ) that affects the rewards generated by each arm at each point in time when this arm is pulled . such a formulation was first introduced in @xcite under parametric assumptions and in a somewhat restricted setting ; see @xcite and @xcite for very different recent approaches to the study of such bandit problems , as well as references therein for further links to antecedent literature . the first work to venture outside the realm of parametric modeling assumptions appeared in @xcite . in particular , the mean response in each arm , conditionally on the covariate value , was assumed to follow a general functional form ; hence one can view their setting as a _ nonparametric _ bandit problem . they propose a variant of the @xmath3-greedy policy ( see , e.g. , @xcite ) and show that the average regret tends to zero as the time horizon @xmath2 grows to infinity . however , it is unclear whether this policy satisfies a more refined notion of optimality , insofar as the magnitude of the regret is concerned , as is the case for ucb - type policies in traditional bandit problems . such questions were partially addressed in @xcite where near - optimal bounds on the regret are proved in the case of a two - armed bandit problem under only two assumptions on the underlying functional form that governs the arms responses . the first is a mild smoothness condition , and the second is a so - called _ margin condition _ that involves a _ margin parameter _ which encodes the `` separation '' between the functions that describe the arms responses . the purpose of the present paper is to extend the setup of @xcite to the @xmath4-armed bandit problem with covariates when @xmath4 may be large . this involves a customized definition of the margin assumption . moreover , the bounds proved in @xcite suffered two deficiencies . first , they hold only for a limited range of values of the margin parameter and second , the upper bounds and the lower bounds mismatch by a logarithmic factor . improving upon these results requires radically new ideas . to that end , we introduce three policies : successive elimination ( se ) is dedicated to the static bandit case . it is the cornerstone of the other policies that deal with covariates . during a first phase , this policy explores the different arms , builds estimates and eliminates sequentially suboptimal arms ; when only one arm remains , it is pulled until the horizon is reached . a variant of se was originally introduced in @xcite . however , it was not tuned to minimize the regret as other measures of performance were investigated in this paper . we prove new regret bounds for this policy that improve upon the canonical papers @xcite and @xcite . binned successive elimination ( bse ) follows a simple principle to solve the problem with covariates . it consists of grouping similar covariates into bins and then looks only at the average reward over each bin . these bins are viewed as indexing `` local '' bandit problems , solved by the aforementioned se policy . we prove optimal regret bounds , polynomial in the horizon but only for a restricted class of difficult problems . for the remaining class of easy problems , the bse policy is suboptimal . adaptively binned successive elimination ( abse ) overcomes a severe limitation of the naive bse . indeed , if the problem is globally easy ( this is characterized by the margin condition ) , the bse policy employs a fixed and too fine discretization of the covariate space . instead , the abse policy partitions the space of covariates in a fashion that adapts to the local difficulty of the problem : cells are smaller when different arms are hard to distinguish and bigger when one arm dominates the other . this adaptive partitioning allows us to prove optimal regrets bounds for the whole class of problems . the optimal polynomial regret bounds that we prove are much larger than the logarithmic bounds proved in the static case . nevertheless , it is important to keep in mind that they are valid for a much more flexible model that incorporates covariates . in the particular case where @xmath5 and the problem is _ difficult _ , these bounds improve upon the results of @xcite by removing a logarithmic factor that is idiosyncratic to the _ exploration vs. exploitation _ dilemma encountered in bandit problems . moreover , it follows immediately from the previous minimax lower bounds of @xcite and @xcite , that these bounds are optimal in a minimax sense and thus can not be further improved . it reveals an interesting and somewhat surprising phenomenon : the price to pay for the partial information in the bandit problem is dominated by the price to pay for nonparametric estimation . indeed the bound on the regret that we obtain in the bandit setup for @xmath5 is of the same order as the best attainable bound in the _ full information _ case , where at each round , the operator receives the reward from only one arm but observes the rewards of both arms . an important example of the full information case is sequential binary classification . our policies for the problem with covariates fall into the family of `` plug - in '' policies as opposed `` minimum contrast '' policies ; a detailed account of the differences and similarities between these two setups in the full information case can be found in @xcite . minimum contrast type policies have already received some attention in the bandit literature with side information , aka _ contextual bandits _ , in the papers @xcite and also @xcite . a related problem online convex optimization with side information was studied in @xcite , where the authors use a discretization technique similar to the one employed in this paper . it is worth noting that the cumulative regret in these papers is defined in a weaker form compared to the traditional bandit literature , since the cumulative reward of a proposed policy is compared to that of the best policy in a certain restricted class of policies . therefore , bounds on the regret depend , among other things , on the complexity of said class of policies . plug - in type policies have received attention in the context of the continuum armed bandit problem , where as the name suggests there are uncountably many arms . notable entries in that stream of work are @xcite and @xcite , who impose a smoothness condition both on the space of arms and the space of covariates , obtaining optimal regret bounds up to logarithmic terms . in this section , it will be convenient for notational purposes , to consider a multi - armed bandit problem with @xmath6 arms . we revisit the successive elimination ( se ) policy introduced in @xcite in the traditional setup of multi - armed bandit problems . as opposed to the more popular ucb policy ( see , e.g. , @xcite ) , it allows us in the next section , to construct an adaptive partition that is crucial to attain optimal rates on the regret for the dynamic case with covariates . in this section , we prove refined regret bounds for the se policy that exhibit a better dependence on the expected rewards of the arms compared to the bounds for ucb that were derived in @xcite . such an improvement was recently attempted in @xcite and also in @xcite for modified ucb policies and we compare these results to ours below . let us recall the traditional setup for the static multi - armed bandit problem ; see , for example , @xcite . let @xmath7 be a given set of @xmath8 arms . successive pulls of arm @xmath9 yield rewards @xmath10 that are i.i.d . random variables in @xmath11 $ ] with expectation given by @xmath12 = f^{(i)}\in[0,1]$ ] . assume without loss of generality that @xmath13 so that @xmath6 is one of the best arms . for simplicity , we further assume that the best arm is _ unique _ since for the se policy , having multiple optimal arms only improves the regret bound . in the analysis , it is convenient to denote this optimal arm by @xmath14 and to define the _ gaps _ traditionally denoted by @xmath15 , by @xmath16 . a _ policy _ @xmath17 is a sequence of random variables @xmath18 indicating which arm to pull at each time @xmath19 , and such that @xmath20 depends only on observations strictly anterior to @xmath21 . the performance of a policy @xmath22 is measured by its ( _ cumulative _ ) _ regret _ at time @xmath2 defined by @xmath23 note that for a data - driven policy @xmath24 , this quantity is random and , in the rest of the paper , we provide upper bounds on @xmath25 . such bounds are referred to as _ regret bounds_. we begin with a high - level description of the se policy denoted by @xmath24 . it operates in rounds that are different from the decision times @xmath26 . at the beginning of each round @xmath27 , a subset of the arms has been eliminated and only a subset @xmath28 remains . during round @xmath29 , each arm in @xmath28 is pulled exactly once ( exploration ) . at the end of the round , for each remaining arm in @xmath28 , we decide whether to eliminate it using a simple statistical hypothesis test : if we conclude that its mean is significantly smaller than the mean of any remaining arm , then we eliminate this arm and we keep it otherwise ( elimination ) . we repeat this procedure until @xmath2 pulls have been made . the number of rounds is random but obviously smaller than @xmath2 . the se policy , which is parameterized by two quantities @xmath30 and @xmath31 and described in policy [ algse ] , outputs an infinite sequence of arms @xmath32 without a prescribed horizon . of course , it can be truncated at any horizon @xmath2 . this description emphasizes the fact that the policy can be implemented without perfect knowledge of the horizon @xmath2 and in particular , when the horizon is a random variable with expected value @xmath2 ; nevertheless , in the static case , it is manifest from our result that , when the horizon is known to be @xmath2 , choosing @xmath33 is always the best choice when possible and that other choices may lead to suboptimal results . set of arms @xmath34 ; parameters @xmath35 ; horizon @xmath2 . @xmath37 , @xmath38 , @xmath39 , @xmath40^k$ ] @xmath41 @xmath42 @xmath43 ( observe @xmath44 ) exploration @xmath45 , @xmath46 @xmath47 $ ] @xmath48 . elimination @xmath49 . note that after the exploration phase of each round @xmath50 each remaining arm @xmath51 has been pulled exactly @xmath29 times , generating rewards @xmath52 . denote by @xmath53 the average reward collected from arm @xmath54 at round @xmath29 that is defined by @xmath55 , where here and throughout this paper , we use the convention @xmath56 . in the rest of the paper , @xmath57 denotes the natural logarithm and @xmath58 . for any positive integer @xmath59 , define also @xmath60 which is essentially a high probability upper bound on the magnitude of deviations of @xmath61 from its mean @xmath62 . the se policy for a @xmath4-armed bandit problem can be implemented according to the pseudo - code of policy [ algse ] . note that , to ease the presentation of sections [ secbse ] and [ secabse ] , the se policy also returns at each time @xmath21 , the number of rounds @xmath63 completed at time @xmath21 and a subset @xmath64 of arms that are active at time @xmath21 , where @xmath65 denotes the power set of @xmath66 . the following theorem gives a first upper bound on the expected regret of the se policy . [ thstatic ] consider a @xmath67-armed bandit problem where horizon is a random variable @xmath68 of expectation @xmath2 that is independent of the random rewards . when implemented with parameters @xmath69 , the se policy @xmath24 exhibits an expected regret bounded , for any @xmath70 , as @xmath71\leq392 \gamma^2 \biggl(1 + \frac{n}{t } \biggr)\frac { k}{\delta}{\overline{\log}}\biggl(\frac{t\delta^2}{18\gamma^2 } \biggr ) + n \delta^-,\ ] ] where @xmath72 is the largest @xmath73 such that @xmath74 if it exists , otherwise @xmath75 . assume without loss of generality that @xmath76 , for @xmath77 since arms @xmath78 such that @xmath79 do not contribute to the regret . define @xmath80 . moreover , for any @xmath81 in the set @xmath28 of arms that remain active at the beginning of round @xmath29 , define @xmath82 . recall that , at round @xmath29 , if arms @xmath83 , then ( i ) the optimal arm @xmath84 eliminates arm @xmath81 if @xmath85 , and ( ii ) arm @xmath81 eliminates arm @xmath84 if @xmath86 . since @xmath87 estimates @xmath88 , the event in ( i ) happens approximately , when @xmath89 , so we introduce the deterministic , but unknown , quantity @xmath90 ( and its approximation @xmath91 ) defined as the solution of @xmath92 note that @xmath93 as well as the bound @xmath94 we are going to decompose the regret accumulated by a suboptimal arm @xmath81 into three quantities : * the regret accumulated by pulling this arm at most until round @xmath95 : this regret is smaller than @xmath96 ; * the regret accumulated by eliminating the optimal arm @xmath84 between round @xmath97 and @xmath95 ; * the regret induced if arm @xmath81 is still present at round @xmath95 ( and in particular , if it has not been eliminated by the optimal arm @xmath84 ) . we prove that the second and third events happen with small probability , because of the choice of @xmath95 . formally , define the following _ good _ events : @xmath98 moreover , define @xmath99 and observe that @xmath100 . for any @xmath101 , the contribution to the regret incurred after time @xmath95 on @xmath102 is at most @xmath103 since each pull of arm @xmath104 contributes to the regret by @xmath105 . we decompose the underlying sample space denoted by @xmath106 into the disjoint union @xmath107 where @xmath108 is chosen later . it implies the following decomposition of the expected regret : @xmath109 define by @xmath110 the complement of an event @xmath111 . note that the first term on the right - hand side of the above inequality can be decomposed as follows : @xmath112\\[-8pt ] & & { } + n\sum _ { i=1}^{k_0}\delta_i { \mathbb p}\bigl ( { \mathcal{b}}_i^c\cap{\mathcal{a}}_i\cap{\mathcal{b}}_{i-1 } \bigr),\nonumber\end{aligned}\ ] ] where the right - hand side was obtained using the decomposition @xmath113 and the fact that @xmath114 . from hoeffding s inequality , we have that for every @xmath115 , @xmath116\\[-8pt ] & \leq&\exp\biggl(-\frac{\tau(\delta_i-\gamma\varepsilon _ \tau ) ^2}{2 } \biggr).\nonumber\end{aligned}\ ] ] on the event @xmath117 , arm @xmath84 has not eliminated arm @xmath81 at @xmath95 . therefore @xmath118 . together with the above display with , it yields @xmath119 where we used the fact that @xmath120 . it remains to bound the first term in the right - hand side of ( [ eqdecompci ] ) . on the event @xmath121 , the optimal arm @xmath84 has not been eliminated before round @xmath122 , but every suboptimal arm @xmath123 has . so the probability that there exists an arm @xmath124 that eliminates @xmath84 between @xmath122 and @xmath95 can be bounded as @xmath125,\end{aligned}\ ] ] where @xmath126 . using lemma [ lempeeling ] , we get @xmath127 . this bound implies that @xmath128 \\ & & \qquad\le\sum_{j=1}^k \sum_{i=1}^{j\wedge k_0 - 1 } \phi_j ( \tau_i ) ( \delta_i -\delta_{i+1 } ) + \sum _ { j=1}^k \phi_{j\wedge k_0}(\tau _ { j\wedge k_0})\delta_{j\wedge k_0 } \\ & & \qquad\leq\frac{4}{t } \sum_{j=1}^k \sum_{i=1}^{j\wedge k_0 - 1 } \tau_i ( \delta_i -\delta_{i+1 } ) + \frac{4}{t}\sum _ { j=1}^k \tau_{j\wedge k_0}\delta_{j\wedge k_0}.\end{aligned}\ ] ] using ( [ eqboundtau ] ) and @xmath129 , the first sum can be bounded as @xmath130.\end{aligned}\ ] ] the previous two displays together with ( [ eqboundtau ] ) yield @xmath131 putting together ( [ eqregret1 ] ) , ( [ eqdecompci ] ) , ( [ eqboundbi ] ) and the above display yield that the expected regret @xmath132of the se policy is bounded above by @xmath133\\[-8pt ] & & \qquad{}+304\frac{\gamma^2n}{t } \frac{k - k_0}{\delta_{k_0}}{\overline{\log}}\biggl(\frac{n\delta^2_{k_0}}{18\gamma^2 } \biggr)+n\delta_{k_0 + 1}.\nonumber\end{aligned}\ ] ] fix @xmath134 and let @xmath135 be such that @xmath136 . an easy study of the variations of the function @xmath137 reveals that @xmath138 for any @xmath139 . using this bound equation ( [ eqregretgen ] ) with @xmath140 and @xmath141 completes the proof . the following corollary is obtained from a slight variations on the proof of theorem [ thstatic ] . it allows us to better compare our results to the extant literature . [ corstatic ] under the setup of theorem [ thstatic ] , the se policy @xmath24 run with parameter @xmath33 and @xmath142 satisfies for any @xmath143 , @xmath144 in particular , @xmath145 note that ( [ eqcorstatic1 ] ) follows from ( [ eqregretgen ] ) . to prove ( [ eqthstatic ] ) , take @xmath146 in ( [ eqcorstatic1 ] ) and @xmath147 in theorem [ thstatic ] , respectively . this corollary is actually closer to the result of @xcite . the additional second term in our bound comes from the fact that we had to take into account the probability that an optimal arm @xmath84 can be eliminated by any arm , not just by some _ suboptimal arm _ with index lower than @xmath135 ; see @xcite , page 8 . it is unclear why it is enough to look at the elimination by those arms , since if @xmath84 is eliminated no matter the arm that eliminated it the hoeffding bound ( [ eqhoeff ] ) no longer holds . the right - hand side of ( [ eqthstatic ] ) is the minimum of two terms . the first term is distribution - dependent and shows that the se policy adapts to the unknown distribution of the rewards . it is very much in the spirit of the original bound of @xcite and of the more recent finite sample result of @xcite . our bound for the se policy is smaller than the aforementioned bounds for the ucb policy by a logarithmic factor . reference @xcite did not provide the first bounds on the expected regret . indeed , @xcite and @xcite had previously derived what is often called _ gap - free _ bound as they hold uniformly over the @xmath88 s . the second term in our bound is such a gap - free bound . it is of secondary interest in this paper and arise as a byproduct of refined distribution dependent bound . nevertheless , it allows us to recover near optimal bounds of the same order as @xcite . they depart from optimal rates by a factor @xmath148 as proved in @xcite . actually , the result of @xcite is much stronger than our gap - free bound since it holds for any sequence of bounded rewards , not necessarily drawn independently . none of the distribution - dependent bounds in corollary [ corstatic ] or the one provided in @xcite is stronger than the other . the superiority of one over the other depends on the set @xmath149 : in some cases ( e.g. , if all suboptimal arms have the same expectation ) the latter is the best while in other cases ( if the @xmath88 are spread ) our bounds are better . this section is dedicated to a detailed description of the nonparametric bandit with covariates . a _ @xmath4-armed bandit machine with covariates _ ( with @xmath4 an integer greater than 2 ) is characterized by a sequence @xmath150 of independent random vectors , where @xmath151 , is a sequence of i.i.d . covariates in @xmath152^d$ ] with probability distribution @xmath153 , and @xmath154 denotes the random reward yielded by arm @xmath81 at time @xmath21 . throughout the paper , we assume that @xmath153 has a density , with respect to the lebesgue measure , bounded above and below by some @xmath155 and @xmath156 , respectively . we denote by @xmath157 the expectation with respect to @xmath153 . we assume that , for each @xmath158 , rewards @xmath159 , are random variables in @xmath11 $ ] with conditional expectation given by @xmath160 = f^{(i)}(x_t),\qquad i=1,\ldots , k , t=1,2 , \ldots,\ ] ] where @xmath161 , are unknown functions such that @xmath162 , for any @xmath163 . a natural example is where @xmath154 takes values in @xmath164 so that the conditional distribution of @xmath154 given @xmath165 is bernoulli with parameter @xmath166 . the _ game _ takes place sequentially on this machine , pulling one of the arms at each time @xmath26 . @xmath17 is a sequence of random functions @xmath167 indicating to the operator which arm to pull at each time @xmath21 , and such that @xmath20 depends only on observations strictly anterior to @xmath21 . the _ oracle policy _ @xmath168 , refers to the strategy that would be run by an omniscient operator with complete knowledge of the functions @xmath169 . given side information @xmath165 , the oracle policy @xmath170 prescribes to pull any arm with the largest expected reward , that is , @xmath171 with ties broken arbitrarily . note that the function @xmath172 is equal to the pointwise maximum of the functions @xmath173 , defined by @xmath174 the oracle rule is used to benchmark any proposed policy @xmath22 and to measure the performance of the latter via its ( _ cumulative _ ) _ regret _ at time @xmath2 defined by @xmath175 without further assumptions on the machine , the game can be arbitrarily difficult and , as a result , expected regret can be arbitrarily close to @xmath2 . in the following subsection , we describe natural regularity conditions under which it is possible to achieve sublinear growth rate of the expected regret , and characterize policies that perform in a near - optimal manner . as usual in nonparametric estimation we first impose some regularity on the functions @xmath176 . here and in what follows we use @xmath177 to denote the euclidean norm on @xmath178 . smoothness condition . we say that the machine satisfies the smoothness condition with parameters @xmath179 if @xmath180 is @xmath179-hlder , that is , if @xmath181 for some @xmath182 $ ] and @xmath183 . now denote the second pointwise maximum of the functions @xmath173 , by @xmath184 ; formally for every @xmath185 such that @xmath186 it is defined by @xmath187 and by @xmath188 otherwise . notice that a direct consequence of the smoothness condition is that the function @xmath189 is @xmath179-hlder ; however , @xmath190 might not even be continuous . the behavior of the function @xmath191 critically controls the complexity of the problem and the hlder regularity gives a local upper bound on this quantity . the second condition gives a lower bound on this function though in a weaker global sense . it is closely related to the margin condition employed in classification @xcite , which drives the terminology employed here . it was originally imported to the bandit setup by @xcite . margin condition . we say that the machine satisfies the margin condition with parameter @xmath192 if there exists @xmath193 , @xmath194 such that @xmath195\le c_0 \delta^\alpha\qquad \forall\delta\in[0 , \delta_0].\ ] ] if the marginal @xmath153 has a density bounded above and below , the margin condition contains only information about the behavior of the function @xmath196 and not the marginal @xmath153 itself . this is in contrast with @xcite where the margin assumption is used precisely to control the behavior of the marginal @xmath153 while that of the reward functions is fixed . a large value of the parameter @xmath197 means that the function @xmath196 either takes value 0 or is bounded away from 0 , except over a set of small @xmath153-probability . conversely , for values of @xmath197 close to 0 , the margin condition is essentially void , and the two functions can be arbitrary close , making it difficult to distinguish them . this reflects in the bounds on the expected regret derived in the subsequent section . intuitively , the smoothness condition and the margin condition work in opposite directions . indeed , the former ensures that the function @xmath196 does not `` depart from zero '' too fast whereas the latter warrants the opposite . the following proposition quantifies the extent of this conflict . [ propalpha - beta ] under the smoothness condition with parameters@xmath179 , and the margin condition with parameter @xmath197 , the following hold : * if @xmath198 , then a given arm is either always or never optimal ; in the latter case , it is bounded away from @xmath199 and one can take @xmath200 ; * if @xmath201 , then there exist machines with nontrivial oracle policies . this proposition is a straightforward consequences of , respectively , the first two points of proposition 3.4 in @xcite . for completeness , we provide an example with @xmath202 , @xmath152 $ ] , @xmath203 and @xmath204 . notice that @xmath205 is @xmath206-hlder if and only if @xmath207 . any oracle policy is nontrivial , and , for example , one can define @xmath208 if @xmath209 and @xmath210 if @xmath211 . moreover , it can be easily shown that the machine satisfies the margin condition with parameter @xmath197 and with @xmath212 . we denote by @xmath213 the class of @xmath4-armed bandit problems with covariates in @xmath152^d$ ] with a machine satisfying the margin condition with parameter @xmath192 , the smoothness condition with parameters @xmath214 and such that @xmath153 has a density , with respect to the lebesgue measure , bounded above and below by some @xmath155 and @xmath156 , respectively . to design a policy that solves the bandit problem with covariates described above , one has to inevitably find an estimate of the functions @xmath176 , at the current point @xmath165 . there exists a wide variety of nonparametric regression estimators ranging from local polynomials to wavelet estimators . both of the policies introduced below are based on estimators of @xmath176 , that are @xmath153 almost surely piecewise constant over a particular collection of subsets , called _ bins _ of the covariate space @xmath215 . we define a partition of @xmath215 in a measure theoretic sense as a collection of measurable sets , hereafter called _ bins _ , @xmath216 such that @xmath217 , @xmath218 and @xmath219 , @xmath220 , up to sets of null @xmath153 probability . for any @xmath221 and any bin @xmath222 , define @xmath223 to define and analyze our policies , it is convenient to reindex the random vectors @xmath224 as follows . given a bin @xmath222 , let @xmath225 denote the @xmath226th time at which the sequence @xmath227 is in @xmath222 and observe that it is a stopping time . it is a standard exercise to show that , for any bin @xmath222 and any arm @xmath81 , the random variables @xmath228 , @xmath229 are i.i.d . with expectation given by @xmath230 $ ] . as a result , the random variables @xmath231 obtained by successive pulls of arm @xmath81 when @xmath232 form an i.i.d . sequence in @xmath11 $ ] with expectation given by @xmath230 $ ] . therefore , if we restrict our attention to observations in a given bin @xmath222 , we are in the same setup as the static bandit problem studied in the previous section . this observation leads to the notion of _ policy on @xmath222_. more precisely , fix a subset @xmath233 , an integer @xmath234 and recall that @xmath235 is the set of chronological times @xmath21 posterior to @xmath236 at which @xmath232 . fix @xmath237 and consider the static bandit problem with arms @xmath238 defined in section [ secstatic ] where successive pulls of arm @xmath239 , at times posterior to @xmath236 , yield rewards @xmath240 that are i.i.d . in @xmath11 $ ] with mean @xmath230 $ ] . the se policy with parameters @xmath241 on this static problem is called _ se policy on @xmath222 initialized at time @xmath236 with initial set of arms @xmath238 and parameters @xmath35_. we first outline a naive policy to operate the bandit machine described in section [ secdesc ] . it consists of fixing a partition of @xmath215 and for each set @xmath222 in this partition , to run the se policy on @xmath222 initialized at time @xmath242 with initial set of arms @xmath66 and parameters @xmath241 to be defined below . the _ binned successive elimination _ ( bse ) policy @xmath243 relies on a specific partition of @xmath215 . let @xmath244 be the regular partition of @xmath245^d$ ] : the collection of hypercubes defined for @xmath246 by @xmath247 in this paper , all sets are defined up to sets of null lebesgue measure . as mentioned in section [ subbin ] , the problem can be decomposed into @xmath248 independent static bandit problems , one for each @xmath249 . denote by @xmath250 the se policy on bin @xmath222 with initial set of arms @xmath66 and parameters @xmath251 . for any @xmath185 , let @xmath252 denote the bin such that @xmath253 . moreover , for any time @xmath254 , define @xmath255 to be the number of times before @xmath21 when the covariate fell in bin @xmath222 . the bse policy @xmath256 is a sequence of functions @xmath257 defined by @xmath258 , where @xmath259 . it can be implemented according to the pseudo - code of policy [ algbse ] . set of arms @xmath34 . parameters @xmath260 . @xmath262 initialize a se policy @xmath250 with parameters @xmath251 . @xmath266 ( observe @xmath267 ) . the following theorem gives an upper bound on the expected regret of the bse policy in the case where the problem is difficult , that is , when the margin parameter @xmath197 satisfies @xmath268 . [ thmain ] fix @xmath182 $ ] , @xmath183 and @xmath269 and consider a problem in @xmath270 . then the bse policy @xmath243 with @xmath271 has an expected regret at time @xmath2 bounded as follows : @xmath272 where @xmath273 is a positive constant that does not depend on @xmath4 . the case @xmath5 was studied in @xcite using a similar policy called ucbogram . unlike in @xcite where suboptimal bounds for the ucb policy are used , we use here the sharper regret bounds of theorem [ thstatic ] and the se policy as a running horse for our policy , thus leading to a better bound than @xcite . optimality for the two - armed case is discussed after theorem [ thmainadap ] . proof of theorem [ thmain ] we assume that @xmath274 where the indexing will be made clearer later in the proof . moreover , to keep track of positive constants , we number them @xmath275 . for any real valued function @xmath276 on @xmath215 and any measurable @xmath277 , we use the notation @xmath278 . moreover , for any @xmath221 , we use the notation @xmath279 . define @xmath280 , and let @xmath281 be the largest integer such that @xmath282 , where @xmath283 is the constant appearing in the margin condition . if @xmath284 , we have @xmath285 so that the result of the theorem holds when @xmath286 is chosen large enough , depending on the constant @xmath287 . in the rest of the proof , we assume that @xmath288 so that @xmath289 . recall that the bse policy @xmath256 is a collection of functions @xmath290 that are constant on each @xmath291 . therefore , the regret of @xmath256 can be decomposed as @xmath292 , where @xmath293 conditioning on the event @xmath294 , it follows from ( [ eqaverage ] ) that @xmath295={\mathbb e}\biggl[\sum _ { s=1}^{n_{j}(n ) } \bigl(\bar f_j^\star -\bar f{}^{(\hat{\pi}_{b_j , s})}_j \bigr ) \biggr],\ ] ] where @xmath296 is defined in ( [ eqdefnbj ] ) ; it satisfies , by assumption , @xmath297\leq\bar{c}nm^{-d}$ ] . let @xmath298 be the regret associated to a static bandit problem with arm @xmath81 yielding reward @xmath299 and where @xmath300 is the largest average reward . it follows from the smoothness condition that @xmath301 so that @xmath302 consider _ well - behaved _ bins on which the expected reward functions are well separated . these are bins @xmath291 with indices in @xmath303 defined by @xmath304 a bin @xmath222 that is not well behaved is called _ strongly ill behaved _ if there is some @xmath305 such that @xmath306 , for all @xmath9 , and _ weakly ill behaved _ otherwise . strongly and weakly ill behaved bins have indices in @xmath307 and @xmath308 respectively . note that for any @xmath9 , the function @xmath309 is @xmath310-hlder . thus for any @xmath311 and any @xmath312 , we have @xmath313 for all @xmath314 so that the inclusion @xmath315 indeed holds . _ first part _ : _ strongly ill behaved bins in @xmath316 . _ recall that for any @xmath317 , any arm @xmath318 , and any @xmath319 , @xmath320 . therefore , @xmath321\\[-8pt ] & \leq & c_1^{1+\alpha } n m^{-\beta(1+\alpha)},\nonumber\end{aligned}\ ] ] where we used the fact that the set @xmath322 does not contribute to the regret . _ second part _ : _ weakly ill behaved bins in @xmath323 . _ the numbers of weakly ill behaved bins can be bounded using @xmath324 on such a bin ; indeed , the margin condition implies that @xmath325 it yields @xmath326 . moreover , we bound the expected regret on weakly ill behaved bins using theorem [ thstatic ] with specific values @xmath327 together with ( [ eqwsfr ] ) , it yields @xmath328.\ ] ] _ third part _ : _ well - behaved bins in @xmath303 . _ this part is decomposed into two steps . in the first step , we bound the expected regret in a given bin @xmath329 ; in the second step we use the margin condition to control the sum of all these expected regrets . fix @xmath330 and recall that there exists @xmath331 such that @xmath332 . define @xmath333 and @xmath334 . we call @xmath335 the set of ( almost ) _ optimal _ arms over @xmath291 and @xmath336 the set of _ suboptimal _ arms over @xmath291 . note that @xmath337 for any @xmath330 . the smoothness condition implies that for any @xmath338 , @xmath339 therefore , @xmath340 on @xmath291 . moreover , for any arm @xmath341 that is not the best arm at some @xmath342 , then necessarily @xmath343 . so for any @xmath314 and any @xmath341 , it holds that either @xmath344 or @xmath345 . it yields @xmath346 thus , for any optimal arm @xmath341 , the reward functions averaged over @xmath291 satisfy @xmath347 , where @xmath348 together with ( [ eqwsfr ] ) , it yields @xmath349 . for any suboptimal arms @xmath350 , ( [ eqinaught ] ) implies that @xmath351 . assume now without loss of generality that the average gaps @xmath352 are ordered in such a way that @xmath353 . define @xmath354 and observe that if @xmath355 is such that @xmath356 , then @xmath341 . therefore , it follows from ( [ eqistar ] ) that @xmath357 for such @xmath81 . applying theorem [ thstatic ] with @xmath358 as above and @xmath142 , we find that there exists a constant @xmath359 such that , for any @xmath360 , @xmath361 hence , @xmath362 step 2 . we now use the margin condition to provide lower bounds on @xmath358 for each @xmath330 . assume without loss of generality that the indexing of the bins is such that @xmath363 and that the gaps are ordered @xmath364 . for any @xmath330 , from the definition of @xmath358 , there exists a suboptimal arm @xmath350 such that @xmath365 . but since the function @xmath366 satisfies the smoothness condition with parameters @xmath367 , we find that if @xmath368 for some @xmath369 , then @xmath370 together with the fact that @xmath371 over @xmath291 for any @xmath330 ( see step 1 above ) , it yields @xmath372\ge\sum _ { k=1}^{j_1 } p_k { \mathbh1}(0<{\underaccent{\bar } } { \delta}_k \le{\underaccent{\bar}}{\delta}_{j})\ge\frac { { \underaccent{\bar}}{c } j}{m^d},\ ] ] where we used the fact that @xmath373 . define @xmath374 to be the largest integer such that @xmath375 . since for any @xmath330 , we have @xmath376 , the margin condition yields for any @xmath377 that @xmath372\le c_\delta(c_1{\underaccent{\bar}}{\delta}_j)^\alpha,\ ] ] where we have used the fact that @xmath378 , for any @xmath379 . the previous two inequalities , together with the fact that @xmath380 for any @xmath330 , yield @xmath381 therefore , using the fact that @xmath382 for @xmath383 , we get from ( [ eqregj ] ) that @xmath384\\[-8pt ] & & \qquad\le c_5 \biggl[\sum _ { j=1}^{j_2 } k \frac{{\overline{\log}}({n}\gamma _ j^2/{m^d } ) } { \gamma_j } + \sum _ { j = j_2 + 1}^{j_1}k\log(n)+ \sum _ { j \in { \mathcal{j } } } nm^{-d-\beta}q_j \biggr].\nonumber\end{aligned}\ ] ] _ fourth part _ : _ putting things together . _ combining ( [ eqstronglyill ] ) , ( [ eqweaklyill ] ) and ( [ eqwellbehaved ] ) , we obtain the following bound : @xmath385.\nonumber\end{aligned}\ ] ] we now bound from above the first sum in ( [ eqreg2 ] ) by decomposing it into two terms . from the definition of @xmath386 , there exists an integer @xmath387 satisfying @xmath388 and such that @xmath389 for @xmath390 and @xmath391 for @xmath392 . it holds @xmath393 and @xmath394^{{2}/{\alpha } } \biggr ) \nonumber\\[-8pt]\\[-8pt ] & \le & c_{10 } m^d \int_{m^{-\alpha\beta}}^{1 } { \overline{\log}}\biggl(\frac{n}{m^d}x^{{2}/{\alpha } } \biggr)x^{-1/\alpha } \,{\mathrm{d}}x . \nonumber\end{aligned}\ ] ] since @xmath395 , this integral is bounded by @xmath396 . the second sum in ( [ eqreg2 ] ) can be bounded as @xmath397\\[-8pt ] & \leq&\frac{m^d}{{\underaccent{\bar}}{c}}{\mathbb p}\bigl\{0 < f^{\star}(x)-f^{\sharp } ( x)\leq c_1m^{-\beta } \bigr \ } \leq\frac{c_1^\alpha}{{\underaccent{\bar}}{c } } m^ { d-\beta\alpha}.\nonumber\end{aligned}\ ] ] putting together ( [ eqreg2])([eqsumqj ] ) , we obtain @xmath398,\end{aligned}\ ] ] and the result follows by choosing @xmath399 as prescribed . we should point out that the version of the bse described above specifies the number of bins @xmath399 as a function of the horizon @xmath2 , while in practice one may not have foreknowledge of this value . this limitation can be easily circumvented by using the so - called _ doubling argument _ ( see , e.g. , page 17 in @xcite ) which consists of `` reseting '' the game at times @xmath400 . the reader will note that when @xmath401 there is a potentially superfluous @xmath1 factor appearing in the upper bound using the same proof . more generally , for any @xmath402 , it is possible to minimize the expression in ( [ eqreg2 ] ) with respect to @xmath399 , but the optimal value of @xmath399 would then depend on the value of @xmath197 . this sheds some light on a significant limitation of the bse which surfaces in this parameter regime : for @xmath2 large enough , it requires the operator to pull each arm at least once in each bin and therefore to incur an expected regret of at least order @xmath248 . in other words , the bse splits the space @xmath215 in `` too many '' bins when @xmath402 . intuitively this can be understood as follows . when @xmath402 , the gap function @xmath403 is bounded away from zero on a large subset of @xmath215 . hence there is no need to carefully estimate it since the optimal arm is the same across the region . as a result , one could use larger bins in such regions reducing the overall number of bins and therefore removing the extra logarithmic term alluded to above . we need the following definitions . assume that @xmath404 and let @xmath405 be the smallest integer such that @xmath406 for any bin @xmath407 , let @xmath408 be the smallest integer such that @xmath409 where @xmath410 is defined in ( [ eqdefu ] ) and @xmath411 . this definition implies that @xmath412 for @xmath413 , because @xmath414 is decreasing for @xmath415 . the abse policy operates akin to the bse except that instead of fixing a partition @xmath416 , it relies on an adaptive partition that is refined over time . this partition is better understood using the notion of rooted tree . let @xmath417 be a tree with root @xmath215 and maximum depth @xmath405 . a node @xmath222 of @xmath417 with depth @xmath418 is a set from the regular partition @xmath419 . the children of node @xmath420 are given by @xmath421 , defined to be the collection of @xmath422 bins in @xmath423 that forms a partition of @xmath222 . note that the set @xmath424 of leaves of each subtree @xmath425 of @xmath417 forms a partition of @xmath215 . the abse policy constructs a sequence of partitions @xmath426 that are leaves of subtrees of @xmath417 . at a given time @xmath26 , we refer to the elements of the current partition @xmath427 as _ live _ bins . the sequence of partitions is nested in the sense that if @xmath428 , then either @xmath429 or @xmath430 . the sequence @xmath426 is constructed as follows . in the initialization step , set @xmath431 , @xmath432 , and the initial set of arms @xmath433 . let @xmath434 be a time such that @xmath435 , and let @xmath436 be the collection of sets @xmath222 such that @xmath437 . we say that the bins @xmath438 are _ born _ at time @xmath21 . for each set @xmath439 , assume that we are given a set of active arms @xmath440 . note that @xmath441 is such a time with @xmath442 and active arms @xmath443 . for each born bin @xmath438 , we run a se policy @xmath250 initialized at time @xmath21 with initial set of arms @xmath444 and parameters @xmath445 , @xmath446 . such a policy is defined in section [ subbin ] . let @xmath447 denote the time at which @xmath250 has reached @xmath408 rounds and let @xmath448 denote the number of times covariate @xmath165 fell in bin @xmath222 while @xmath222 was a live @xmath222 . at time @xmath449 , we replace the node @xmath222 by its children @xmath421 in the current partition . namely , @xmath450 . moreover , to each bin @xmath451 , we assign the set @xmath452 of arms that were left active by policy @xmath250 on its parent @xmath222 at the end of the @xmath408 rounds . this procedure is repeated until the horizon @xmath2 is reached . the intuition behind this policy is the following . the parameters of the se policy @xmath250 run at the birth of bin @xmath222 are chosen exactly such that arms @xmath81 with average gap @xmath453 are eliminated by the end of @xmath408 rounds with high probability . the smoothness condition ensures that these eliminated arms satisfy @xmath454 for all @xmath305 so that such arms are uniformly suboptimal on bin @xmath222 . among the kept arms , none is uniformly better than another , so bin @xmath222 is burst and the process is repeated on the children of @xmath222 where other arms may be uniformly suboptimal . the formal definition of the abse is given in policy [ algabse ] ; it satisfies the following theorem . set of arms @xmath455 . parameters @xmath456 . @xmath458 , @xmath459 , @xmath460 . initialize a se policy @xmath461 with parameters @xmath462 and arms @xmath463 . @xmath464 . @xmath465 . @xmath265 . @xmath466 ( observe @xmath467 ) . @xmath468 @xmath469 @xmath470initialize se policy @xmath471 with @xmath472 and arms @xmath473 . @xmath474.@xmath475 @xmath476 [ thmainadap ] fix @xmath182 $ ] , @xmath183 , @xmath192 , assume that @xmath404 and consider a problem in @xmath270 . if @xmath477 , then the abse policy @xmath478 has an expected regret at time @xmath2 bounded by @xmath479 where @xmath273 does not depend on @xmath4 . if @xmath200 , then @xmath480 . note that the bounds given in theorem [ thmainadap ] are optimal in a minimax sense when @xmath5 . indeed , the lower bounds of @xcite and @xcite imply that the bound on expected regret can not be improved as a function of @xmath2 except for a constant multiplicative term . the lower bound proved in @xcite implies that any policy that received information from _ both _ arms at each round has a regret bound at least as large as the one from theorem [ thmainadap ] , up to a multiplicative constant . as a result , there is no price to pay for being in a partial information setup and one could say that the problem of nonparametric estimation dominates the problem associated to making decisions sequentially . note also that when @xmath200 , proposition [ propalpha - beta ] implies that there exists a unique optimal arm over @xmath215 and that all other arms have reward bounded away from that of the optimal arm . as a result , given this information , one could operate as if the problem was static by simply discarding the covariates . theorem [ thmainadap ] implies that in this case , one recovers the traditional regret bound of the static case without the knowledge that @xmath200 . proof of theorem [ thmainadap ] we first consider the case where @xmath481 , which implies that @xmath482 ; see proposition [ propalpha - beta ] . we keep track of positive constants by numbering them @xmath483 , yet they might differ from previous sections . on each newly created bin @xmath222 , a new se policy is initialized , and we denote by @xmath484 , the rewards obtained by successive pulls of a remaining arm @xmath81 . their average after @xmath29 rounds / pulls is denoted by @xmath485 for any integer @xmath226 , define @xmath486 , where @xmath410 is defined in ( [ eqdefu ] ) . for any @xmath487 , define the unique _ parent _ of @xmath222 by @xmath488 and @xmath489 . moreover , let @xmath490 and for any @xmath491 define recursively @xmath492 . then the set of _ ancestors _ of any @xmath493 is denoted by @xmath494 and defined by @xmath495 denote by @xmath496 the regret incurred by the abse policy @xmath478 when covariate @xmath165 fell in a _ live _ bin @xmath497 , where we recall that @xmath427 denotes the current partition at time @xmath21 . it is defined by @xmath498 { \mathbh1}(x_t \in b){\mathbh1}(b\in{\mathcal{l}}_t).\ ] ] we also define @xmath499 to be the set of bins that were born at some time @xmath500 . we denote by @xmath501 the regret incurred when covariate @xmath165 fell in such a bin . it is defined by @xmath502 { \mathbh1}(x_t \in b){\mathbh1}(b \in{\mathcal{b}}_t).\ ] ] observe that if we define @xmath503 , we have @xmath504 since @xmath505 and @xmath506 for all @xmath21 . note that for any @xmath493 , @xmath507 denote by @xmath508 the set of arms left active by the se policy @xmath250 on @xmath222 at the end of @xmath408 rounds . moreover , define the following reference sets of arms : @xmath509 @xmath510 define the event @xmath511 on which the remaining arms have a gap of the correct order and observe that ( [ eqdecompborn ] ) implies that @xmath512 let @xmath513 denote the set of leaves of @xmath514 , that is the set of bins @xmath222 such that @xmath515 . in what follows , we adapt the convention that @xmath516 . we are going to treat regret incurred on live nonterminal nodes and live leaves separately and differently . as a result , the quantity we are interested in is decomposed as @xmath517 where @xmath518 is the regret accumulated on live nonterminal nodes , and @xmath519 is regret accumulated on live leaves . our proof relies on the following events : @xmath520 . _ first part _ : _ control of the regret on the nonterminal nodes . _ fix . on @xmath521 , we have @xmath522 so that any active arm @xmath523 satisfies @xmath524 . moreover , regret is only incurred at points where @xmath525 , so defining @xmath526 and conditioning on events @xmath527 yields @xmath528\le{\mathbb e}\bigl[\tilde n_b(n ) \bigr]c_1|b|^\beta q_b \le c_1 k \ell_b |b|^\beta q_b,\ ] ] where @xmath529 and @xmath530 is defined in ( [ eqtildesb ] ) . we can always assume that @xmath2 is greater than @xmath531 , defined by @xmath532 and let @xmath533 be the smallest integer such that @xmath534 . indeed , if @xmath535 , the result is true with a constant large enough . applying the same argument as in ( [ eqsumqj ] ) yields the existence of @xmath536 such that , for any @xmath537 , @xmath538 indeed , for @xmath539 one can define @xmath540 , and the same equation holds with @xmath541 if @xmath542 . summing over all depths @xmath543 , we obtain @xmath544\nonumber\\[-10pt]\\[-10pt ] & & \qquad\leq c_1c_2c_\ell k \sum_{k=0}^{k_0 - 1 } 2^{k(d+\beta-\alpha\beta)}\log \bigl(n2^{-k(2\beta + d ) } \bigr).\nonumber\end{aligned}\ ] ] on the other hand , for every bin @xmath545 , one also has @xmath546\le c_1 n |b|^\beta q_bp_x(b ) { \mathbb p}\bigl({\mathcal{g}}_b\cap{\mathcal{a}}_b^c\bigr).\ ] ] it remains to control the probability of @xmath547 ; we define @xmath548 . on @xmath521 , the event @xmath549 can occur in two ways : by eliminating an arm @xmath550 at the end of the at most @xmath408 rounds played on bin @xmath222 . these arms satisfy @xmath551 ; this event is denoted by @xmath552 . by not eliminating an arm @xmath553 within the at most @xmath408 rounds played on bin @xmath222 . these arms satisfy @xmath554 ; this event is denoted by @xmath555 . we use the following decomposition : @xmath556 we first control the probability of making error ( i ) . note that for any @xmath557 and any arms @xmath558 , it holds @xmath559 therefore , if an arm @xmath550 is eliminated , that is , if there exists @xmath560 such that @xmath561 for some @xmath557 , then either @xmath562 or @xmath563 does not belong to its respective confidence interval @xmath564 $ ] or @xmath565 $ ] for some @xmath557 . therefore , since @xmath566 , @xmath567 where in the second inequality , we used lemma [ lempeeling ] . next , we treat error ( ii ) . for any @xmath553 , there exists @xmath568 such that @xmath569 . let @xmath570 be any arm such that @xmath571 ; the smoothness condition implies that @xmath572\\[-9pt ] & \ge&\bar f^{(i)}_b + 6c_0|b|^\beta\geq\bar f^{(i)}_b + \tfrac{3}{2 } \varepsilon_{b,\ell _ b}.\nonumber\end{aligned}\ ] ] on the event @xmath573 , no arm in @xmath574 , and in particular any of the arms @xmath575 , has been eliminated until round @xmath408 . therefore , the event @xmath576 occurs if there exists @xmath577 such that @xmath578 . in view of ( [ eqgapg2 ] ) and ( [ eqlbeps ] ) , it implies that there exists @xmath523 such that @xmath579 hence , the probability of error ( ii ) can be bounded by @xmath580\\[-10pt ] & \leq&2k \frac{\ell _ b}{n|b|^d},\nonumber\end{aligned}\ ] ] where the second inequality follows from ( [ eqhoeffding ] ) . putting together ( [ eqdecompg ] ) , ( [ eqboundg1 ] ) , ( [ eqboundg2 ] ) and ( [ eqboundlb ] ) , we get @xmath581 together with ( [ eqrn1 ] ) , it yields for @xmath582 that @xmath583\le c_3 k @xmath584 is such that @xmath585 , then any bin @xmath222 such that @xmath586 satisfies @xmath587\le c_4 k\log n$ ] . if @xmath584 is such that @xmath588 , then the above display together with the margin condition yield @xmath589 \le c_5 k 2^{k(\beta+d-\alpha\beta)}\log\bigl(n2^{-k(2\beta+d)}\bigr).\ ] ] summing over all depths @xmath418 and using ( [ eqhatrn1 ] ) , we obtain @xmath590\le c_6 k \sum _ { k=0}^{k_0 - 1}2^{k(\beta+d-\alpha\beta)}\log\bigl(n2^{-k(2\beta+d ) } \bigr).\vadjust{\goodbreak}\ ] ] we now compute an upper bound on the right - hand side of the above inequality . fix @xmath591 and define @xmath592 observe that @xmath593^{-({d+2\beta } ) /({d+\beta-\beta\alpha } ) } \bigr),\ ] ] where @xmath594 . therefore , ( [ eqregnodes ] ) can be rewritten as @xmath595\nonumber\\ & & \qquad\le c_6k \biggl[\sum _ { k=1}^{k_0 - 1 } ( s_{k}-s_{k-1})\log\bigl(n[c_{7 } s_{k}+1]^{- ( { d+2\beta})/({d+\beta-\beta\alpha } ) } \bigr ) + \log n \biggr ] \nonumber \\ & & \qquad\le c_6k \biggl[\int_{0}^{s_{k_0 - 1 } } \log\bigl(n[c_{7 } x+1]^{- ( { d+2\beta})/({d+\beta-\beta\alpha } ) } \bigr ) \,{\mathrm{d}}x + \log n \biggr ] \\ & & \qquad\le c_8k \bigl[2^{k_0(d+\beta-\beta\alpha ) } \log\bigl ( n2^{-k_0(d+2\beta ) } \bigr ) + \log n \bigr ] \nonumber \\ & & \qquad\le c_9 n \biggl(\frac{n}{k\log(k ) } \biggr)^{-{\beta ( 1+\alpha)}/({d+2\beta})},\nonumber\end{aligned}\ ] ] where we used ( [ eqdefk0 ] ) in the last inequality and the fact that @xmath596 is dominated by @xmath597 since @xmath482 . _ second part _ : _ control of the regret on the leaves . _ recall that the set of leaves @xmath513 is composed of bins @xmath222 such that @xmath515 . proceeding in the same way as in ( [ eqrn1 ] ) , we find that for any @xmath598 , it holds @xmath599\le c_1 n |b|^\beta p_x\bigl(0<f^\star - f^\sharp \le c_1 |b|^\beta , x \in b\bigr).\ ] ] since @xmath600 , then @xmath601 and using the margin assumption , we find @xmath602 & \le & c_1 n 2^{-k_0\beta(1+\alpha ) } \nonumber\\[-8pt]\\[-8pt ] & \le & c_1 n \biggl ( \frac{n}{k\log ( k ) } \biggr)^{-{\beta(1+\alpha)}/({d+2\beta})},\nonumber\end{aligned}\ ] ] where we used ( [ eqdefk0 ] ) in the second inequality . the theorem follows by summing ( [ eqregnodes2 ] ) and ( [ eqregleaves ] ) . if @xmath603 , then the same proof holds except that @xmath596 dominates @xmath604 in ( [ eqregnodes2 ] ) . the following lemma is central to our proof of theorem [ thstatic ] . we recall that a process @xmath605 is a martingale difference sequence if @xmath606=0 $ ] . moreover , if @xmath607 and if we denote the sequence of averages by @xmath608 , then hoeffding azuma s inequality yields that , for every integer @xmath609 , @xmath610 the following lemma is a generalization of this result : define @xmath612 . recall first the hoeffding azuma maximal concentration inequality . for every @xmath613 and every integer @xmath254 , @xmath614 using a peeling argument , one obtains @xmath615 hence the result .

we describe in this paper analysis procedures that can answer two of the most important questions of neutrino research . how can one determine the total solar neutrino fluxes ( @xmath12 , and @xmath10 ) for comparison with solar model predictions ? how can one determine the sterile contribution to the total solar neutrino fluxes ? our answers allow for the possibility of an arbitrary mixture in solar neutrino oscillations of active and sterile neutrinos , but require the correctness of the lma solution of the solar neutrino problems and careful attention to all the sources of error ( theoretical as well as experimental).be and @xmath0b neutrinos are unchanged and the numerical results have not been affected significantly , but the present version is more up - to - date with respect to the input data . we have also added , inspired by the sage discussion , a detailed analysis of what one can learn about @xmath10 neutrinos before there is a dedicated experiment to measure just the @xmath10 neutrino flux . ] we focus first on determining total fluxes by comparing charged current ( cc ) observables measured in different experiments ; this method yields results as independent as possible of uncertainties due to the presence of sterile neutrinos . we then describe how similar techniques can be applied to determine total solar neutrino fluxes using a cc experiment plus a neutrino - electron scattering experiment [ or a neutral current ( nc ) measurement ] , which yields results that depend more on the sterile neutrino mix but which can nevertheless be relatively accurate . the numerical values we estimate for the expected precision with which different quantities can be measured rely upon simulations of the performance of the relevant experiments . therefore the accuracies that we quote are illustrative ; the actual accuracies that are obtainable can only be determined once the experimental uncertainties are known . neutrinos change flavors as they travel to the earth from the center of the sun . this flavor change was seen directly by the comparison of the sudbury neutrino observatory ( sno ) measurement @xcite of the charged current reaction for @xmath0b solar neutrinos with the super - kamiokande measurement @xcite of the neutrino - electron scattering rate ( charged plus neutral current ) . even more clearly , flavor change has been demonstrated by comparing the neutral current measurement by sno with the sno cc measurement @xcite . the conclusion that flavor changes occur among solar neutrinos , if based solely upon the comparison of the sno and super - kamiokande event rates , is valid statistically at about the @xmath13 confidence level @xcite . the neutral current measurement of sno increases the significance level for flavor changes among solar neutrinos to the @xmath14 confidence level . the sno and super - kamiokande results demonstrate simply that new physics is required to resolve the long - standing solar neutrino problem @xcite , i.e. , to understand the origin of the discrepancy between the predictions of the standard solar model @xcite and the observed solar neutrino event rates @xcite . if one includes the results of the chlorine @xcite , kamiokande @xcite , sage @xcite , gallex @xcite , and gno @xcite experiments together with the sno ( cc ) and super - kamiokande results , then the combined measurements require @xcite new physics at @xmath15 and , if the relative temperature scaling of the @xmath8be and @xmath0b neutrino production reactions is taken into account , at @xmath16 . helioseismological measurements confirm the predicted sound speeds of the standard solar model to better than @xmath17% and show that stellar physics can not account for the discrepancies between standard predictions and the observed solar neutrino rates @xcite . the combination of the charged current ( cc ) and the charged plus neutral - current measurement with super - kamiokande has been used by several groups @xcite to determine the flux of active @xmath0b neutrinos independent of the solar model . these model - independent determinations of the active flux exploit the similarity between the response functions in the sno and super - kamiokande detectors @xcite . in addition , if one includes all the experimental data in a global oscillation solution in which the @xmath0b flux is a free parameter , one obtains a similar ( but slightly smaller ) allowed range for the @xmath0b neutrino flux @xcite . the measurement of the nc rate by sno @xcite provides an independent determination of the active @xmath0b neutrino flux . all of the analyses yield the same result : if electron neutrinos oscillate into only active neutrinos , then the total @xmath0b neutrino flux is in excellent agreement with the flux predicted by the standard solar model . this close agreement of the active @xmath0b neutrino flux with the total flux predicted by the standard solar model ( ssm ) @xcite is , if the flux of sterile neutrinos is small , an important confirmation of the quantitative theory of stellar evolution . we summarize below the current best - estimates and the associated @xmath2 uncertainties for the active @xmath0b neutrino flux , @xmath18 . * standard solar model ( bp00 ) prediction @xcite : @xmath19 * comparison of sno and super - kamiokande event rates ( cf . esp . ref . @xcite ) : @xmath20 * sno neutral current measurement ( assuming undistorted spectrum ) @xcite . @xmath21 * global neutrino oscillation analysis ( see table [ tab : fbranges ] this paper and ref . @xcite ) : @xmath22 the agreement , summarized in eqs . ( [ eq : itemfb8ssm])([eq : itemfb8global ] ) , between the ssm calculated flux and the measured active flux suggests that the sterile neutrino contribution to the @xmath0b neutrino flux may be small . in this paper , we ignore this tempting suggestion and instead concentrate on developing methods to determine experimentally the total @xmath0b and @xmath8be neutrino fluxes emitted by the sun , independent of the active - sterile mixture ( for a discussion of earlier investigations of sterile neutrinos see refs . @xcite ) . if we want to understand the particle physics implications of solar neutrino research , we must determine if sterile neutrinos are present in the solar neutrino flux . moreover , the original and still valid goal of solar neutrino research was @xcite to compare solar model predicted and experimentally measured ( total ) solar neutrino fluxes . what is the situation if sterile neutrinos exist ? the total flux of @xmath0b neutrinos could in this case be much larger than the standard solar model prediction ; a major fraction of the total flux that reaches the earth could arrive in a form that is not detected in solar neutrino experiments . the existing data disfavor ( at @xmath23 cl ) oscillation into purely sterile neutrinos . nevertheless , a large sterile component is allowed @xcite if oscillations occur into a combination of active and sterile neutrino states ( see refs . @xcite for a description of the formalism adopted here ) . the flux of sterile neutrinos could in principle be large enough to destroy the apparently excellent agreement between the flux predicted by the ssm and the true flux of @xmath0b neutrinos [ which is assumed to be pure active neutrinos in the comparison shown above in eqs . ( [ eq : itemfb8ssm])-([eq : itemfb8global ] ) ] . a measurement of the total @xmath0b solar neutrino flux , including the sterile component ( if any ) , will provide information that is important for astrophysics and for particle physics . the motivation for investigating sterile neutrinos is not dependent upon the lsnd @xcite results that might suggest the existence of sterile neutrinos . of course , the lsnd results are not supported by other experimental results ( cf . @xcite ) nor by theoretical predictions that have been precisely confirmed ( like the helioseismological verifications of the standard solar model ) , as is the case for the inference of flavor change based upon the sno - super - kamiokande comparison . fortunately , the kamland reactor neutrino experiment @xcite , when combined with the sno measurement of the cc flux , is capable of providing a precise determination of the total , i.e. the active plus the sterile , @xmath0b neutrino flux . we assume throughout this paper the correctness of the currently favored large mixing angle ( lma ) solution to the solar neutrino problem . if the lma solution is not correct , then all global analyses of the available solar and reactor data indicate that either the mass difference , @xmath24 , or the vacuum mixing angle , @xmath25 , will be too small to produce a measurable effect in the kamland experiment ( see e.g. refs . @xcite ) . in this case , kamland will be unable to provide the information required to determine the total @xmath0b neutrino flux . here is the basic physical idea of the method we propose for measuring the total @xmath0b solar neutrino flux . for the kamland @xcite experiment , one will know accurately the flux of anti - neutrinos from the @xmath26 reactors that contribute significantly to the measured anti - neutrino events . from measurements of the total event rate and the energy spectrum induced by the surviving @xmath27 , the kamland experimentalists can determine with precision @xcite the anti - neutrino propagation parameters , @xmath28 and @xmath29 . both the kamland measurement and the cc sno measurements are disappearance experiments for neutrinos ( or anti - neutrinos ) of similar energies . for the cc measurement made with sno , one does not know the total @xmath0b neutrino flux created in the sun . but , assuming conservation of cpt , one can use the propagation parameters @xmath30 and @xmath31 determined by kamland and the measured ( by sno ) cc rate to solve for the flux that gives the observed result . summarizing , for the kamland experiment one knows the total flux but not the propagation parameters , which are measured . for the sno cc experiment , one will know ( from kamland ) the propagation parameters and therefore can measure the total flux . figure [ fig : solarvskamlandreal ] shows the results of a refined global solution for the solar neutrino oscillation parameters that was made ( see sec . [ sec : presentknowledge ] and the appendix ) using all the available solar and reactor data . the figure displays the allowed solar neutrino oscillation contours at @xmath2 , @xmath32 , and @xmath33 . the results are obtained by the procedures described most recently in ref . @xcite , where we have used in the present paper the analysis strategy ( a ) ( of ref . @xcite ) including the 1496 day super - kamiokande data sample @xcite as well as the sno cc , nc and day - night observations @xcite . we have also made some improvements ( see the appendix ) in the treatment of the uncertainties in the neutrino cross sections and in the correlation of errors . for comparison , we also show in fig . [ fig : solarvskamlandreal ] the small size of the expected allowed region for kamland if this reactor anti - neutrino experiment observes a signal corresponding to the current best - fit point of the solar neutrino analysis . in calculating the kamland allowed region , we have made a conservative estimate ( see sec . [ subsec : precisionb8 ] ) , following the principles discussed in refs . @xcite . figure [ fig : solarvskamlandreal ] shows clearly that kamland has the potential for making a precise measurement of the solar neutrino oscillation parameters , provided that the lma is the correct oscillation solution . the only complication involved in determining the total @xmath0b flux from a comparison of the kamland and sno cc measurements results from the fact that for the favored large mixing angle ( lma ) solar neutrino oscillation solution matter effects in the sun and the earth can be significant . matter effects are unimportant for the kamland reactor experiment . the role of matter effects in solar neutrino experiments depends somewhat upon the _ a priori _ unknown active - sterile mixture , which introduces a calculable uncertainty in the inferred total @xmath0b neutrino flux . in summary , the combination of the sno cc neutrino measurement and the kamland anti - neutrino measurement will determine the total ( active plus sterile ) flux , @xmath34 , of @xmath0b solar neutrinos . by subtracting the previously determined flux ( see above ) , @xmath18 , from @xmath35 , one can determine the flux of sterile solar neutrinos . using similar reasoning , we shall also show that the combined super - kamiokande and kamland measurements can be analyzed to yield an accurate value for the total @xmath0b flux , although in this case the results are somewhat more sensitive to the active - sterile admixture . the flux of @xmath8be solar neutrinos , which in the ssm is predicted@xcite to be @xmath36 , can be determined in a model independent way from a global analysis of the solar neutrino data assuming only active neutrino oscillations . for example , the latest analysis by garzelli and giunti @xcite yields @xmath37 at 99% cl . we shall also show in this paper that one can extract the value of the @xmath8be neutrino flux from measurements of the gallium solar neutrino experiments , gallex , sage , and gno and the results of the sno and kamland measurements . the value of the @xmath8be flux that will be derived in this way is relatively insensitive to the assumed neutrino oscillation parameters , although it does depend somewhat on the assumed contributions of the cno , @xmath38 , and @xmath10 neutrino fluxes which we adopt from the standard solar model . the constraint provided by the chlorine experiment is not very useful for determining the total @xmath8be neutrino flux . one can obtain an independent measurement of the total @xmath8be solar neutrino flux by comparing data from the kamland experiment with data from the borexino @xcite solar neutrino experiment . if both the kamland and the borexino detectors work as expected , then this method will be more accurate than the methods involving the gallium and chlorine radiochemical detectors . the determination of the total solar neutrino fluxes , and even more so the determination of the sterile components of these neutrino fluxes , requires precision in both the experimental measurements and the theoretical calculations and analyses . we present in the appendix a refined discussion of the theoretical errors , and their correlations , for the absorption cross for the gallium and chlorine solar neutrino experiments . the outline of this paper is as follows . in sec . [ sec : presentknowledge ] , we describe the current experimental knowledge of the @xmath0b solar neutrino flux . our results are summarized in table [ tab : fbranges ] both for the special case of oscillations between purely active neutrinos and for the general case of oscillations between electron neutrinos and an active - sterile neutrino admixture . we limit our analysis to the allowed lma region of solar neutrino oscillations . we show in sec . [ sec : determiningb8total ] how one can use the cc measurements with sno and kamland to determine an accurate total @xmath0b solar neutrino flux including experimental and theoretical uncertainties and the possibility of an appreciable active - sterile admixture . we switch to the @xmath8be flux in sec . [ sec : be7 ] and evaluate how well one can determine the total @xmath8be solar neutrino flux by also using the results of the gallium ( gallex , sage , and gno ) solar neutrino experiments or the chlorine experiment . in sec . [ sec : klandes ] , we investigate how well the total @xmath0b and @xmath8be neutrino fluxes can be determined using the combined measurements of kamland and @xmath39 scattering observed in the super - kamiokande ( sec . [ subsec : kamlandsuperkb8 ] ) and borexino ( sec . [ subsec : kamlandborexino ] ) detectors . we show that even in the presence of active - sterile admixtures the total @xmath8be solar neutrino flux may be measured with relatively high accuracy by comparing results from the kamland and the borexino experiments . we describe and analyze in sec . [ sec : determiningpp ] three strategies for determining the total @xmath10 solar neutrino flux in the absence of a dedicated experiment that measures separately the @xmath10 neutrinos . we summarize and discuss our principal conclusions in sec . [ sec : discuss ] . we urge the reader to turn first to sec . [ sec : discuss ] and read there the summary and discussion of our main results and their implications . the rest of the paper can then be understood more easily . we generalize in this section the determination of the @xmath0b neutrino flux to the case in which @xmath40 oscillates into a state that is a linear combination of active ( @xmath41 ) and sterile ( @xmath42 ) neutrino states , @xmath43 where @xmath44 is the parameter that describes the active - sterile admixture . this admixture arises naturally in the framework of 4-@xmath45 mixing @xcite . the total @xmath0b neutrino flux can be written @xmath46 where @xmath47 . clearly , the larger the sterile component , the larger the value of @xmath48 that is inferred from the experimental data . we have performed a global analysis of the solar neutrino data treating the total @xmath0b neutrino flux as a free parameter . the details of the analysis procedure are the same as those used in ref . @xcite except where we explicitly state otherwise . we concentrate here on the lma region , @xmath49 , @xmath50 . to take account of the possibility of oscillations into sterile neutrinos , we determine the allowed regions in the parameter space defined by @xmath24 , @xmath51 , and a third parameter , @xmath52 , that is defined by eq . ( [ eq : linearcombination ] ) . it is convenient to introduce the dimensionless parameter , @xmath53 where @xmath54 is the total @xmath0b neutrino flux in units of the predicted standard solar model flux . .the allowed range of the total @xmath0b neutrino flux . the table presents the allowed range of @xmath55 [ defined by eq . ( [ eq : defnfb ] ) ] that was found in a global solution of all the currently available solar and reactor neutrino data . the second column gives the allowed range of @xmath55 for an arbitrary mixture of active and sterile neutrinos and the third column gives the range for the case where only active neutrinos are considered . the results shown were obtained using eq . ( [ eq : fb8determination3 ] ) . [ tab : fbranges ] [ cols="^,^,^",options="header " , ] the largest uncertainties in determining @xmath56 using the chlorine data are caused by the experimental errors in the chlorine absorption rate ( @xmath57 ) and the sno cc absorption rate ( @xmath58 ) . the uncertainty in the chlorine neutrino absorption cross sections are also significant ( @xmath59 to @xmath60 , depending upon the neutrino oscillation parameters ) . to be explicit , the uncertainties in @xmath56 due to the theoretical uncertainties in calculating the chlorine neutrino absorption cross sections that were used in constructing table [ tab : fbecl ] were evaluated from the following equation , @xmath61 ^ 2 + \left[\sum_{{\rm j= ^8b},hep}(f_{\rm be}({\rm c.s.}_j\pm 1 \sigma_{\rm c.s._j } ) - f_{\rm be}(i)\right]^2 \label{appex : chlorineferrors}~.\ ] ] ( see the appendix for more details regarding the treatment of the cross section errors . ) even the cno fluxes ( @xmath62 ) and the sno cc absorption cross section ( @xmath63 ) contribute non - negligible errors . the uncertainty from sterile neutrinos is small ( @xmath64 ) and does not significantly affect the total uncertainty . remarkably , the individual uncertainties in @xmath56 which arise from several different sources are larger than the current best - fit values of @xmath56 ( see table [ tab : fbecl ] ) . for the chlorine based determination , we have computed a representative ( average ) shift in @xmath56 by a monte carlo sampling of the allowed kamland oscillation region shown in fig . [ fig : solarvskamlandreal ] . we find @xmath65 this uncertainty is so large as to render not very useful the determination of @xmath56 with the aid of chlorine data . in this section , we show how data from kamland can be combined with @xmath39 scattering data obtained with the super - kamiokande and borexino experiments in order to determine , respectively , the total @xmath0b ( sec . [ subsec : kamlandsuperkb8 ] ) and @xmath8be ( sec . [ subsec : kamlandborexino ] ) solar neutrino fluxes . as discussed before , the advantage of using purely cc measurements to determine the fluxes is that the answers depend only mildly on the active - sterile admixture . on the other hand , @xmath39 scattering measurements have the advantage of smaller uncertainties in the interaction cross sections . moreover , for the determination of the @xmath8be flux , borexino depends less strongly on other neutrino fluxes predicted by the standard solar model than do the radiochemical experiments , chlorine , gallex , sage , and gno . in this section , we show how data from the super - kamiokande and kamland experiments can be combined to measure the total @xmath0b neutrino flux . we shall see that the determination using super - kamiokande and kamland yields , on average , a value for the total flux than is comparable in precision to what is expected to be obtained using sno and kamland . the systematic uncertainties are different in the two experiments , sno and super - kamiokande . therefore , it will be important to compare the total @xmath0b neutrino flux that is inferred using super - kamiokande and kamland with the value that is obtained using sno and kamland . if the uncertainties are correctly estimated , then the two methods should agree within their quoted errors . the advantage of using purely charged current measurements with sno and kamland to determine the total @xmath0b flux is that the answer depends only very slightly upon the unknown active - sterile mixture ( as discussed in sec . [ sec : determiningb8total ] ) . the principal advantages of the super - kamiokande experiment in the present context is that the neutrino interaction cross section is accurately known @xcite and the statistical and systematic errors have already been extensively investigated @xcite . however , as we shall see below , when using super - kamiokande there is a significant uncertainty ( @xmath67% and @xmath68% ) in the total @xmath0b flux due to the active - sterile mixture , at least with our present knowledge of @xmath44 . the average survival probability for @xmath0b solar neutrinos can be written in the form @xmath69 where @xmath70 is the probability of oscillating into active neutrinos and @xmath71 is the ratio of the the @xmath72 and @xmath73 elastic scattering cross - sections . the expected sensitivity of @xmath54 to the principal sources of errors can be calculated from the following equation : @xmath74 we suppose that kamland will observe ( with associated uncertainties that we simulate with a monte carlo ) the rate predicted for the global best - fit point shown in fig . [ fig : solarvskamlandreal ] . for purely active oscillations , we find that @xmath75= 1\left[1 \pm 0.054\right ] , ~~\cos^2 \eta = 1.0 , \label{eq : skfboactive}\ ] ] where the first error corresponds to the super kamiokande experimental uncertainty and the second error is caused by the finite size of the allowed kamland region in oscillation parameter space . the combined kamland and super - kamiokande measurement will , for purely active neutrinos , yield a determination for the total @xmath0b flux that is more accurate than can be obtained with the sno cc measurement and kamland . within the 1 @xmath76 lma region , the average uncertainty in the combined kamland and super - kamiokande measurement for the total active @xmath0b neutrino flux is expected to be @xmath77 for purely active neutrinos , super - kamiokande and kamland may provide us with a more accurate determination of the total @xmath0b neutrino flux than sno cc and kamland [ cf . ( [ eq : averagefbsigma ] ) and eq . ( [ eq : activesigmab8kamlandsk ] ) ] . allowing for the currently allowed @xmath2 ( @xmath78% ) sterile admixture , we find for the best fit point @xmath79,~~\cos^2 \eta \geq 0.75 \label{eq : borexinofbosterile}\ ] ] the last error in eq . ( [ eq : borexinofbosterile ] ) corresponds to the present 1@xmath76 uncertainty from the active - sterile admixture @xmath80 . the average uncertainty in the combined kamland and super - kamiokande measurement for @xmath81 is expected to be @xmath82% , within the 1 @xmath76 lma region and if the sterile admixture is as large as currently allowed . in this section , we show how data from the kamland and borexino experiments can be combined to measure the total @xmath8be neutrino flux . the principal advantage of using the borexino experiment for this purpose is that the signal in the borexino experiment @xcite is predicted to be dominated by @xmath8be neutrinos , whereas @xmath8be solar neutrinos are expected to contribute only a relatively small ( and unlabeled ) fraction to the observed event rate in the gallium and chlorine radiochemical measurements ( see table [ tab : garates ] ) scattering . the radiochemical detectors do not have energy resolution , only an energy threshold . ] . however , unlike the cases involving the gallium and chlorine experiments that were discussed in sec . [ sec : be7 ] , which include only @xmath40 ( cc ) absorption , the borexino experiment detects both @xmath83 scattering and @xmath84 and @xmath85 scattering . one must consider in the present case the extent to which the active - sterile neutrino admixture influences the detected event rate for each set of oscillation parameters , @xmath86 , determined by kamland . as we shall see quantitatively in the following discussion , this uncertainty regarding the sterile admixture does not decrease significantly the overall accuracy of the inferred total @xmath8be neutrino flux . the average survival probability for active neutrinos can be written conveniently in the form @xmath87 where @xmath88 is the oscillation probability of neutrino fluxes of source @xmath89 into active neutrinos and @xmath90 is the ratio of the the @xmath91 and @xmath73 elastic scattering cross - sections . the expression for @xmath56 has the same form for the case in which the kamland and borexino experiments are considered together as for the previously discussed cases involving the gallium experiments [ see eq . ( [ eq : fbe ] ) ] and the chlorine experiment [ see eq . ( [ eq : fbechlorine ] ) ] . using the average survival probability defined in eq . ( [ eq : survivalscattering ] ) , we can write @xmath92 \label{eq : fborexino}\end{aligned}\ ] ] in the last term of eq . ( [ eq : fborexino ] ) , we include the contributions from @xmath10 , and cno neutrinos . the contribution from @xmath0b neutrinos is negligible because the observed @xmath0b neutrino flux is about a factor of @xmath93 smaller than the predicted ssm @xmath8be neutrino flux and because @xmath0b neutrinos primarily produce high energy recoils electrons that the borexino detector can distinguish from the low energy recoil electrons produced by @xmath8be neutrinos . the expected sensitivity of @xmath56 to different sources of errors is given by @xmath94 in order to estimate the accuracy of this method for determining @xmath56 , we suppose that kamland will observe ( with associated uncertainties that we simulate with a monte carlo ) the rate predicted for the global best - fit point , see ref . @xcite . ] shown in fig . [ fig : solarvskamlandreal ] and that borexino also will observe a signal corresponding to this best - fit point ( with associated uncertainties ) . in the absence of any published data , we estimate that - based upon experience in previous solar neutrino experiments @xcite - that borexino will achieve a systematic uncertainty of between @xmath95% and @xmath4% . we estimate in this way that the combined experiments will yield a determination of @xmath56 that is , for purely active oscillations , @xmath96= 1\left[1 \pm { 0.057 } ( \pm { 0.103)}\right ] , ~~\cos^2 \eta = 1.0 , \label{eq : borexinofbeactive}\ ] ] where the first error corresponds to the borexino experimental uncertainty , the second to the uncertainty in the reconstructed kamland region , and the last error is due to the theoretical uncertainty in the predicted cno fluxes . within the 1 @xmath76 lma region , the average uncertainty in the combined kamland and borexino measurement for the total active @xmath8be neutrino flux is expected to be @xmath97 if we allow for a 25% sterile admixture , we find @xmath98 \label{eq : borexinofbesterile}~.\ ] ] the last error in eq . ( [ eq : borexinofbesterile ] ) corresponds to the present 1@xmath76 uncertainty from the active - sterile admixture @xmath80 . the numbers in parentheses in eq . ( [ eq : borexinofbeactive ] ) and eq . ( [ eq : borexinofbesterile ] ) correspond to assuming the larger systematic uncertainty , @xmath99% , for the borexino measured rate . for @xmath100 , the average uncertainty in the combined kamland and borexino measurement is expected to be @xmath101% , within the 1 @xmath76 lma region and if the sterile admixture is as large as currently allowed , in this section , we analyze three strategies for determining the total @xmath10 solar neutrino flux without requiring a dedicated experiment that measures only the @xmath10 neutrinos . we first describe in sec . [ subsec : existingdata ] how one can make a crude determination of the @xmath10 neutrino flux using the measured gallium , chlorine , and sno ( cc ) event rates @xcite and the standard solar model predictions for all but the @xmath0b , @xmath8be , and @xmath10 solar neutrino fluxes . this part of the discussion is similar to the analysis described in ref . @xcite , although we evaluate explicitly the uncertainty caused by the finite size of the allowed region in oscillation parameter space . we then determine in sec . [ subsec : galliumsno ] how well one can infer the @xmath10 flux using just the gallium and the sno measurements and the bp00 predictions for the other neutrino fluxes , especially the @xmath8be neutrino flux . the unknown sterile - active mixture contributes only a negligible error using the strategies described in sec . [ subsec : existingdata ] and sec . [ subsec : galliumsno ] . next , we show in sec . [ subsec : ppkamlandborexino ] how the precision of the determination of the @xmath10 flux can be improved by using , in the future , the results from kamland ( to constrain the neutrino oscillation parameters ) and from borexino ( to constrain the @xmath8be neutrino flux ) . the principle of this strategy has also been discussed by the sage collaboration @xcite . the theoretical error on the @xmath10 neutrino flux is @xmath103% ( see ref . @xcite ) , which is an order of magnitude smaller than the estimated errors that we find in this section on the empirically derived values of @xmath104 . to achieve a precision of @xmath4% or better in the determination of the solar neutrino @xmath10 flux will require a dedicated and accurate experiment devoted to measuring the @xmath10 flux . throughout this section , we treat the @xmath10 flux and the @xmath38 flux as a single variable because they are so closely linked physically @xcite . the two reactions are linked because the @xmath38 reaction is obtained from the @xmath10 reaction by exchanging a positron in the final state with an electron in the initial state . thus the rates for the two reactions are proportional to each other to high accuracy @xcite . for convenience , we denote the sum of the @xmath10 and @xmath38 fluxes as simply @xmath10 . in this subsection , we show how data from the chlorine , gallium , and sno experiments can be combined with the bp00 predictions for the cno and @xmath105 fluxes to determine the total @xmath10 neutrino flux . the reduced @xmath10 solar neutrino flux , @xmath104 defined ( by analogy with @xmath54 ) with respect to the predicted standard solar model sum of the @xmath10 and @xmath38 fluxes , can be written as : @xmath106 & & - f_{\rm be } r^{\rm ^7be , ssm}_{\rm ga } \langle p_{ee}(\delta m^2,\tan^2\theta)\rangle^{\rm ^7be}_{\rm ga } -r^{\rm^8b , ssm}_{\rm ga } \frac{r^{\rm cc , exp}_{\rm sno } } { r^{\rm cc , ssm}_{\rm sno } } \frac{\langle p_{ee}(\delta m^2,\tan^2\theta)\rangle^{\rm ^8b}_{\rm ga } } { \langle p_{ee}(\delta m^2,\tan^2\theta)\rangle^{\rm ^8b}_{\rm sno } } \big ] .\label{eq : fpp}\end{aligned}\ ] ] here the sum over @xmath107 in eq . ( [ eq : fpp ] ) refers to the three cno neutrino sources and the @xmath105 neutrinos ( see table [ tab : garates ] ) . we insert in eq . ( [ eq : fpp ] ) the expression for @xmath56 in terms of the chlorine and sno rates and the standard solar model cno and @xmath105 neutrino fluxes . explicitly , @xmath108 & & -r^{\rm ^8b , ssm}_{\rm cl } \frac{r^{\rm cc , exp}_{\rm sno } } { r^{\rm cc , ssm}_{\rm sno}}\frac { \langle p_{ee } ( \delta m^2,\tan^2\theta)\rangle^{\rm ^8b}_{\rm cl } } { \langle p_{ee } ( \delta m^2,\tan^2\theta)\rangle_{\rm sno } } \big ] . \label{eq : fppbechlorine}\end{aligned}\ ] ] for the special case of the best fit point in the lma solution region , we find @xmath109 the first error in eq . ( [ eq : fppbestlma ] ) , ( @xmath110 ) , results from the weighted average experimental error for the gallium experiment @xcite . the second error , ( @xmath111 ) , reflects the uncertainties in the calculated neutrino absorption cross sections on gallium @xcite . the third error , ( @xmath112 ) , is caused by the uncertainties in the calculated standard solar model cno fluxes @xcite . the fourth error , ( @xmath113 ) , derives from the measurement errors in the sno cc experimental data @xcite and neutrino absorption cross section @xcite . the fifth error , ( @xmath114 ) , results from the experimental error for the chlorine event rate @xcite and the sixth error , ( @xmath115 ) , reflects the uncertainties in the calculated neutrino absorption cross sections for chlorine @xcite . the physical constraint , @xmath116 , was imposed in obtaining these errors . within the @xmath2 lma allowed region , we find that the central value of @xmath104 can vary in the range @xmath117 and the best - fit value can be determined with an average uncertainty of @xmath118% . the range of central values of @xmath104 always exceeds the bp00 value because the observed chlorine rate implies a low value of @xmath119 . since the largest contributions to the gallium rate are from @xmath10 neutrino and @xmath8be neutrinos , a low @xmath8be flux implies a relatively high @xmath10 flux . we can summarize as follows the results of the simulations carried out within the @xmath2 allowed lma region for the @xmath10 flux : @xmath120 the first error in eq . ( [ eq : fppsummaryknown ] ) results from the allowed range of neutrino oscillation parameters and the second error results from the uncertainties in all other recognized sources of errors , combined quadratically . the result given in eq . ( [ eq : fppsummaryknown ] ) should be compared with the quoted estimate by the sage collaboration @xcite , @xmath121 . this agreement is very welcome since the sage paper points out that they made several approximations ... whose nature can not be easily quantified . " in this subsection , we determine the range of allowed values for @xmath104 using the bp00 predictions ( and uncertainties ) for the @xmath8be neutrinos as well as the cno and @xmath105 neutrinos . we use in this subsection data only from the gallium and sno solar neutrino experiments . following the same line of reasoning as in section [ subsec : existingdata ] , we find for the best fit point in the lma allowed oscillation region , @xmath122 just as in eq . ( [ eq : fppbestlma ] ) , the first error in eq . ( [ eq : fbbestlmabe7 ] ) , ( @xmath123 ) , is the experimental error from the weighted average event rate in the gallium experiments , the second error,(@xmath124 ) , is due to the neutrino absorption cross section on gallium , the third error , ( @xmath125 ) , is due to the uncertainties in the bp00 predictions of the cno neutrino fluxes , the fourth error , ( @xmath126 ) , contains the uncertainty in the sno cc experimental rate and the calculated absorption cross sections on deuterium . the uncertainty in the bp00 prediction for the @xmath8be neutrino flux contributes the last error , ( @xmath127 ) , in eq . ( [ eq : fbbestlmabe7 ] ) . within the @xmath2 allowed lma region , we find the central value of @xmath104 varies in the range @xmath128 ; the best - fit value of @xmath104 can be determined with an average uncertainty of @xmath129% . we can summarize the determination of @xmath104 as follows : @xmath130 where the first error is the uncertainty due to the neutrino oscillation parameters and the second error contains the uncertainties from all other sources . in this subsection , we show how the uncertainty in @xmath104 could be improved by using the kamland data to determine the neutrino oscillation parameters and the borexino data , together with the kamland oscillation parameters , to constrain @xmath131 . we start with eq . ( [ eq : fpp ] ) but now use @xmath56 as determined from borexino and kamland data [ eq . ( [ eq : fborexino ] ) ] . we find that for the best - fit lma solution ( assuming that borexino finds the rate expected for this best fit point ) , @xmath132 the first error shown in eq . ( [ eq : fppkamlandborexino ] ) , ( @xmath133 ) , is the experimental error on the measured gallium event rate , the second error , ( @xmath124 ) , represents the uncertainties from the calculated gallium cross sections . the third error , @xmath134 , is from the predicted cno fluxes and the fourth error , ( @xmath126 ) , contains the uncertainties from the sno experimental data and calculated deuterium absorption cross sections . the range of oscillation parameters within the kamland reconstructed region gives the fifth error , ( @xmath135 ) , and the sixth error , ( @xmath136 [ @xmath137 ) , results from the uncertainty in the measured borexino event rate [ which we take to be 5% ( 10% ) ] . the unknown active - sterile admixture ( which goes in the direction of lowering @xmath104 ) contributes the last error shown in eq . ( [ eq : fppkamlandborexino ] ) . on average , the the precision expected using the gallium , kamland , and sno experiments is @xmath138 . \label{eq : avfppborexinokamland}\ ] ] the dominant sources of error in eq . ( [ eq : avfppborexinokamland ] ) are the experimental error in the gallium rate and the uncertainties in the calculated gallium absorption cross section . in this section , we review and discuss our principal conclusions . we begin in sec . [ subsec : discusscurrentlyallowed ] by summarizing the experimentally - allowed range that currently exists for the total @xmath0b solar neutrino flux , taking account of the possibility that sterile neutrinos exist . we then summarize in sec . [ subsec : discussb8 ] how well the total @xmath0b neutrino flux , and separately the sterile @xmath0b neutrino flux , can be obtained by combining kamland and sno measurements . next we describe in sec . [ subsec : discussbe7 ] how well the total @xmath8be flux can be determined using data from the kamland , gallium ( gallex , sage , and gno ) , and sno experiments , i.e. , using only cc disappearance experiments . in this same subsection , we summarize the measurement accuracy that can be obtained using data from just the kamland and the borexino ( @xmath39 scattering ) experiments . we describe three strategies ( making use of existing and future solar neutrino measurements and guidance from the bp00 solar model ) for determining the @xmath10 solar neutrino flux without the help of an experiment that measures separately the @xmath10 neutrinos . we point out in sec . [ subsec : discusscrosssections ] ( and in the appendix ) that in order to determine the total @xmath0b or @xmath8be neutrino flux , and _ a fortiori _ to determine the sterile contributions to these fluxes , the correlations among theoretical errors for neutrino absorption cross sections and for solar neutrino fluxes must be treated more accurately than in previous discussions . section [ subsec : discussfocus ] summarizes the main focus of the present paper . we concentrate on procedures to determine experimentally the total solar @xmath0b neutrino flux and the total solar @xmath8be neutrino flux in a universe in which sterile neutrinos might exist . our methods can work only if the lma solution of the solar neutrino problem is correct . the numerical values given here for the precision with which different quantities can be determined are obtained using simulations of how well different experiments may perform . therefore , the numerical values are intended only as illustrative guides to what may be possible . the combined sno cc data and the super - kamiokande @xmath83 scattering data together yield a widely acclaimed agreement between the @xmath0b solar neutrino flux that is predicted by the standard solar model and the empirically inferred @xmath0b neutrino flux . however , this agreement between solar model prediction and solar neutrino measurement is not a unique interpretation of the existing measurements if one allows for the possibility that the incident solar neutrino flux could contain a significant component of sterile neutrinos . we show in table [ tab : fbranges ] and in sec . [ sec : presentknowledge ] that if one takes account of the possibility that sterile neutrinos may exist then the total solar @xmath0b neutrino flux could be as large as @xmath139 times the flux predicted by the standard solar model . in principle , the existing solar neutrino data could be inconsistent with the standard solar model predictions . the total @xmath0b solar neutrino flux , active plus sterile neutrinos , can be determined with a typical @xmath2 accuracy of about @xmath4% by comparing the charged current measurements from the kamland reactor experiment and the sno experiments ( see sec . [ sec : determiningb8total ] ) . the active @xmath0b neutrino flux has been measured by comparing the sno cc flux ( which measures @xmath40 ) with the super - kamiokande neutrino - electron scattering rate ( which measures @xmath40 plus , with less sensitivity , @xmath140 ) . the sno neutral current measurement will provide an additional and , ultimately , more accurate measurement of the total active @xmath0b solar neutrino flux . by subtracting the independently - measured active @xmath0b neutrino flux from the total ( active plus sterile ) @xmath0b neutrino flux , one can determine empirically the sterile component of the solar neutrino flux . the active @xmath0b neutrino flux will eventually be determined accurately by the sno neutral current measurement @xcite . we estimate therefore that the procedure described here has the potential of measuring , or setting an upper limit on , the sterile component of the @xmath0b neutrino flux that is as small as @xmath141% of the total @xmath0b solar neutrino flux . the measurement of the total @xmath0b neutrino flux , and the sterile component of this flux , are independent of solar model considerations . in order to establish the quantitative conclusions , we have performed detailed simulations of the accuracy of the kamland reactor experiment in determining neutrino oscillation parameters ( see fig . [ fig : solarvskamlandreal ] ) and have evaluated the theoretical and experimental uncertainties that affect the different flux determinations ( see sec . [ subsec : precisionb8 ] and the appendix ) . the combined measurements of the super - kamiokande and kamland experiments can be used to determine independently a value for the total @xmath0b neutrino flux . this determination may be as accurate as @xmath6% for purely active neutrinos . with the current limits on the active - sterile admixture , the total @xmath0b neutrino flux could be inferred to an accuracy of @xmath7% or better , as described in sec . [ subsec : kamlandsuperkb8 ] . it will be important to compare the value of the total @xmath0b neutrino flux inferred by combining the kamland and sno charged current measurements with the value obtained using the kamland and super - kamiokande experiments . this comparison will be an important test of whether the systematic uncertainties in the experiment and in the analyses are understood . the total @xmath8be solar neutrino flux , active plus sterile neutrinos , can be determined to a @xmath2 accuracy of about @xmath142% by combining measurements from kamland , sno , and the gallium experiments ( see sec . [ sec : be7 ] ) . unlike the purely empirical determination that is possible for the @xmath0b flux , the measurement of the total @xmath8be solar neutrino flux requires some assumption regarding the cno solar neutrino fluxes . in our estimates , we have assumed that the standard solar model predictions for the cno fluxes , and their uncertainties , are at least approximately valid . table [ tab : fbe ] shows that as long as the cno fluxes are not a factor of three or more larger than the standard solar model predictions , then they will not significantly limit the accuracy with which the total @xmath8be neutrino flux can be determined . the measured capture rate in the gallium experiments ( gallex , sage , and gno ) currently constitutes the largest recognized uncertainty in the determination of the total @xmath8be flux by the method described here ( see table [ tab : fbe ] ) . the constraints provided by the chlorine experiment are not very useful in providing an accurate determination of the @xmath8be neutrino flux [ see table [ tab : fbecl ] and eq . ( [ eq : averagebe7shiftcl ] ) ] . one can also determine the allowed range of the total @xmath8be solar neutrino flux using the data from the kamland reactor experiment and the borexino solar neutrino experiment . we show in sec . [ subsec : kamlandborexino ] that with this method one may hope to obtain a @xmath2 accuracy of @xmath1% or better for the total @xmath8be solar neutrino flux , which is more accurate than we estimate will be possible with the gallium and chlorine radiochemical experiments . one can determine an upper limit to the sterile component of the @xmath8be solar neutrino flux by combining the measured rate in the neutrino - electron scattering experiment , borexino , with the @xmath8be total flux inferred from measurements of the kamland , sno , and gallium experiments . if one assumes that the entire signal measured in borexino is due to @xmath40 , then one obtains a minimum value for the active component of the @xmath8be neutrino flux . subtracting this minimum value from the total @xmath8be flux , one will obtain an upper limit to the sterile component of the flux . it seems unlikely that the procedure described here has the sensitivity to measure a value for the sterile component unless the sterile flux is larger than @xmath142% of the total @xmath8be neutrino flux . however , limits on the sterile neutrino admixture can be obtained from the analysis of @xmath0b and kamland neutrino measurements described in sec . [ subsec : discussb8 ] . in order to make a direct and precise measurement of the sterile component of the @xmath8be solar neutrino flux , we need a charged current measurement of the @xmath8be flux . a @xmath8be solar neutrino absorption experiment , e.g. , with a lithium target @xcite or with lens @xcite , would make possible an accurate determination of the sterile @xmath8be neutrino flux by providing a set of experimental constraints that is analogous to what will exist for the sno , super - kamiokande , and kamland experiments . the theoretical uncertainty in the calculated @xmath10 solar neutrino flux is estimated to be only @xmath143 @xcite . therefore , a precise determination of the @xmath10 solar neutrino flux will be of great interest as a crucial test of the theory of stellar evolution . the measurement of the @xmath10 neutrino flux will also provide a critical test of whether the neutrino oscillation theory , which works well at energies above @xmath95 mev , also describes accurately the lower energy neutrino phenomena ( energies less than @xmath144 mev ) . we show in sec . [ sec : determiningpp ] that the @xmath10 flux can be determined with modest accuracy , of order @xmath145% to @xmath146% , using a combination of existing experimental data and some guidance from the bp00 standard solar model . in the future , it should be possible to determine the @xmath10 neutrino flux to an accuracy of @xmath11% using experimental data from the borexino , kamland , and sno solar neutrino experiments and predictions ( in a non - critical way ) from the standard solar model ( see eq . [ eq : avfppborexinokamland ] ) . the unknown flux of sterile neutrinos does not significantly affect the quoted estimates on the accuracy with which the @xmath10 flux can be determined . to measure the @xmath10 flux with an accuracy sufficient to test stringently the standard solar model prediction will require a dedicated and accurate experiment that measures separately the @xmath10 neutrino flux . in order to determine the total @xmath0b or @xmath8be solar neutrino flux , and their sterile components , one must evaluate carefully all known sources of error . in the course of this investigation , we realized that the evaluations of the neutrino absorption cross section uncertainties in previous neutrino oscillation studies , including our own , have not properly taken account of the correlations among the theoretical uncertainties in the cross section calculations ( for an insightful discussion of this point , see ref . we discuss in the appendix how the cross section errors can be treated more correctly . we also emphasize here that it is necessary to treat as fully correlated the uncertainties in the principal cno neutrino fluxes that are obtained from standard solar model predictions . these effects are small compared to other uncertainties in determining solar neutrino oscillation parameters ( the principal goal of nearly all previous oscillation studies ) . the correlations among cross section uncertainties become important only when one wants to make accurate inferences regarding the neutrino fluxes themselves . the main focus in this paper is on determining experimentally the total @xmath0b , @xmath8be , and @xmath10 solar neutrino fluxes in order to make possible more precise tests of solar model predictions . we have also shown that the contribution of sterile neutrinos to the total flux can be measured , or a useful upper limit can be set , for @xmath0b solar neutrinos and for @xmath8be solar neutrinos . jnb acknowledges support from nsf grant no . mcg - g is supported by the european union marie - curie fellowship hpmf - ct-2000 - 00516 . this work was also supported by the spanish dgicyt under grants pb98 - 0693 and fpa2001 - 3031 . we determine , as is conventional for many analyses of solar neutrino data , the allowed regions in the neutrino oscillation space using a @xmath147 function that includes all the relevant data . in the construction of the @xmath147 function , we have followed closely the prescription of ref . @xcite ( see also ref . @xcite ) , but we have included some modifications to this prescription in order to account in more detail for the energy dependence and the correlation of the cross section errors for the chlorine and gallium solar neutrino experiments . for a given experiment @xmath148 ( for example , gallium or chlorine ) , the expected number of events can be written as @xmath149 where @xmath150 labels the solar neutrino fluxes , @xmath89 , and @xmath151 labels the energy bins of energy @xmath152 . the quantity @xmath153 is the cross section for the interaction of a neutrino of energy @xmath152 in the experiment @xmath148 ; @xmath154 is the survival probability of @xmath40 for a given energy @xmath152 at a specified point , @xmath155 , in neutrino oscillation space , because different neutrino species have different probability distributions for the location of their production within the sun . for matter oscillations , the survival probability obviously depends upon where in the sun the neutrino was produced . ] . the error matrix for the neutrino absorption cross sections can be derived using eq . ( [ eqapp : rate ] ) . since the errors are uncorrelated between different experiments , we need to evaluate the following expression for a specified experiment : @xmath156 where @xmath157 is best - estimate for the ( gallium or chlorine ) experimental capture rate and the brackets indicate an average over the probability distribution of uncertainties in the neutrino cross sections . writing out the various terms in eq . ( [ appeq : defnsigma ] ) , we find @xmath158 in the standard ref . @xcite , the cross section error matrix for each experiment is given as : @xmath159 where @xmath160 is the average uncertainty in the interaction cross section of neutrinos of flux type @xmath107 in experiment @xmath148 . equation ( [ sigma2 ] ) is obtained from eq . ( [ eqapp : rate ] ) by neglecting the correlations of the cross section errors among the different neutrino fluxes and by neglecting the energy dependence of the cross section errors . for the gallium detector , neither of these assumptions is correct and even for chlorine the cross section errors for different neutrino sources are strongly correlated . for all but the @xmath0b and @xmath105 neutrino fluxes , either ground state to ground state transitions are the only energetically possible transitions ( which is the case for the chlorine detector ) or the ground state to ground state transitions dominate ( which is the case for the gallium detector ) . thus all the cross sections for the lower energy neutrinos move up or down together , proportional to the square of the dominant matrix element . when considering the gallium experiments , we include the energy dependence of the cross section errors and assume full correlations among the cross section errors of the different neutrino sources that contribute at a specific energy . specifically , we use the results from ref . @xcite for the cross section errors . for continuum sources , we take the fractional error of the interaction cross section in the energy bin @xmath151 to be @xmath161,\ ] ] where @xmath162 are the 3@xmath76 upper and lower limit cross sections given in tables iii and iv of ref . @xcite . for the line sources , @xmath8be and @xmath38 , we use the errors given in eqs . ( 41 ) and ( 42 ) of ref . @xcite . to be complete , we have checked that the shapes of these neutrino lines ( see ref . @xcite ) do not affect significantly the error estimates . the gallium contribution to the cross section error matrix is therefore given by @xmath163 where @xmath164 is the correlation matrix for the cross section errors of the different energy bins . for low energy neutrino sources , the dominant process is the transition to the germanium ground state and for higher energy neutrino sources ( @xmath165 mev ) transitions to excited states dominate . therefore , we assume the cross section errors to be fully correlated between energy bins either below or above @xmath166 mev ( @xmath167 for @xmath168 mev or @xmath169 mev ) . we take the errors to be uncorrelated between one energy bin below @xmath170 mev and one energy bin above @xmath166 mev ( @xmath171 for @xmath172 mev and @xmath173 mev , or @xmath172 mev and @xmath173 mev ) . for chlorine , we take as fully correlated the cross section errors for the @xmath10 , @xmath38 , @xmath8be , and cno neutrinos . the cross section errors for the @xmath0b and @xmath105 neutrinos are uncorrelated with the errors for the lower energy neutrinos but are fully correlated with each other . we can neglect for chlorine the energy dependence of the chlorine to argon cross section errors because the uncertainty for the lower energy neutrinos is determined almost entirely from the ground - state to ground - state matrix element . ( forbidden corrections are unimportant for these low energy neutrinos . ) for the @xmath0b and @xmath105 neutrinos , the absorption cross sections are also dominated by a single ( but different ) transition , in this case the super - allowed transition @xcite , and therefore we can also neglect the energy dependence for the higher energy neutrinos . we adopt the values of the averaged chlorine cross section errors given in refs . @xcite . to be explicit , the chlorine contribution to the cross section error matrix is given by @xmath174 where @xmath175 for @xmath176 or @xmath177 . also , @xmath178 for @xmath179 and @xmath180 . 99 q. r. ahmad _ et al . _ , sno collaboration , nucl - ex/0204008 . q. r. ahmad _ et al . _ , sno collaboration , nucl - ex/0204009 . j. n. abdurashitov _ et al . _ , sage collaboration , astro - ph/0204245 . q. r. ahmad _ et al . _ , sno collaboration , phys . lett . * 87 * , 071301 ( 2001 ) . s. fukuda _ et al . _ , super - kamiokande collaboration , phys.rev . lett . * 86 * , 5651 ( 2001 ) . g. l. fogli , e. lisi , d. montanino , and a. palazzo , phys . d * 64 * , 093007 ( 2001 ) . c. giunti , phys . d * 65 * , 033006 ( 2002 ) ; m. v. garzelli and c. giunti , phys . d * 65 * , 093005 ( 2002 ) . g. fiorentini , f. l. villante , and b. ricci , hep - ph/0109275 . j. n. bahcall , phys . c. * 65 * , 015802 ( 2002 ) . j. n. bahcall , phys . * 12 * , 300 ( 1964 ) ; r. davis jr . * 12 * , 303 ( 1964 ) ; j. n. bahcall and r. davis , science * 191 * , 264 ( 1976 ) . j. n. bahcall , m. h. pinsonneault , and s. basu , astrophys . j. * 555 * , 990 ( 2001 ) . b. t. cleveland , t. daily , r. davis , jr . , j. r. distel , k. lande , c. k. lee , and p. s. wildenhain , astrophys . j. * 496 * , 505 ( 1998 ) . y. fukuda _ _ , kamiokande collaboration , phys . lett . * 77 * , 1683 ( 1996 ) . j. n. abdurashitov _ et al . _ , sage collaboration , phys . c * 60 * , 055801 ( 1999 ) . w. hampel _ et al . _ , gallex collaboration , phys . b * 447 * , 127 ( 1999 ) . e. belloti , nucl . b ( proc . suppl . ) * 91 * , 44 ( 2001 ) . ( ` http://nu2000.sno.laurentian.ca ` ) . j. n. bahcall , s. basu , and m. pinsonneault , phys . b * 433 * , 1 ( 1998 ) . w. kwong and s. p. rosen , phys . d * 54 * , 2043 ( 1996 ) . f. l. villante , g. fiorentini , and e. lisi , phys . d * 59 * , 013006 ( 1999 ) . j. n. bahcall , m. c. gonzalez - garcia , and c. pena - garay , j. high energy phys . * 08 * , 014 ( 2001 ) . p. i. krastev and a. y. smirnov , phys . d * 65 * , 073022 ( 2002 ) . j. n. bahcall , m. c. gonzalez - garcia , and c. pena - garay , j. high energy phys . * 04 * , 007 ( 2002 ) . a. r. junghans , e. c. mohrmann , k. a. snover , t. d. steiger , e. g. adelberger , j. m. casandjian , h. e. swanson , l. buchmann , s. h. park , and a. zyuzin , phys . * 88 * , 041101 ( 2002 ) . s. m. bilenky , c. giunti , and w. grimus , prog . * 43 * , 1 ( 1999 ) . m. c. gonzalez - garcia and y. nir , hep - ph/0202058 . v. barger , d. marfatia , and k. whisnant , phys . lett . * 88 * , 011302 ( 2002 ) ; v. barger , d. marfatia , and k. whisnant , phys . b * 509 * , 19 ( 2001 ) . d. dooling , c. giunti , k. kang , and c. w. kim , phys . d * 61 * , 073011 ( 2000 ) . c. giunti , m. c. gonzalez - garcia , and c. pena - garay , phys . d * 62 * , 013005 ( 2000 ) . j. aguilar _ et al . _ , lsnd collaboration , phys . rev . d * 64 * , 112007 ( 2001 ) . k. eitel , karmen collaboration , nucl . b ( proc . suppl . ) * 91 * , 191 ( 2001 ) . a. piepke , kamland collaboration , nucl . b ( proc . suppl . ) * 91 * , 99 ( 2001 ) ; j. busenitz _ et al . , _ , kamland collaboration , `` proposal for us participation in kamland , '' march 1999 ( unpublished ) . may be downloaded from:`http://kamland.lbl.gov/ ` ; k. ishihara for the kamland collaboration , talk at the nufact01 workshop in tsukuba , japan ( may 2430 , 2001 ) . transparencies at:`http://psux1.kek.jp/~nufact01/index.html ` . a. bandyopadhyay , s. choubey , s. goswami , and k. kar , phys . b * 519 * , 83 ( 2001 ) . v. barger , d. marfatia , and b. p. wood , phys . b * 498 * , 53 ( 2001 ) . r. barbieri and a. strumia , j. high energy phys . * 12 * , 016 ( 2000 ) . h. murayama and a. pierce , phys . d * 65 * , 013012 ( 2002 ) . j. n. bahcall , m. c. gonzalez - garcia , and c. pena - garay , hep - ph/0204314 . m. smy , proceedings of the noon2001 workshop , kashiwa , japan , december 5 - 8 , 2001 , hep - ex/0202020 a. de gouvea and c. pena - garay , phys . d * 64 * , 113011 ( 2001 ) . m. c. gonzalez - garcia and c. pena - garay , phys . b * 527 * , 199 ( 2002 ) . g. alimonti _ et al . _ , borexino collaboration , astropart . * 16 * , 205 ( 2002 ) ; a. piepke _ et al . _ , kamland collaboration , nucl . suppl . ) * 91 * , 99 ( 2001 ) . s. nakamura , t. sato , s. ando , t .- s . park , f. myhrer , v. gudkov , and k. kubodera , nucl - th/0201062 . see also , a. kurylov , m. j. ramsey - musolf , and p. vogel , phys . rev . c * 65 * , 055501 ( 2002 ) ; j. f. beacom and s. j. parke , phys . rev . d * 64 * , 091302 ( 2001 ) . m. butler , j .- w . chen , and x. kong , phys . c * 63 * , 035501 ( 2001 ) . r. davis jr . , d. s. harmer , and k. c. hoffman , phys . lett . * 20 * , 1205 ( 1968 ) . j. n. bahcall , phys . c * 56 * , 3391 ( 1997 ) . j. n. bahcall phys . lett . * 71 * , 2369 ( 1993 ) ; phys . d * 49 * , 3923 ( 1994 ) . j. n. bahcall and h. a. bethe , phys . lett . * 65 * , 2233 ( 1990 ) . j. n. bahcall , m. kamionkowski , and a. sirlin , phys . d * 51 * , 6146 ( 1995 ) . j. n. bahcall , _ neutrino astrophysics _ ( cambridge university press , cambridge , 1989 ) . j. n. bahcall and r. m. may , astrophys . j. * 155 * , 501 ( 1969 ) . j. n. bahcall , e. lisi , d. e. alburger , l. de braeckeleer , s. j. freedman , and j. napolitano , phys . c * 54 * , 411 ( 1996 ) . j. n. bahcall , p. i. krastev , and a. y. smirnov , phys . d * 62 * , 093004 ( 2000 ) . m. galeazzi , g. gallinaro , f. gatti , p. meunier , s. vitale , a. v. kopylov , v. v. petukhov , e. a. yanovich , and g. t. zatsepin , phys . b * 398 * , 187 ( 1997 ) ; a. kopylov and v. petukhov , hep - ph/0201042 . r. s. raghavan , phys . lett . * 78 * , 3618 ( 1997 ) ; r. s. raghavan , hep - ex/0106054 . m. v. garzelli and c. giunti , phys . b * 488 * , 339 ( 2000 ) . g. l. fogli and e. lisi , astropart . phys . * 3 * , 185 ( 1995 ) with updated uncertainties as discussed in g. l. fogli , e. lisi , d. montanino , and a. palazzo , phys . d * 62 * , 013002 ( 2000 ) . j. n. bahcall , phys . c * 56 * , 3391 ( 1997 ) .

the present paper is part of the local cosmology from isolated dwarfs ( lcid ) project . we have obtained deep _ hubble space telescope _ ( hst ) photometry , reaching the oldest main sequence turn - off points , of six isolated dwarf galaxies in the local group : ic 1613 , leo a , cetus , tucana , , and phoenix . five galaxies were observed with the advanced camera for surveys ( acs , * ? ? ? * ) , while phoenix was observed with the wide field and planetary camera-2 ( wfpc2 , * ? ? ? * ) . the main goal of the lcid project is to derive the star formation histories ( sfhs ) , age - metallicity relations ( amrs ) , variable star populations , and stellar population gradients of this sample of galaxies . our objective is to study their evolution at early epochs and to probe effects of cosmological processes , such as the cosmic uv background subsequent to the onset of star formation in the universe or physical processes such as the gas removal by supernovae ( sne ) feedback . our sample consists of field dwarfs which were chosen in an effort to study systems as free as possible from environmental effects due to strong interactions with a host , massive galaxy . the sfh is a powerful tool to derive fundamental properties of dwarf galaxies and their evolution @xcite , but to study the earliest epochs of star formation , deep cmds , reaching the oldest main sequence turn - offs , are required ( cf . our impressions of the evolution of dwarf galaxies are biased by studies of the nearby , gas - poor dsph companions of the milky way . because of their proximity , deep cmds are obtainable from ground - based observatories . in contrast , the gas - rich , dirr galaxies , which are found at greater distances , have few studies with comparable photometric depth . the magellanic clouds represent the one exception to this generalization , but , because of they are substantially more massive than the typical dwarf galaxy and have potentially complex histories due to their current interactions ( see * ? ? ? * and references therein ) , they present less than ideal targets for study . because of the larger distances to the dirrs , until the lcid project , none have had resolved star studies which reach down to the oldest main sequence turnoff stars . thus , our view of the sfhs of dwarf irregular galaxies has been shaped by relatively indirect measures . the work by @xcite represents a seminal contribution to our understanding of the sfhs of dirrs . based on galaxy mass estimates , blue luminosities , and h@xmath1 luminosities , they demonstrated that most irregular galaxies were consistent with nearly constant star formation over their lifetimes ( as opposed to the larger spiral galaxies which showed declining star formation rates , hereafter sfrs ) . most importantly , they pointed out that `` the constant sfr history implies that the simple classical model in which star formation is proportional to gas density in a closed system can not be correct for irregular galaxies . '' modern observations of nearby dirrs have been collected and summarized by @xcite , and indicate that , on average , sfrs are higher at early times , but that there are some dirrs for which constant sfrs are a good approximation . recently , there have been theoretical papers emphasizing the difficulty in making models of galaxies which have nearly constant sfrs . @xcite , @xcite , @xcite , and @xcite have all highlighted the difficulty of producing dwarf galaxies with properties comparable to those observed in the current universe . the degree of failure is greatest in the amount of mass converted into stars , which is of order one magnitude too large in cosmological simulations . this overproduction of stars in dwarfs is an extreme symptom of the `` over - cooling problem '' ( c.f . * ) faced by all galaxy modeling . without deep imaging of resolved stars in dirr galaxies , we lack sufficient time resolution to study sfhs at the earliest times . for example , the earliest time bin used by @xcite is 4 gyr in duration . at this time resolution , we can not distinguish between sfrs that are constant at all times , or sfrs that show considerable variation ( most importantly evidence for an early dominant episode of sf ) . the initial time bin of 4 gyr covers the time range from before re - ionization up to a redshift of @xmath22 ( which is approaching the peak of in the history of universal star formation , e.g. , * ? ? ? * ) , and represents this entire important range with a single average number . the observations of ic 1613 presented here resolve this initial period and represent a small step into this relatively unexplored territory . here we present our analysis of a deep hst / acs observation of a field in ic 1613 . as we discuss in section [ repfield ] , the sfh that we derive for this field is likely a good representation for the entire galaxy . ic 1613 is a low - luminosity , local group , dirr galaxy which is one of the nearest gas - rich irregular galaxies ( for a review of the properties of ic 1613 see * ? ? ? because of its proximity , ic 1613 offers the opportunity to reconstruct a detailed sfh of a relatively isolated and non - interacting dwarf irregular galaxy . ic 1613 also has very low foreground and internal reddening ( although it lies within 5 degrees of the ecliptic ) . there have been several determinations of its distance based on cepheid and rr lyr variable stars and the tip of the red giant branch . using all three methods , @xcite derived a distance of 730 kpc . in a comparison of all of the literature values ( which included their own measurement using rr lyrae and cepheid data of 770 kpc ) , @xcite determine a mean distance of 760 kpc . the most recent determination is by @xcite who derived a distance of 740 kpc using cepheids . for consistency with the other lcid studies , we will adopt the distance estimate from @xcite which corresponds to scales of 221 pc arcminute@xmath3 , 3.7 pc arcsecond@xmath3 , and @xmath2 0.2 pc pixel@xmath3 . ic 1613 is not considered to be a satellite of either the milky way or m31 . @xcite included ic 1613 in the diffuse `` local group cloud , '' and @xcite determined a distance of 517 kpc and a velocity of @xmath490 km s@xmath3 relative to the local group barycenter . as such , it is located very close to the zero velocity surface for m 31 and well within the zero velocity surface for the local group @xcite . given its isolated position and velocity , ic 1613 has not had any recent interactions , although interactions with other galaxies long ago can not be ruled out . to date , there are no proper motion studies of ic 1613 , which would be very valuable in determining its potential interaction history . the physical parameters of ic 1613 were summarized in @xcite . these properties are normal for an i m v galaxy with a moderate luminosity ( @xmath515.2 ) and a small value of the maximum amplitude of the rotation curve ( v@xmath6 @xmath7 25 km s@xmath3 ; * ? ? ? its sfr of 0.003 @xmath8 yr@xmath3 @xcite is also normal for its type and luminosity . the most recent ism oxygen abundance measurement was made by @xcite , who obtained a spectrum where the @xmath94363 auroral line of [ ] was detected resulting in a measurement of 12 @xmath10 log ( o / h ) @xmath7 7.62 @xmath11 0.05 . this corresponds to 8.5% of the solar oxygen abundance ( as determined by * ? ? ? * ) , which is slightly less than that of the smc , and normal for a galaxy of its luminosity ( e.g. , * ? ? ? * ; * ? ? ? this combination of proximity and normality makes ic 1613 one of the best opportunities to study the properties of a dwarf star - forming galaxy that is relatively isolated ( as is typical for i m v galaxies ) . the overall structure of ic 1613 is also typical of an i m v galaxy . ground - based studies have provided an overview of the stellar population distributions . @xcite studied the distribution of luminous cool stars from j and k - band imaging and found agb stars covering a wide range in age in all of their inner galaxy fields . @xcite conducted a wide field survey of ic 1613 for c and m stars , and found the ( intermediate age ) c stars extended out to 15 arcminutes , well beyond the regions where star formation currently is active . recently , @xcite have conducted a wide field optical survey of ic 1613 , and trace red giant branch ( rgb ) stars out to radii @xmath12 16.5 arcminutes ( @xmath2 3.6 kpc ) , showing the galaxy to be more extended than previously thought . the resolved stellar populations of ic 1613 have been studied with the hst twice in the past , both times using the wfpc2 camera . @xcite studied a central field and found ic 1613 to be a smoothly evolving galaxy with a relatively constant sfr over the last gyr . horizontal branch ( hb ) stars were detected , indicating the presence of an old population . @xcite obtained deep imaging for a field located 7.4 arcminutes southwest of the center . while that imaging was not quite deep enough to reach to the oldest main sequence turnoff stars , greatly limiting the time resolution at the oldest ages , the derived sfrs were constant within a factor of three over the entire lifetime of the galaxy . in this paper , we present the sfh of ic 1613 obtained from observations with the acs on the hst . the photometry reaches the oldest main sequence turn - offs of the galaxy , allowing us to obtain an accurate sfh even for the oldest stellar populations . @xcite have already used these observations to conduct a study of the variable star content of ic 1613 . the structure of the paper is as follows : in [ secred ] the observations and data reduction are discussed and the cmd is presented . the derived sfh of ic 1613 is presented in [ secsfhresult ] and is compared with those of other lcid galaxies in [ seclcid ] . the implications of the sfh of ic 1613 for galaxy modeling , and , in particular , the over - cooling problem are discussed in [ seccosmo ] . the main conclusions of the work are summarized in [ seccon ] . as with the previous lcid papers , cosmological parameters of @xmath13 , @xmath14 , and a flat universe with @xmath15 are assumed ( i.e. , * ? ? ? * ) . ) at 45 ( @xmath2 1.0 kpc ) and the half - light radius ( r@xmath16 ) at 65 ( @xmath2 1.4 kpc ) as can be seen from the figure , the hst acs field is located between the two . also indicated are the positions of the two previous hst wfpc2 fields ( chevrons ) from @xcite ( inner ) and @xcite ( outer ) . [ f1],width=325 ] the acs observations of ic 1613 were obtained between august 18 and 20 , 2006 . the f475w and f814w bands were selected as the most efficient combination to trace age differences at old ages , since they provide the smallest relative error in age and metallicity in the main - sequence and sub - giant regions ( see c. gallart et al . 2013 , in prep ) . total integration times were 31,489 s in f475w and 27,119 s in f814w . the observations were organized into 12 visits of 2 orbits each , and each orbit was split into one f475w and one f814w exposure ( in order to maximize sampling of variable star light curves ) . the observing log is reported in @xcite . dithers of a few pixels between exposures were introduced to minimize the impact of pixel - to - pixel sensitivity variations ( `` hot pixels '' ) in the ccds . the observed field of ic 1613 is shown in figure [ f1 ] . at the adopted distance to ic 1613 , the footprint of the acs covers 0.56 kpc@xmath17 . the optical scale length of ic 1613 is 29 @xcite and the stellar distribution can be traced out beyond 15 arcminutes , so the 34 @xmath18 34 format of the acs covers only a fraction of the area of ic 1613 ( @xmath2 9% ) . we analyzed the images taken directly from the stsci pipeline ( bias , flat - field , and image distortion corrected ) . two psf - fitting photometry packages , daophot / allframe @xcite and dolphot @xcite , were used independently to obtain the photometry of the resolved stars . non - stellar objects and stars with discrepant and large uncertainties were rejected based on estimations of profile sharpness and goodness of fit . see @xcite for more details about the photometry reduction procedures . individual photometry catalogs were calibrated using the equations provided by @xcite . the zero - point differences between the two sets of photometry are small ( @xmath19 0.04 mag ) and typical for obtaining hst photometry with different methods @xcite . we direct the reader to extensive photometry reduction comparisons of lcid observations discussed in @xcite and @xcite . for simplicity , the rest of this paper is based on only the dolphot photometry dataset which contains 165,572 stars . signal - to - noise limitations , detector defects , and stellar crowding can all impact the quality of the photometry of resolved stars with the resulting loss of stars , changes in measured stellar colors and magnitudes , and systematic uncertainties . to characterize these observational effects , we injected @xmath20 artificial stars in the observed images and obtained their photometry in an identical manner as for the real stars . @xcite and @xcite provide detailed descriptions of the procedures we adopt for the characterization and simulation of these observational effects . , evenly spaced by factors of 2 . the left axis shows observed magnitudes corrected for extinction . the right axis shows absolute magnitudes . a distance modulus of @xmath21 and an extinction @xmath22 , @xmath23 have been used . the lines across the bottom of the right panel show the 0.25 , 0.50 , 0.75 and 0.90 completeness levels . three isochrones from the basti stellar evolution library have been over plotted for comparison . [ f2],width=325 ] the cmd of ic 1613 is shown in figure [ f2 ] . individual stars are plotted in the left panel and density levels are shown in the right panel . the left axis shows magnitudes in the acs photometric system corrected for extinction . absolute magnitudes are given on the right axis using the adopted values for the distance modulus ( @xmath24 ) and extinctions ( @xmath25 and @xmath26 mag ) from @xcite . the completeness factor , @xmath27 , the rate of recovered artificial stars as a function of magnitude and color , is shown in figure [ f2 ] , right panel , at @xmath28 and 0.90 . finally , in order to highlight the main features of the cmd , three isochrones from the basti stellar evolution library @xcite are also shown in the right panel as labeled in the inset . as shown in figure [ f2 ] , our photometry reaches below the oldest main - sequence turn - off at the 50% completeness limit , allowing for very strong constraints on the oldest epochs of star formation . these observations are @xmath29 mag fainter than the deepest cmd previously obtained for this galaxy @xcite . since there are no early isolated episodes of elevated sfr in the sfh , there is no obvious indication of the time resolution in the early sfh of ic 1613 shown in figure [ f2 ] . however , extensive modeling with similar lcid datasets has shown that observations of this quality yield a time resolution of @xmath2 1 gyr @xcite . consistent with earlier observations , a comparison of the observations with the over - plotted isochrones and the presence of an extended horizontal branch indicate that a very old , very low metallicity stellar population is present in the galaxy . the gap produced in the hb by the rr - lyrae variables and a red clump ( rc ) at the red end of the hb are clearly defined . an ms with stars younger than 200 myr is clearly apparent . some blue and red core helium - burning stars might also be present above the rc ( @xmath30 ; @xmath31 ) . the rgb bump can be also observed at @xmath32 and @xmath33 mag ; this broad feature can be explained by an extended sfh @xcite . as can be seen from figure [ f1 ] , the hst acs field of view covers only a small fraction of ic 1613 ( @xmath2 9% ) . the position of the field was chosen mainly as a balance between optimizing the number of stars but minimizing the effects of crowding at the photometric limit . however , the resulting position lies just inside of the half - light radius of ic 1613 , which is quite fortunate . ideally , we would like to characterize the global sfh of ic 1613 from the observed field . stellar population gradients are universal in dwarf galaxies and have been observed over a wide range of physical characteristics . in all cases , the gradients are in the sense that the mean age of the stellar population is younger toward the center of the galaxy ( see discussion in * ? ? ? * for further details ) . since the observed field is close to the half - light radius , we make a type of `` mean value '' assumption and assume that the sfh derived from the observed field is a relatively good representation of the global sfh for ic 1613 . the balance between outer fields ( with relatively more older stars ) and inner fields ( with relatively more younger stars ) should not be too different from the sfh of a field at the half - light radius . for the rest of the paper the discussion will be based on this assumption and the sfh history of the observed field and the global sfh of ic 1613 will be used interchangeably . this assumption is supported by the comparison of hst observations of ic 1613 from different fields presented in [ comp3 ] . note that our main result , the absence of a dominant star formation episode at early times , would only strengthen by the addition of more younger stars at smaller radii . the opposite possibility ( a relative surplus of older stars in the inner parts of the galaxy ) is constrained by the direct observation of the number of rgb stars in the central field from the earlier hst / wfpc2 observations ( * ? ? ? * see discussion in section [ comp3 ] ) . following the other lcid studies , we first use the iac method , consisting of iac - star @xcite , iac - pop @xcite , and minniac @xcite in order to solve for the sfh and amr . details of the methodology can be found in @xcite . specifically , the solution has been obtained using the basti @xcite stellar evolutionary libraries and the bolometric corrections of @xcite , and the uncertainties are the statistical uncertainties and do not included certain systematics ( e.g. , the use of a different stellar library ) . , of ic 1613 derived as described in the text . the sfh as a function of age @xmath34 and metallicity @xmath35 are shown projected on the @xmath36 and @xmath37 planes , respectively . dashed lines give the statistical error intervals . the age - metallicity relationship is the projection onto the look - back time metallicity plane . [ f3],width=325 ] figure [ f3 ] shows a plot of the sfh of ic 1613 , @xmath38 , as a function of both @xmath39 and @xmath40 . @xmath34 and @xmath35 are also shown in the @xmath41 and @xmath42 planes , respectively . the projection of @xmath38 on the age - metallicity plane shows the amr , including the metallicity dispersion . figure [ f3 ] shows the presence of star formation at all times , and the gradual build up of metallicity with time . note especially the relatively constant sfr over the first 6 gyr of the history of ic 1613 ( variations of @xmath2 30% or less in sfr ) . while the iac - pop solution suggests a slight ( factor of 2 ) decrease in average sfr about 8 gyr ago ( @xmath43 @xmath44 1 ) , there is no evidence for a dominant episode of star formation before the end of reionization ( @xmath2 12.8 gyr , or @xmath43 @xmath44 6 , * ? ? ? * ) or around the age of the peak sfr density of the universe ( @xmath2 10 gyr , or @xmath43 @xmath44 2 , * ? ? ? we emphasize that there is no evidence of a dominant episode of star formation in very early times ( lookback time @xmath12 11 gyr ) , despite our ability to resolve such a feature . ( upper panel ) , metallicity ( middle panel ) , and cumulative stellar mass fraction ( lower panel ) of ic 1613 . thin lines give the uncertainties . vertical dotted - dashed lines indicate the times corresponding to the 10@xmath45 , 50@xmath45 and 95@xmath45 percentiles of @xmath34 , i.e. , the times for which the cumulative fraction of mass converted into stars was 0.1 , 0.5 and 0.95 of the current total . the cumulative total mass of stars formed is @xmath46 @xmath47 m@xmath48 . the mean metallicity , @xmath49[m / h]@xmath50 , is @xmath51 . the average star formation rate , @xmath52 , is @xmath53 m@xmath48 yr@xmath3 . a redshift scale is given in the upper axis . [ f4],width=325 ] in figure [ f4 ] we show @xmath34 , the amr ( with their associated errors ) , and the cumulative stellar mass fraction of ic 1613 , as a function of time . three vertical dashed lines show the ages of the 10th , 50th , and 95th - percentile of the integral of @xmath34 . the main features of the sfh of ic 1613 are nearly continuous star formation and an accompanying smooth increase in the stellar metallicity . the sfr varies by at most a factor of two from its mean value over the entire lifetime of the galaxy . most importantly , there is no strong initial epoch of star formation , as is seen in the lcid dsph galaxies and transition ( dsph / dirr ) galaxies . this is reflected in the cumulative stellar mass fraction plot which is close to the diagonal line of a constant sfr . . the residuals are in units of poisson uncertainties.[f5],width=325 ] figure [ f5 ] shows the observed cmd , the best - fit cmd , and the corresponding residuals . residuals are given in units of poisson uncertainties , obtained as @xmath54 , where @xmath55 and @xmath56 are the number of stars in bin @xmath57 of an uniform grid defined on the observed and calculated cmds , respectively . because the solution is heavily determined by the main sequence stars ( i.e. , the rgb and hb stars are not used in the solution , see discussion in * ? ? ? * ) , the residuals along the ms are negligible . there are significant residuals at the faintest levels ( where the incompleteness is large ) and in the horizontal branch ( which is typically not well fit by models ) . , width=325 ] as for the other galaxies in the lcid sample , we have obtained the sfh of ic 1613 using the match @xcite and cole @xcite methods in addition to the iac method . these two methods use @xcite for the stellar evolution library for this comparison ( for a description of the main features of these methods and the particulars of their application to lcid galaxies , see * ? ? ? in this exercise , we have allowed all parameters ( e.g. , distance , reddening , etc . ) to be solved for optimally independently . that is , we are not trying to compare codes , but rather , trying to make an assessment of the systematic errors that may arise due to choices of different stellar libraries , codes , etc . there are three main sources of uncertainties in deriving sfhs ( see , e.g. , discussion in * ? ? ? * ) : the input observational uncertainties , the statistical uncertainties of the solution , and the systematic uncertainty of the limitations of our knowledge of stellar evolution . in a specific test of creating synthetic photometry from one stellar library and using a different stellar library to derive a sfh , @xcite showed that the uncertainties associated with our knowledge of stellar evolution dominated those of the numerical methodology or the observation uncertainties . they recommend always using more than one stellar evolution library when analyzing a real population to test for these effects . @xcite and @xcite have also shown that , for sufficiently deep photometry , the stellar evolution uncertainties , as approximated by differences between stellar evolution model and their resulting libraries , represent the dominant systematic uncertainty for derived sfhs . ( see also the discussion in * ? ? ? * concerning the effects of varying heavy element abundance patterns . ) thus , using different stellar libraries is integral to this comparison . the sfh derived using match features uncertainties calculated following the prescriptions in @xcite . in this case , random uncertainties were generated using a hybrid monte carlo ( hmc ) process @xcite , with implementation details as described by @xcite . the result of this markov chain monte carlo routine is a sample of 10,000 sfhs whose density is proportional to the probability density . ( that is , the density of samples is highest near the maximum likelihood point . ) upper and lower random error bars for any given value ( e.g. , cumulative stellar mass fraction at a particular point in time ) are calculated by identifying the boundaries of the highest - density region containing 68% of the samples , with the value 68% adopted as it is the percentage of a normal distribution falling between the @xmath11 1 @xmath58 bounds . figure [ f6 ] shows the cumulative stellar mass fraction of ic 1613 as obtained with the three different methods . the agreement between the three methods is very good and all three are best described as consistent with nearly constant star formation . the one small difference between the sfhs is the offset at intermediate ages between the solution based on the basti models and the two based on the padua models . a comparison using the same code with two different stellar libraries shows that about half of this difference is due to the differences between the stellar libraries ( which have differences in physical inputs and differ in the treatment of core convective overshoot during the central h burning stage ) and the other half is likely attributable to how different sections of the cmd are binned and weighted . this comparison strongly supports the conclusion that the star formation in ic 1613 has been relatively constant for the lifetime of the galaxy . the similar behavior of the three solutions allows us to be confident that our conclusions are independent of the stellar evolution libraries and sfh solution method . most importantly , an early dominant star formation event is clearly ruled out . note also that ic 1613 has clearly formed the vast majority of its stars after the epoch of reionization which occurred in the first gyr of the history of the universe ( cf . * ) . , width=325 ] , wfpc2-center corresponds to the 9 orbit observation presented in @xcite , wfpc2-outer corresponds to the 24 orbit observation presented in @xcite , and acs - outer corresponds to new acs observations presented here . note that the deeper acs observations have allowed significantly smaller uncertainties at the earliest times . all three sfhs are consistent with a constant sfr over the lifetime of the galaxy , but only the deeper acs observations clearly rule out a dominant episode of early star formation . [ f8],width=325 ] relatively deep imaging of ic 1613 has been obtained with the hst three times : a central field obtained with the wfpc2 camera ( 9 orbits ) with a depth in i - band of @xmath2 26.5 mag . presented by @xcite , a deep ( 24 orbits ) outer field ( radius @xmath44 1.4 kpc ) obtained with the wfpc2 camera with a depth in i - band of @xmath2 27.3 presented by @xcite , and our new imaging of an outer field ( radius @xmath44 1.2 kpc ) obtained with the acs camera with a depth in i - band of @xmath2 28.7 . figure [ f7 ] presents a comparison of the photometry for these three observations . in making figure [ f7 ] , identical quality cuts were made on the data , so that they are directly comparable ( although the wfpc observations were obtained with an f555w filter and the present acs observations were obtained with an f475w filter ) . figure [ f8 ] shows the cumulative stellar mass fraction of ic 1613 as obtained from the three different observations . the sfhs for the wfpc2 fields are newly derived from the archival data @xcite . all three sfhs were obtained using the match code @xcite and the padua @xcite stellar evolution library . the errorbars were derived following the prescription outlined in @xcite . this comparison to sfhs derived from shallower data needs to be approached carefully . clearly , the larger uncertainties inherent in the shallower observations are reflected by larger errorbars on the individual points . however , since the dominant component of the uncertainties are systematic , and since these can only be estimated and not calculated directly , these should be interpreted as `` best estimates '' . predicting early sfhs from photometry which does not reach the oldest main sequence turn - offs is vulnerable to systematic biases . there is no substitute for the deep observations with the resulting strong constraints on the old age sfh . nonetheless , this comparison allows us to test our assumption of the representative nature of the acs observation for the global sfh of ic 1613 . the main impression from figure [ f8 ] is that the sfhs for all three observations are best described as consistent with nearly constant star formation , but only the deeper acs observations clearly rule out a dominant episode of early star formation . thus , figure [ f8 ] does provide strong support for the assumption that the sfh derived from the acs observations can be assumed to be representative of the global sfh for ic 1613 . given the caveats from the previous paragraph we will refrain from any further discussion of this comparison . in this section , we will compare the sfh of ic 1613 with that of leo a ( the other dirr galaxy of the lcid sample , * ? ? ? * ) , phoenix and ( the two dsph / dirr transition galaxies , * ? ? ? * ; * ? ? ? * ) and with those of cetus and tucana ( the two dsphs , * ? ? ? * ; * ? ? ? * ) . note that the lcid sample covers a range in dynamical masses . from @xcite the dynamical masses of ic 1613 , leo a , lgs-3 , cetus , and tucana are : 1.1 @xmath11 0.2 @xmath18 10@xmath59 @xmath8 , 1.5 @xmath60 @xmath18 10@xmath61 @xmath8 , 2.7 @xmath62 @xmath18 10@xmath61 @xmath8 , 4.0 @xmath63 @xmath18 10@xmath61 @xmath8 , and 7.1 @xmath11 1.2 @xmath18 10@xmath61 @xmath8 , respectively . for phoenix , we derive a value of @xmath22.7 @xmath18 10@xmath61 @xmath8 , assuming a velocity dispersion of @xmath28 km s@xmath3 @xcite and a half - light radius of 454 pc @xcite . , width=325 ] figure [ f9 ] compares the sfh and the amr of ic 1613 with the other lcid galaxies . in the top panel of figure [ f9 ] , the sfrs ( @xmath64 ) have been normalized such that the area under each curve is 1 ( i.e. , the integral of the sfr over the lifetime of the galaxy is 1 ) . the differences between the sfhs as a function of morphological type is quite remarkable . both dirr galaxies lack the dominant initial episode of star formation seen in both the dsph and transition type galaxies . although the sfhs are the main topic of this study , the amrs are also solved for in deriving the sfhs and are included here for completeness . both dirr galaxies also show relatively little metal enrichment during the first @xmath2 6 gyr , with gradual enrichment following . note that @xcite highlighted the lack of early chemical enrichment in the transition galaxy lgs-3 and hypothesized that the differences between the transition galaxies and dsphs ( which show early chemical enrichment ) might be understood if the dsphs were initially more massive systems than the transition galaxies . under this hypothesis , the delayed metal enrichment in the transition galaxies is due to higher losses of metals during times of higher sfrs for the lower mass transition galaxies . in this light , it is interesting that ic 1613 ( and also leo a ) show delayed metal enrichment , but with no initial high sfr . in this regard , further comparisons of the sfh and amrs of transition galaxies and dirr galaxies would be be great interest . , width=325 ] before comparing to theoretical models , we introduce one last comparative figure . in figure [ f10 ] , we show a comparison between the sfhs of the six lcid galaxies as cumulative stellar mass fractions . in the upper panel , only the statistical uncertainties are shown . in the lower panel , the systematic uncertainties are included following the methodology of @xcite . because all of the lcid galaxies have been observed to comparable depth , systematics in the models should have similar impacts to all of the galaxies . thus , it is likely appropriate to make comparisons using the upper panel . nonetheless , the lower panel shows the larger uncertainties encountered when trying to account for systematics and , even with the larger uncertainties , the six galaxies are shown to each have distinctive features in their sfhs . portraying the sfhs as cumulative stellar mass fractions ( as opposed to sfr as a function of time ) is the optimal way to compare observations to theoretical models for several reasons . variations in observed sfrs can be strongly affected by time binning and the changing time resolution as a function of lookback time . often , it is possible to have very different impressions of a single sfh simply by changing the time binning . it is possible to match the observational time binning by reducing the resolution in the models , but using the cumulative stellar mass fraction as the diagnostic avoids this problem altogether . it is also possible to compare galaxies at any arbitrary value of the cumulative stellar mass fraction , as opposed to choosing particular values to focus on . in the comparisons that follow , we will use the cumulative stellar mass fraction as the sole diagnostic . note that there is one obvious failing of the cumulative stellar mass fraction as the sole diagnostic , and that is the lack of information about the absolute masses of the systems . in the following comparisons , we will provide information about the masses of both the observed and modeled systems . recently , @xcite , @xcite , @xcite , and @xcite have all highlighted the difficulty of producing dwarf galaxies in simulations with properties comparable to those observed in the current universe . together , the introductions to their papers give a comprehensive overview of the problems with modeling dwarf galaxy evolution . to summarize , there are two major problems . the first problem is the observed abundance of low - mass galaxies . the observed slope of the low - mass galaxy luminosity function is shallow relative to the slope of the halo mass function and this difference seems to be a result of an extreme inefficiency of galaxy - scale star formation over cosmic times . the well known problem of `` the missing satellites '' @xcite is one manifestation of this problem at very low luminosities / halo masses . the second problem relates to _ when _ stars are formed in galaxies . a natural assumption is that the time - scale for global star formation is related to the time - scale of baryonic accretion onto galaxies . however , low - mass halos assemble almost all of their ( dark matter ) mass at high redshift ( e.g. , * ? ? ? * ) , while essentially all field dwarf galaxies show star formation continuing to the present day ( e.g. , * ? ? ? ic 1613 and leo a are extreme examples of this , with essentially a constant sf rate across cosmic time for ic 1613 and delayed star formation in the case of leo a. in order to suppress the abundance of low - mass galaxies , most theoretical models impose strong feedback in small halos ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? two processes can dramatically affect the formation and evolution of dwarf - sized halos : heating from the ultraviolet radiation arising from cosmic reionization and feedback from internal supernovae . both processes are , in principle , capable of completely halting the star formation in a dwarf halo and even fully removing all of the galaxy s gas . employing these feedback mechanisms while tying star formation to the collection of baryons has the effect of predicting that essentially all star formation in low - mass halos happens at early times . both semi - analytic models and hydrodynamical models fail to satisfactorily reproduce the evolution of low mass galaxies in the sense that stellar mass is over - produced per dark matter halo mass . stars are produced too quickly at early times , resulting in stellar mass fractions that are too high by an order of magnitude . model galaxies usually do not have the high gas mass fractions commonly observed in present day dwarfs ( e.g. , * ? ? ? * ) , and , as a result , star formation falls off too fast and the colors of simulated dwarf galaxies are too red at @xmath43 @xmath7 0 . this is a manifestation of the `` over - cooling '' problem which is a challenge for all modeling efforts , but which is exaggerated at low masses . the missing satellites problem is largest at the lowest masses , and ic 1613 is massive enough that it could be expected to emerge from reionization with its gas intact . thus , ic 1613 does not provide as strong a test of models for this problem as the other , less massive , galaxies in the lcid sample . however , our deep hst observations have produced sfhs for ic 1613 and leo a with small uncertainties even at earliest times and have completely ruled out the possibility of an early , dominant phase of star formation . _ this represents a real challenge for any model where star formation follows mass assembly . _ @xmath7 0 of @xmath2 10@xmath65 m@xmath66 , which is comparable to the estimated halo mass for ic 1613 . the difference in the early sfhs between the models and ic 1613 is very clear . , width=325 ] @xcite have highlighted the shortcomings of models for dwarf galaxies where star formation follows directly from mass assembly . they presented a set of high - resolution hydrodynamical simulations of the formation and evolution of isolated dwarf galaxies including the most relevant physical effects , namely metal - dependent cooling , star formation , feedback from type ii and ia sne , and uv background radiation . their results are very useful for a direct comparison with observations . they identify and study six halos with present day dark matter halo masses of @xmath44 10@xmath65 m@xmath48 with stellar masses ranging between @xmath67 m@xmath48 and @xmath68 m@xmath48 . for comparison , ic 1613 has an estimated stellar mass of @xmath2 10@xmath59 m@xmath48 @xcite , and the observed maximum rotation velocity of 25 km s@xmath3 @xcite . if we associate this peak rotation velocity with the maximum circular velocity of ic 1613 s host dark matter halo , this corresponds to a virial mass . ] of @xmath69 . this is likely to be a lower limit , however , as the peak of the circular velocity curve for the dark matter halo may be attained at radii beyond the extent of the stellar or gaseous tracers . abundance matching models find that galaxies with stellar masses of @xmath70 should be hosted by halos with virial masses of @xmath71 ( e.g. , * ? ? ? * ) . given these estimates , the models of sawala et al . provide an excellent sample to compare to the derived sfh of ic 1613 . as figure [ f11 ] demonstrates clearly , the early evolution of the model galaxies all depart significantly from that of ic 1613 . in essence , the models all have sfrs which depend directly on the gas content of the galaxies . as highlighted by @xcite , since , in the current paradigm , most of the mass of a low mass galaxy is in place well before @xmath43 @xmath7 1 ( e.g. , * ? ? ? * ) , then any prescription in which star formation follows gas content is going to build up most of the stellar mass before @xmath43 @xmath7 1 . this discrepancy between models and observations , while demonstrated here for ic 1613 , is beginning to appear to be the norm . although the numbers are very limited , so far , all gas - rich dwarfs for which there exist sufficiently deep hst observations ( i.e. , leo a , leo t , smc : * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) show no evidence for a dominant early episode of star formation . there have been various attempts to improve models to better accommodate the inefficient star formation in dwarf galaxies . @xcite used sph + n - body simulations and showed that supernova feedback could disrupt enough gas to temporarily quench star formation . episodic star formation follows from the cycling of gas into a hot halo with subsequent cooling and infall . @xcite account for the prolonged star formation observed in dwarfs by adding a stochastic variation in the density threshold of the star formation law . essentially , this simply reduces the efficiency of star formation by hand , but the result is a significantly improved match to the observed sfhs of dwarfs . recently , @xcite have demonstrated that the thermal feedback from early star formation can effectively decouple star formation from mass assembly , thus producing more realistic sfhs . @xcite and @xcite have proposed a different physical approach by investigating the conversion of atomic to molecular gas and its affect on the efficiency of star formation . this is well motivated by the observational work of @xcite and @xcite , which clearly demonstrate that star formation follows the molecular gas content and not the total gas content . in the specific case of dwarf galaxies , @xcite show that models which incorporate a star formation prescription regulated by the local abundance of molecular hydrogen lead to the less efficient star formation that is desired . unfortunately , their first attempts are not completely successful as they state , `` like most cosmological galaxy formation simulations to date , our simulated galaxies suffer from the so - called baryonic overcooling problem , resulting in unrealistically high central densities ( and hence strongly peaked circular velocity curves ) and stellar mass fractions in our high - mass halos that are too large compared to observations . '' however , this appears to be a very promising avenue for future exploration ( see also * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite used a semi - analytic galaxy formation model to investigate the properties of the satellites of milky way - like galaxies and were able to match the star formation histories of several dwarf satellites . the extended sfhs of these satellites were a result of a gas density threshold for star formation . galaxies could have large reservoirs of gas lying just below the threshold , leading to inefficient star formation ( requiring a minor accretion or interaction event ) . @xmath7 0 of @xmath2 4 @xmath18 10@xmath65 m@xmath66 and @xmath2 1 @xmath18 10@xmath65 m@xmath66 , respectively , comparable to the estimated halo mass for ic 1613 . `` bashful '' and `` doc '' have nearly constant star formation and lack a dominant early episode of star formation as is observed in ic 1613 . the two model galaxies with a delayed onset of star formation are `` dopey '' and `` grumpy '' with smaller halo masses at @xmath43 @xmath7 0 of @xmath2 3 @xmath18 10@xmath72 m@xmath66 and @xmath2 2 @xmath18 10@xmath72 m@xmath66 , respectively , ( which are considerably less massive than ic 1613 , but might be more directly comparable to leo a ) . , width=325 ] @xcite have produced new simulations including supernova feedback , a star formation recipe based on a high gas density threshold , metal - dependent radiative cooling , turbulent diffusion of metals and thermal energy , and uv background radiation . they reproduce the observed stellar mass and cold gas content , the sfhs , and the metallicities typical of field dwarfs in the local volume . in figure [ f12 ] , we reproduce the cumulative stellar mass fraction for four of the model galaxies in @xcite . two of the galaxies ( `` bashful '' and `` doc '' ) have virial masses which are comparable to ic 1613 . these two models show nearly constant star formation with no dominant early episode of star formation , as observed in ic 1613 . in detail , the sfhs of these two models start later and rise faster during intermediate ages . the slower start of the models as compared to that of the measured sfhs may be attributed to differences in how time is defined in the two methods . that is , the simulations start at a fixed , given time , whereas the observationally derived ages come from stellar evolution libraries , and therefore depend on their underlying physics ( e.g. , nuclear reaction rates ) . the two lower mass galaxies ( `` dopey '' and `` grumpy '' ) show no early star formation at all , and only start forming stars at intermediate ages . these two models correspond to halo masses significantly less than that of ic 1613 , but might be more comparable to that of leo a ( cf . figure [ f10 ] ) . note , however , that the stellar masses are even more discrepant . that is , dopey and grumpy have stellar masses of 10@xmath73 and 5 @xmath18 10@xmath73 m@xmath66 compared to 10@xmath59 and 6 @xmath18 10@xmath74 m@xmath66 for ic 1613 and leo a. naively , the late onset of star formation in the dopey and grumpy models would appear to be in contradiction to observations in which all dwarf galaxies observed to date show evidence for an old ( @xmath12 10 gyr ) population , particularly as evidenced by the presence of rr lyrae stars . for example , rr lyrae stars have been observed in leo a @xcite . however , @xcite hypothesize that such systems may be the very metal poor systems necessary for the production of extremely metal deficient galaxies such as i zw 18 ( see , e.g. , discussion and references in * ? ? ? one of the successes of recent modeling of dwarfs has been the ability of simulations to solve the cusp / core problem for dwarf galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the dark matter halos of observed dwarf galaxies show nearly constant density cores , while steep central dark - matter profiles are expected from cdm models ( e.g. , * ? ? ? the resolution to this conundrum lies in strong outflows from supernovae to remove low - angular - momentum gas from the centers of dwarf galaxies . this solves two problems as bulgeless dwarf galaxies are formed with shallow central dark - matter profiles , and a large fraction of the original baryons are lost from the galaxy , resulting in the lower baryon fractions ( relative to the universal fraction ) which are observed ( e.g. , * ? ? ? what is not clear is whether the current ideas concerning how to produce inefficient star formation also lead to cored dark matter halos . clearly the strong outflows need to occur when the mass assembly is taking place , which is early on . if the vast majority of the star formation is delayed until later times , after the mass assembly is virtually finished , then the central outflow solution to the cusp / core problem may not arise self - consistently . a related point is the energetic requirements for creating bulgeless galaxies and forming cores in state - of - the - art hydrodynamic simulations . simulations that are currently successful require ( 1 ) highly efficient coupling between the energy released in a supernova explosion and the ism of a galaxy values typically adopted are 0.41 @xcite and ( 2 ) an artificial prevention of gas cooling for approximately @xmath75 years in a region surrounding a supernova explosion , meant to capture processes that may occur at sub - grid scales and can not be directly included in the simulation @xcite . the success of simulations in producing bulgeless , cored dwarf galaxies with relatively low - level , extended star formation can be therefore seen as predictions for the effective coupling of supernova explosions with the ism and for the efficacy of additional feedback processes ( such as stellar winds ) at heating the ism on small scales . these predictions of very efficient coupling may be confirmed or refuted with future generations of simulations that model the relevant processes directly . finally , although we have emphasized the difficulty of producing a nearly constant sfr as seen in ic 1613 , observations of dwarf galaxies in the local volume point toward a large diversity of sfhs . determining the critical parameters which drive that diversity is paramount . how can it be that dwarfs with halos of similar mass , which presumably have similar accretion histories and early formation times , have vastly different sfhs ? is environment and interaction history the main driver of the diversity ( see , e.g. , * ? ? ? * ; * ? ? ? * ) ? from an observational point of view , strong constraints will depend on accurate determinations of the early sfhs of a larger number of galaxies within and beyond the local group from both deep photometry as presented here and spectroscopic studies of individual stars ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? clearly the sample of one presented here , and the very small number of comparably observed galaxies presents a great limitation to our knowledge of the evolution of dwarf galaxies . we have presented the sfh of the dirr galaxy ic 1613 , based on deep hst photometry obtained with the acs . * the sfh with relatively small uncertainties has been obtained for the entire lifetime of the galaxy . the solution shows that the sfh of ic 1613 is consistent with a constant sfr over its entire lifetime . * most or all the star formation was produced in ic 1613 after the reionization epoch , assumed to occur @xmath76 gyr ago . there is no evidence of an early dominant episode of star formation in ic 1613 . * a comparison of the derived sfh of ic 1613 with the models of @xcite reinforce their observation that models where star formation follows mass assembly form too many stars too early . this well known aspect of the so called `` over - cooling problem '' appears to be universal , but now , with the deep hst photometry of ic 1613 , we see that the problem reaches back to the very earliest times in the evolution of galaxies . * there are proposed solutions to the over - cooling problem for dwarf galaxies . the solutions discussed in this paper rely on an efficient coupling of supernova feedback to the ism . the predictions of very efficient coupling may be confirmed or refuted with future generations of simulations that model the relevant processes directly and through future observations of dwarf galaxy sfhs with similar quality to the lcid studies . we would like to thank till sawala for sharing the numerical results of his modeling , and greg stinson , piero madau , alyson brooks , and the referee for helpful comments . support for this work was provided by nasa through grant go-10515 from the space telescope science institute , which is operated by aura , inc . , under nasa contract nas5 - 26555 . support for drw is provided by nasa through hubble fellowship grant hst - hf-51331.01 awarded by the space telescope science institute . the computer network at iac operated under the condor software license has been used . authors sh , aa , cg , and mm are funded by the iac ( grant p3/94 ) and , with sc , ps , and eb , by the science and technology ministry of the kingdom of spain ( grant aya2007 - 3e3507 ) and economy and competitiveness ministry of the kingdom of spain ( grant aya2010 - 16717 ) . this research has made use of nasa s astrophysics data system bibliographic services and the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration .

many vibrating mechanical systems experience undesirable impacts that cause wear or sub - optimal performance . occasional impacts may be permissible if they result from running components of the system at high speeds for greater efficiency , and some impacts are unavoidable such as those due to random or rare events . in these cases it is important to have a clear understanding of the dynamical behaviour that impacts may induce . impacting dynamics is often complicated or chaotic because impacts are highly nonlinear phenomena @xcite . for instance tubes in heat exchangers vibrate at high fluid velocities and impact against baffles in place to guide the fluid flow . simple mathematical models of heat exchangers reveal that chaotic dynamics may be created at the onset of recurring impacts @xcite . rotating cutters spun at high speeds experience repeated contact loss with the material being cut . the resulting impacts between the cutter and the material may similarly induce chaotic dynamics @xcite . in contrast , some mechanical systems use impacts to achieve their function . atomic force microscopes measure surface topography and the chemical properties of a sample on the nanoscale by gently hitting the sample with a vibrating cantilever . in this context it is important to understand the complex impacting dynamics so that the inverse problem of describing the sample can be performed effectively @xcite . impacts can often be modelled accurately by carefully describing the deformations that components of the system undergo during impacts @xcite . however , for the purposes of understanding vibro - impacting dynamics , such a modelling approach is too cumbersome and a low degree - of - freedom ode model can be more useful . despite the low - dimensionality , such models have been shown to quantitatively match the experimental data of a variety of impacting systems . examples include a cam - follower system involving occasional detachments between the cam and the follower @xcite , a pendulum experiencing near - instantaneous impacts with a solid wall @xcite , and compliant impacts of a steel block with an elastic beam @xcite . ( 15,3.8 ) ( 0,0 ) , where @xmath0 is a system parameter . [ fig : grazbifschem ] , title="fig:",height=136 ] ( 5.1,0 ) , where @xmath0 is a system parameter . [ fig : grazbifschem ] , title="fig:",height=136 ] ( 10.2,0 ) , where @xmath0 is a system parameter . [ fig : grazbifschem ] , title="fig:",height=136 ] ( .5,3.6)*a * ( 5.6,3.6)*b * ( 10.7,3.6)*c * ( 2,3.6)@xmath1 ( 7.1,3.6)@xmath2 ( 12.2,3.6)@xmath3 ( 4.48,.2)@xmath4 ( 9.58,.2)@xmath4 ( 14.68,.2)@xmath4 using a low degree - of - freedom ode model , the evolution of the system between impacts is tracked in phase space . periodic behaviour without impacts corresponds to a periodic orbit in phase space that does not reach the switching manifold , @xmath4 , fig . [ fig : grazbifschem]-a . physically , @xmath4 corresponds to locations or instances where mechanical components come into contact , or lose contact . as parameters vary , the system may transition from an impact - free regime to the repeated ( though not necessarily regular ) occurrence of impacts . in phase space , the transition occurs when the periodic orbit of the ode model attains an intersection with @xmath4 , fig . [ fig : grazbifschem]-b . this is known as a _ grazing bifurcation_. in this paper we study grazing bifurcations of the three - dimensional piecewise - smooth ode system , @xmath5 where @xmath6 and @xmath7 are smooth functions , @xmath0 is a parameter , and the coordinates @xmath8 are chosen so that @xmath4 is simply the coordinate plane @xmath9 . we assume that for @xmath1 , there exists an attracting periodic orbit describing non - impacting dynamics located entirely in the region @xmath10 , and that the periodic orbit grazes @xmath4 at the origin when @xmath2 . in the context of vibro - impacting systems , @xmath10 corresponds to not - in - contact dynamics governed by @xmath6 , and @xmath11 corresponds to in - contact dynamics governed by @xmath7 . impacts may instead be modelled as instantaneous events with energy loss and velocity reversal , in which case a map is usually defined on the switching manifold to describe the action of an impact @xcite . theoretical studies of piecewise - smooth and hybrid dynamical systems have led to a useful classification of grazing bifurcations @xcite . this paper concerns regular grazing bifurcations . the grazing bifurcation of ( [ eq : f ] ) at @xmath2 is said to be _ regular _ if @xmath12 for all @xmath13 in a neighbourhood of @xmath14 . this condition arises naturally in mechanical systems with compliant impacts , and implies that @xmath4 is neither attracting nor repelling at any point . as indicated in fig . [ fig : grazbifschem]-c , the steady - state dynamics of ( [ eq : f ] ) for @xmath3 is often complicated . for this reason it is valuable to study the oscillatory dynamics using a return map based on the points on a poincar section . a normal form for such a map for regular grazing in @xmath15 is the nordmark map @xmath16 where @xmath17 with @xmath18 , and @xmath19 , as determined by the sign of certain coefficients , see [ sec : nordmark ] . the coordinates @xmath20 represent points on a poincar section of ( [ eq : f ] ) , with @xmath21 the bifurcation parameter and the grazing bifurcation occurring at @xmath22 . the nordmark map ( [ eq : n ] ) , applicable also to models with instantaneous impacts @xcite , includes only the leading order terms of the return map , so is valid only for dynamics close to the grazing bifurcation @xcite . each iteration of ( [ eq : n ] ) corresponds to one oscillation of ( [ eq : f ] ) near the grazing periodic orbit . the utility of ( [ eq : n ] ) lies in the fact that the nature of the dynamics can be identified by the location of the corresponding points in the @xmath23-plane . specifically , ( [ eq : n ] ) is formulated so that if @xmath24 then the oscillation lies entirely in @xmath10 , and if @xmath25 then the oscillation enters @xmath11 . the non - impacting , attracting periodic orbit shown in fig . [ fig : grazbifschem]-a corresponds to the fixed point of @xmath26 , @xmath27 with @xmath28 and @xmath29 . more generally , a periodic orbit of ( [ eq : f ] ) appears as a finite set of points in the @xmath23 plane , as in fig . [ fig : detbifdiag ] which shows a typical bifurcation diagram of ( [ eq : n ] ) . this figure shows a period - incrementing cascade , corresponding to several different periodic orbits , and apparently chaotic dynamics , as indicated by a cloud of points . for alternate values of @xmath30 and @xmath31 , the fixed point @xmath32 may bifurcate directly to chaos @xcite . the square - root term in ( [ eq : n ] ) is an artifact of the tangency between the periodic orbit and the switching manifold of ( [ eq : f ] ) at the grazing bifurcation , and is responsible for the distinctive shape of the bifurcation diagram near @xmath22 . for some vibro - impacting systems it is more appropriate for the return map to be piecewise - linear and either continuous @xcite or discontinuous @xcite . such maps predict fundamentally different bifurcation structures to those of ( [ eq : n ] ) . ( 12,6 ) ( 0,0 ) ) with @xmath33 , @xmath34 , and @xmath35 . these parameter values correspond to the vibro - impacting system described in [ sec : oscillator ] , with @xmath36 . the fixed point of ( [ eq : n ] ) for @xmath29 , given by ( [ eq : xlyl ] ) , corresponds to an attracting , non - impacting periodic orbit of period @xmath37 . the map ( [ eq : n ] ) has attracting @xmath38 and @xmath39-cycles for certain values of @xmath40 as shown , as well as an attracting @xmath41-cycle in the approximate range , @xmath42 . these correspond to periodic orbits of period approximately equal to @xmath43 , for @xmath44 , that experience one impact per period . there also appears to be a chaotic attractor for the approximate range , @xmath45 . [ fig : detbifdiag ] , title="fig:",height=226 ] ( 6.2,0)@xmath46 ( 0,3.8)@xmath47 in order to properly explain complicated vibro - impacting dynamics , the effects of randomness and uncertainties needs to be taken into account . mechanical systems are subject to background vibrations and other sources of noise . experimentally measured parameters involve error , and some physical features are left unmodelled . for instance , one - degree - of - freedom models do not capture high frequency modes that are often excited by impacts @xcite . to quantitatively describe stochastic impacting dynamics , stochastic averaging methods have proved useful for vibro - impacting systems that experience a wide range of impact velocities @xcite . if only low - velocity impacts are relevant , then it is useful to study ( [ eq : n ] ) . in his phd thesis @xcite , griffin studied ( [ eq : n ] ) in the presence of additive white noise . he found that noise blurs bifurcation diagrams and washes out high - period solutions in the same manner as for smooth maps , such as the logistic map @xcite . recently it was shown that white noise added to ( [ eq : f ] ) translates to additive white noise in ( [ eq : n ] ) , @xcite . such a noise formulation may be sensible for vibro - impacting systems for which a forcing term or external fluctuations represent a significant source of uncertainty . however , impact events themselves constitute a substantial source of randomness . the purpose of this paper is to construct and analyse stochastic versions of ( [ eq : n ] ) for which randomness stems purely from impact events . we consider three different types of impact noise for ( [ eq : f ] ) using an ornstein - uhlenbeck process with stationary density @xmath48 $ ] , where @xmath49 represents the noise amplitude and @xmath50 is the correlation time . we first consider uncertainties in @xmath4 , then uncertainties in @xmath7 , and lastly uncertainties in @xmath7 in the white noise ( zero correlation time ) limit . here let us indicate the forms of the stochastic maps that we obtain . coloured noise in @xmath4 leads to random perturbations in both the map and the switching condition as @xmath51 where @xmath52 are gaussian random variables , and @xmath53 is a constant . for coloured noise in the impacting dynamics we obtain @xmath54 for a particular nonlinear function @xmath55 . for white noise ( @xmath56 ) in the impacting dynamics the map takes the form @xmath57 where @xmath58 and @xmath59 are random variables , and @xmath60 and @xmath61 are nonlinear functions . notice that @xmath62 is stochastic for @xmath24 , whereas @xmath63 and @xmath64 are not . this is because noise in @xmath4 generates anomalous crossings of @xmath65 . consequently @xmath62 exhibits stochastic dynamics for @xmath29 , while @xmath63 and @xmath64 do not . @xmath63 and @xmath64 involve noise terms proportional to @xmath66 , and for this reason exhibit increasing variability for larger values of @xmath40 . to obtain a more detailed comparison of @xmath62 , @xmath63 and @xmath64 , we first carefully derive ( [ eq : n ] ) in [ sec : nordmark ] . in [ sec : oscillator ] we introduce a prototypical compliant vibro - impacting system to illustrate our results . in [ sec : randomness ] we add randomness and derive ( [ eq : na2])-([eq : nc2 ] ) . since ( [ eq : na2])-([eq : nc2 ] ) involve fundamentally different noise terms , they exhibit different sensitivities to the noise amplitude @xmath67 . therefore we use different values of @xmath67 in each of the different models , in order to make appropriate comparisons . for each @xmath68 ( @xmath69 ) we write @xmath70 and identify the appropriate value @xmath71 , so that @xmath62 , @xmath63 and @xmath64 display roughly the same dynamics for the vibro - impacting system with the parameters of fig . [ fig : detbifdiag ] and @xmath72 ( chosen for illustration ) and @xmath73 . this enables us to quantitatively compare @xmath62 , @xmath63 and @xmath64 in [ sec : dynamics ] . conclusions are presented in [ sec : conc ] . ( 15,7.5 ) ( 0,0 ) ) for the three - dimensional piecewise - smooth system ( [ eq : f ] ) . the solid curve represents an orbit of ( [ eq : f ] ) . the dashed curves show virtual extensions of this orbit into @xmath11 as governed by @xmath6 . [ fig : grazschem3d ] , title="fig:",height=283 ] ( 14.5,3.25)@xmath74 ( 13.65,5.75)@xmath75 ( 9.7,7.1)@xmath76 ( 1.2,5.8)@xmath77 ( 12.1,1.6)@xmath78 ( 11.3,6.2)@xmath4 ( 1.78,4.92)@xmath79 ( 11.94,3.7)@xmath80 ( 10.7,5.84)@xmath81 ( 8.54,1.12)@xmath82 ( 10.7,2.64)@xmath83 ( 3.02,4.48)@xmath84 ( 11.2,3.5)@xmath85 in this section we derive a return map for the generic deterministic system ( [ eq : f ] ) valid near the grazing bifurcation @xmath2 and provide an explicit coordinate change that transforms the map to ( [ eq : n ] ) . such a derivation is given in @xcite . we provide an explicit construction of ( [ eq : n ] ) here in order to provide a basis for deriving the stochastic maps in [ sec : randomness ] . throughout this section we write @xmath86 for points of ( [ eq : f ] ) in @xmath15 . as discussed in [ sec : intro ] , we assume that with @xmath2 the system ( [ eq : f ] ) has a periodic orbit that intersects the origin , but is otherwise contained in @xmath10 . this implies that @xmath87 for @xmath6 at the origin with @xmath2 ( i.e. @xmath88 ) . for simplicity , we assume that we further have @xmath89 for all @xmath90 in a neighbourhood of @xmath14 , which can usually be imposed by an appropriate coordinate change . this assumption is particularly useful in the case that randomness is present in the switching condition , see [ sub : s1 ] . the key to deriving the nordmark map ( [ eq : n ] ) is selecting a convenient poincar section and constructing a _ discontinuity map _ that accounts for the difference between @xmath6 and @xmath7 , that is , the difference between impacting and non - impacting dynamics . we let @xmath77 represent a generic poincar section of ( [ eq : f ] ) that lies in @xmath10 and intersects the grazing periodic orbit transversally , and let @xmath78 be the coordinate plane @xmath91 , see fig . [ fig : grazschem3d ] . given @xmath92 , we define @xmath93 as the next intersection of the forward orbit governed by @xmath6 ( i.e. ignoring the switching condition at @xmath9 ) with @xmath78 . if @xmath11 for the point @xmath80 , then @xmath80 does not represent the true intersection of the orbit of ( [ eq : f ] ) with @xmath77 . nevertheless , we study the return map on @xmath78 ( @xmath94 ) rather than the return map on @xmath77 ( @xmath95 ) , because the two maps are conjugate and the map on @xmath78 has a simpler form than the map on @xmath77 . to derive the return map on @xmath78 , we must consider three additional points , @xmath81 , @xmath82 , and @xmath83 , see fig . [ fig : grazschem3d ] . the points @xmath81 and @xmath82 correspond to the entry and exit locations of the orbit with the impacting region , @xmath11 . then @xmath80 is obtained by travelling forward from the entry point @xmath81 to @xmath78 using @xmath6 , whilst @xmath83 is obtained by travelling backward from the exit point @xmath82 to @xmath78 using @xmath6 . the discontinuity map is defined as @xmath96 . if @xmath97 for the point @xmath80 , then @xmath98 . in order to derive an explicit expression for the discontinuity map , we consider the three steps , @xmath99 , @xmath100 , and @xmath101 , individually , and expand @xmath6 and @xmath7 about @xmath2 and the origin by writing @xmath102 where @xmath103 , and @xmath104 . in ( [ eq : alphabetagamma ] ) , and throughout this section , @xmath105 is short - hand big - o notation for @xmath106 . notice that @xmath107 is assumed to be of the same order as @xmath75 , @xmath76 , and @xmath108 , which is appropriate in view of the tangency between the grazing periodic orbit and @xmath4 . we assume @xmath109 such that the tangency is quadratic for both @xmath6 and @xmath7 , and has the orientation depicted in fig . [ fig : grazschem3d ] . from series expansions of the orbits governed by @xmath6 and @xmath7 , see @xcite , we obtain the following formulas for the three steps in the discontinuity map in the case @xmath110 ( writing @xmath111 ) . for @xmath99 , @xmath112 for @xmath100 , @xmath113 and for @xmath101 , @xmath114 by combining ( [ eq : v2w2])-([eq : u4w4 ] ) we get @xmath115 in terms of @xmath116 , @xmath117 where @xmath118 the discontinuity map is then @xmath119 to complete the map ( [ eq : n ] ) , which represents @xmath94 as shown in fig . [ fig : grazschem3d ] , we must combine @xmath120 with the global return map @xmath121 . @xmath122 depends on global properties of @xmath6 and is smooth , so we can write @xmath123 for some @xmath124 where each @xmath125 . then the desired return map on @xmath78 is the composition @xmath126 . finally we apply a coordinate change to convert the map to the normal form ( [ eq : n ] ) that involves only three parameters , @xmath30 , @xmath31 and @xmath127 . under @xmath128 and with higher order terms omitted , @xmath126 transforms to ( [ eq : n ] ) with @xmath129 since the nordmark map keeps only leading order terms for @xmath90 near @xmath14 , terms that are linear in @xmath47 are omitted since they are dominated by @xmath130 , refer to @xcite for a further discussion . in ( [ eq : coordinatechange ] ) we require @xmath131 and @xmath132 . these represent non - degeneracy conditions for the grazing bifurcation . ( 15,5 ) ( 0,0 ) ) . this system exhibits a _ regular _ grazing bifurcation with a square - root singularity because with @xmath133 and @xmath134 the equations of motion are discontinuous at the grazing point . [ fig : vibroimpactcompliant ] , title="fig:",height=188 ] ( 6.78,.08)@xmath135 ( 8.44,.08)@xmath136 ( 9.64,.08)@xmath9 ( 3.78,3.38)@xmath137 ( 3.78,1.4)@xmath138 ( 11.58,3.64)@xmath139 ( 11.58,1.18)@xmath140 ( 9.34,.63)@xmath141 ( 6.58,3.88)block ( 11.02,4.12)support ( 10.38,4.64)stop to motivate and illustrate our results for stochastic nordmark maps , we consider the prototypical vibro - impacting system shown in fig . [ fig : vibroimpactcompliant ] and studied in @xcite . this system consists of a harmonically forced one - degree - of - freedom linear oscillator that experiences compliant ( or soft ) impacts with a support , and we use the following non - dimensionalised equations to model the dynamics : @xmath142 here @xmath143 denotes the location of the block , which has the equilibrium position @xmath144 . a rigid stop prevents the support reaching a position with @xmath10 , and prestresses the support by a distance , @xmath134 . the constants @xmath137 , @xmath138 , @xmath139 and @xmath140 represent non - dimensionalised spring and damping coefficients for the oscillator and support . we neglect the mass of the support , ignore energy loss at impacts , and assume that whenever the block is not in contact with the support , the support is located at @xmath9 . experiments of simple vibro - impacting systems with compliant impacts have shown that low - dimensional models such as ( [ eq : impactoscillator ] ) can quantitatively match the physically observed dynamics near grazing bifurcations @xcite . here we treat the forcing amplitude , @xmath145 , as the primary bifurcation parameter , and assume @xmath146 . the steady - state solution ( behaviour in the limit @xmath147 ) to ( [ eq : impactoscillator ] ) with @xmath10 is @xmath148 when @xmath149 , where @xmath150 the maximum value of @xmath151 over one period is negative , and so @xmath151 is an attracting non - impacting periodic orbit of ( [ eq : impactoscillator ] ) . the critical value , @xmath152 , is a grazing bifurcation at which @xmath151 has unit amplitude and attains the value @xmath153 at times @xmath154 , for @xmath155 , where @xmath156 and @xmath157 . to convert ( [ eq : impactoscillator ] ) to the general form ( [ eq : f ] ) , we define @xmath158 here the phase space of ( [ eq : impactoscillator ] ) with ( [ eq : vw ] ) is isomorphic to @xmath159 , rather than @xmath15 , but this does not affect the bifurcation structure near grazing . for ( [ eq : impactoscillator ] ) with ( [ eq : vw ] ) , the coefficients in ( [ eq : alphabetagamma ] ) , which describe the behaviour of the system near the grazing point , are given by @xmath160 and by evaluating ( [ eq : c ] ) with ( [ eq : odecoeffsosc ] ) we obtain @xmath161 in addition , from the general solution to ( [ eq : impactoscillator ] ) we find that the coefficients in the global map ( [ eq : g ] ) are given by @xmath162 [ eq : mapcoeffosc ] to model noise and uncertainties we use the one - dimensional ornstein - uhlenbeck process @xmath163 where @xmath164 are constants and @xmath165 is a standard brownian motion . given an initial value @xmath166 , at any positive time @xmath167 is a gaussian random variable with mean and variance @xmath168 = \xi_0 \,{\rm e}^{\frac{-t}{\nu } } \ ; , \qquad { \rm var } \left [ \xi(t ) | \xi(0 ) = \xi_0 \right ] = \frac{\ee^2}{2 \nu } \left ( 1 - { \rm e}^{\frac{-2 t}{\nu } } \right ) \;. \label{eq : meanvarxi}\ ] ] in the limit @xmath147 , @xmath169 $ ] , where @xmath170 $ ] denotes the gaussian distribution of mean @xmath46 and variance @xmath171 . the _ correlation time _ of ( [ eq : xi ] ) , defined as @xmath172}{{\rm var } \left [ \xi(0 ) \right ] } \,dt$ ] , with @xmath173 $ ] , is equal to @xmath174 . in our context , @xmath167 is coloured noise and the parameter @xmath67 governs the size of the noise . unlike white noise , @xmath167 has an inherent time - scale , @xmath174 , and is suitable for modelling various types of uncertainties in mechanical systems , such as background vibrations @xcite . in the white noise limit , forcing by @xmath167 becomes a diffusion process @xmath175 . we first consider the following stochastic perturbation of ( [ eq : f ] ) , @xmath176 where @xmath167 is given by ( [ eq : xi ] ) . in ( [ eq : fa ] ) randomness is present in the switching condition , while evolution between switching events remains deterministic . we expect ( [ eq : fa ] ) to be applicable to a wide variety of piecewise - smooth systems . for the vibro - impacting system of [ sec : oscillator ] , @xmath167 may capture uncertainties in the point at which contact between the block and support occurs or is lost . for switched control systems , @xmath167 may correspond to measurement errors that produce variability in evaluations of switching rules @xcite . here we consider orbits of ( [ eq : fa ] ) that are close to the grazing periodic orbit of ( [ eq : f ] ) . orbits of ( [ eq : fa ] ) near grazing only spend short lengths of time in the region @xmath11 while passing near the origin , and for simplicity we suppose that the value of @xmath174 is large compared to such times . in this case it is reasonable to approximate @xmath167 by a constant while an orbit is near the origin . with this approximation , the sum @xmath177 does not switch sign more than twice as the orbit passes near the origin , which substantially simplifies our calculations below . we let @xmath178 denote the value of @xmath167 during the @xmath179 instance that the orbit of ( [ eq : fa ] ) passes near the origin . the time between between consecutive traversals near the origin is well - approximated by the period of the grazing periodic orbit , call it @xmath180 . with this approximation , @xmath181 \;. \label{eq : xiiab}\ ] ] to derive the stochastic version of ( [ eq : n ] ) for ( [ eq : fa ] ) with ( [ eq : xiiab ] ) , we first derive the induced stochastic discontinuity map . here condition ( [ eq : pipcondition ] ) is useful , as it implies that an orbit governed by @xmath6 attains a local maximum value of @xmath74 at an intersection with @xmath78 . for @xmath182 , we conclude that @xmath183 as the orbit passes near the origin , so that @xmath184 . if instead @xmath185 , then the discontinuity map @xmath186 is given by ( [ eq : u4w42 ] ) except that @xmath187 appears inside the square root because this quantity represents the distance from the switching condition . that is , @xmath188 where @xmath189 . by combining ( [ eq : u4w42a ] ) with the global map @xmath122 , applying the coordinate change ( [ eq : coordinatechange ] ) , and dropping higher order terms , we obtain @xmath190 @xmath62 is the stochastic nordmark map corresponding to ( [ eq : fa ] ) . notice that randomness in the switching condition of ( [ eq : fa ] ) has translated to randomness in both the switching condition of ( [ eq : n ] ) and in the image of the map with @xmath25 . in contrast , a piecewise - linear map for which randomness is present purely in the switching condition is studied in @xcite . in order to fairly compare @xmath62 with other stochastic versions of ( [ eq : n ] ) in [ sec : dynamics ] , we estimate the effective size of the stochastic contribution for our illustrative parameters values of the prototypical system ( [ eq : impactoscillator ] ) and a representative value of @xmath72 . this motivates us to express @xmath67 in terms of a scaled parameter @xmath191 , and to obtain comparable stochastic contributions for fixed @xmath192 in the different cases . the square - root term of @xmath62 is @xmath193 , where @xmath194 . if @xmath195 is large relative to @xmath196 , then this term is well - approximated by @xmath197 , and so @xmath198 estimates the additive stochastic contribution to @xmath62 . for the impact oscillator with the parameter values of fig . [ fig : detbifdiag ] ( here @xmath199 and @xmath200 ) and using @xmath201 ( corresponding to the value of @xmath25 in fig . [ fig : detbifdiag ] for @xmath72 ) , this quantity is approximately @xmath202 . with @xmath203 ( used in [ sec : dynamics ] ) , the standard derivation of the stochastic contribution is approximately @xmath204 . therefore , for @xmath191 , where @xmath205 the standard deviation of the stochastic contribution is approximately @xmath206 when @xmath73 . ( 6.8,5.1 ) ( 0,0 ) , ( [ eq : na ] ) , with a numerical solution to ( [ eq : fa ] ) . the three groups of purple dots were obtained by numerically solving ( [ eq : fa ] ) with ( [ eq : xi ] ) , @xmath203 and @xmath207 , for the vibro - impacting system ( [ eq : impactoscillator ] ) with ( [ eq : vw ] ) and @xmath36 ( as in fig . [ fig : detbifdiag ] ) and @xmath208 ( which corresponds to @xmath72 ) . more precisely , @xmath209 points on @xmath78 ( labelled @xmath80 in fig . [ fig : grazschem3d ] ) were obtained by numerically solving ( [ eq : fa ] ) , and these were transformed to @xmath20-coordinates by applying ( [ eq : coordinatechange ] ) and ( [ eq : vw ] ) to produce the purple dots . the three groups of black dots are @xmath209 iterates of @xmath62 with ( [ eq : xiiab ] ) and parameter values chosen to match the vibro - impacting system ( specifically , @xmath203 , @xmath207 , @xmath210 , @xmath72 , @xmath33 , @xmath34 , @xmath35 , @xmath199 and @xmath200 ) . the deterministic @xmath38-cycles of ( [ eq : fa ] ) and ( [ eq : na ] ) are shown with triangles . [ fig : comparea_03 ] , title="fig:",height=192 ] ( 3.76,0)@xmath47 ( 0,3.5)@xmath211 to illustrate the accuracy of @xmath62 , fig . [ fig : comparea_03 ] compares iterates of @xmath62 ( black dots ) with a numerical solution to ( [ eq : fa ] ) for the vibro - impacting system of [ sec : oscillator ] ( purple dots ) using @xmath207 . for the given parameter values , the system has an attracting @xmath38-cycle in the absence of noise . for this reason , both sets of points are grouped about the @xmath38-cycle . the two sets of points are slightly separated . this is because the form of the deterministic nordmark map does not include higher order terms of the true return map , as observed by the separation of the values taken by the deterministic @xmath38-cycles shown in fig . [ fig : comparea_03 ] . the size and shape of the spread of the two sets of randomly generated points are similar , as is their location relative to the deterministic values of the map . this demonstrates that @xmath62 can accurately capture the stochastic dynamics of ( [ eq : fa ] ) . a more precise characterisation of the accuracy of @xmath62 is beyond the scope of this paper . next we consider the case where randomness and uncertainty in ( [ eq : f ] ) is associated with @xmath7 . for mechanical systems with impacts , this corresponds to variability in the evolution of the system during an impact . for simplicity we include noise in only the @xmath75-component of @xmath7 , that is @xmath212 where @xmath167 is given by ( [ eq : xi ] ) . indeed , for the vibro - impacting system of fig . [ fig : vibroimpactcompliant ] , if noise is incorporated into the force on the block when it is in contact with the support , then the equations of motion may be put in the form ( [ eq : fbc ] ) . with noise added to the @xmath74-component of @xmath7 ( or @xmath6 ) , orbits may cross @xmath153 many times in a short time frame which makes the system substantially more difficult to analyse . we leave such considerations for future work . as in [ sub : s1 ] , we consider near - grazing orbits and assume that the value of @xmath174 is much larger than the time each orbit spends in the region @xmath11 . in this case it is reasonable to treat @xmath167 as constant while @xmath11 . during the @xmath179 instance that an orbit passes near the origin , we approximate @xmath167 by @xmath178 , distributed according to ( [ eq : xiiab ] ) . in this scenario the three components of the discontinuity map ( [ eq : v2w2])-([eq : u4w4 ] ) are unchanged except that @xmath213 is replaced by @xmath214 in ( [ eq : v3w3 ] ) ( because the @xmath75-component of the system with @xmath11 is given by @xmath215 , see ( [ eq : alphabetagamma ] ) ) . by combining ( [ eq : v2w2])-([eq : u4w4 ] ) we find that for @xmath110 the discontinuity map is given by @xmath216 and therefore the corresponding stochastic nordmark map is @xmath217 notice that with @xmath218 , @xmath63 is identical to ( [ eq : we can write the stochastic component of @xmath63 as @xmath219 , where @xmath220 . with the parameter values of the impact oscillator ( [ eq : odecoeffsosc ] ) , and @xmath221 , we have @xmath222 . therefore the noise provides a multiplicative stochastic contribution of approximately @xmath223 , ignoring signs . in order to compare the effect of the noise to the other cases , we write @xmath224 , choosing @xmath225 so that the standard deviation of the stochastic contribution is @xmath206 when @xmath73 . for @xmath201 and @xmath203 ( as in [ sub : s1 ] ) , the standard deviation of @xmath223 is approximately @xmath226 , so we therefore choose @xmath227 ( 6.8,5.1 ) ( 0,0 ) with a numerical solution to ( [ eq : fbc ] ) . the three groups of purple dots were obtained by numerically solving ( [ eq : fbc ] ) with ( [ eq : xi ] ) , @xmath203 and @xmath228 , for the vibro - impacting system ( [ eq : impactoscillator ] ) with ( [ eq : vw ] ) using the same parameter values as in fig . [ fig : comparea_03 ] . the three groups of black dots are @xmath209 iterates of @xmath63 with ( [ eq : xiiab ] ) and parameter values matching those of the vibro - impacting system ( refer to the caption of fig . [ fig : comparea_03 ] ) . the deterministic @xmath38-cycles of ( [ eq : fa ] ) and ( [ eq : na ] ) are shown with triangles . [ fig : compareb_03 ] , title="fig:",height=192 ] ( 3.76,0)@xmath47 ( 0,3.5)@xmath211 fig . [ fig : compareb_03 ] compares iterates of @xmath63 to intersections with @xmath78 of a numerical solution to ( [ eq : fbc ] ) using @xmath228 . as expected the two sets of points are similarly distributed about the deterministic @xmath38-cycle . lastly we consider ( [ eq : fbc ] ) in the white noise limit , @xmath229 . in this case ( [ eq : fbc ] ) reduces to a diffusion process forced by white noise , specifically @xmath230 is replaced by @xmath175 . by using ( [ eq : alphabetagamma ] ) to expand @xmath7 , ( [ eq : fbc ] ) for @xmath11 may be written as the three - dimensional stochastic differential equation @xmath231 to obtain a stochastic nordmark map corresponding to this scenario , we first derive the stochastic version of the middle component of the discontinuity map ( [ eq : v3w3 ] ) . given an initial point @xmath232 , where @xmath233 and @xmath234 are small , the desired values of @xmath235 and @xmath236 are given by the point @xmath237 of _ first return _ for the stochastic process ( [ eq : frc ] ) to @xmath153 . first return or first passage problems are an important class of theoretical problems in stochastic calculus with applications traditionally in finance and chemical kinetics @xcite . we approximate ( [ eq : frc ] ) by keeping only the leading order contributions , i.e. @xmath238 together with @xmath239 . with this approximation we are able to provide an explicit expression for the joint probability density function of the return location and time . a formal justification for the omission of the higher order terms is left for future work . we introduce the change of variables @xmath240 this puts ( [ eq : frc2 ] ) in a standard form studied in @xcite , @xmath241 where @xmath242 specifically , @xmath243 is a diffusion process with constant drift , and @xmath244 . therefore @xmath245 may be interpreted as _ integrated brownian motion with constant drift_. we let @xmath246 denote the joint probability density function for the first return of ( [ eq : dpdq ] ) to @xmath247 , at a time @xmath248 , and location @xmath249 . in @xcite , mckean derived an explicit expression for @xmath250 in the case of zero drift by computing the inverse kontorovich - lebedev transform of the corresponding renewal equation . in @xcite , atkinson and clifford extended this result to the case of non - zero drift by applying the radon - nikodyn derivative . specifically @xmath251 \right ) { \rm erf } \left ( \frac{\sqrt{6 h}}{\sqrt{\varrho r } } \right ) \ ; , \label{eq : f}\ ] ] where @xmath252 is the error function . the constant @xmath253 governs the shape of @xmath250 . the limit @xmath254 corresponds to the deterministic case , for which @xmath255 and @xmath256 . with a small value of @xmath257 , @xmath250 is roughly gaussian . in contrast , the limit @xmath258 corresponds to the case of no drift , as in @xcite , or to the limit @xmath259 . in this limit the marginal probability density function for @xmath260 is asymptotically proportional to @xmath261 , for large @xmath260 , and so is long - tailed @xcite . in view of the scaling ( [ eq : pqs ] ) , the stochastic version of ( [ eq : v3w3 ] ) corresponding to ( [ eq : frc ] ) is given by @xmath262 where @xmath58 and @xmath59 have the joint probability density function ( [ eq : f ] ) . by combining ( [ eq : v3w3c ] ) with ( [ eq : v2w2 ] ) and ( [ eq : u4w4 ] ) we obtain @xmath263 which represents the stochastic version of the discontinuity map for points with @xmath110 . then using ( [ eq : u4w4c ] ) we arrive at the following stochastic nordmark map @xmath264 in ( [ eq : nc ] ) we treat each pair @xmath265 as a two - dimensional stochastic random variable with probability density function @xmath266 , where @xmath267 which results from combining ( [ eq : varrho ] ) with @xmath268 , ( [ eq : v2w2 ] ) , and @xmath269 , ( [ eq : coordinatechange ] ) . we now estimate the size of the stochastic contribution in @xmath64 . the leading order stochastic component of @xmath64 is @xmath270 , where @xmath271 . with ( [ eq : odecoeffsosc ] ) and @xmath221 , we can write @xmath272 , where the second term represents the multiplicative stochastic contribution , ignoring the sign of @xmath127 , because the deterministic values of @xmath59 and @xmath58 are @xmath273 and @xmath274 respectively . with a small value of @xmath257 ( @xmath275 is suitable ) , @xmath246 is approximately gaussian because the effective noise amplitude in ( [ eq : dpdq ] ) is small . by ( [ eq : varrho2 ] ) , this approximation is valid when , roughly speaking , @xmath67 is not too large and @xmath195 is not too small . from ( [ eq : f ] ) we determine the covariance matrix of the gaussian approximation to be @xmath276 and it follows that in this approximation the linear combination @xmath277 has standard deviation @xmath278 . the standard deviation of the stochastic contribution in @xmath64 is therefore approximately @xmath279 . following the previous cases , we write @xmath280 , and choose @xmath281 so that the standard deviation of the stochastic contribution is @xmath206 when @xmath73 . for @xmath64 , this quantity is @xmath282 , using ( [ eq : varrho2 ] ) , the parameter values from fig . [ fig : detbifdiag ] , and @xmath201 . therefore we let @xmath283 here @xmath284 when @xmath73 , and so for these values the gaussian approximation to ( [ eq : f ] ) is justified . ( 6.8,5.1 ) ( 0,0 ) with a numerical solution to ( [ eq : fbc ] ) . the three groups of purple dots were obtained by numerically solving ( [ eq : fbc ] ) with ( [ eq : xi ] ) , @xmath229 ( in which case @xmath230 is replaced by @xmath285 ) and @xmath286 , for the vibro - impacting system ( [ eq : impactoscillator ] ) with ( [ eq : vw ] ) using the same parameter values as in fig . [ fig : comparea_03 ] . the three groups of black dots are @xmath209 iterates of @xmath64 with ( [ eq : f ] ) and ( [ eq : varrho2 ] ) and parameter values matching those of the vibro - impacting system . the deterministic @xmath38-cycles of ( [ eq : fa ] ) and ( [ eq : na ] ) are shown with triangles . [ fig : comparec_03 ] , title="fig:",height=192 ] ( 3.76,0)@xmath47 ( 0,3.5)@xmath211 fig . [ fig : comparec_03 ] compares iterates of @xmath64 to intersections with @xmath78 of a numerical solution to ( [ eq : fbc ] ) using @xmath286 . as with the previous two figures , this shows that @xmath64 can accurately capture the stochastic dynamics of ( [ eq : fbc ] ) . in this section we explore the dynamics of the three stochastic nordmark maps , @xmath62 , @xmath63 and @xmath64 , and discuss how the different forms of these maps is evident in their dynamical behaviour . to briefly summarise , @xmath62 applies to the system with stochastic switching ( [ eq : fa ] ) , whereas @xmath63 and @xmath64 apply to ( [ eq : fbc ] ) . for @xmath62 and @xmath63 it is assumed that the value of @xmath174 ( the correlation time of the noise ) is large relative to the times that orbits spend in @xmath11 . as these times are rarely larger than @xmath287 for the parameter values considered here , we take @xmath203 in @xmath62 and @xmath63 to ensure that the correlation time is large enough . the values of @xmath178 in @xmath62 and @xmath63 are distributed according to ( [ eq : xiiab ] ) . @xmath64 corresponds to the limit @xmath56 . in @xmath64 , @xmath58 and @xmath59 are distributed according to ( [ eq : f ] ) and depend on the value of @xmath288 ( [ eq : varrho2 ] ) . for each @xmath68 we have written @xmath289 where the @xmath71 are given by ( [ eq : tildeee1 ] ) , ( [ eq : tildeee2 ] ) and ( [ eq : tildeee3 ] ) . these values have been chosen such that for a given value of @xmath192 , the size of the stochastic contribution in @xmath62 , @xmath63 and @xmath64 is roughly the same , at least when @xmath72 . for @xmath46 close to @xmath290 , the stochastic contributions are noticeably different for these choices of @xmath71 . in [ sub : mudependence ] we look at stochastic bifurcation diagrams in order to obtain a basic understanding of how the stochastic dynamics differs with the value of @xmath46 . in the subsequent parts of this section we study two - dimensional invariant densities in order to gain a deeper understanding of the dynamics . ( 12,18.4 ) ( 0,12.4 ) ( panel a ) , @xmath63 ( panel b ) , and @xmath64 ( panel c ) , with @xmath33 , @xmath34 , and @xmath35 . these parameter values correspond to the vibro - impacting system ( [ eq : impactoscillator ] ) , with @xmath36 . the black dots are iterates of @xmath62 , @xmath63 and @xmath64 with transient points omitted . the noise amplitudes are given by ( [ eq : tildeee1 ] ) , ( [ eq : tildeee2 ] ) and ( [ eq : tildeee3 ] ) with @xmath73 , and @xmath203 for panels a and b ( panel c corresponds to @xmath229 ) . in each panel the deterministic bifurcation diagram ( fig . [ fig : detbifdiag ] ) is shown in blue . [ fig : noisybifdiag ] , title="fig:",height=226 ] ( 0,6.2 ) ( panel a ) , @xmath63 ( panel b ) , and @xmath64 ( panel c ) , with @xmath33 , @xmath34 , and @xmath35 . these parameter values correspond to the vibro - impacting system ( [ eq : impactoscillator ] ) , with @xmath36 . the black dots are iterates of @xmath62 , @xmath63 and @xmath64 with transient points omitted . the noise amplitudes are given by ( [ eq : tildeee1 ] ) , ( [ eq : tildeee2 ] ) and ( [ eq : tildeee3 ] ) with @xmath73 , and @xmath203 for panels a and b ( panel c corresponds to @xmath229 ) . in each panel the deterministic bifurcation diagram ( fig . [ fig : detbifdiag ] ) is shown in blue . [ fig : noisybifdiag ] , title="fig:",height=226 ] ( 0,0 ) ( panel a ) , @xmath63 ( panel b ) , and @xmath64 ( panel c ) , with @xmath33 , @xmath34 , and @xmath35 . these parameter values correspond to the vibro - impacting system ( [ eq : impactoscillator ] ) , with @xmath36 . the black dots are iterates of @xmath62 , @xmath63 and @xmath64 with transient points omitted . the noise amplitudes are given by ( [ eq : tildeee1 ] ) , ( [ eq : tildeee2 ] ) and ( [ eq : tildeee3 ] ) with @xmath73 , and @xmath203 for panels a and b ( panel c corresponds to @xmath229 ) . in each panel the deterministic bifurcation diagram ( fig . [ fig : detbifdiag ] ) is shown in blue . [ fig : noisybifdiag ] , title="fig:",height=226 ] ( 6.2,12.4)@xmath46 ( 6.2,6.2)@xmath46 ( 6.2,0)@xmath46 ( 0,16.2)@xmath47 ( 0,10)@xmath47 ( 0,3.8)@xmath47 ( 1.7,18.1)*a * ( 1.7,11.9)*b * ( 1.7,5.7)*c * fig . [ fig : noisybifdiag ] shows stochastic versions of the bifurcation diagram shown in fig . [ fig : detbifdiag ] for the three stochastic nordmark maps . as expected , the noise blurs the bifurcation diagram . in panel a , which corresponds to the map @xmath62 , for values of @xmath46 very close to zero ( say @xmath291 ) the points are relatively highly spread . this is because here the deterministic map has an attractor near @xmath292 , so in @xmath62 the sign of @xmath195 is often different to the sign of @xmath293 . that is , the choice of the half - map of @xmath62 is regularly determined by @xmath178 rather than @xmath195 . furthermore , as shown in [ sub : s1 ] , the leading order component of the stochastic contribution to the right half - map of @xmath62 is inversely proportional to @xmath66 , which for very small @xmath46 is large relative to its value for @xmath46 away from zero . in contrast , with @xmath294 say , the underlying attracting @xmath38-cycle is sufficiently far from @xmath292 so that the sign of @xmath195 rarely differs from that of @xmath293 . the points are randomly distributed about the @xmath38-cycle due to noise in the right half - map of @xmath62 . the size of the deviation decreases with increasing @xmath46 , because the strength of attraction of the @xmath38-cycle increases with @xmath46 . in panels b and c , which correspond to @xmath63 and @xmath64 respectively , the bifurcation diagrams show no variability for @xmath29 . this is because for @xmath29 , @xmath32 ( [ eq : xlyl ] ) is a fixed point of @xmath63 and @xmath64 . for @xmath295 , the size of deviations about the underlying attracting @xmath39-cycle increases with @xmath46 . this is primarily because the coefficient of the @xmath130-term of @xmath63 and @xmath64 is random , and for the @xmath39-cycle this value of @xmath47 increases with @xmath46 . for @xmath294 , the size of deviations varies little with @xmath46 because the increased variability caused by a larger value of @xmath47 is balanced by the fact that the strength of attraction of the @xmath38-cycle increases . panels b and c are similar , suggesting that the value of @xmath174 has little effect on the long - term dynamics , although panel c shows slightly more variability for very small values of @xmath40 . ( 13.8,10.2 ) ( 0,4.7 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . the value of the density is indicated by colour ( dark red the maximum value of the density ; dark blue zero ) . panel b shows the invariant density at three times the noise amplitude as panel a. panel c plots @xmath296 the fraction of instances that points return to @xmath297 in @xmath298 iterations ( [ eq : sigma ] ) against the noise amplitude . [ fig : invdensitya_03 ] , title="fig:",height=192 ] ( 7,4.7 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . the value of the density is indicated by colour ( dark red the maximum value of the density ; dark blue panel b shows the invariant density at three times the noise amplitude as panel a. panel c plots @xmath296 the fraction of instances that points return to @xmath297 in @xmath298 iterations ( [ eq : sigma ] ) against the noise amplitude . [ fig : invdensitya_03 ] , title="fig:",height=192 ] ( 2.4,0 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . the value of the density is indicated by colour ( dark red the maximum value of the density ; dark blue panel b shows the invariant density at three times the noise amplitude as panel a. panel c plots @xmath296 the fraction of instances that points return to @xmath297 in @xmath298 iterations ( [ eq : sigma ] ) against the noise amplitude . [ fig : invdensitya_03 ] , title="fig:",height=170 ] ( 3.42,10)@xmath73 ( 10.42,10)@xmath299 ( 3.76,4.7)@xmath47 ( .3,8.2)@xmath211 ( 10.76,4.7)@xmath47 ( 7.3,8.2)@xmath211 ( 7.2,0)@xmath192 ( 8,3.44)@xmath300 ( 8.87,1.45)@xmath301 ( 9.33,2)@xmath302 ( .5,9.9)*a * ( 7.5,9.9)*b * ( 2.4,4.2)*c * ( 13.8,10.2 ) ( 0,4.7 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . panel b shows the invariant density with @xmath299 , and panel c is a plot of the fractions @xmath296 ( [ eq : sigma ] ) . [ fig : invdensityb_03 ] , title="fig:",height=192 ] ( 7,4.7 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . panel b shows the invariant density with @xmath299 , and panel c is a plot of the fractions @xmath296 ( [ eq : sigma ] ) . [ fig : invdensityb_03 ] , title="fig:",height=192 ] ( 2.4,0 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . panel b shows the invariant density with @xmath299 , and panel c is a plot of the fractions @xmath296 ( [ eq : sigma ] ) . [ fig : invdensityb_03 ] , title="fig:",height=170 ] ( 3.42,10)@xmath73 ( 10.42,10)@xmath299 ( 3.76,4.7)@xmath47 ( .3,8.2)@xmath211 ( 10.76,4.7)@xmath47 ( 7.3,8.2)@xmath211 ( 7.2,0)@xmath192 ( 8,3.44)@xmath300 ( 6.66,1.64)@xmath301 ( 8.78,2)@xmath302 ( .5,9.9)*a * ( 7.5,9.9)*b * ( 2.4,4.2)*c * ( 13.8,10.2 ) ( 0,4.7 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . panel b shows the invariant density with @xmath299 , and panel c is a plot of the fractions @xmath296 ( [ eq : sigma ] ) . [ fig : invdensityc_03 ] , title="fig:",height=192 ] ( 7,4.7 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . panel b shows the invariant density with @xmath299 , and panel c is a plot of the fractions @xmath296 ( [ eq : sigma ] ) . [ fig : invdensityc_03 ] , title="fig:",height=192 ] ( 2.4,0 ) with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath72 . panel b shows the invariant density with @xmath299 , and panel c is a plot of the fractions @xmath296 ( [ eq : sigma ] ) . [ fig : invdensityc_03 ] , title="fig:",height=170 ] ( 3.42,10)@xmath73 ( 10.42,10)@xmath299 ( 3.76,4.7)@xmath47 ( .3,8.2)@xmath211 ( 10.76,4.7)@xmath47 ( 7.3,8.2)@xmath211 ( 7.2,0)@xmath192 ( 7.3,3.55)@xmath300 ( 8,1.7)@xmath301 ( 8.93,2.03)@xmath302 ( .5,9.9)*a * ( 7.5,9.9)*b * ( 2.4,4.2)*c * figs . [ fig : invdensitya_03]-[fig : invdensity_m002 ] show two - dimensional invariant densities of @xmath62 , @xmath63 and @xmath64 . by assuming ergodicity , invariant densities were computed on a @xmath303 grid of @xmath47 and @xmath211 values from @xmath304 consecutive iterates of a single orbit with transient points omitted . let us first explain panel c of figs . [ fig : invdensitya_03]-[fig : invdensityc_03 ] . given a sample orbit of @xmath62 , @xmath63 or @xmath64 , for each point with @xmath25 we let @xmath305 be the smallest positive integer for which @xmath306 , as in @xcite . @xmath305 represents the number of iterations required for a return to @xmath307 from the point @xmath308 . numerically we can compute a large number of values of @xmath305 ( say @xmath309 of them ) . then for each @xmath310 , we let @xmath296 denote the fraction of the @xmath305 that are equal to @xmath298 , i.e. @xmath311 where @xmath312 if @xmath313 , and @xmath314 otherwise . [ fig : invdensitya_03]-[fig : invdensityc_03 ] correspond to @xmath72 for which there is an underlying attracting @xmath38-cycle . therefore with small noise , @xmath315 , and for all @xmath316 , @xmath317 . [ fig : invdensitya_03 ] corresponds to the map @xmath62 . in panel a , the size of the noise is relatively small , so iterates of @xmath62 follow close to the @xmath38-cycle . the invariant density is well - approximated by a scaled sum of three gaussian densities centred at each point of the @xmath38-cycle . about the point with @xmath318 , the density is stretched substantially more in @xmath47-direction than in the @xmath211-direction . this is because points with @xmath318 have likely just undergone an iteration under the right half - map of @xmath62 which is stochastic in the @xmath47-component but not the @xmath211-component . the stretching around other iterates is then a consequence of iterating under @xmath62 with @xmath319 . with larger values of @xmath192 , it is relatively common for the orbit to return to @xmath307 in a number of iterations other than three , fig . [ fig : invdensitya_03]-c . for this reason the invariant density displays additional characteristics . for instance with @xmath299 , the orbit returns to @xmath307 in two iterations almost @xmath320 of the time . consequently , a substantial part of the invariant density centred roughly about the point of the @xmath38-cycle with @xmath321 , lies in @xmath307 , fig . [ fig : invdensitya_03]-b . the invariant density in panel b also has a small component with @xmath322 . this is due to points of the orbit with small values of @xmath307 mapping under the left half - map of @xmath62 due to the noise ( i.e. returning to @xmath307 in only one iteration ) . [ fig : invdensityb_03 ] illustrates @xmath63 using the same parameter values . again with small noise the invariant density is roughly gaussian about each point of the @xmath38-cycle , whereas for relatively large noise iterates often cross into @xmath307 prematurely causing the invariant density to take an irregular shape . when @xmath299 , points of the orbit that do not return to @xmath307 in three iterations , usually return to @xmath307 in two iterations . [ fig : invdensityc_03 ] corresponds to @xmath64 and is similar to the previous figure . this indicates that the correlation time @xmath174 has little effect on these pictures , although the invariant density has a slightly different shape when @xmath299 . ( 13.8,10.4 ) ( 0,5.3)-cycle of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath323 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , using the same parameter values as fig . [ fig : noisybifdiag ] and @xmath323 . [ fig : invdensity_001 ] , title="fig:",height=192 ] ( 7,5.3)-cycle of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath323 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , using the same parameter values as fig . [ fig : noisybifdiag ] and @xmath323 . [ fig : invdensity_001 ] , title="fig:",height=192 ] ( 0,0)-cycle of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath323 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , using the same parameter values as fig . [ fig : noisybifdiag ] and @xmath323 . [ fig : invdensity_001 ] , title="fig:",height=192 ] ( 7,0)-cycle of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath323 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , using the same parameter values as fig . [ fig : noisybifdiag ] and @xmath323 . [ fig : invdensity_001 ] , title="fig:",height=192 ] ( 3.76,5.3)@xmath47 ( .3,8.8)@xmath211 ( 10.76,5.3)@xmath47 ( 7.3,8.8)@xmath211 ( 3.76,0)@xmath47 ( .3,3.5)@xmath211 ( 10.76,0)@xmath47 ( 7.3,3.5)@xmath211 ( .6,10.3)*a * ( 7.6,10.3)*b * ( .6,5)*c * ( 7.6,5)*d * fig . [ fig : invdensity_001 ] shows invariant densities of @xmath62 , @xmath63 and @xmath64 for parameter values closer to the grazing bifurcation than the previous three figures , specifically @xmath323 . at this value of @xmath46 , there is an underlying attracting @xmath39-cycle , panel a. the noise amplitudes are given by ( [ eq : tildeee1 ] ) , ( [ eq : tildeee2 ] ) and ( [ eq : tildeee3 ] ) with @xmath73 . recall , these amplitudes were chosen such that the size of the noise response of the three maps is roughly the same for larger values of @xmath46 . here , however , the size of noise response differs substantially . in panels c and d , which correspond to @xmath63 and @xmath64 respectively , the invariant density is approximately a scaled sum of four gaussians about each point of the @xmath39-cycle . the invariant density in panel d , corresponding to @xmath229 , is noticeably larger than that of panel c. in panel b , which corresponds to @xmath62 , the noise has a substantial effect because the switching condition of @xmath62 is stochastic , and many points of @xmath62 fall close to @xmath292 . the invariant density has a small @xmath324-shaped component in @xmath297 corresponding to consecutive points of the orbit mapping under the left half - map of @xmath62 . the part of the invariant density for @xmath319 and @xmath325 corresponds to images of points under the right half - map of @xmath62 , and is bimodal because the invariant density has roughly two components in @xmath307 . ( 13.8,10.4 ) ( 0,5.3)-cycle and a numerically computed attractor of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath326 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath326 . [ fig : invdensity_0145 ] , title="fig:",height=192 ] ( 7,5.3)-cycle and a numerically computed attractor of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath326 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath326 . [ fig : invdensity_0145 ] , title="fig:",height=192 ] ( 0,0)-cycle and a numerically computed attractor of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath326 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath326 . [ fig : invdensity_0145 ] , title="fig:",height=192 ] ( 7,0)-cycle and a numerically computed attractor of ( [ eq : n ] ) with the same parameter values as fig . [ fig : detbifdiag ] and @xmath326 . panels b , c and d show the invariant densities of @xmath62 , @xmath63 and @xmath64 , respectively , with the same parameter values as fig . [ fig : noisybifdiag ] and @xmath326 . [ fig : invdensity_0145 ] , title="fig:",height=192 ] ( 3.76,5.3)@xmath47 ( .3,8.5)@xmath211 ( 10.76,5.3)@xmath47 ( 7.3,8.5)@xmath211 ( 3.76,0)@xmath47 ( .3,3.2)@xmath211 ( 10.76,0)@xmath47 ( 7.3,3.2)@xmath211 ( .4,10.3)*a * ( 7.4,10.3)*b * ( .4,5)*c * ( 7.4,5)*d * ( 13.8,5.1 ) ( 0,0 ) ( [ eq : xlyl ] ) and the attracting @xmath38-cycle of @xmath327 with the same parameter values as fig . [ fig : detbifdiag ] , except @xmath328 and @xmath329 . here @xmath330 , @xmath34 and @xmath35 . panel b shows the invariant density of the corresponding map @xmath62 , with @xmath203 and @xmath331 . [ fig : invdensity_m002 ] , title="fig:",height=192 ] ( 7,0 ) ( [ eq : xlyl ] ) and the attracting @xmath38-cycle of @xmath327 with the same parameter values as fig . [ fig : detbifdiag ] , except @xmath328 and @xmath329 . here @xmath330 , @xmath34 and @xmath35 . panel b shows the invariant density of the corresponding map @xmath62 , with @xmath203 and @xmath331 . [ fig : invdensity_m002 ] , title="fig:",height=192 ] ( 3.76,0)@xmath47 ( .3,3.4)@xmath211 ( 10.76,0)@xmath47 ( 7.3,3.4)@xmath211 ( .4,5)*a * ( 7.4,5)*b * with @xmath326 in fig . [ fig : detbifdiag ] , there is an attracting @xmath39-cycle and an apparently chaotic attractor . these are shown in fig . [ fig : invdensity_0145]-a . as with smooth maps @xcite , in the presence of noise orbits commonly dwell near the attractors for relatively long periods of time , and switch between attractors quickly . invariant densities of @xmath62 , @xmath63 and @xmath64 are shown in panels b , c and d. in each case the bulk of the density is centred about the two underlying attractors . with white noise ( panel d ) there is no gap in the invariant density around @xmath332 due to randomness in both the @xmath47 and @xmath211-components of @xmath64 . lastly , fig . [ fig : invdensity_m002 ] illustrates stochastic dynamics with @xmath29 . this figure corresponds to @xmath328 ( different to fig . [ fig : detbifdiag ] ) and @xmath329 at which @xmath327 has an attracting @xmath38-cycle as well as the attracting fixed point @xmath32 ( [ eq : xlyl ] ) . these are shown in panel a. the dynamics of @xmath63 and @xmath64 for @xmath319 is deterministic , hence @xmath32 is a fixed point of these maps . given an initial point @xmath333 near the @xmath38-cycle , sample orbits of @xmath63 and @xmath64 eventually reach @xmath32 . in contrast , @xmath62 has an invariant density concentrated about the two attractors , panel b. the part of the density with @xmath334 and @xmath335 corresponds to points of the orbit repeatedly following the left half - map of @xmath62 ( with @xmath24 and @xmath336 ) . this paper concerns grazing bifurcations for which the associated dynamics is described by the nordmark map ( [ eq : n ] ) . the potential influence of randomness and uncertainties on the dynamics of ( [ eq : n ] ) was described in @xcite by studying ( [ eq : n ] ) in the presence of additive white gaussian noise . such a noise formulation is suitable if the nature of the randomness in the ode system is practically independent to the state of the system , such as if there is a random forcing term . in this paper we considered the alternate scenario that randomness is present in the switching condition of the ode system , and in @xmath7 the part of the vector field opposite to the tangential intersection of the grazing periodic orbit . these cases are especially relevant for vibro - impacting systems for which impact events represent the primary source of uncertainty . we derived three different stochastic versions of ( [ eq : n ] ) . these are the maps @xmath62 ( [ eq : na ] ) , which corresponds to a noisy switching condition in the ode system , @xmath63 ( [ eq : nb ] ) , which corresponds to noise in @xmath7 with a large correlation time , and @xmath64 ( [ eq : nc ] ) , which corresponds to white noise in @xmath7 . in each case the noise is nonlinear and non - additive . this indicates that some diligence should be taken when formulating stochastic return maps for grazing bifurcations of piecewise - smooth systems . the stochastic dynamics of @xmath62 , @xmath63 and @xmath64 differs in many ways to that of ( [ eq : n ] ) with additive noise , described in @xcite . for @xmath62 , @xmath63 and @xmath64 , dynamics prior to the grazing bifurcation is deterministic , and beyond the grazing bifurcation two - dimensional invariant densities are often skewed dramatically so that they appear almost one - dimensional . near the grazing bifurcation , @xmath62 exhibits a large noise response relative to @xmath63 and @xmath64 . this suggests that if experimental data of a physical system shows relatively high variability near a grazing bifurcation , then it is likely to be most appropriate to include randomness in the switching condition of a mathematical model . invariant densities of @xmath62 near grazing are highly irregular due to the randomness in the switching condition . for @xmath63 and @xmath64 , the size of the noise response increases , for most part , with the distance ( in parameter space ) beyond the grazing bifurcation . the maps @xmath63 and @xmath64 exhibit qualitatively similar invariant densities , which implies that the correlation time has little effect . indeed the correlation time only influences the short - time dynamics of ( [ eq : fbc ] ) with @xmath11 , and the precise nature of these dynamics has a negligible effect on the invariant densities , which relate to long - time dynamics . m. oestreich , n. hinrichs , k. popp , and c.j . budd . analytical and experimental investigation of an impact oscillator . in _ proceedings of the asme 16th biennal conf . on mech . vibrations and noise . _ , pages 111 , 1997 . e. knobloch and j.b . effect of noise on discrete dynamical systems with multiple attractors . in m.f . mcclintock and p.v.e . moss , editors , _ noise in nonlinear dynamical systems . theory of noise induced processes in special applications . _ , volume 2 , pages 6586 . cambridge university press , new york , 1989 .

in this paper we consider the complexity of problems related to the one - player combinatorial game flood - it , introduced by arthur , clifford , jalsenius , montanaro and sach in @xcite . the original game is played on a board consisting of an @xmath1 grid of coloured squares , each square given a colour from some fixed colour - set , but we can more generally regard the game as being played on a vertex - coloured graph . a move then consists of picking a vertex @xmath2 and a colour @xmath3 , and giving all vertices in the same monochromatic component as @xmath2 colour @xmath3 . the goal is to make the entire graph monochromatic with as few such moves as possible . when the game is played on a planar graph , it can be regarded as modelling repeated use of the flood - fill tool in microsoft paint . implementations of the game , played on a square grid , are widely available online , and include a flash game @xcite as well as popular smartphone apps @xcite . there also exist implementations using a hexagonal grid : mad virus @xcite is the same one - player game described above , while the honey bee game @xcite is a two player variant , and has been studied by fleischer and woeginger @xcite . all these implementations are based on the `` fixed '' version of the game , where all moves must be played at the same fixed vertex ( usually the vertex corresponding to the top left square when the board is an @xmath1 grid ) . for any coloured graph , we define the following problems . * free - flood - it is the problem of determining the minimum number of moves required to flood the graph , if we are allowed to make moves anywhere in the graph . * fixed - flood - it is the same problem when all moves must be played at a single specified vertex . * @xmath4-free - flood - it and @xmath4-fixed - flood - it respectively are the variants of free - flood - it and fixed - flood - it in which only colours from some fixed set of size @xmath4 are used . note that we can trivially flood an @xmath5-vertex graph with @xmath6 moves , and that if @xmath4 colours are present in the initial colouring we require at least @xmath7 moves . these problems are known to be computationally difficult in many situations . in @xcite , arthur , clifford , jalsenius , montanaro and sach proved that @xmath4-free - flood - it is np - hard in the case of an @xmath1 grid , for every @xmath8 , and that this result also holds for the fixed variant . lagoutte , noual and thierry @xcite showed that the same result holds when the game is played instead on a hexagonal grid , as in mad virus or a one - player version of the honey bee game . fleischer and woeginger @xcite proved that @xmath4-fixed flood it remains np - hard when restricted to trees , for every @xmath9 , and fukui , nakanishi , uehara , uno and uno @xcite demonstrated that this result can be extended to show the hardness @xmath4-free flood it under the same conditions . a few positive results are known , however . 2-free - flood - it is solvable in polynomial time on arbitrary graphs , a result shown independently by clifford et . @xcite , lagoutte @xcite and meeks and scott @xcite . it is also known that fixed - flood - it and free - flood - it are solvable in polynomial time on paths @xcite and cycles @xcite , and more generally on any graph with only a polynomial number of connected subgraphs @xcite . meeks and scott also show that the number of moves required to create a monochromatic component containing an arbitrary , bounded - size subset of the vertices can be computed in polynomial time , even when the number of colours is unbounded @xcite . a major focus of previous research has been the restriction of the game to rectangular boards of fixed height . although an additive approximation for @xmath4-free - flood - it can be computed in polynomial time @xcite , solving either @xmath4-free - flood - it or @xmath4-fixed - flood - it exactly remains np - hard on @xmath10 boards , whenever @xmath9 @xcite . however , clifford et . al . @xcite give a linear time algorithm for fixed - flood - it on @xmath0 boards . they also raise the question of the complexity of the free variant in this setting . here we address this remaining case of ( @xmath4-)free - flood - it restricted to @xmath0 boards , which turn out to be a particularly interesting class of graphs on which to analyse the game . the majority of the paper describes an algorithm to demonstrate that @xmath4-free - flood - it , restricted to @xmath0 boards , is fixed parameter tractable with parameter @xmath4 . to do this we exploit some general results from @xcite about the relationship between the number of moves required to flood a graph and its spanning trees . on the other hand , we also show that free - flood - it remains np - hard in this setting . this is a somewhat surprising result , as it gives the first example of a class of graphs on which the complexity of fixed - flood - it and free - flood - it has been shown to be different . the rest of the paper is organised as follows . we begin with notation and definitions in section [ notation ] , before giving our algorithm for @xmath4-free - flood - it in section [ fpt ] . finally , in section [ nphard ] , we show that the problem remains np - hard when the number of colours used is unbounded . although the original flood - it game is played on a square grid , and our main results here concern the game restricted to a rectangular grid , it is convenient to consider the generalisation of the game to an arbitrary graph @xmath11 , equipped with an initial colouring @xmath12 using colours from the _ colour - set _ @xmath13 . then each move @xmath14 consists of choosing some vertex @xmath15 and a colour @xmath16 , and assigning colour @xmath3 to all vertices in the same monochromatic component as @xmath2 . the goal is to give every vertex in @xmath17 the same colour , using as few moves as possible . given any connected graph @xmath17 , equipped with a colouring @xmath12 ( not necessarily proper ) , we define @xmath18 to be the minimum number of moves required in the free variant to give all its vertices colour @xmath3 , and @xmath19 to be @xmath20 . if @xmath21 is a sequence of moves played on a graph @xmath17 with initial colouring @xmath12 , we denote by @xmath22 the new colouring obtained by playing @xmath21 in @xmath17 . note that , if the initial colouring @xmath12 of @xmath17 is not proper , we may obtain an equivalent coloured graph @xmath23 ( with colouring @xmath24 ) by contracting monochromatic components of @xmath17 with respect to @xmath12 . let @xmath25 be any subset of @xmath26 . we denote by @xmath27 the set of colours assigned to vertices of @xmath25 by @xmath12 . we say a move @xmath28 is _ played in _ @xmath25 if @xmath29 , and that @xmath25 is _ linked _ if it is contained in a single monochromatic component . subsets @xmath30 are _ adjacent _ if there exists @xmath31 with @xmath32 and @xmath33 . when we consider the game played on a rectangular board @xmath34 , we are effectively playing the game in a corresponding coloured graph @xmath17 , obtained from the planar dual of @xmath34 ( in which there is one vertex corresponding to each square of @xmath34 , and vertices are adjacent if they correspond to squares which are either horizontally or vertically adjacent in @xmath34 ) by giving each vertex the colour of the corresponding square in @xmath34 . we identify areas of @xmath34 with the corresponding subgraphs of @xmath17 , and may refer to them interchangeably . we define a _ border _ of @xmath34 to be a union of edges of squares on the original board @xmath34 that forms a path from the top edge of the board to the bottom ( but not including any edges that form the top or bottom edge of the board ) . thus , a border in @xmath34 corresponds to an edge - cut in the corresponding graph . observe that a border is uniquely defined by the points at which it meets the top and bottom of the board , so there are @xmath35 borders in total . we denote by @xmath36 and @xmath37 the borders corresponding to the left - hand and right - hand edges of the board respectively . given two borders @xmath38 and @xmath39 , we write @xmath40 if and only if @xmath38 meets both the top and bottom of the board to the left of ( or at the same point as ) @xmath39 , and write @xmath41 if @xmath40 and @xmath42 . note that if @xmath40 then @xmath38 lies entirely to the left of @xmath39 ( the two borders may meet but never cross ) ; this is a special property of @xmath0 boards and does not hold for @xmath43 boards for @xmath44 . if @xmath17 is the graph corresponding to the @xmath0 board @xmath34 , we say that a vertex ( or subgraph ) is _ incident _ with a border @xmath45 if the vertex ( or some vertex in the subgraph ) corresponds to a square on @xmath34 whose edge forms part of @xmath45 . if @xmath41 are borders , we denote the subgraph induced by vertices lying between @xmath38 and @xmath39 by @xmath46 $ ] , and we say @xmath46 $ ] is a _ section _ if it is connected . finally , given any tree @xmath47 , we denote by @xmath48 the subtree obtained by deleting all leaves of @xmath47 , and given any @xmath49 we set @xmath50 to be the unique path from @xmath51 to @xmath52 in @xmath47 . in this section , we give an algorithm to solve @xmath4-free - flood - it on @xmath0 boards . more specifically , we prove the following result , which shows that @xmath4-free - flood - it , restricted to @xmath0 boards , is fixed parameter tractable , parameterised by @xmath4 . this answers an open question of clifford , jalsenius , montanaro and sach @xcite . when restricted to @xmath0 boards , @xmath4-free - flood - it can be solved in time @xmath53 . [ 2xn - fpt ] we begin with some background and auxiliary results in section [ background ] , and then describe the algorithm in section [ algorithm ] . before describing our algorithm in the next section , we need a number of results which will be used to prove its correctness . we begin with some previous results from @xcite . meeks and scott prove that it suffices to consider spanning trees in order to determine the minimum number of moves required to flood a graph . for any connected graph @xmath17 , let @xmath54 denote the set of all spanning trees of @xmath17 . let @xmath17 be a connected graph with colouring @xmath12 from colour - set @xmath13 . then , for any @xmath16 , @xmath55 [ spanning - tree ] for any @xmath16 , we say that @xmath47 is a _ @xmath3-minimal _ spanning tree for @xmath17 if @xmath56 . in the remainder of this section , we prove that in the special case in which @xmath17 corresponds to a @xmath0 board , there is always a @xmath3-minimal spanning tree @xmath47 such that @xmath48 is a path . in doing so , and in proving the correctness of our algorithm in the next section , we make use of a corollary of theorem [ spanning - tree ] , again proved in @xcite , which shows that the number of moves required to flood a graph is bounded above by the sum of the numbers of moves required to flood connected subgraphs which cover the vertex - set . let @xmath17 be a connected graph , with colouring @xmath12 from colour - set @xmath13 , and let @xmath25 and @xmath34 be subsets of @xmath57 such that @xmath58 and @xmath59 , g[b]$ ] are connected . then , for any @xmath16 , @xmath60 [ non - interference ] a key step used to prove theorem [ spanning - tree ] in @xcite is to prove a special case of corollary [ non - interference ] , where the underlying graph @xmath17 is a tree and @xmath25 and @xmath34 are disjoint . we will need the following result , proved using an extension of part of this proof from @xcite . let @xmath47 be a tree , with colouring @xmath12 from colour - set @xmath13 , let @xmath25 and @xmath34 be disjoint subsets of @xmath61 such that @xmath62 and @xmath63 , t[b]$ ] are connected , and let @xmath51 be the unique vertex of @xmath34 with a neighbour in @xmath25 . suppose that * the sequence @xmath64 floods @xmath63 $ ] with colour @xmath65 , * the sequence @xmath66 floods @xmath67 $ ] with colour @xmath68 , * at least one move of @xmath66 changes the colour of @xmath51 , and * playing @xmath64 in @xmath47 changes the colour of @xmath51 . then @xmath69 [ compatible - ordering ] we proceed by induction on @xmath70 . note that we may assume without loss of generality that @xmath12 gives a proper colouring of @xmath34 ; otherwise we may contract monochromatic components . suppose @xmath71 . then @xmath64 must change the colour of the only vertex in @xmath34 ( linking it to some @xmath32 ) , and so playing @xmath64 in @xmath47 makes the whole tree monochromatic with colour @xmath65 . thus @xmath72 , and @xmath73 as required , since by assumption @xmath74 . now suppose @xmath75 , so @xmath34 is not monochromatic initially , and assume that the result holds for smaller @xmath34 . set @xmath76 to be the initial segment of @xmath66 , up to and including the move that first makes @xmath34 monochromatic ( in any colour @xmath77 ) , so any final moves that simply change the colour of @xmath34 are omitted . we may , of course , have @xmath78 ( and so @xmath79 ) , if @xmath34 is not monochromatic before the final move of @xmath66 . suppose that @xmath80 does not change the colour of @xmath51 ( which is only possible in the case @xmath81 ) . then playing @xmath80 in @xmath47 to make @xmath34 monochromatic can not change the colour of any vertex in @xmath25 , so if we play @xmath80 in @xmath47 and then play @xmath64 , this will still flood @xmath25 with colour @xmath65 . moreover , as playing @xmath80 has not changed the colour of @xmath51 , playing @xmath64 will still change the colour of @xmath51 , thus linking all of @xmath34 to @xmath25 and so flooding @xmath47 with colour @xmath65 . hence , in this case , we have @xmath82 and so , as we must in this case have @xmath83 , @xmath84 as required . suppose now that @xmath80 does change the colour of @xmath51 . before the final move of @xmath76 there are @xmath85 monochromatic components in @xmath34 ( all but one of which have colour @xmath77 ) , with vertex - sets @xmath86 . for @xmath87 , set @xmath88 to be the subsequence of @xmath76 consisting of moves played in @xmath89 , and note that these subsequences partition @xmath76 . observe also that playing @xmath88 in @xmath90 $ ] gives @xmath89 colour @xmath77 , so @xmath91 . let @xmath92 be the unique component adjacent to @xmath25 , and set @xmath93 $ ] . note that @xmath64 floods @xmath94 $ ] with colour @xmath65 , and @xmath95 floods @xmath96 $ ] with colour @xmath77 . moreover , as playing @xmath64 in @xmath47 changes the colour of @xmath51 , playing @xmath64 in @xmath97 must also change the colour of @xmath51 . also , at least one move from @xmath66 changes the colour of @xmath51 , the unique vertex of @xmath92 with a neighbour in @xmath25 , and this move must belong to @xmath95 . thus we can apply the inductive hypothesis to see that @xmath98 now suppose without loss of generality that @xmath99 is adjacent to @xmath92 . we can then apply corollary [ non - interference ] to @xmath100 $ ] to see that @xmath101 continuing in this way , each time adding an adjacent component , we see that @xmath102 now , if @xmath78 , this immediately gives the desired result , as @xmath79 . otherwise , note that @xmath103 and so @xmath104 as required . in the next result , we exploit this lemma to give a strengthening of corollary [ non - interference ] under additional assumptions . this can be applied to show that , in certain situations , we may assume that _ no _ optimal sequence to flood a subtree can change the colour of any vertex outside the subtree , when played in a larger tree . let @xmath47 be a tree , with colouring @xmath12 from colour - set @xmath13 , and let @xmath105 and @xmath106 be disjoint sutbrees of @xmath47 such that @xmath107 $ ] is connected , and such that * there is a sequence @xmath108 of @xmath109 moves that floods @xmath105 with some colour @xmath110 , * there is a sequence @xmath111 of @xmath112 moves that floods @xmath106 with colour @xmath3 , and that changes the colour of the unique vertex @xmath113 , and * playing @xmath108 in @xmath47 changes the colour of at least one vertex in @xmath106 . then , setting @xmath114 and @xmath115 , we have @xmath116 [ strong - non - int ] note that @xmath108 must change the colour of @xmath2 , so we can apply lemma [ compatible - ordering ] to see that @xmath117,\omega , d ) \leq |s_x| + |s_y| = \alpha + \beta.\ ] ] corollary [ non - interference ] then gives @xmath118 , \omega , d ) + m(t',\omega',d ) \\ & \leq m(t',\omega',d ) + \alpha + \beta,\end{aligned}\ ] ] as required . before proving the main result of this section , we need one further result , relating the number of moves required to flood the same graph with different initial colourings . let @xmath17 be a connected graph , and let @xmath12 and @xmath24 be two colourings of the vertices of @xmath17 ( from colour - set @xmath13 ) . let @xmath119 be the set of all monochromatic components of @xmath17 with respect to @xmath24 , and for each @xmath120 let @xmath121 be the colour of @xmath25 under @xmath24 . then , for any @xmath16 , @xmath122 [ change - colouring ] we proceed by induction on @xmath123 . note that if @xmath124 then the result is trivially true : in this case @xmath119 contains a single monochromatic component @xmath17 , with colour @xmath3 , so we have @xmath125 suppose now that @xmath126 , and let @xmath21 be an optimal sequence of moves to flood @xmath17 with colour @xmath3 , when the initial colouring is @xmath24 . we proceed by case analysis on the final move , @xmath109 , of @xmath21 . first suppose that @xmath17 is already monochromatic before @xmath109 , so this final move just changes the colour of the entire graph to @xmath3 from some colour @xmath110 . in this case , @xmath127 , and so we may apply the inductive hypothesis to see that @xmath128 as required . now suppose that @xmath17 is not monochromatic before @xmath109 , and so this move links monochromatic components @xmath129 . we may assume that @xmath109 changes the colour of @xmath130 from @xmath77 to @xmath3 , and that all the components @xmath131 have colour @xmath3 before @xmath109 . let @xmath88 denote the subsequence of @xmath21 consisting of moves played in @xmath132 , and observe that playing @xmath88 in the isolated subgraph @xmath132 must flood this graph with colour @xmath3 , so @xmath133 . note that , as no move can split a monochromatic component , the sets @xmath134 ( for @xmath87 ) partition @xmath119 . observe that , for @xmath135 , @xmath136 , and so we may apply the inductive hypothesis to see that @xmath137 similarly , the inductive hypothesis gives @xmath138 and so , as @xmath139 , we have @xmath140 now we can apply corollary [ non - interference ] to see that @xmath141 and so @xmath142 completing the proof . using the previous results , we are now ready to prove the key result of this section . let @xmath17 with colouring @xmath12 ( from colour - set @xmath13 ) be the graph corresponding to a @xmath0 flood - it board @xmath34 , let @xmath143 be a connected induced subgraph of @xmath17 , and let @xmath144 and @xmath145 be vertices lying in the leftmost and rightmost columns of @xmath143 respectively . then , for any @xmath16 , there exists a @xmath3-minimal spanning tree @xmath47 for @xmath143 such that @xmath146 . [ leafy - path ] we proceed by induction on @xmath147 . note that the result is trivially true if @xmath148 as the graph is initially monochromatic with colour @xmath3 and so any spanning tree will do . suppose then that @xmath149 . let @xmath21 be an optimal sequence to flood @xmath143 with colour @xmath3 , and suppose that the last move of @xmath21 is @xmath109 . if @xmath143 is monochromatic in some colour @xmath110 before @xmath109 is played , and so this final move just changes the colour of the whole graph to @xmath3 , we see that @xmath150 . thus we may apply the inductive hypothesis to obtain a @xmath77-minimal spanning tree @xmath47 for @xmath143 such that @xmath146 . but then @xmath151 and so @xmath47 is also a @xmath3-minimal spanning tree for @xmath143 . thus we may assume that @xmath143 is not monochromatic immediately before @xmath109 is played . this means that @xmath109 must change the colour of a monochromatic component @xmath25 from some @xmath110 to @xmath3 , where @xmath152 is nonempty and has colour @xmath3 before @xmath109 is played . since @xmath143 is a connected induced subgraph of a @xmath0 board , @xmath152 has at most one component @xmath153 which contains vertices lying in columns to the left of all columns containing a vertex of @xmath25 , and correspondingly at most one component @xmath154 containing vertices lying in columns entirely to the right of @xmath25 . there may additionally be some components @xmath129 of @xmath152 which contain only vertices which lie in the same column as some vertex of @xmath25 . a possible structure for @xmath143 is illustrated in figure [ h - cpts ] . in the remainder of the proof , we will exploit the structure of @xmath152 to define a @xmath3-minimal spanning tree @xmath47 for @xmath143 whose non - leaf vertices lie on @xmath155 . observe that we may have @xmath156 , as illustrated in figure [ l = r ] ; we will deal with this case later , so for the moment we assume that @xmath157 . set @xmath2 ( respectively @xmath158 ) to be any vertex lying in the leftmost ( respectively rightmost ) column of @xmath25 that has at least one neighbour in @xmath153 ( respectively @xmath154 ) ; if @xmath153 ( respectively @xmath154 ) is empty , we set @xmath159 ( respectively @xmath160 ) . if two vertices of @xmath153 lie in the rightmost column of @xmath153 , one of these must be adjacent to @xmath2 , in which case we set this vertex to be @xmath161 ; otherwise @xmath161 is defined to be the unique vertex of @xmath153 that lies in the rightmost column . we define @xmath162 symmetrically , so that @xmath162 lies in the leftmost column of @xmath154 , and so that if there is a choice of vertices of @xmath154 in this column then @xmath162 is the vertex adjacent to @xmath158 . note that @xmath163 and so , by the inductive hypothesis , there exists a @xmath3-minimal spanning tree @xmath164 for @xmath153 such that @xmath165 . similarly , there exists a @xmath3-minimal spanning tree @xmath166 for @xmath154 such that @xmath167 , and a @xmath77-minimal spanning tree @xmath168 for @xmath25 such that @xmath169 . let @xmath64 be an optimal sequence of moves to flood @xmath168 with colour @xmath77 , and @xmath170 and @xmath171 be optimal sequences to flood @xmath164 and @xmath166 respectively with colour @xmath3 . observe that , as well as containing vertices that lie in columns to the left ( respectively right ) of @xmath25 , @xmath153 ( respectively @xmath154 ) may additionally contain some vertices that lie in the same column as a vertex of @xmath25 . we set @xmath172 to be the subtree of @xmath164 induced only by those vertices in @xmath153 that lie in the same column as or to the left of the leftmost vertex of @xmath25 , and define @xmath173 symmetrically . we further define @xmath174 ( respectively @xmath175 ) to be the subsequence of @xmath170 ( respectively @xmath171 ) consisting of moves that change the colour of at least one vertex in @xmath172 ( respectively @xmath173 ) , and note then that @xmath174 ( respectively @xmath175 ) floods @xmath172 ( respectively @xmath173 ) with colour @xmath3 ( implying that @xmath176 , and @xmath177 ) . now set @xmath178 to be the spanning tree for @xmath179 obtained from @xmath168 by adding an edge from every vertex @xmath180 of this subgraph that does _ not _ lie in @xmath25 to the vertex of @xmath25 that lies in the same column as @xmath180 ( and observe that @xmath181 ) . finally , we obtain a spanning tree @xmath47 for @xmath143 by connecting @xmath172 , @xmath173 and @xmath178 . if @xmath182 , we use the edge @xmath183 to connect @xmath172 and @xmath178 ; otherwise we use the edge of @xmath164 with exactly one endpoint in @xmath172 . in either case we must have @xmath184 ) \subseteq p(t , u , v')$ ] . similarly , if @xmath185 then we connect @xmath173 and @xmath178 with @xmath186 , and otherwise use the edge of @xmath166 with exactly one endpoint in @xmath173 . the construction of @xmath47 is illustrated in figure [ t - construction ] . it is clear from the construction that @xmath47 is a spanning tree for @xmath143 , and that @xmath146 ; we will argue that in fact @xmath47 is a @xmath3-minimal spanning tree for @xmath143 . for the rest of the argument , it will be useful to identify two important vertices of @xmath47 . we set @xmath51 to be the last vertex on the path @xmath155 before @xmath25 , and @xmath52 the first vertex after @xmath25 , when this path is traversed from left to right ( as illustrated in figure [ t - construction ] ) ; if @xmath153 ( respectively @xmath154 ) is empty then @xmath51 ( respectively @xmath52 ) is not defined . assuming @xmath153 ( respectively @xmath154 ) is nonempty , let @xmath187 ( respectively @xmath188 ) be the neighbour in @xmath25 of @xmath51 ( respectively @xmath52 ) . note that @xmath189 and @xmath190 . having defined the spanning tree @xmath47 for @xmath143 , we now consider how to flood @xmath47 with colour @xmath3 . first , observe that @xmath191 we will say that a colour @xmath192 is _ autonomous _ either if @xmath193 appears in the initial colouring in one or more component @xmath132 , or else if a move of @xmath170 or @xmath171 is is played in a monochromatic component of colour @xmath193 that does not intersect @xmath172 or @xmath173 . let @xmath194 denote the set of all autonomous colours . note then that , for each @xmath195 that does not have colour @xmath3 initially , at least one of the following must hold in order for @xmath2 to be given colour @xmath3 : 1 . @xmath196 , or 2 . either initially , or after some move of @xmath170 , @xmath51 has colour @xmath197 . let @xmath198 be the set of vertices @xmath195 such that the first statement holds , so @xmath199 if and only if @xmath196 . we then set @xmath200 , and note that the second statement must hold for every @xmath201 . we can apply exactly the same reasoning to @xmath202 ( replacing @xmath51 with @xmath52 ) , and define @xmath203 and @xmath204 analogously . observe that @xmath205 , and @xmath206 . thus , by ( [ |s| ] ) , we see that @xmath207 in order to flood @xmath47 with colour @xmath3 , we will first play @xmath64 , flooding @xmath25 , and then repeatedly change the colour of @xmath25 to cycle through all colours in @xmath194 . note that these first @xmath208 moves create a monochromatic component @xmath209 containing @xmath210 . there are now three cases to consider , depending on whether none , both or one of @xmath211 and @xmath203 are non - empty . first suppose that @xmath212 . note in this case that our first @xmath208 moves make @xmath178 monochromatic in some colour , so @xmath213 . thus we can apply corollary [ non - interference ] to see that @xmath214 as required . now suppose that exactly one of @xmath211 and @xmath203 is nonempty , and without loss of generality suppose that @xmath215 . we claim that playing @xmath64 in @xmath47 does not change the colour of any vertex in @xmath172 . indeed , if this sequence does change the colour of a vertex in @xmath172 , it must change the colour of @xmath51 , and this colour change will be due to moves in @xmath64 changing the colour of @xmath187 . thus , if we played @xmath64 in the tree @xmath97 , obtained by connecting @xmath164 , @xmath168 and @xmath166 with the edges @xmath216 and @xmath217 , it would would still change the colour of @xmath51 ( which is the unique vertex of @xmath164 adjacent to @xmath168 ) . however , as @xmath215 , we know that @xmath170 must change the colour of @xmath51 , and so by proposition [ strong - non - int ] ( setting @xmath218 , @xmath219 , @xmath220 and @xmath221 ) we would have @xmath222 , implying ( by theorem [ spanning - tree ] ) that @xmath223 , a contradiction . we may further assume that then cycling @xmath25 through all colours in @xmath194 does not change the colour of any vertex in @xmath172 : if @xmath224 we can choose this to be the last colour we play in @xmath25 , and so our sequence will link @xmath25 to @xmath51 but will not change the colour of @xmath51 ( or therefore of any other vertex in @xmath172 ) . next , if playing @xmath64 and cycling through the colours of @xmath194 has not already linked @xmath209 to @xmath51 , we play one further move to give @xmath209 the same colour as @xmath51 . since the sequence of moves we play up to this point does not change the colour of any vertex in @xmath172 , we can now play the sequence @xmath174 to give every vertex in @xmath172 colour @xmath3 . as @xmath51 is in the same monochromatic component as @xmath209 this will also give all vertices in @xmath209 colour @xmath3 . moreover , playing this sequence will at some point give @xmath51 , and hence @xmath209 , every colour in @xmath225 , and so will link every vertex in @xmath203 to @xmath209 and ultimately give these vertices colour @xmath3 . thus , playing @xmath64 , cycling through @xmath194 , if necessary linking @xmath51 to @xmath209 , and then playing @xmath174 will flood all the vertices of @xmath226 with colour @xmath3 ( as @xmath227 ) , so we see that @xmath228 but then , once again , we can apply corollary [ non - interference ] to see that @xmath229 as required . for the final subcase , we suppose that @xmath230 . we begin once again by playing @xmath64 , cycling @xmath25 through all colours in @xmath194 , and then ( if required ) playing an additional move to change the colour of the monochromatic component containing @xmath25 to be the same as @xmath51 ; as before we may assume that these initial moves do not change the colour of any vertex in @xmath174 . note that , as @xmath215 , the colour of @xmath51 must change at least once when we play @xmath174 in @xmath172 . set @xmath112 to be the last move in @xmath174 to change the colour of @xmath51 , and note then that @xmath112 must change the colour of some component @xmath231 , containing @xmath51 , to @xmath3 . set @xmath232 , and let @xmath233 be the subsequence of @xmath174 consisting of moves played in @xmath231 ( so @xmath233 floods @xmath231 with colour @xmath3 , and @xmath112 is the final move of @xmath233 ) . as @xmath231 is monochromatic before @xmath112 , playing @xmath234 in @xmath231 must flood this component with some colour @xmath235 . observe also that the sequence @xmath236 must , when played in the forest @xmath237 , give every vertex of @xmath237 colour @xmath3 . suppose that , after playing @xmath64 and linking @xmath25 to @xmath51 , we then play @xmath234 . this will ensure that @xmath51 and hence @xmath25 at some point receives every colour in @xmath238 ( except possibly d ) , so every vertex in @xmath211 is either linked to @xmath25 or has colour @xmath3 ( in which case it will certainly end up with colour @xmath3 , as its colour can only change if it is linked to another vertex which will ultimately be given colour @xmath3 ) . note that we now have a monochromatic component @xmath34 that contains @xmath25 , @xmath231 and all vertices of @xmath239 that do not initially have colour @xmath3 . we claim that the sequence of moves we play up to this point can not change the colour of any vertex in @xmath173 . to prove the validity of this claim , first observe that @xmath175 , played in @xmath166 , floods a subtree @xmath240 of @xmath166 with colour @xmath3 , where @xmath241 . now set @xmath242 to be the spanning tree for @xmath143 obtained by connecting @xmath240 and @xmath243 with the edge @xmath217 . it is clear that , if our sequence of moves so far changes the colour of any vertex in @xmath173 when played in @xmath47 , playing the same sequence in @xmath242 would change the colour of @xmath244 . however , as @xmath245 , we also know that @xmath171 changes the colour of @xmath52 . note also that all vertices of @xmath246 that do not belong to @xmath237 have colour @xmath3 initially , so @xmath247 moves suffice to flood @xmath246 with colour @xmath3 . we can now apply proposition [ strong - non - int ] , setting @xmath248 , @xmath108 to be the sequence of moves we have played up to this point , @xmath249 and @xmath250 to see that @xmath251 theorem [ spanning - tree ] would then imply that @xmath252 a contradiction . if the monochromatic component containing @xmath25 does not already have the same colour as @xmath52 , we now play one further move to link it to this vertex ( and note that such a move will not change the colour of any vertex in @xmath173 ) . hence , if we now play @xmath175 , this will flood @xmath173 with colour @xmath3 ; as @xmath25 and @xmath52 lie in the same monochromatic component before these moves are played , this sequence will also give every vertex in the same monochromatic component as @xmath25 colour @xmath3 . moreover , linking @xmath25 to @xmath52 and playing @xmath175 will at some point give @xmath52 , and hence @xmath25 , every colour in @xmath225 , and so all vertices in @xmath203 will be linked to @xmath25 and thus end up with colour @xmath3 . hence this sequence of moves gives every vertex in @xmath253 colour @xmath3 , and so we have @xmath254 finally , we apply corollary [ non - interference ] to give @xmath255 as required . this completes the proof in the case that @xmath157 . it remains to consider the case in which @xmath156 , as in figure [ l = r ] . we define @xmath47 exactly as before ( as shown in figure [ t , l = r ] ) , and again identify the important vertices @xmath51 and @xmath52 . the previous reasoning only fails in the case @xmath156 because it is not necessarily true , in this case , that @xmath174 floods @xmath172 with colour @xmath3 and @xmath175 floods @xmath173 with colour @xmath3 . however , by considering more carefully the sequence of moves that floods @xmath152 , we are able to deal with this problem . if @xmath51 and @xmath52 belong to the same monochromatic component @xmath256 of @xmath257 , with colour @xmath258 , under the initial colouring @xmath12 , then we can flood @xmath259 $ ] by playing @xmath64 and then changing the colour of @xmath25 to @xmath258 , so @xmath260 , \omega , d_{xy } ) \leq |s_a| + 1 $ ] . let @xmath24 be the colouring of @xmath47 which agrees with @xmath12 on every vertex in @xmath164 , and gives every vertex in @xmath25 colour @xmath258 . then @xmath47 with colouring @xmath24 is equivalent ( when monochromatic components are contracted ) to @xmath164 with colouring @xmath12 , implying that @xmath261 . we can then apply lemma [ change - colouring ] to give @xmath262 , \omega , d_{xy } ) \\ & \leq |s_l| + |s_a| + 1 \\ & = |s| \\ & = m(h,\omega , d).\end{aligned}\ ] ] so we may assume that @xmath51 and @xmath52 do not belong to the same monochromatic component initially . let @xmath263 be the initial segment of @xmath170 up to and including the move that first links @xmath51 and @xmath52 ; let @xmath256 be the monochromatic component of @xmath264 that contains @xmath51 and @xmath52 at this point , and suppose that it has colour @xmath265 and that @xmath266 moves of @xmath263 are played in @xmath256 . we now consider flooding the subtree @xmath259 $ ] with colour @xmath265 . note that the subsequence of @xmath263 consisting of moves played in @xmath267 must in this case flood @xmath267 with colour @xmath265 , and similarly the subsequence consisting of moves played in @xmath268 must flood @xmath268 with colour @xmath265 . thus , applying exactly the same arguments as in the case that @xmath157 , we see that @xmath269 now set @xmath24 to be the colouring of @xmath61 that agrees with @xmath270 on @xmath164 and gives all vertices of @xmath25 colour @xmath265 . note that @xmath47 with colouring @xmath24 is equivalent ( when monochromatic components are contracted ) to @xmath164 with colouring @xmath270 , and so @xmath271 . if @xmath119 is the set of monochromatic components of @xmath47 with respect to @xmath24 , and each @xmath120 has colour @xmath121 under this colouring , then observe that @xmath272 } } m(a,\omega , c_a ) \leq |s'| - k,\ ] ] and so @xmath273 we can now apply lemma [ change - colouring ] to see that @xmath274 in this case also , completing the proof . in our analysis of the algorithm in the next section , we will need one additional result : we show in the next lemma that any tree can be flooded by an optimal sequence in which no moves are played at leaves . let @xmath47 be any tree , and @xmath12 a colouring of the vertices of @xmath47 . then there exists a sequence of moves @xmath21 , of length @xmath275 , which makes @xmath47 monochromatic and in which all moves are played in @xmath48 . [ no - leaf - moves ] let @xmath276 be any optimal sequence to flood @xmath47 , and set @xmath277 to be the subsequence of @xmath276 consisting of moves that change the colour of a vertex in @xmath48 . note that we may assume without loss of generality that all moves of @xmath277 are played in @xmath48 . note further that @xmath278 contains only moves played at leaves , and let @xmath279 be the set of leaves in which moves of @xmath278 are played . observe that playing @xmath277 in @xmath47 will make @xmath280 monochromatic , and so we can flood the entire tree by playing a sequence @xmath21 which consists of @xmath277 followed by a further @xmath281 moves , cycling through the colours still present in leaves of @xmath47 ( playing all moves in @xmath48 ) . thus @xmath282 however , it is clear that @xmath283 , as @xmath278 contains at least one move played at each vertex in @xmath279 . hence we see that @xmath284 , and so @xmath21 is an optimal sequence to flood @xmath47 in which all moves are played in @xmath48 , as required . in this section we describe our algorithm to solve @xmath4-free - flood - it on @xmath0 boards , and use results from the previous section to prove its correctness . we begin with some further definitions . for any section @xmath46 $ ] , we define @xmath285 $ ] to be the set of all spanning trees for @xmath46 $ ] . given any @xmath0 flood - it board @xmath34 , corresponding to a graph @xmath17 with colouring @xmath12 from colour - set @xmath13 , we define a set of vectors @xmath286 , where @xmath287 \text { is a section } , \\ & r_1 , r_2 \in v(b[b_1,b_2 ] ) , \\ & \exists t \in \mathcal{t}[b_1,b_2 ] \text { such that } \operatorname{trunk}(t ) \subseteq p(t , r_1,r_2 ) , \\ & r_1 \text { incident with } b_1 , r_2 \text { incident with } b_2 , \\ & d \in c , \\ & i \subseteq c \}.\end{aligned}\ ] ] note that there always exists a tree @xmath288 $ ] such that @xmath289 unless one of the following holds : 1 . there is more than one vertex of @xmath46 $ ] lying strictly to the left of @xmath290 or strictly to the right of @xmath291 , or 2 . there is exactly one vertex of @xmath46 $ ] lying strictly to the left of @xmath290 ( respectively to the right of @xmath291 ) , which is not adjacent to @xmath290 ( respectively @xmath291 ) and whose neighbour in the same column as @xmath290 ( respectively @xmath291 ) has no neighbour in @xmath46 $ ] other than @xmath290 ( respectively @xmath291 ) . thus we can check whether this condition is satisfied in constant time . we now introduce a function @xmath292 which is closely related to the minimum number of moves required to flood a @xmath0 board . for any @xmath293 we define @xmath294 to be the minimum , taken over all @xmath288 $ ] such that @xmath289 , of the number of moves that must be played in @xmath295 to flood @xmath295 with colour @xmath3 , and link to @xmath295 all leaves of @xmath47 that do not have colours from @xmath296 . it follows immediately from lemmas [ leafy - path ] and [ no - leaf - moves ] that @xmath297 our algorithm in fact computes recursively a function @xmath298 , with the same parameters as @xmath292 . we will argue that , for every @xmath299 , @xmath300 and hence that it suffices to compute all values of @xmath298 in order to calculate @xmath19 . the first step of the algorithm is to initialise certain values of @xmath298 to zero . we set @xmath301 if and only if , under the initial colouring , there exists a @xmath290-@xmath291 path of colour @xmath3 in @xmath46 $ ] , and all vertices in @xmath46 $ ] that do not lie on this path are adjacent to the path and have colours from @xmath302 . all other values of @xmath303 are initially set to infinity . note that , under this definition , @xmath304 , and that for each @xmath299 we can easily determine in time @xmath305 whether @xmath298 should be initialised to zero or infinity . in order to define further values of @xmath298 , we introduce two more functions . first , for any @xmath306 , we set @xmath307 we also define , for any @xmath299 , @xmath308 finally , we set @xmath309 for the reasoning below , it will be useful to introduce another function @xmath310 , taking the same parameters as @xmath292 and @xmath298 . for any @xmath306 , we define @xmath311 in the following two lemmas , we show that @xmath300 for all @xmath299 , as claimed . we begin by demonstrating that @xmath303 gives an upper bound for @xmath294 . let @xmath17 with colouring @xmath12 ( from colour - set @xmath13 ) be the coloured graph corresponding to a @xmath0 flood - it board @xmath34 . then @xmath312 for all @xmath306 . [ f*>=f ] we proceed by induction on @xmath313 . recall that we have equality between @xmath294 and @xmath303 whenever @xmath314 , so certainly the base case for @xmath315 must hold . assume therefore that @xmath316 , and that the result holds for all @xmath317 with @xmath318 . since @xmath316 , we must have @xmath319 . suppose first that @xmath320 . then , for some @xmath110 , @xmath321 but then we know , by definition of @xmath292 , that there exists @xmath288 $ ] and a sequence @xmath21 of @xmath322 moves , all played in @xmath295 , which , when played in @xmath47 , floods @xmath323 with colour @xmath77 and links all leaves to @xmath295 except possibly those with colours from @xmath302 . by appending one further move to @xmath21 , which changes the colour of @xmath295 to @xmath3 , we obtain a sequence @xmath263 of length @xmath324 ( with all moves played in @xmath295 ) which , when played in @xmath47 , floods @xmath295 with colour @xmath3 and is such that all leaves of @xmath47 not linked to @xmath295 by @xmath21 have colours from @xmath296 . hence @xmath325 , and so @xmath326 , as required . now suppose that @xmath327 . then , by definition of @xmath328 , there exists a border @xmath45 with @xmath329 and an edge @xmath330 such that @xmath331 and @xmath332 note that @xmath333 and @xmath334 are both strictly smaller than @xmath335 , so by the inductive hypothesis we have @xmath336 by definition of @xmath292 , there exist trees @xmath337 $ ] and @xmath338 $ ] , and sequences @xmath95 and @xmath339 of length @xmath340 and @xmath341 respectively , such that ( for @xmath342 ) all moves of @xmath88 are played in @xmath343 and @xmath88 floods @xmath343 with colour @xmath3 , additionally linking all leaves of @xmath344 to @xmath343 except possibly those with colours from @xmath296 . now set @xmath345 . it is clear that @xmath288 $ ] , and moreover that @xmath289 . suppose @xmath346 and @xmath347 are the subtrees of @xmath97 and @xmath242 respectively that are given colour @xmath3 by @xmath95 and @xmath339 , and set @xmath348 . note that @xmath349 and that @xmath350 , so @xmath351 . we can then apply corollary [ non - interference ] to see that @xmath352 completing the proof . next we show that the reverse inequality also holds . let @xmath17 with colouring @xmath12 ( from colour - set @xmath13 ) be the coloured graph corresponding to a @xmath0 flood - it board @xmath34 . then @xmath353 for all @xmath306 . [ f*<=f ] we proceed by induction on @xmath294 , noting again that we have equality in the base case for @xmath354 . suppose that @xmath355 , and that the result holds for @xmath317 whenever @xmath356 . by definition , there exists a tree @xmath288 $ ] and a sequence @xmath21 of length @xmath357 such that @xmath289 , all moves of @xmath21 are played in @xmath295 , and @xmath21 floods @xmath295 with colour @xmath3 , leaving only leaves with colours from @xmath296 not linked to @xmath295 . we proceed by case analysis on @xmath109 , the final move of @xmath21 . suppose first that @xmath295 is already monochromatic before @xmath109 , and that this final move just changes its colour to @xmath3 from some @xmath110 ( possibly flooding some additional leaves of colour @xmath3 in the process ) . in this case it is clear that @xmath358 and so we can apply the inductive hypothesis to see that @xmath359 . but then , by definition of @xmath360 , we know that @xmath361 as required . so we may assume that @xmath295 is not monochromatic before @xmath109 : it may have either two or three monochromatic components . suppose first that @xmath295 has exactly three monochromatic components before @xmath109 is played , @xmath362 , @xmath363 and @xmath364 ; we may assume that @xmath362 and @xmath364 have colour @xmath3 before @xmath109 , and that this final move gives @xmath363 colour @xmath3 to flood the entire path . for @xmath365 , set @xmath88 to be the subsequence of @xmath366 consisting of moves played in @xmath367 , and set @xmath368 to be @xmath367 together with all leaves of @xmath47 that lie in the same column as a vertex of @xmath367 or whose only neighbour on @xmath295 is in @xmath367 . note that that @xmath369 , @xmath370 and @xmath371 partition the vertex set of @xmath47 , and that @xmath95 , @xmath339 and @xmath372 partition @xmath366 . we may assume without loss of generality that @xmath373 and @xmath374 . observe that there must exist borders @xmath45 and @xmath375 , with @xmath376 , such that @xmath377 $ ] , @xmath378 $ ] and @xmath379 $ ] . set @xmath380 to be the edge of @xmath47 such that @xmath381 , @xmath382 and @xmath383 the edge of @xmath47 such that @xmath384 and @xmath385 . note that @xmath386 \in \mathcal{t}[b_1,b]$ ] , @xmath387 \in \mathcal{t}[b , b']$ ] , and @xmath388 \in \mathcal{t}[b',b_2]$ ] , and moreover that we have @xmath389 ) \subseteq p(t[a_1],r_1,x_1)$ ] , @xmath390 ) \subseteq p(t[a_2],x_2,y_1)$ ] and @xmath391 \subseteq p(t[a_3],y_2,r_2)$ ] . observe also that @xmath95 is a sequence of moves played in @xmath392,r_1,x_1)$ ] that floods @xmath392,r_1,x_1)$ ] with colour @xmath3 and links all leaves , except possibly those with colours from @xmath296 , to @xmath392,r_1,x_1)$ ] , so we must have @xmath393 . similarly , we see that @xmath394 and @xmath395 . since @xmath396 , we can apply the inductive hypothesis to see that @xmath397 and @xmath398 the inductive hypothesis also gives @xmath399 and we can then apply the definition of @xmath360 to see that @xmath400 now we can apply the definition of @xmath298 to see that @xmath401 as required . for the remaining case , in which @xmath295 has exactly two monochromatic components before @xmath109 , we can use the same reasoning as in the previous case for three components to show that we must once again have @xmath402 , completing the proof . the final step to is to show that all values of @xmath298 can be computed in time @xmath403 . for any @xmath0 flood - it board @xmath34 , the function @xmath303 can be computed , for all @xmath299 , in time @xmath404 . [ complexity ] we compute values of @xmath298 recursively using a dynamic programming technique . our table has one entry for each pair of borders , for each possible vertex incident with each of the borders , for each colour in the colour - set and for each possible subset of colours , so the total number of entries is at most @xmath405 the table is initialised by setting all values to either zero or infinity , and for each entry we can determine which of these values it should take in time at most @xmath305 , so we can initialise the entire table in time @xmath406 . the next step is to apply the recursive definition of @xmath298 repeatedly to all entries in the table that are not already set to zero . each time we apply this definition to a single entry , we take the minimum of at most @xmath407 values ( one for each choice of colour , plus one for each combination of a border and a pair of adjacent vertices on either side ) , each a combination of at most two other entries in the table , so each entry can be calculated in time @xmath407 . we can therefore perform one iteration in which we apply the definition to each non - zero entry in the table in time @xmath408 . note that once we have initialised the table , we have the correct value of @xmath303 for any @xmath409 such that @xmath410 . moreover , the value of @xmath303 depends only on values of @xmath411 where @xmath318 , so after @xmath266 iterations we will have correctly computed the value of @xmath303 for all @xmath409 with @xmath412 . note that for every @xmath299 , @xmath413 , as there are at most @xmath414 borders in total , and no more than @xmath415 moves can be required to flood a graph with at most this many vertices , so @xmath416 iterations are sufficient to guarantee we have computed all values of @xmath298 correctly . thus , we can compute all values of @xmath303 for @xmath299 in time @xmath417 , as required . we now combine the previous three results to give the proof of our main theorem . recall that , from the definition of @xmath292 and lemmas [ leafy - path ] and [ no - leaf - moves ] , @xmath297 thus , in order to compute @xmath19 in time @xmath417 , it suffices to compute all relevant values of @xmath292 in time @xmath417 . however , we know from lemmas [ f*>=f ] and [ f*<=f ] that @xmath418 for all @xmath299 , and from proposition [ complexity ] we know that we can compute @xmath303 for all @xmath299 in time @xmath417 . this completes the proof of the theorem . in this section we prove the following theorem . free - flood - it remains np - hard when restricted to @xmath0 boards . [ 2xnnp ] this is somewhat surprising , as we have seen in the previous section that @xmath4-free - flood - it can be solved in polynomial time on @xmath0 boards , while @xcite gives a linear time algorithm to solve fixed flood it in this situation . we demonstrate here that the problem is almost certainly not in * p * if we remove both these restrictions ( that moves are always played at the same vertex , or the number or colours is bounded ) . this is the first class of graphs for which such a result has been shown . the proof is by means of a reduction from vertex cover , shown to be np - hard by karp in @xcite . given a graph @xmath11 , we construct a @xmath0 flood - it board @xmath419 as follows . suppose @xmath420 . for each edge @xmath421 we construct the gadget @xmath422 , as illustrated in figure [ gadget - g_e ] . we will refer to the single - square components incident with the bottom edge in @xmath423 as _ @xmath423 is then embedded in the larger gadget @xmath424 , as shown in figure [ gadget - g_e ] . distinct colours @xmath425 are used for each @xmath426 , where @xmath427 . we then obtain the board @xmath419 by placing these gadgets @xmath424 in a row , as illustrated in figure [ board-2xn - inf - b ] . observe that we can take @xmath428 . let us also set @xmath429 . [ h ] , title="fig : " ] [ h ] , title="fig : " ] [ h ] in the two following lemmas , we show that we can flood this board @xmath34 in @xmath430 steps if and only if @xmath17 has a vertex cover of size at most @xmath266 . if @xmath17 has a vertex cover of size at most @xmath266 , then we can flood the board @xmath419 in @xmath430 steps . [ vc=>strat ] first observe that , if @xmath431 , then with @xmath432 moves we can flood the gadget @xmath424 , except for a single island of colour @xmath433 , so that it is monochromatic in colour @xmath434 : first play a single move to make all of @xmath423 except for a single island monochromatic , then play colours @xmath425 in this central component . ignoring the islands for the moment , the components corresponding to each @xmath424 now have distinct colours , so we can link these components with a minimum of @xmath435 moves . finally , we need to flood the islands , and this requires exactly @xmath436 moves . but we know that @xmath17 has a vertex cover of size at most @xmath266 , say @xmath437 . by the definition of a vertex cover , if the gadget @xmath423 uses colours @xmath144 and @xmath2 , then at least one of @xmath438 . so for each @xmath424 , we may choose to leave an island of colour @xmath3 where @xmath439 . following this strategy , we are left in the final stage with islands of at most @xmath266 distinct colours , and can flood these in @xmath266 steps ( by cycling through each colour in turn in the external monochromatic component ) . hence we can flood @xmath419 in @xmath430 steps . if we can flood @xmath419 in @xmath440 steps ( for some @xmath441 ) , then @xmath17 has a vertex cover of size at most @xmath266 . [ strat=>vc ] suppose the sequence @xmath21 floods @xmath419 , where @xmath442 . observe that , if we contract monochromatic components of the coloured graph corresponding to @xmath419 , we obtain a tree @xmath47 . let @xmath443 be the unique path in @xmath443 joining the two vertices in @xmath47 that correspond to the monochromatic components incident with opposite ends of the board and note that , by lemma [ no - leaf - moves ] , we may assume that all moves of @xmath21 are played in @xmath444 . moreover , @xmath21 must flood @xmath443 when played in this isolated path . we will say that a component of colour @xmath3 is _ eliminated _ by the move @xmath109 if @xmath109 changes the colour of that component , linking it to an adjacent component of colour @xmath445 . we say that @xmath109 _ eliminates the colour @xmath3 _ if it removes the last component of colour @xmath3 remaining in the graph . suppose that , for some @xmath15 , a single move @xmath446 eliminates two components @xmath362 and @xmath363 of @xmath443 that both have colour @xmath2 initially . @xmath362 and @xmath363 can not belong to the same gadget @xmath424 so , for @xmath109 to eliminate them both , the moves played in @xmath21 before @xmath109 must create a single monochromatic component @xmath25 containing both @xmath362 and @xmath363 . such a component @xmath25 must contain @xmath447 different colours under the original colouring , and so @xmath21 must include at least @xmath448 moves played in this section of the path before @xmath109 ( so in total at least @xmath449 moves of @xmath21 are played in @xmath25 ) . but there are at least @xmath450 colours on the path outside @xmath25 ( as at least @xmath451 of the colours in @xmath25 must also appear outside @xmath25 ) , and all but one of these must be eliminated by moves played outside @xmath25 . this gives @xmath452 a contradiction . so we may assume that no move in @xmath21 eliminates more than one component that originally has colour @xmath15 . now consider the leaves of @xmath47 , and let @xmath453 be the set of moves in @xmath21 that eliminate the second leaf in each @xmath423 . suppose that one leaf in @xmath423 has already been eliminated , and that the move @xmath454 removes the second leaf . since one leaf has already been eliminated , no components in @xmath455 which originally had a colour @xmath15 still have colour @xmath2 . suppose that @xmath109 reduces the number of monochromatic components on @xmath443 . by the reasoning above , if @xmath109 links @xmath423 to another component outside @xmath424 that originally had colour @xmath2 , we would have a contradiction with @xmath456 , so in fact @xmath109 must link @xmath423 to a component in @xmath424 whose colour was previously changed to @xmath2 by some move @xmath112 ; such a move @xmath112 could not decrease the number of monochromatic components of @xmath443 . thus , for every @xmath454 , there is at least one move of @xmath21 that does not decrease the number of monochromatic components of @xmath443 . hence we see that @xmath457 and so @xmath458 . however , we know that @xmath453 eliminates at least one leaf from every @xmath423 , and clearly each move in @xmath453 can eliminate leaves of only one colour . hence there exists some set @xmath459 such that @xmath460 and at least one leaf in each @xmath423 has a colour from @xmath461 . in other words , @xmath462 , and for every edge @xmath463 , @xmath464 , so @xmath461 is in fact a vertex cover for @xmath17 of size at most @xmath266 . the reduction from vertex cover is immediate from lemmas [ vc=>strat ] and [ strat=>vc ] . we have demonstrated an algorithm which shows that the problem @xmath4-free - flood - it , restricted to @xmath0 boards , is fixed parameter tractable with parameter @xmath4 , and on the other hand we have shown that free - flood - it remains np - hard in this setting . this answers an open question from @xcite , in which clifford , jalsenius , montanaro and sach showed that fixed - flood - it can be solved in time @xmath305 on such boards . our results therefore give the first example of a class of graphs on which the complexity status of the fixed and free versions of the game differ . david arthur , raphal clifford , markus jalsenius , ashley montanaro , and benjamin sach , the complexity of flood filling games , in paolo boldi and luisa gargano , editors , _ fun _ , volume 6099 of _ lecture notes in computer science _ , springer , isbn 978 - 3 - 642 - 13121 - 9 , 2010 , pages 307 - 318 . kitty meeks and alexander scott , spanning trees and the complexity of flood - filling games , in kranakis , krizanc and luccio , editors , _ fun _ , volume 7288 of _ lecture notes in computer science _ , springer , isbn 978 - 3 - 642 - 30346 - 3 , 2012 , pages 282 - 292 .

in the context of quantum information and quantum technology the idea of reading classical data by means of quantum states arises quite naturally @xcite . in general , the standard implementations of reading are based on optical technologies : the task is the readout of a digital optical memory , where information is stored by means of the optical properties of the memory cells that are in turn probed by shining light , e.g. a laser beam , on them . the probing light is usually denoted as the _ transmitter_. interesting features arise in the regime in which the transmitter has to be treated quantum mechanically . the maximum rate of reliable readout defines the quantum reading capacity @xcite . the latter can overcome the classical reading capacity , obtained by probing with classical light , in several relevant settings . the ( possibly quantum ) transmitter that is needed to extract the encoded information is prepared in some initial state . by scanning a particular cell the transmitter changes its properties in a way depending on the cell . the task is to recognize which cell occurs based on the output state of the transmitter after it has been detected and measured . therefore the problem of reading reduces to the problem of distinguishing the output states of the transmitter . in such optical settings one needs to consider two main coding protocols depending on the trade off between energy and coherence of the transmitters and the channels that are being used the first protocol is the so - called _ amplitude shift keying _ ( ask ) in which the changes in the state of the transmitter are caused by the cell - dependent losses in the intensity of the transmitted signal @xcite . the second main protocol is the so - called _ phase shift keying _ ( psk ) @xcite . this is a type of coding which does not produce energy dissipation . on the other hand , it requires a very high coherence of the transmitter , a feature that might be realized in realistic implementations @xcite . if the transmitter is quantum , the cells play the role of effective quantum channels . the ask protocol then corresponds to a dissipative channel coding , while the psk protocol is a particular case of unitary coding corresponding to a unitary channel . within the ask protocol , it can be shown that in the low - energy regime there is an energy threshold above which the maximally entangled transmitter , i.e. a two - mode squeezed state , yields a better reading efficiency than any of the classical states with the same energy @xcite . the general result is still valid in the presence of some noise - induced decoherence . within the ask protocol , coding is then realized by local channels corresponding to cells with different reflectivities . in the psk protocol , the coding is realized by means of local unitary operations , specifically local phase shifts @xcite . in the ideal , noise - free protocol the transmitter is taken to be in a pure gaussian quasi - bell state , i.e. a bell - type superposition of quasi - orthogonal coherent states . in this scheme , the resulting quantum advantage is absolute , in the sense that quantum reading of the classical information encoded via a phase shift of @xmath0 is achieved with vanishing error , while any classical state of the transmitter always yields a finite error probability . in both the ask and psk protocols the transmitter is assumed to be a bipartite system such that only one part of it scans the memory cell . this choice is motivated by the fact that it maximizes distinguishability at the output when the state of the transmitter is quantum . as already mentioned , the reading efficiency is characterized by the probability of error . information is encoded in binary memory cells with indices @xmath1 and @xmath2 . it is thus written using only two local channels that are assumed to occur with equal _ a priori _ probabilities . given the bi - partite input transmitter @xmath3 , the two possible output states will be denoted by @xmath4 and @xmath5 . the probability of error in distinguishing the two output states when reading a memory cell by means of the same input @xmath3 is given by the well - known helstrom formula @xcite : @xmath6 where @xmath7 is the trace distance , with @xmath8 . with our normalization convention , the trace distance ranges from @xmath1 to a maximum of @xmath9 for orthogonal pure states . in the original reading protocols the goal is to minimize @xmath10 over the set of possible transmitter states @xmath3 at fixed encoding in the memory cells @xcite . the problem is thus dependent on the type of memory device being used . here instead we wish to provide a device - independent characterization of a given transmitter by considering the worst - case scenario that maximizes the probability of error @xmath10 over all possible codings . once the worst - case scenario is identified , one can then compare different classes of transmitter states to identify the ones that minimize the maximum probability of error @xmath11 . we will show that the maximum probability of error @xmath11 is a monotonically decreasing function of the amount of quantum correlations present in the transmitter state @xmath3 , as quantified by a recently introduced measure of quantum correlations , the so - called discord of response @xcite in its gaussian version @xcite ( for general reviews on quantum correlations and distinguishability of quantum states and on discord - like correlations see @xcite ) . as a consequence , every state with non - vanishing discord of response is able to read any type of memory device with maximal @xmath12 . on the other hand , for each classical transmitter , i.e. for transmitter with vanishing discord of response , there will always exist at least one memory device which is completely invisible , i.e. for which @xmath13 . in these considerations we exclude the situations when two channels are chosen to be arbitrarily similar . in this case the probability of error always approaches @xmath14 independently on the chosen transmitter . in sec . [ errorprobdiscord ] we derive the exact analytical relation between the maximum probability of error and the gaussian discord of response . in sec . [ gausstrans ] we discuss the properties of classical and quantum gaussian transmitters , comparing squeezed - thermal , thermal - squeezed and coherent - thermal states ; in subsection [ gausstransa ] we derive upper and lower bounds on the maximum probability of error , and in subsection [ phaseshiftreading ] we identify the unitary coding that maximizes the quantum chernoff bound , namely the upper bound on the maximum probability of error , for the classes of quantum gaussian transmitters considered . the unitary channel which maximizes the bounds on the probability of error turns out to be a particular psk coding , namely the unique traceless one , which is realized for a @xmath15 phase shift . in sec . [ comparisonclass ] we compare the performance of gaussian quantum states of light with classical states ( coherent thermal states ) . we show that strongly discordant squeezed thermal states possess a higher reading efficiency than the corresponding classical states of light , that is non - discordant gaussian coherent thermal transmitters with the same total number of photons ( fixed energy ) . this realizes an important instance of noise - enhanced quantum advantage over the corresponding noisy classical resources . in sec . [ comparisonquant ] we compare different families of discordant gaussian states , the squeezed thermal and the thermal squeezed states at fixed total number of photons or squeezing . while for both classes of states the entanglement obviously decreases with increasing thermal noise , we show that for squeezed thermal states the discord is an increasing function of the number of thermal photons at fixed squeezing , while the opposite holds for thermal squeezed states . moreover , for squeezed thermal transmitters , the quantum chernoff bound is independent of thermal noise . as a consequence , this type of transmitter plays a privileged role in the considered class of quantum gaussian resources because the associated quantum efficiency is either enhanced or unaffected by increasing the thermal noise . thus squeezed thermal transmitters realize an instance of noise - enhanced or noise - independent quantum resources at fixed squeezing . both in the classical - to - quantum comparison and in the comparison of different quantum resources , the key enhancement of quantum advantage is realized in the situation of strongest asymmetry of the distribution of thermal noise among the field modes : _ local _ noise enhancement leads to _ global _ enhancement of quantum correlations . the main results are summarized and some outlook perspectives for future work are discussed in sec . [ summary ] . detailed calculations and auxiliary reasonings are reported in four appendixes . in protocols of quantum reading with unitary coding the two local channels acting on the input probe state @xmath3 are unitary , and denoted as @xmath16 and @xmath17 . therefore , in this type of protocol @xmath18 and @xmath19 , so that the probability of error reads @xmath20 since the trace norm is invariant under local unitary transformations , one has equivalently @xmath21 where @xmath22 and @xmath23 is still a local unitary transformation acting on the transmitted subsystem @xmath24 . the absolute upper bound for the probability of error is thus @xmath14 , corresponding to a situation in which there is no way to distinguish the two output states and therefore the memory device becomes completely invisible to the transmitter . in general , computing the trace distance proves to be extremely challenging @xcite , even more so for gaussian states of infinite - dimensional continuous - variable systems @xcite . therefore one has to look for analytically computable _ a priori _ upper and lower bounds . a natural upper bound on the probability of error , eq . ( [ helstrom ] ) , in distinguishing two states @xmath25 and @xmath26 occurring with the same probability is provided by the quantum chernoff bound @xmath27 @xcite : @xmath28 \ ; . \label{qcb}\ ] ] if the states @xmath25 and @xmath26 are not arbitrary , but they are qubit - qudit states related by a local , single - qubit unitary transformation , as in the case of the quantum reading protocol with unitary coding , for which @xmath29 and @xmath30 , then the quantum chernoff bound @xmath27 is achieved for @xmath31 in eq . ( [ qcb ] ) , as discussed in appendix [ chbm ] : @xmath32 \ ; , \label{qcblu}\ ] ] and its expression coincides with the quantum bhattacharyya coefficient @xcite , which provides an upper bound on @xmath27 for arbitrary quantum states . the same result , eq . ( [ qcblu ] ) , holds for gaussian states related by traceless local symplectic transformations , as discussed in the following sections and in appendix [ chbm ] . next , considering the uhlmann fidelity yields a complete hierarchy of lower and upper bounds @xcite : @xmath33 where the lower bound on the probability of error @xmath34 , and the uhlmann fidelity @xmath35 between two quantum states @xmath36 is defined as @xmath37 . for the quantum reading protocol with unitary coding , let us consider the maximum probability of error in distinguishing the output of a binary memory cell encoded using one identity and one arbitrary unitary channel @xmath38 chosen in the set of local unitary operations with non - degenerate harmonic spectrum . the latter is the spectrum of the complex roots of the unity and its choice is motivated by observing that it excludes unambiguously the identity from the set of possible operations : indeed , unitary operations with harmonic spectrum are orthogonal ( in the hilbert - schmidt sense ) to the identity . we further assume that the coding is unbiased , that is the two channels are equiprobable . the worst - case scenario is defined by the probability of error eq . ( [ proberr ] ) being the largest possible : @xmath39 let us now consider a recently introduced measure of quantum correlations , the so - called discord of response @xcite : @xmath40 where the index @xmath41 denotes the possible different types of well behaved , contractive metrics under completely positive and trace - preserving ( cptp ) maps . the normalization factor @xmath42 depends on the given metrics and is chosen in such a way to assure that @xmath43 varies in the interval @xmath44 $ ] . finally , the set of local unitary operations @xmath45 includes all and only those local unitaries with harmonic spectrum . in the following , we will need to consider both the probability of error and different types of upper and lower bounds on it . therefore we will be concerned with three different discords of response corresponding to three types of contractive distances : trace , hellinger , and bures . the trace distance @xmath46 between any two quantum states @xmath25 and @xmath26 is defined as : @xmath47 \ , . \label{tracedistance}\ ] ] the bures distance , directly related to the fidelity @xmath35 , is defined as : @xmath48 finally , the hellinger distance is defined as : @xmath49 } \ ; \ , . \label{helldistance}\ ] ] for each discord of response , trace , hellinger , and bures , the normalization factor in eq . ( [ discofresp ] ) is , respectively : @xmath50 , @xmath51 . if the two states @xmath25 and @xmath26 are bipartite gaussian states related by local traceless symplectic transformations or bipartite qubit - qudit states related by a local single - qubit unitary operation , that is @xmath29 and @xmath52 , then , by exploiting eq . ( [ qcblu ] ) , it is straightforward to show that the quantum chernoff bound is a simple , monotonically non - increasing simple function of the hellinger distance : @xmath53 it is then immediate to show that the maximum of @xmath27 over the set of local unitary operations @xmath54 with completely non - degenerate harmonic spectrum is a simple linear function of the hellinger discord of response : @xmath55 the discord of response quantifies the response of a quantum state to least - disturbing local unitary perturbations and satisfies all the basic axioms that must be obeyed by a _ bona fide _ measure of quantum correlations @xcite : it vanishes if and only if @xmath3 is a classical - quantum state ; it is invariant under local unitary operations ; by fixing a well - behaved metrics such as trace , bures , or hellinger , it is contractive under cptp maps on subsystem @xmath56 , i.e. the subsystem that is not perturbed by the local unitary operation @xmath57 ; and reduces to an entanglement monotone for pure states , for one of which it also assumes the maximum possible value ( @xmath2 ) . by comparing eqs . ( [ perrmax ] ) and ( [ discofresp ] ) with @xmath58 , it is immediate to relate the maximum probability of error @xmath59 to the trace discord of response @xmath60 : @xmath61 from eq . ( [ perrdiscord ] ) it follows that half of the square root of the trace discord of response yields the difference between the absolute maximum of the probability of error ( i.e. @xmath14 ) and the maximum probability of error at fixed transmitter state @xmath3 . a vanishing trace discord of response implies that there exists at least one memory that can not be read by classical - quantum transmitters . a maximum trace discord of response ( @xmath62 ) implies that , irrespective of the coding , the maximally entangled transmitter will read any memory without errors : indeed , any local unitary operation with harmonic spectrum transforms a maximally entangled state into another maximally entangled state orthogonal to it , and therefore yields perfect distinguishability at the output . in the following , in order to compare the efficiency of classical ( non - discordant ) and quantum noisy sources of light in reading protocols , we will consider two - mode gaussian states of the electromagnetic field . the states with vanishing first moments of the quadratures are fully described by their covariance matrix @xmath63 @xcite : @xmath64 the range of values of @xmath65 and @xmath66 for which the corresponding states are physical ( i.e. correspond to positive density matrices ) is determined by the heisenberg uncertainty relation stated in symplectic form : @xmath67 where @xmath68 is the symplectic form . in all the paper , if we refer to the symmetric states we mean @xmath69 . in the following we will focus on two rather general classes of ( undisplaced ) gaussian states , the squeezed thermal states ( sts ) and the thermal squeezed states ( tss ) . the former are defined by two - mode squeezing @xmath70 applied on , possibly non - symmetric , two - mode thermal states . notice that the denomination sts is sometimes used in the literature to denote any gaussian states characterized by the covariance matrix eq . ( [ sqthcorm ] ) with @xmath71 . in the present work we adopt the convention that stss describe a physically rather frequent situation in which the thermal noise acts possibly non - symmetrically on the two modes , that is @xmath72 , and thus the total number of thermal photons is @xmath73 . here @xmath74 is the two - mode squeezing parameter and @xmath75 are the annihilation operators in each of the two modes ( @xmath76 ) . the diagonal and off - diagonal covariance matrix elements for these states , respectively @xmath77 , @xmath78 and @xmath79 , read : where @xmath80 where @xmath81 is the number of squeezed photons . thermal squeezed states ( tsss ) describe the reverse physical situation : an initially two - mode squeezed vacuum is allowed to evolve at later times in a noisy channel and eventually thermalizes with an external environment characterized by a total number of thermal photons @xmath73 . the covariance matrix elements of tsss , respectively @xmath82 , @xmath83 and @xmath84 , are : @xmath85 the same covariance matrix , eq . ( [ sqthcorm ] ) , also describes classical uncorrelated tensor product states , which we assume to be gaussian . thermal states are obtained letting @xmath86 , @xmath87 and @xmath88 . these gaussian states are classical in the sense that they can be written as convex combinations of coherent states and , moreover , they are the only gaussian states with vanishing discord @xcite . notice that in the standard quantum optics terminology the wording classical states is used to denote any state with positive glauber - sudarshan @xmath89-representation . in the following , without loss of generality , we will identify party @xmath24 with mode @xmath90 and party @xmath56 with mode @xmath91 . for unitary - coding protocols with gaussian transmitters , gaussian local ( single - mode ) unitary operations acting on an infinite - dimensional hilbert space are implemented by local ( single - mode ) symplectic transformations acting on the covariance matrix @xmath63 of two - mode gaussian input states @xmath92 . in the following we will consider only traceless transformations . the traceless condition must be imposed in order to exclude trivial coding by two identical channels , for which the maximum probability of error is always @xmath14 . moreover , imposing tracelessness allows to investigate and determine the correspondence between reading efficiency and quantum correlations , as will be shown in the following . denoting by @xmath93 the local traceless symplectic transformations acting on mode @xmath24 , the two local unitary operations implementing the encodings of the binary memory cells are the identity @xmath94 and @xmath95 . in order to assess the performance of quantum and classical gaussian resources in the unitary - coding quantum reading protocol we need to evaluate the upper and lower bounds , eqs . ( [ qcb ] ) and ( [ errorineq ] ) , on the maximum probability of error @xmath59 , eqs . ( [ perrmax ] ) and ( [ perrdiscord ] ) , for gaussian two - mode transmitters @xmath92 . to this end , we introduce first the _ gaussian _ discord of response @xcite , i.e the discord of response obtained by minimizing over local unitaries restricted only to the subset of local symplectic , traceless , transformations @xmath93 : @xmath96 where the index @xmath97 stands for trace , hellinger , or bures distance with the same normalization factors @xmath98 as before , and @xmath99 is the transpose of the symplectic matrix @xmath93 and @xmath100 . the gaussian discord of response provides an upper bound to the true discord of response of gaussian states and vanishes on and only on gaussian classical states ( subset of separable states that are in product form ) . the main properties of the gaussian discord of response are reported in appendix [ appdiscord ] . in complete analogy with eq . ( [ perrdiscord ] ) the maximum probability of error in discriminating two gaussian transmitters related by a local symplectic transformation can be expressed as a simple function of the trace gaussian discord of response : @xmath101 specializing the bounds given by eq . ( [ errorineq ] ) to the maximum probability of error in distinguishing gaussian states , one has : @xmath102 where the lower bound @xmath103 is a simple monotonically non - increasing function of the bures gaussian discord of response : @xmath104 and the upper bound @xmath27 is a simple linear , monotonically non - increasing function of the hellinger gaussian discord of response : @xmath105 therefore , for increasing gaussian discords of response the bounds on the probability of error decrease correspondingly . the explicit expressions of the quantum chernoff bound @xmath27 , the hellinger gaussian discord of response , the uhlmann fidelity , and the bures gaussian discord of response are derived in appendices [ uhlmanngaussian ] and [ chernoffgaussian ] . the probability of error in distinguishing @xmath92 from @xmath100 is given by eq . ( [ proberr ] ) with the local symplectic transformations @xmath93 replacing @xmath38 . among the local unitary operations @xmath93 which can implement the unitary - coding reading protocol , an important subset includes the single - mode phase shifts @xmath106 acting on mode @xmath90 , parameterized by the angle parameter @xmath107 : @xmath108 . under a local phase shift the local mode @xmath90 is transformed as follows : @xmath109 , while the two - mode covariance matrix @xmath110 transforms according to @xmath111 , where the symplectic matrix @xmath112 reads @xmath113 for the maximum probability of error , eq . ( [ gaussperrdiscord ] ) , the upper bound is achieved , from eqs . ( [ gaussbounds ] ) and ( [ qcbmax ] ) , in terms of a simple linear function of the hellinger gaussian discord of response . the latter , in turn , is obtained by minimizing the hellinger distance over the entire set of local unitary operations implemented on the covariance matrix by local symplectic , traceless , transformations . for squeezed thermal and thermal squeezed states one finds that this minimum is realized by the @xmath15 phase shift @xmath114 , that is the only possible traceless phase shift . therefore the extremal unitary - coding protocol in the ensemble of local traceless symplectic operations is realized by a particular psk coding , the phase shift @xmath15 , which is the only traceless psk coding . the details of the proof are reported in appendix [ chbm ] . on the other hand the quantity @xmath115 , eq . ( [ lbpmax ] ) , evaluated at @xmath15 , may not be optimal but certainly still provides a lower bound on the maximum probability of error : @xmath116 since for a @xmath15 phase shift the corresponding transformation is implemented by the traceless symplectic matrix @xmath117 , the expectation values of the canonical quadrature operators @xmath41 and @xmath118 transform as follows : @xmath119 and @xmath120 . therefore , undisplaced thermal gaussian states ( @xmath121 ) are left invariant , and the worst - case psk coding @xmath122 is completely invisible to classical transmitters ( thermal states ) since the @xmath123 shift does not change their covariance matrix . the probability of error @xmath10 for every such classical transmitter always achieves the absolute maximum @xmath14 . viceversa , the very same coding can always be read by any quantum gaussian transmitter with nonvanishing gaussian discord of response . as a consequence , quantum transmitters always outperform undisplaced classical transmitters in device - independent , worst - case scenario quantum reading . the situation changes when we consider displaced thermal states , as displacement unavoidably increases distinguishability . indeed the coherent and thermal coherent states are very efficient in detecting phase shift transformations . nonetheless , in the next section we will show that thermal coherent transmitters are outperformed by noisy quantum ones provided that the distribution of the thermal noise among the modes in the quantum resource is strongly _ non - symmetric_. we have seen that without displacement classical transmitters ( thermal states ) are completely blind to reading . introducing displacement enhances the distinguishability of output states and turns classical states ( thermal coherent states ) in useful transmitters . it is straightforward to show that distinguishability and the reading efficiency increase by implementing a single - mode displacement rather than a two - mode one with equal single - mode amplitudes . let us then consider a scenario in which one compares discordant quantum transmitters with displaced classical ones . we will show that in this case , that is comparing noisy quantum resources with distinguishability - enhanced noisy classical ones , discordant transmitters can outperform classical ones , and that the quantum advantage increases with increasing ( thermal ) noise . stated precisely , given the same coding @xmath124 acting locally on the first mode , we want to identify the regimes in which the probability of error associated to a quantum transmitter is smaller than the probability of error associated to a thermal coherent one . from eqs . ( [ qcb ] ) and ( [ errorineq ] ) this is equivalent to identifying the regimes in which the upper bound @xmath27 on the probability of error using squeezed thermal transmitters , denoted by @xmath125 , is smaller than the lower bound @xmath126 using thermal coherent states , that will be denoted by @xmath127 . obviously , only a constrained comparison at given fixed physical quantities is meaningful . we will thus compare squeezed thermal states and displaced thermal states at fixed purity and fixed total number of photons . we will observe that the quantum advantage is achieved provided the covariance matrix is not symmetric with respect to exchange of the modes . with these notations , the requirement for a _ bona fide _ quantum advantage reads as follows : @xmath128 both the coherent thermal and squeezed thermal states are two extremal classes of the general family of states which can be described as squeezed displaced thermal states ( sdts ) , defined as : @xmath129 where @xmath130 is the two - mode squeezing operator and we assume that squeezing parameter @xmath74 is real . here @xmath131 is the single - mode displacement operator and @xmath132 is the non - symmetric two - mode thermal state , where @xmath133 . the purity @xmath134 of the sdts is a function of the covariance matrix @xmath63 and depends only on the number of thermal photons : @xmath135 the total number of photons , @xmath136 , in the sdts reads : @xmath137 where @xmath138 is the number of squeezed photons . putting @xmath139 , sdtss reduce to thermal coherent states @xmath140 with total number of photons @xmath141 . decreasing the displacement amplitude @xmath142 the distinguishability of coherent thermal transmitters is reduced . we want to investigate whether this loss of distinguishability can be compensated by the quantum contribution due to increase of @xmath74 keeping @xmath143 and the purity fixed . in the limiting situation when @xmath144 the corresponding quantum state @xmath145 is a squeezed thermal state ( sts ) . in the following we will show that for stss @xmath145 and thermal coherent states @xmath140 with equal total number of photons @xmath143 , ineq . ( [ uplow ] ) is satisfied for some ranges of @xmath146 and @xmath147 . the condition of equal total number of photons @xmath143 implies @xmath148 . in order to evaluate the uhlmann fidelity @xmath35 and the quantum chernoff bound @xmath27 in eq . ( [ uplow ] ) we need to know how the phase shift @xmath149 transforms the transmitters that we wish to compare : the squeezed thermal states and the thermal coherent states . the dependence of @xmath35 and @xmath27 on the displacement vector and on the covariance matrix of general gaussian states is reported in appendices [ uhlmanngaussian ] and [ chernoffgaussian ] . the uhlmann fidelity providing the lower bound on @xmath10 for thermal coherent states depends only on the displacement vector , since the covariance matrix of thermal coherent states is unaffected by the action of the symplectic transformation @xmath150 , where @xmath151 . without loss of generality , the displacement vector of a thermal coherent state can be written as @xmath152^t$ ] . under a @xmath15 phase shift the difference @xmath153 between the final and the initial displacement vectors reads as follows : @xmath154 the uhlmann fidelity of a thermal coherent state is then @xmath155 where @xmath156 . the @xmath27 of non - symmetric , undisplaced squeezed thermal state depends only on the covariance matrix , eq . ( [ sqthcorm ] ) , with entries eqs . ( [ ccvsqth ] ) , and its explicit expression is reported in appendix [ chernoffgaussian ] . in fig . [ boundsenergy ] , upper panel , we report the exact values of @xmath157 for a squeezed vacuum with squeezing @xmath74 , @xmath158 for a coherent state @xmath159 , and @xmath160 for a squeezed displaced vacuum with squeezing @xmath161 and displacement @xmath162 , in the absence of noise , @xmath163 , and at fixed total photon number @xmath164 . the coherent states outperform the quantum resources given by the undisplaced squeezed vacuum . the coherent transmitters are then compared with squeezed displaced vacua of the same energy . even if the latter include a classical contribution due to displacement and a quantum contribution due to squeezing they are still outperformed by the classical coherent states . the quantum efficiency converges to the classical one in the high - energy limit . for completeness , in fig . [ boundsenergy ] we also report the quantum chernoff bound @xmath165 for the squeezed - vacuum transmitters . , of the probability of error @xmath10 in the absence of thermal noise ( @xmath163 ) . blue dotted line : probability of error @xmath157 of squeezed vacuum states . black dashed line : @xmath158 of coherent states . green dot - dashed line : @xmath160 of squeezed displaced vacuum states with displacement @xmath166 . red solid line : quantum chernoff bound @xmath165 of squeezed vacuum states yielding the upper bound on @xmath10 with quantum transmitters . no quantum gain is not observed in this regime . central panel : behavior of quantum and classical bounds on @xmath10 as functions of @xmath143 at fixed asymmetric thermal noise : @xmath167 , @xmath168 . red solid line : quantum upper bound @xmath125 on @xmath10 with undisplaced squeezed thermal states . black dashed line : classical lower bound @xmath127 on @xmath10 with thermal coherent states . lower panel : same as central panel , but with stronger thermal noise : @xmath169 , @xmath168 . with increasing @xmath143 the quantum upper bound goes below the corresponding classical figures of merit and quantum transmitters certainly outperform classical ones.,width=264 ] in the presence of symmetric thermal noise , @xmath170 , there is no improvement in the quantum efficiency relative to the classical one . introducing non - symmetric thermal noise , e.g. @xmath171 , the gaussian discords of response , that are intrinsically asymmetric quantities with respects to the subsystems in a given bipartition , increase dramatically , and so does the corresponding quantum reading efficiency . as a consequence , for sufficiently strong non - symmetric thermal noise the quantum resources outperform the classical ones . in the presence of non - symmetric noise exact expressions for @xmath10 are no longer available . therefore , in the central and lower panels of fig . [ boundsenergy ] we report the exact lower and upper bounds on @xmath10 based on the uhlmann fidelity @xmath35 and on the quantum chernoff bound @xmath27 . we observe that at intermediate values of the total number of photons @xmath143 the quantum upper bound @xmath125 on @xmath10 is strictly lower than the classical lower bound @xmath127 , assuring that the quantum resources outperform the classical ones . the classical transmitters ( thermal coherent states ) recover the quantum efficiency for large values of the total photon number . moreover , comparing the central and the lower panels in fig . [ boundsenergy ] , we observe that as the number of thermal photons @xmath146 is increased , the range of values of the total photon number @xmath143 for which one has a quantum advantage increases . in fig . [ rconstans ] we provide a plot of the contour lines for the differences @xmath172 for different asymmetries : @xmath168 ( upper panel ) and @xmath173 ( lower panel ) as functions of the total photon number @xmath143 and of the purity ( or , equivalently of the number of thermal photons @xmath146 ) . when these differences become negative , ineq . ( [ uplow ] ) is satisfied and the quantum resources certainly outperform the classical ones . from the upper panel of fig . [ rconstans ] , for @xmath168 comparing noisy quantum transmitters with noisy coherent ones , one observes that @xmath174 in a large region of parameters . fixing the squeezing , so that the change in the total photon number @xmath143 is due only to the change in the number of thermal photons @xmath146 , corresponds to a straight line in the plane ( in the figure , drawn at @xmath175 ) . remarkably , for these iso - squeezed states the quantum advantage increases with increasing number of thermal photons . this is an instance of noise - enhanced quantum efficiency that will be discussed further in sec . [ comparisonquant ] . in the lower panel of fig . [ rconstans ] we decrease the asymmetry @xmath176 . we observe that the quantum gain is also achieved but in the range of much higher @xmath177 . again , fixing the squeezing , e.g. at @xmath175 , we notice that the quantum advantage increases with thermal noise . we remark that these results are obtained in a scenario in which we compare the minimum quantum efficiency ( upper bound on the error probability using quantum transmitters ) with the maximum classical efficiency ( lower bound on the error probability using coherent thermal transmitters ) . therefore the actual quantum advantage will be even larger . for @xmath168 ( upper panel ) and for @xmath173 ( lower panel ) as functions of the total photon number @xmath178 and of the number of thermal photons @xmath146 . the region in which these quantities assume negative values corresponds to quantum transmitters certainly outperforming coherent thermal ones . the red dashed curve identifies its boundary . the straight solid yellow lines in both panels corresponds to a fixed degree of squeezing @xmath175 . moving along this lines in the direction of increasing number of thermal photons @xmath146 one observes that as noise grows there is a growing advantage in using quantum transmitters over classical ones . this behavior provides an instance of noise - enhanced quantum performance . the effect is reduced when the asymmetry @xmath179 decreases.,width=264 ] the quantum advantage disappears in the symmetric situation @xmath180 . indeed , the inequality @xmath181 expresses the condition that the mode which passes through the coding channels is more noisy . this condition is unfavorable for thermal coherent states and favorable for stss . namely , in stss with fixed finite squeezing , increasing the number of thermal photons in the first mode certainly increases the discord of response and , as a consequence , increases also the reading efficiency for this type of transmitters . this phenomenon is further analyzed in the following sections on the comparison of different quantum transmitters . these two concatenated effects cause the advantage of quantum states over the classical transmitters in the protocol of quantum reading with noisy transmitters . the asymmetry between the local thermal noise terms is the crucial element for realizing the enhancement of the reading efficiency . as we will see in the next section , the behavior of stss with increasing number of thermal photons in the symmetric situation @xmath180 , although not sufficient to realize a quantum advantage over classical resources , favors stss among other noisy quantum transmitters in the previous section we compared classical and quantum transmitters and for worst - case scenario we identified the regimes in which noisy but discordant quantum resources outperform classical thermal coherent ones . we also observed that the quantum advantage can increase , at fixed squeezing , with increasing thermal noise . we will now compare the behavior of squeezed - thermal and thermal - squeezed states in order to investigate how thermal noise affects the quantum efficiency of different classes of quantum transmitters . we shall compare symmetric squeezed thermal and thermal squeezed transmitters either at fixed number of thermal photons or at fixed squeezing . we will then consider how non - symmetric noise further enhances the quantum efficiency by suppressing the upper bound on the probability of error . finally , we will investigate how the quantum efficiency of different quantum transmitters improves when multiple reading operations are implemented at fixed thermal noise . let us start by comparing quantum reading with symmetric squeezed thermal and symmetric thermal squeezed transmitters at fixed number of thermal photons and its performance as a function of the total number of photons . this comparison is motivated by the fact that the interplay between quantum and thermal fluctuations is very different for these two classes of quantum states . squeezed thermal states ( stss ) are obtained by applying on thermal states , namely states that have already thermalized ( e.g. at the output of a noisy channel ) a purely quantum operation , two - mode squeezing , that can be interpreted as a re - quantization of the thermal vacuum . viceversa , thermal squeezed states ( tsss ) are realized by letting pure squeezed vacua evolve and eventually thermalize in a noisy channel . both squeezed thermal and thermal squeezed states are two extremal classes of the very general family of squeezed thermal squeezed displaced states ( stsdss ) which are defined as follows : @xmath182 s(r)^{\dagger } \ , . \nonumber\end{aligned}\ ] ] here @xmath183 and @xmath184 are two - mode squeezing operators with different squeezing parameters @xmath74 and @xmath161 , @xmath185 is a single - mode displacement operator , @xmath186 is a noisy channel introducing @xmath146 and @xmath147 thermal photons respectively in the first and in the second mode . the channel acts on a given gaussian state adding the number of thermal photons to the diagonal entries of its covariance matrix . finally , @xmath187 denotes the two - mode vacuum state . we study this family of states at constant fixed values of the parameters @xmath146 and @xmath147 . the total number of photons in a stsds is : @xmath188 consider first the situation without displacement , @xmath144 , and with symmetric thermal noise @xmath189 . putting @xmath139 , stsdss reduces to thermal squeezed states tsts @xmath190 . decreasing @xmath161 and correspondingly increasing @xmath74 while keeping @xmath143 and @xmath191 fixed , in the limit @xmath192 one recovers the squeezed thermal states stss @xmath193 . let us compare these two extremal classes of quantum gaussian transmitters that coincide for @xmath194 and differ for @xmath195 or , in non symmetric situations , when either @xmath196 and/or @xmath197 . in fig . [ quantumquantum ] , upper left panel , we observe that for nonvanishing but small number of thermal photons @xmath191 the upper bound on the probability of error @xmath10 using stss still remains above the lower bound on @xmath10 using tsss . by further increasing thermal noise , as shown in the upper right panel of fig . [ quantumquantum ] , all bounds with stss are below all bounds with tsss and the stss certainly outperform the tsss . due to the different effects of the noise in stss and tsss we observe the clear advantage of using stss over tsss in a quantum reading protocol . the lower panels of fig . [ quantumquantum ] show the comparison done for displaced thermal squeezed states @xmath198 and displaced squeezed thermal states @xmath199 . at fixed total number of photons we observe that single - mode displacement always increases the reading efficiency for both classes of states , while decreasing the squeezing reduces and eventually wipes out the quantum advantage of stss over tsss these two classes of states coincide in the limiting case @xmath200 in which they both recover the classical thermal coherent states . as we have seen in the previous section , the advantage of stss over classical states is recovered by considering non - symmetric thermal noise . , of the upper and lower bounds on the probability of error @xmath10 using either undisplaced squeezed thermal states ( stss ) or undisplaced thermal squeezed states ( tsss ) with fixed , symmetric , thermal noise : @xmath201 . upper right panel : the same but with @xmath202 . in this case one observes that beyond a threshold value of @xmath177 the stss certainly outperform the tsss . lower panels similar to the upper panels but with nonvanishing displacement @xmath203 . the upper bounds on the probability of error are given here by the quantum bhattacharyya coefficient @xmath204 , which for states that include displacement does not necessary coincide with the quantum chernoff bound @xmath27 . the displacement increases the efficiency of the reading , however it does not guarantee with certainty the noise - enhanced performance of stss with respect to tsss.,width=340 ] when comparing the behavior of the uhlmann fidelity , quantum chernoff bound , and the gaussian discords of response under variations of the classical noise at a fixed level of quantum fluctuations ( squeezing ) , we expect a radically diverging behaviors of the stss with respect to the tsss . on intuitive grounds , since fidelity , chernoff bound , and discord are measures of distinguishability between an input state and the corresponding output after a local disturbance , if we compare stss and tsss we notice from the structure of their covariance matrices , see eqs . ( [ cthsq ] ) , that as @xmath191 increases the correlation part of the stss increases , while it remains constant in tsss . indeed , the quantum chernoff bound @xmath27 and of the uhlmann fidelity @xmath205 for any two gaussian states of the form eq . ( [ sqthcorm ] ) with @xmath69 and @xmath206 , related by a @xmath15 phase shift , take the form : @xmath207 @xmath208^{2 } } \ ; . \label{derivfidel}\ ] ] in fig . [ e18 ] we report the behavior of the upper bound on the probability of error @xmath27 , eq . ( [ derivchernoff ] ) , and of the lower bound @xmath126 ( which is a monotonic increasing function of the uhlmann fidelity @xmath205 , eq . ( [ derivfidel ] ) ) for the squeezed thermal and thermal squeezed states as functions of the number of thermal photons at fixed squeezing . we observe that for tsss @xmath205 and @xmath27 both increase with increasing thermal noise , converging asymptotically to the absolute maximum ( @xmath14 ) of the probability of error . therefore , the quantum efficiency of tsss is suppressed by increasing the thermal noise . on the contrary , for stss @xmath27 remains constant and @xmath126 decreases . this behavior guarantees that the probability of error , at fixed squeezing , is bound to vary in a restricted interval below @xmath209 . in the given example , the squeezing amplitude @xmath74 has been fixed at a relatively low value @xmath210 . increasing the level of squeezing will further reduce the maximum value achievable by the probability of error . in conclusion , the quantum advantage associated to squeezed thermal states is paramount at fixed , even moderate , squeezing , and increases monotonically with increasing thermal noise . and of the lower bound on the probability of error @xmath126 as functions of the number of thermal photons @xmath191 , at fixed number of squeezed photons @xmath211 , for thermal squeezed and squeezed thermal states . blue dashed line : @xmath27 for thermal squeezed states . dotted red line : @xmath126 for thermal squeezed states . solid black line : @xmath27 for squeezed thermal states . orange dot - dashed line : @xmath126 for squeezed thermal states . the colored areas between the upper and lower bounds denote the admissible intervals of variation for the probability of error @xmath10 . increasing thermal noise suppresses the efficiency of thermal squeezed transmitters and increases the efficiency of squeezed thermal ones.,width=298 ] a more detailed understanding of these opposite behaviors can be gained by looking at the variation of the measures of distinguishability with respect to the variations of the thermal noise and of the parameters of the covariance matrix . consider a generic measure of distinguishability denoted by @xmath212 where @xmath213 , among others , includes the uhlmann fidelity @xmath205 and the quantum chernoff bound @xmath27 . consider then the _ total _ derivative of @xmath213 with respect to @xmath191 , keeping @xmath74 constant : @xmath214 specializing to either @xmath205 or @xmath27 we obtain the explicit expressions of their derivatives , as reported in appendix [ derivatives ] . from these explicit expressions it follows that it is always @xmath215 irrespective of the type of quantum transmitter considered . hence , if @xmath213 represents either the uhlmann fidelity or the quantum chernoff bound , the derivative @xmath216 . this behavior agrees with the intuition that the operation of increasing the diagonal entries of the covariance matrix and keeping the off - diagonal entries constant acts like a thermal channel which makes the initial state and the final state after the phase shift less distinguishable . the behavior @xmath217 for both @xmath205 and @xmath27 is also intuitively clear , since changes in @xmath63 under @xmath149 are the greater the larger the off - diagonal entries when keeping the diagonal @xmath218 constant . let us now discuss the state - dependent derivatives : for stss and tsss the partial derivatives @xmath219 and @xmath220 are non - negative , therefore they can not oppose the behavior of the state - independent part . for stss they are given by @xmath221 and @xmath222 respectively , while for tsss @xmath223 and @xmath224 . the behavior of @xmath225 or @xmath27 with increasing @xmath191 depends then on the ratio of the positive and negative parts on the left hand side of eq . ( [ derivative ] ) . for tsss there is only a positive contribution in eq . ( [ derivative ] ) and both @xmath205 and @xmath27 increase with increasing number of thermal photons . as a consequence , both the lower and the upper bounds on the probability of error must increase , as observed in fig . [ e18 ] . on the other hand , for stss the negative contribution always prevails when considering the uhlmann fidelity , while the positive and negative contributions always cancel exactly when considering the quantum chernoff bound , leading to a constant upper bound on the probability of error , as observed in fig . [ e18 ] . the constant behavior of @xmath27 as a function of thermal noise for stss can be also seen directly from eq . ( [ derivchernoff ] ) . this equation can be rewritten straightforwardly only in terms of @xmath226 . indeed , this ratio for stss does not depend on @xmath191 . in this section we have considered reading protocols with binary coding given by the identity and the phase shift @xmath15 , and transmitters implemented by symmetric stss . this is actually a worst - case scenario in two respects . on the one hand , the phase shift @xmath15 provides the worst possible coding among all traceless local symplectic operations ( maximum probability of error , device - independent reading ) . on the other hand , the _ symmetric _ stss provide the worst possible transmitters among general stss . indeed , in the next subsection we will show that non - symmetric stss provide much larger quantum efficiencies and even effectively suppress the probability of error . one might speculate that the increment of the bures discord of response for increasing thermal noise and the corresponding decrement of the lower bound on the probability of error are due to the particular relation with the bures metrics induced by the uhlmann fidelity . however , this is not the case . we will now show that if one considers non - symmetric two - mode stss then also the hellinger discord of response increases under increasing local thermal noise and therefore the corresponding upper bound on the probability of error decreases as well . this is a strong indication that the true probability of error decreases as well with increasing thermal noise and thus that the use of discordant , non - symmetric stss yields an absolute advantage , even over the use of entangled pure states , namely two - mode squeezed vacua with the same amount of squeezing as in the corresponding stss . the covariance matrix of non - symmetric two - mode stss is given in eq . ( [ sqthcorm ] ) with the parameters given in eqs . ( [ ccvsqth ] ) . the corresponding @xmath27 achieves its maximum for the @xmath15 phase shift , as proven in appendix [ chbm ] , and the exact expression of @xmath27 for non - symmetric stss related by a @xmath15 phase shift is : @xmath227 let us consider the variation @xmath228 of the quantum chernoff bound , at constant squeezing @xmath74 and constant number of thermal photons @xmath147 in the second mode , whose analytical expression is provided in appendix [ derivatives ] . from this expression it is clear that there is a range of values of @xmath146 and @xmath147 , namely @xmath181 , for which @xmath229 . therefore , in this regime @xmath27 decreases with increasing @xmath146 . on the other hand , @xmath27 increases with increasing @xmath146 if @xmath230 . henceforth , in the symmetric situation @xmath189 the quantum chernoff bound is maximum and constant , independent of @xmath191 , as discussed in the previous section . , as a function of @xmath146 . in each panel the number of thermal photons @xmath147 in the second mode is fixed at a constant value . upper left panel : @xmath231 . upper right panel : @xmath232 . lower left panel : @xmath233 . lower right panel : @xmath233 and extended range of values of @xmath146 , in order to show the asymptotic vanishing of @xmath27 with increasing local thermal noise . for all panels the two - mode squeezing is fixed at @xmath234 . the maximum of @xmath27 is achieved for symmetric stss and provides the upper bound on the maximum probability of error @xmath235 of the worst - case scenario.,width=340 ] in fig . [ qcbnth1 ] we report the behavior of @xmath27 as a function of @xmath146 for different fixed values of @xmath147 and fixed squeezing @xmath74 . in this physical situation the quantum chernoff bound decreases with increasing _ local _ thermal noise and vanishes asymptotically for @xmath236 . therefore the probability of error in a gaussian quantum reading protocol can be made arbitrarily small by using non - symmetric stss transmitters with very large _ local _ thermal noise . this very remarkable result may look at first quite counter - intuitive . in fact , the crucial point is that this feature is obtained by the _ global quantum _ operation of two - mode squeezing applied to a two - mode thermal state with very strong asymmetry in the _ local _ thermal noises affecting the two field modes . it is therefore not entirely unexpected that the consequences can be dramatic . while entanglement certainly decreases , the operation of squeezing a larger amount of noise can increase quantum state distinguishability by _ `` orthogonalizing '' _ on a larger portion of hilbert space with respect to the thermal states . collecting all the previous results we are finally in the position to compare the best resources of device - independent gaussian quantum reading , namely the noisy and discordant non - symmetric stss , to the best absolute resources of gaussian quantum reading , namely pure entangled two - mode squeezed vacuum states ( tmsvss ) . in the limit of infinite squeezing the tmsvss are maximally entangled pure einstein - podolsky - rosen ( epr ) states whose probability of error in a quantum reading protocol vanishes identically . in absolute terms , tmsvss are certainly the best among classical and quantum resources in a reading protocol with continuous variables . indeed , in fig . [ sqthsqvac ] we report the behavior of the exact probability of error for tmsvss and the lower bound on it for non - symmetric stss as functions of the total number of photons at fixed thermal noise , that is for _ arbitrarily increasing squeezing _ as the total number of photons increases . one observes that the lower bound on the probability of error for non - symmetric stss is always above the exact probability of error for tmsvss , converging towards it only asymptotically . of the probability of error @xmath157 for two - mode squeezed vacuum transmitters ( tmsvs ) and of the lower bound on the probability of error @xmath237 for two - mode squeezed thermal transmitters ( stss ) with @xmath238 and @xmath168 . tmsvs have better reading efficiency than stss . the squeezing in tmsvs is larger than the one in stss at each fixed value of @xmath143 . the two efficiencies converge asymptotically with increasing total number of photons.,width=302 ] on the other hand , it is also important to compare tmsvss and non - symmetric stss in terms of the concrete use of resources in realistically feasible experimental scenarios . in fig . [ stsvstmsvs ] and fig . [ stsvstmsvs1 ] we report the behavior of the exact probability of error @xmath157 associated to tmsvs transmitters compared to the upper and lower bounds @xmath125 and @xmath237 for non - symmetric sts transmitters as functions of the total number of photons . in the case of tmsvss the total number of photons obviously depends only on the squeezing and the behavior of the probability of error is the same as the one reported in fig . [ sqthsqvac ] . however , at variance with fig . [ sqthsqvac ] , in fig . [ stsvstmsvs ] and [ stsvstmsvs1 ] we compare it with the lower and upper bounds for st transmitters at a _ fixed _ finite value , low and comparably easy to produce experimentally , of the squeezing . in this case , the total number of photons in stss varies only with the amount of thermal photons . [ stsvstmsvs ] shows the behavior of the exact probability of error for tmsvs transmitters and the bounds on the probability of error for non - symmetric stss , as functions of the total number of photons @xmath143 and constant squeezing parameter fixed at @xmath234 . for comparison , in fig . [ stsvstmsvs1 ] we report the same quantities but for a larger fixed two - mode squeezing @xmath239 . indeed , the higher the squeezing , the better the bounds on the probability of error using sts transmitters approximate the exact probability of error for tmsvs transmitters . the crucial difference is that in real - world experimental setups it is comparatively much easier and less resource - demanding to implement a scheme relying on non - symmetric stss with enhanced thermal noise and quantum discord than to produce pure ( noise - free ) tmsvss with enhanced squeezing and entanglement . therefore , at _ fixed _ squeezing , we can compare the two classes of transmitters for _ different _ values of the total number of photons @xmath143 and ask for the threshold value of @xmath146 above which the discordant stss certainly perform better than the entangled tmsvss at the same fixed level of squeezing ( the noise on the second mode being also fixed at a given reference value , say e.g. @xmath168 ) . this threshold is thus determined by the condition @xmath240 . we give here two numerical examples for two different realistic values of the two - mode squeezing achievable in the laboratory with current technologies . for @xmath234 , we have that @xmath241 as soon as @xmath242 . for @xmath239 , we have that @xmath241 as soon as @xmath243 . therefore , the higher the fixed level of squeezing , the lower is the level of thermal noise and quantum discord required to match the performance of pure entangled tmsvss . alternatively , increasing @xmath146 further above the threshold , we can also look for the complementary information on the minimum threshold values of @xmath74 ( more easily realizable in the laboratory ) above which stss match or surpass the performance of tmsvss at higher values of the squeezing ( harder to achieve experimentally ) . in other words , we can introduce the concept of _ effective _ squeezing @xmath244 associated to the value @xmath245 such that for @xmath246 stss perform better than tmsvss with given squeezing @xmath247 . in conclusion , device - independent quantum reading is a remarkable protocol of quantum technology with noisy resources for which the best transmitters are discordant non - symmetric squeezed thermal states whose performance is optimized by realizing a fine trade - off between increased local thermal noise and fixed global two - mode squeezing , yielding noise - enhanced quantum correlations and state distinguishability . of the probability of error @xmath157 for two - mode squeezed vacuum transmitters , and of the lower and upper bounds on the probability of error @xmath237 and @xmath125 for non - symmetric two - mode squeezed thermal transmitters . the latter two quantities are plotted for variable @xmath146 at _ fixed _ squeezing @xmath234 , as well as fixed reference thermal noise in the second field mode @xmath248.,width=302 ] of the probability of error @xmath157 for two - mode squeezed vacuum transmitters , and of the lower and upper bounds on the probability of error @xmath237 and @xmath125 for non - symmetric two - mode squeezed thermal transmitters . the latter two quantities are plotted for variable @xmath146 at _ fixed _ squeezing @xmath239 , as well as fixed reference thermal noise in the second field mode @xmath248.,width=302 ] let us now analyze the case in which the total number of photons can vary by considering many copies of the transmitter , that is repeating the reading protocol many times independently . using @xmath249 copies of the system the uhlmann fidelity and the quantum chernoff bound decrease as powers of @xmath249 . therefore , the probability of error can decrease both in the case of squeezed thermal and thermal squeezed states . the interesting question which arises here is how many copies we need in both cases to achieve a given level of probability of error . the number of copies defines for instance the time needed for reading one bit of information in the given coding . therefore this process is interesting from the point of view of assessing the reading time and the strength of the sources of squeezed light that one needs . and @xmath250 . blue dashed line : @xmath27 for thermal squeezed states . dotted red line : @xmath126 for thermal squeezed states . solid black line : @xmath27 for squeezed thermal states . orange dot - dashed line : @xmath126 for squeezed thermal states . in order to achieve @xmath251 it is enough to take at most @xmath252 copies of stss , while the needed number of copies of tsss is at least @xmath253.,width=302 ] let us for instance assume that we require a value of the probability of error @xmath254 , having for each copy of the squeezed thermal transmitter the thermal noise fixed at @xmath250 and the weak squeezing fixed at @xmath255 , see fig . looking at the upper bound ( worst - case scenario ) , the number of copies which are needed , in order to achieve the desired level of probability of error , is at most @xmath252 . taking instead the thermal squeezed transmitter with the same squeezing and thermal noise in each copy , we see from fig . [ en ] that one needs , considering the lower bound ( best - case scenario ) , at least @xmath253 copies . these behaviors illustrate very clearly the advantage of using noise - enhanced quantum correlations . indeed , comparing figs . [ en ] and [ e18 ] , we see that by keeping a fixed level of squeezing and increasing the thermal noise , the number of copies of squeezed thermal transmitters needed to achieve a given level of precision stays constant , while the number of copies of thermal squeezed transmitters must increase . we have investigated gaussian quantum reading protocols realized by weak optical sources in the worst - case scenario for quantum transmitters with respect to classical ( thermal coherent ) ones . for protocols that involve local unitary operations in the process of reading by continuous - variable gaussian optical fields , we have showed that the maximum probability of error in reading binary memory cells is directly related to the amount of quantum correlations in a given transmitter , as quantified by the trace gaussian discord of response . this relation allows to quantify the reading efficiency in terms of quantum correlations , providing a natural operational interpretation to the gaussian discord of response . indeed , the latter is a well - defined measure of quantum state distinguishability under the action of local unitary operations . therefore , the more discordant is the transmitter , the smaller is the maximum probability of error when using quantum resources . this relation then allows to determine the physical regimes of state purity and signal strength for which one has a net advantage in using quantum resources over classical thermal coherent ones . since the trace distance is in general uncomputable for gaussian states , we have introduced exact upper and lower bounds on the maximum probability of error . we have showed that these bounds are expressed in terms of other type of quantum discords . in particular , the lower bound is expressed in terms of the bures gaussian discord of response , while the upper bound , provided by the quantum chernoff bound maximized over the set of possible local unitary operations , is expressed in terms of the hellinger gaussian discord of response for squeezed thermal states and thermal squeezed states . both bounds decrease with an increasing amount of quantum correlations , providing a precise quantitative estimate of the quantum advantage obtained by using discordant resources over the corresponding thermal coherent ones . moreover , the bures and hellinger discords of response are of further independent interest , as they play a central role in other quantum protocols studied recently , ranging from the assessment and use of local quantum uncertainty in optimal phase estimation @xcite , the efficiency of black - box quantum metrology @xcite , and the quantum advantage of discordant resources in the protocol of quantum illumination @xcite . after comparing quantum and classical resources , we have discussed two fundamental classes of gaussian quantum transmitters : symmetric squeezed thermal states ( stss ) and symmetric thermal squeezed states ( tsss ) . we have shown that the actual beneficial or detrimental effects of environmental noise depend on the type of quantum state being considered . considering stss as quantum transmitters , the upper and lower bounds on the probability of error decrease with increasing thermal noise and therefore the quantum reading efficiency increases . the opposite behavior is observed when considering tsss : in this case both the upper and the lower bounds on the maximum probability of error increase and therefore the quantum reading efficiency decreases with increasing thermal noise . finally , we went a step further and investigated the use of non - symmetric stss . for such transmitters , also the quantum chernoff bound decreases when the local thermal noise increases in one mode and remains fixed in the second mode . indeed , the quantum chernoff bound vanishes asymptotically with very large local thermal noise and therefore the probability of error must also vanish . in other words , non - symmetric two - mode stss with imbalanced thermal noise between the two modes achieve an asymptotically vanishing probability of error for very large values of the noise imbalance . for such asymptotic states the hellinger and bures gaussian discords of response attain their maximum value , and the quantum chernoff bound and uhlmann fidelity vanish . as a consequence , all upper and lower bounds on the probability of error vanish , the probability of error itself vanishes , and perfect reading is approached asymptotically . since the quantum reading efficiency of non - symmetric two - mode squeezed thermal states is a non - decreasing function of thermal noise , there is no evident advantage in using pure - state squeezed transmitters , the two - mode squeezed vacuum states , or low - noise ones over non - symmetric two - mode stss with large noise imbalance between the field modes , _ as long as _ the squeezing is kept _ fixed _ at a realistic , finite constant value achievable in concrete experiments with currently available technology . hence , noisy stss transmitters _ can _ provide a better quantum efficiency at _ fixed _ two - mode squeezing , provided thermal noise ( number of thermal photons ) is enhanced beyond the threshold value above which the upper bound on the probability of error ( quantum chernoff bound ) for stss goes below the exact probability of error for tmsvss with the same , fixed level of squeezing . this remarkable phenomenon of noise - assisted quantum correlations and quantum efficiency is eventually due to the fact that quantum state distinguishability is intimately related to the concept of geometric quantum correlations , as measured by the discords of response , and the observation that the former can increase under increasing thermal noise . in particular , maximum _ local _ noise enhancement leads to maximum _ global _ enhancement of quantum correlations . in forthcoming studies we will provide a general characterization and quantification of noise - suppressed vs. noise - enhanced quantum correlations for different classes of quantum states @xcite , and we will investigate the relations between different types of quantum correlations according to states , metrics , and operations @xcite . f.i . acknowledges valuable discussions with gerardo adesso . the authors acknowledge financial support from the italian ministry of scientific and technological research under the prin 2010/2011 research fund , and from the eu fp7 strep projects iqit , g.a . 270843 , and equam , g.a . here we discuss the gaussian discord of response , given by eq . ( [ intdr ] ) , and prove that it is a _ bona fide _ measure of quantum correlations . more general discussion can be found in @xcite . the minimal set of axioms with universal consensus includes the following : @xmath256 invariance under local unitary transformations , @xmath257 contractivity under the action of completely positive and trace preserving ( cptp ) maps acting on mode @xmath56 , @xmath258 vanishing of quantum correlations if and only if the state is classically - quantum correlated , i.e. with block - diagonal covariance matrix , @xmath259 reduction to an entanglement monotone for pure states . the first condition is guaranteed by unitary invariance of the chosen distance and the procedure of minimization . the second condition is satisfied due to the fact that we consider only contractive distances in order to define the discord of response . the third condition is verified as follows . it is known that classical - quantum two - mode gaussian states are those and only those which can be represented by the tensor product @xmath260 of single - mode gaussian states @xcite . up to displacements , such states are characterized by the block diagonal covariance matrices @xmath261 . let us consider the local traceless symplectic transformation @xmath93 which can be decomposed as @xmath262 , where @xmath263 and @xmath264 is a symplectic matrix which diagonalize @xmath265 , i.e. @xmath266 . here @xmath267 are two equal symplectic eigenvalues of @xmath265 . the transformation given by @xmath149 is symmetry - preserving and therefore @xmath262 leaves @xmath265 invariant . this shows that if the state is classically correlated there exists at least one local traceless transformation @xmath93 that leaves the state invariant . we now prove the reverse statement that only in the case of classically correlated states there exists such a symplectic traceless transformation that leaves the state invariant . assume that the covariance matrix left invariant by a traceless transformation @xmath268 has the form @xmath269 . local symplectic transformation can bring the covariance matrix in the so called normal form in which @xmath270 . if the state is not changed by the local transformation we have @xmath271 . since @xmath272 is reversible we obtain that @xmath273 which contradicts the assumption on the spectrum of @xmath93 . this shows that condition @xmath258 is satisfied . condition @xmath259 is guaranteed by the fact that for pure states the gaussian discord of response reduces to the gaussian entanglement of response @xcite ( the gaussian counterpart of the entanglement of response @xcite ) which is a _ bona fide _ measure of entanglement . the uhlmann fidelity for two - mode gaussian states can be computed as follows @xcite . let us define the matrix of the symplectic form @xmath274 the displacement vector is the vector of the averages of the amplitude and phase field quadratures @xmath41 and @xmath118 i.e. @xmath275 , where @xmath276 stands for transposition . denote the difference of the displacement vectors of two gaussian states @xmath25 and @xmath26 by @xmath277 . we need also the auxiliary formulas defined using the covariance matrices @xmath278 and @xmath279 of the respective gaussian states : @xmath280 , \\ \lambda & = & 2 ^ 4 \det(\sigma_1+\frac{i}{2}\omega)\det(\sigma_2+\frac{i}{2}\omega).\end{aligned}\ ] ] the uhlmann fidelity for two mode gaussian states is then @xmath281 } \notag \\ & \times&\left[(\sqrt{\gamma}+\sqrt{\lambda})-\sqrt{(\sqrt{\gamma}+ \sqrt{\lambda})^2-\delta}\right]^{-1}. \label{gfidelity}\end{aligned}\ ] ] any @xmath249-mode gaussian state can be represented in its normal mode decomposition parameterized by @xmath282 in which @xmath283 is the vector of the averages of the quadratures and @xmath284u_{{\langle}u{\rangle},s}^{\dagger},\ ] ] where @xmath285 is a thermal state with mean photon number @xmath286 and @xmath287 are the eigenstates of the operator of the number of photons in mode @xmath288 . the set @xmath289 identifies the symplectic spectrum . in this way the covariance matrix is decomposed as @xmath290 for two arbitrary gaussian states with normal mode decompositions @xmath291 and @xmath292 , assuming that @xmath293 , we have @xcite @xmath294^{-1}\delta\ } } , \label{appchs}\end{aligned}\ ] ] where @xmath295 } } \label{qubar}\ ] ] and @xmath296 moreover @xmath297s_1^t,\label{v1 } \\ v_2(1-t)&=&s_2\left[\bigoplus_{k=1}^n\lambda_{1-t}(\beta_k)\mathbbm{1}_k\right]s_2^t,\label{v2}\end{aligned}\ ] ] where @xmath298 the quantum chernoff bound for arbitrary states @xmath25 and @xmath26 is @xmath299 which for gaussian states is expressed by means of eq . ( [ appchs ] ) , i.e. @xmath300 . let us discuss the extremizations of the quantum chernoff bound , @xmath27 , between two states related by a local unitary transformation . lemma 1 in ref . @xcite shows that in the finite - dimensional case if @xmath301 , where @xmath302 is a hermitian matrix , the infimum is achieved for @xmath31 in eq . ( [ appchernoff ] ) . the same proof can be applied as well to @xmath302 one - qubit traceless unitary matrix , since it is hermitian . for gaussian states of infinite - dimensional , continuous - variable systems , we are able to formulate and prove the following theorem : for two - mode gaussian states @xmath303 with covariance matrix @xmath63 of the form eq . ( [ sqthcorm ] ) with @xmath71 , the @xmath27 for the pair @xmath304 , where @xmath264 is any traceless local symplectic transformation , is achieved for @xmath31 , namely : @xmath305 first let us notice that for any two quantum states @xmath25 and @xmath26 function @xmath306 is convex in @xmath307 , which is proven in @xcite . we will show that if the two states @xmath25 and @xmath26 satisfy the assumptions of the theorem , @xmath306 is symmetric with respect to exchange @xmath308 . these two properties imply the theorem . let us show the symmetry with respect to exchange @xmath308 . from @xcite we know that for the states with vanishing first moments @xmath306 is given in eq . ( [ qubar ] ) . the numerator of this formula is already symmetric with respect to exchange @xmath308 for the states related by any unitary transformation . to proof the theorem we only need to show the symmetry of the determinant in denominator of eq . ( [ qubar ] ) , @xmath309 , for the states which have the covariance matrices @xmath310 and @xmath311 . consider the determinant from the above formula @xmath312 the last equality is implied by the following argument . the form of the most general single - mode traceless symplectic transformation ( euler decomposition ) @xcite is : @xmath313 where @xmath314 is positive . the traceless condition is obtained by imposing @xmath315 . immediate check gives us that @xmath316 . the minus sign is irrelevant in expression @xmath317 and can be omitted . this completes the proof . to express the upper bound on the maximum probability of error for undisplaced gaussian states of the covariance matrix given in eq . ( [ sqthcorm ] ) with @xmath71 and its counterpart related to it by traceless symplectic transformations @xmath264 we maximize the quantum chernoff bound over the set of these transformations . formula eq . ( [ appchs ] ) with @xmath31 is maximized if @xmath318}$ ] is minimized over the set @xmath319 . the most general gaussian single - mode unitary transformation is given in eq . ( [ apps ] ) . the determinant @xmath318}$ ] does not depend on @xmath320 and achieves its minimum for @xmath321 . this can be proved by direct check of the first and second derivatives . substituting @xmath321 in eq . ( [ apps ] ) yields the transformation which maximizes the quantum chernoff bound , namely @xmath322 which is the transformation corresponding to a local @xmath15 phase shift @xmath114 . here we discuss the derivatives of the distinguishability functions and their behavior . for the quantum chernoff bound , @xmath27 , the derivatives over the entries of the covariance matrix , eq . ( [ sqthcorm ] ) , read @xmath323 it is immediate to verify that the partial derivative of @xmath27 , eq . ( [ dqda ] ) , is always positive and the partial derivative , eq . ( [ dqdc ] ) , is always negative . the derivatives of the uhlmann fidelity are over the entries @xmath218 and @xmath325 of the covariance matrix , eq . ( [ sqthcorm ] ) , in the range of values corresponding to physical states , eq . ( [ uncert ] ) . the partial derivative @xmath326 is always positive while @xmath327 is always negative.,width=188 ] also in the case of @xmath205 the derivative over the diagonal entry of the covariance matrix at constant off - diagonal elements is positive , while the derivative over the off - diagonal entries at constant diagonal entries is negative . in the case of non - symmetric @xmath328 discussed in sec . [ noiseenhancedqcb ] , the partial derivative of @xmath27 , eq . ( [ derivchernoffas ] ) , with respect to @xmath146 at constant @xmath147 and @xmath74 reads @xmath329 where since @xmath331 the derivative , eq . ( [ derivovernth1 ] ) , is positive only if @xmath332 , negative only if @xmath181 , and vanishes identically for symmetric stss , namely for @xmath170 , yielding a noise - independent @xmath27 . from the above results it follows that in the range @xmath181 the @xmath27 is a monotonically decreasing function of the number of thermal photons in the first mode ( increasing local thermal noise ) and vanishes asymptotically , together with the probability of error , as @xmath236

astrophysicists have always been keen to exploit technology to better understand the universe . @xmath0-body simulations predate the use of digital computers with the use of light bulbs and light intensity measurements as an analog of gravity for manual simulations of a many - body self - gravitating system @xcite . astrophysical objects including planets , individual stars , interstellar clouds , star clusters , galaxies , accretion disks , clusters of galaxies through to large scale structure have all the been the subject of numerical investigations . the most challenging extreme is probably the evolution of space - time itself in computational general relativistic simulations of colliding neutron stars and black holes . since the advent of digital computers , improvements in storage and processing power have dramatically increased the scale of achievable simulations . this , in turn , has driven remarkable progress in algorithm development . increasing problem sizes have forced simulators who were once content with @xmath1 algorithms to pursue more complex @xmath2 and , with limitations , even @xmath3 algorithms and adaptivity in space and time . in this context we present gasoline , a parallel @xmath0-body and gasdynamics code , which has enabled new light to be shed on a range of complex astrophysical systems . the discussion is presented in the context of future directions in numerical simulations in astrophysics , including fundamental limitations in serial and parallel . we begin by tracing past progress in computational astrophysics with an initial focus on self - gravitating systems ( @xmath0-body dynamics ) in section [ gravity ] . gasoline evolved from the pkdgrav parallel @xmath0-body tree code designed by @xcite . the initial modular design of pkdgrav and a collaborative programming model using cvs for source code management has facilitated several simultaneous developments from the pkdgrav code base . these include inelastic collisions , gas dynamics ( gasoline ) and star formation . in section [ grav ] we summarize the essential gravity code design , including the parallel data structures and the details of the tree code as applied to calculating gravitational forces . we complete the section with a brief examination of the gravitational force accuracy . in section [ gas ] we examine aspects of hydrodynamics in astrophysical systems to motivate smoothed particle hydrodynamics ( sph ) as our choice of fluid dynamics method . we describe the gasoline sph implementation in section [ sph ] , including neighbour finding algorithms and the cooling implementation . interesting astrophysical systems usually exhibit a large range of time scales . tree codes are very adaptable in space ; however , time - adaptivity has become important for leading edge numerical simulations . in section [ stepping ] we describe our hierarchical timestepping scheme . following on in section [ performance ] we examine the performance of gasoline when applied to challenging numerical simulations of real astrophysical systems . in particular , we examine the current and potential benefits of multiple timesteps for time adaptivity in areas such as galaxy and planet formation . we present astrophysically oriented tests used to validate gasoline in section [ tests ] . we conclude by summarizing current and proposed applications for gasoline . gravity is the key driving force in most astrophysical systems . with assumptions of axisymmetry or perturbative approaches an impressive amount of progress has been made with analytical methods , particularly in the areas of solar system dynamics , stability of disks , stellar dynamics and quasi - linear aspects of the growth of large scale structure in the universe . in many systems of interest , however , non - linear interactions play a vital role . this ultimately requires the use of self - gravitating @xmath0-body simulations . fundamentally , solving gravity means solving poisson s equation for the gravitational potential , @xmath4 , given a mass density , @xmath5 : @xmath6 where @xmath7 is the newtonian gravitational constant . in a simulations with discrete bodies it is common to start from the explicit expression for the acceleration , @xmath8 on a given body in terms of the sum of the influence of all other bodies , @xmath9 where the @xmath10 and @xmath11 are the position and masses of the bodies respectively . when attempting to model collisionless systems , these same equations are the characteristics of the collisionless boltzmann equation , and the bodies can be thought of as samples of the distribution function . in practical work it is essential to soften the gravitational force on some scale @xmath12 to avoid problems with the integration and to minimize two - body scattering in cases where the bodies represent a collisionless system . early @xmath0-body work such as studies of relatively small stellar systems were approached using a direct summation of the forces on each body due to every other body in the system @xcite . this direct @xmath1 approach is impractical for large numbers of bodies , @xmath0 , but has enjoyed a revival due to incredible throughput of special purpose hardware such as grape @xcite . the grape hardware performs the mutual force calculation for sets of bodies entirely in hardware and remains competitive with other methods on more standard floating hardware up to @xmath13 . a popular scheme for larger @xmath0 is the particle - mesh ( pm ) method which has long been used in electrostatics and plasma physics . the adoption of pm was strongly related to the realization of the existence of the @xmath2 fast fourier transform ( fft ) in the 1960 s . the fft is used to solve for the gravitational potential from the density distribution interpolated onto a regular mesh . in astrophysics sub - mesh resolution is often desired , in which case the force can be corrected on sub - mesh scales with local direct sums as in the particle - particle particle - mesh ( p@xmath14 m ) method . pm is popular in stellar disk dynamics , and p@xmath14 m has seen widespread adoption in cosmology . pm codes have similarities with multigrid and other iterative schemes . however , working in fourier space is not only more efficient , but it also allows efficient force error control through optimization of the green s function and smoothing . fourier methods are widely recognised as ideal for large , fairly homogeneous , periodic gravitating simulations . multigrid has some advantages in parallel due to the local nature of the iterations . the particle - particle correction can get expensive when particles cluster in a few cells . both multigrid and p@xmath14 m can adapt to address this via a hierarchy of sub - meshes . with this approach the serial slow down due to heavy clustering tends toward a fixed multiple of the unclustered run speed . in applications such as galactic dynamics where high resolution in phase space is desirable and particle noise is problematic , the smoothed gravitational potentials provided by an expansion in modes is useful . pm does this with fourier modes ; however , a more elegant approach is the self - consistent field method ( scf ) @xcite . using a basis set closely matched to the evolving system dramatically reduces the number of modes to be modelled ; however , the system must remain close to axi - symmetric and similar to the basis . scf parallelizes well and is also used to generate initial conditions such as stable individual galaxies that might be used for merger simulations . a current popular choice is to use tree algorithms which are inherently @xmath2 . this approach recognises that details of the remote mass distribution become less important for accurate gravity with increasing distance . thus the remote mass distribution can be expanded in multipoles on the different size scales set by a tree - node hierarchy . the appropriate scale to use is set by the opening angle subtended by the tree - node bounds relative to the point where the force is being calculated . the original barnes - hut @xcite method employed oct - trees but this is not especially advantageous , and other trees also work well @xcite . the tree approach can adapt to any topology , and thus the speed of the method is somewhat insensitive to the degree of clustering . once a tree is built it can also be re - used as an efficient search method for other physics such as particle based hydrodynamics . a particularly useful property of tree codes is the ability to efficiently calculate forces for a subset of the bodies . this is critical if there is a large range of time - scales in a simulation and multiple independent timesteps are employed . at the cost of force calculations no longer being synchronized among the particles substantial gains in time - to - solution may be realized . multiple timesteps are particularly important for current astrophysical applications where the interest and thus resolution tends to be focused on small regions within large simulated environments such as individual galaxies , stars or planets . dynamical times can become very short for small numbers of particles . p@xmath14 m codes are faster for full force calculations but are difficult to adapt to calculate a subset of the forces . in order to treat periodic boundaries with tree codes it is necessary to effectively infinitely replicate the simulation volume which may be approximated with an ewald summation @xcite . an efficient alternative which is seeing increasing use is to use trees in place of the direct particle - particle correction to a particle - mesh code , often called tree - pm @xcite . the fast multipole method ( fmm ) recognises that the applied force as well as the mass distribution may be expanded in multipoles . this leads to a force calculation step that is @xmath3 as each tree node interacts with a similar number of nodes independent of @xmath0 and the number of nodes is proportional to the number of bodies . building the tree is still @xmath2 but this is a small cost for simulations up to @xmath15 @xcite . the method used spherical harmonic expansions where the desired accuracy is achieved solely by changing the order of the expansions . for the majority of astrophysical applications the allowable force accuracies make it much more efficient to use fixed order cartesian expansions and an opening angle criterion similar to standard tree codes @xcite . this approach has the nice property of explicitly conserving momentum ( as do pm and p@xmath14 m codes ) . the prefactor for cartesian fmm is quite small so that it can outperform tree codes even for small @xmath0 @xcite . it is a significantly more complex algorithm to implement , particularly in parallel . one reason widespread adoption has not occurred is that the speed benefit over a tree code is significantly reduced when small subsets of the particles are having forces calculated ( e.g. for multiple timesteps ) . gasoline is built on the pkdgrav framework and thus uses the same gravity algorithms . the pkdgrav parallel @xmath0-body code was designed by stadel and developed in conjunction with quinn beginning in the early 90 s . this includes the parallel code structure and core algorithms such as the tree structure , tree walk , hexadecapole multipole calculations for the forces and the ewald summation . there have been additional contributions to the gravity code by richardson and wadsley in the form of minor algorithmic modifications and optimizations . the latter two authors became involved as part of the collisions and gasoline extensions of the original pkdgrav code respectively . we have summarized the gravity method used in the production version of gasoline without getting into great mathematical detail . for full technical details on pkdgrav the reader is referred to . gasoline is fundamentally a tree code . it uses a variant on the k - d tree ( see below ) for the purpose of calculating gravity , dividing work in parallel and searching . designed pkdgrav from the start as a parallel code . there are four layers in the code . the master layer is essentially serial code that orchestrates overall progress of the simulation . the processor set tree ( pst ) layer distributes work and collects feedback on parallel operations in an architecture independent way using mdl . the machine dependent layer ( mdl ) is a relatively short section of code that implements remote procedure calls , effective memory sharing and parallel diagnostics . all processors other than the master loop in the pst level waiting for directions from the single process executing the master level . directions are passed down the pst in a tree based @xmath16 procedure that ends with access to the fundamental bulk particle data on every node at the pkd level . the parallel k - d ( pkd ) layer is almost entirely serial but for a few calls to mdl to access remote data . the pkd layer manages local trees for gravity and particle data and is where the physics is implemented . this modular design enables new physics to be coded at the pkd level without requiring detailed knowledge of the parallel framework . pkdgrav departed significantly from the original @xmath0-body tree code designs of by using 4th ( hexadecapole ) rather than 2nd ( quadrupole ) order multipole moments to represent the mass distribution in cells at each level of the tree . this results in less computation for the same level of accuracy : better pipelining , smaller interaction lists for each particle and reduced communication demands in parallel . the current implementation in gasoline uses reduced moments that require only @xmath17 terms to be stored for the @xmath18 moment . for a detailed discussion of the accuracy and efficiency of the tree algorithm as a function the order of the multipoles used see @xcite and @xcite . the original k - d tree @xcite was a balanced binary tree . gasoline divides the simulation in a similar way using recursive partitioning . at the pst level this is parallel domain decomposition and the division occurs on the longest axis to recursively divide the work among the remaining processors . even divisions occur only when an even number of processors remains . otherwise the work is split in proportion to the number of processors on each side of the division . thus , gasoline may use arbitrary numbers of processors and is efficient for flat topologies without adjustment . at the pst level each processor has a local rectangular domain within which a local binary tree is built . the structure of the lower tree is important for the accuracy of gravity and efficiency of other search operations such as neighbour finding required for sph . oct - trees are traditional in gravity codes . in contrast , the key data structures used by gasoline are spatial binary trees . one immediate gain is that the local trees do not have to respect a global structure and simply continue from the pst level domain decomposition in parallel . the binary tree determines the hierarchical representation of the mass distribution with multipole expansions , of which the root node or cell encloses the entire simulation volume . the local gravity tree is built by recursively bisecting the longest axis of each cell which keeps the cells axis ratios close to one . in contrast , cells in standard k - d trees can have large axis ratios which lead to large multipoles and with correspondingly large gravity errors . at each level the dimensions of the cells are squeezed to just contain the particles . this overcomes the empty cell problem of un - squeezed spatial bisection trees . the sph tree is currently a balanced k - d tree ; however , testing indicates that the efficiency gain is slight and it is not worth the cost of an additional tree build . the top down tree build process is halted when @xmath19 or fewer particles remain in a cell . stopping the tree with @xmath19 particles in a leaf cell reduces the storage required for the tree and makes both gravity and search operations more efficient . for these purposes @xmath20 is a good choice . once the gravity tree has been built there is a bottom - up pass starting from the buckets and proceeding to the root , calculating the center of mass and the multipole moments of each cell from the center of mass and moments of each of its two sub - cells . gasoline calculates the gravitational accelerations using the well known tree - walking procedure of the algorithm , except that it collects interactions for entire buckets rather than single particles . this amortizes the cost of tree traversal for a bucket over all its particles . in the tree building phase , gasoline assigns to each cell of the tree an _ opening radius _ about its center - of - mass . this is defined as , @xmath21 where @xmath22 is the maximum distance from a particle in the cell to the center - of - mass of the cell . the _ opening angle _ , @xmath23 , is a user specified accuracy parameter which is similar to the traditional @xmath23 parameter of the barnes - hut code ; notice that decreasing @xmath23 in equation [ eq:1 ] , increases @xmath24 . the opening radii are used in the _ walk _ phase of the algorithm as follows : for each bucket @xmath25 , gasoline starts descending the tree , opening those cells whose @xmath24 intersect with @xmath25 ( see figure [ fig:2.2 ] ) . if a cell is opened , then gasoline repeats the intersection - test with @xmath25 for the cell s children otherwise , the cell is added to the _ particle - cell interaction list _ of @xmath25 . when gasoline reaches the leaves of the tree and a _ bucket _ @xmath26 is opened , all of @xmath26 s particles are added to the _ particle - particle interaction _ list of @xmath25 . once the tree has been traversed in this manner we can calculate the gravitational acceleration for each particle of @xmath25 by evaluating the interactions specified in the two lists . gasoline uses the hexadecapole multipole expansion to calculate particle - cell interactions . the particle - particle interactions are softened to lessen two - body relaxation effects that compromise the attempt to model continuous fluids , including the collisionless dark matter fluid . in gasoline the particle masses are effectively smoothed in space using the same spline form employed for sph in section [ sph ] . this means that the gravitational forces vanish at zero separation and return to newtonian @xmath27 at a separation of @xmath28 where @xmath29 is the gravitational softening applied to each particle . in this sense the gravitational forces are well matched to the sph forces . a disadvantage of tree codes is that they must deal explicitly with periodic boundary conditions ( as are usually required for cosmology ) . gasoline incorporates periodic boundaries via the ewald summation technique where the force is divided into short and long range components . the gasoline implementation differs from that of in using a new technique due to based on an hexadecapole moment expansion of the fundamental cube to drastically reduce the work for the long range ewald sum that approximates the infinite periodic replications . for each particle the computations are local and fixed , and thus the algorithm scales exceedingly well . there is still substantial additional work in periodic simulations because particles interact with cells and particles in nearby replicas of the fundamental cube . the tree opening criteria places a bound on the relative error due to a single particle - cell interaction . as gasoline uses hexadecapoles the error bound improves rapidly as the opening angle , @xmath23 , is lowered . the relationship between the typical ( e.g. rms ) relative force error and opening angle is not a straight - forward power - law in @xmath23 because the net gravitational forces on each particle result from the cancellation of many opposing forces . in figure [ accfig ] , we show a histogram of the relative acceleration errors for a cosmological gasoline simulations at two different epochs for a range of opening angles . we have plotted the error curves as cumulative fractions to emphasize the limited tails to high error values . for typical gasoline simulations we commonly use @xmath30 which gives an _ rms _ relative error of 0.0039 for the clustered final state referred to by the left panel of figure [ accfig ] . the errors relative to the mean acceleration ( figure [ accfigabs ] ) are larger ( rms 0.0083 ) but of less interest for highly clustered cases . = 5.5 in = 5.5 in as a simulated medium becomes more uniform the net gravitational accelerations approach zero . in a tree code the small forces result from the cancellation of large opposing forces . this is the case in cosmology at early times when the perturbations ultimately leading to cosmological structures are still small . in this situation it is essential to tighten the cell - opening criterion to increase the relative accuracy so that the net accelerations are sufficiently accurate . for example in the right hand panels of figures [ accfig ] and [ accfigabs ] the errors are larger , and at @xmath31 , the _ rms _ relative error is 0.041 for @xmath30 . however , here the absolute errors are lower by nearly a factor of two in rms ( 0.026 ) as shown in figure [ accfigabs ] . at early times when the medium fairly homogeneous growth is driven by local gradients , and thus acceleration errors normalized to the mean acceleration provide the better measure of accuracy as large relative errors are meaningless for accelerations close to zero . to ensure accurate integrations we switch to values such as @xmath32 before @xmath33 , giving an _ rms _ relative error of 0.0077 and an _ rms _ error of 0.0045 normalized to the mean absolute acceleration for the @xmath31 distribution ( a reasonable start time for a cosmological simulation on this scale ) . astrophysical systems are predominantly at very low physical densities and experience wide - ranging temperature variations . most of the material is in a highly compressible gaseous phase . in general this means that a perfect adiabatic gas is an excellent approximation for the system . microscopic physical processes such as shear viscosity and diffusion can usually be neglected . high - energy processes and the action of gravity tend to create large velocities so that flows are both turbulent and supersonic : strong shocks and very high mach numbers are common . radiative cooling processes can also be important ; however , the timescales can often be much longer or shorter than dynamical timescales . in the latter case isothermal gas is often assumed for simplicity . in many areas of computational astrophysics , particularly cosmology , gravity tends to be the dominating force that drives the evolution of the system . visible matter , usually in the form of radiating gas , provides the essential link to observations . radiative transfer is always present but may not significantly affect the energy and thus pressure of the gas during the simulation . fluid dynamics solvers can be broadly classified into eulerian or lagrangian methods . eulerian methods use a fixed computational mesh through which the fluid flows via explicit advection terms . regular meshes provide for ease of analysis and thus high order methods such as ppm @xcite and tvd schemes have been developed . the inner loops of mesh methods can often be pipelined for high performance . lagrangian methods follow the evolution of fluid parcels via the full ( comoving ) derivatives . this requires a deforming mesh or a mesh - less method such as smoothed particle hydrodynamics ( sph ) @xcite . data management is more complex in these methods ; however , advection is handled implicitly and the simulation tends to naturally adapt to follow density contrasts . large variations in physical length scales in astrophysics have limited the usefulness of eulerian grid codes . adaptive mesh refinement ( amr ) @xcite overcomes this at the cost of data management overheads and increased code complexity . in the cosmological context there is the added complication of dark matter . there is more dark matter than gas in the universe so it dominates the gravitational potential . perturbations present on all scales in the dark matter guide the formation of gaseous structures including galaxies and the first stars . a fundamental limit to amr in computational cosmology is matching the amr resolution to the underlying dark matter resolution . particle based lagrangian methods such as sph are well matched to this constraint . a useful feature of lagrangian simulations is that bulk flows ( which can be highly supersonic in the simulation frame ) do not limit the timesteps . particle methods are well suited to rapidly rotating systems such as astrophysical disks where arbitrarily many rotation periods may have to be simulated ( e.g. sph explicitly conserves angular momentum ) . a key concern for all methods is correct angular momentum transport . smoothed particle hydrodynamics is an approach to hydrodynamical modelling developed by and . it is a particle method that does not refer to grids for the calculation of hydrodynamical quantities : all forces and fluid properties are found on moving particles eliminating numerically diffusive advective terms . the use of sph for cosmological simulations required the development of variable smoothing to handle huge dynamic ranges @xcite . sph is a natural partner for particle based gravity . sph has been combined with p@xmath14 m @xcite , adaptive p@xmath14 m , grape @xcite and tree gravity @xcite . parallel codes using sph include hydra mpi , parallel treesph @xcite and the gadget tree code @xcite . the basis of the sph method is the representation and evolution of smoothly varying fluid quantities whose value is only known at disordered discrete points in space occupied by particles . particles are the fundamental resolution elements comparable to cells in a mesh . sph functions through local summation operations over particles weighted with a smoothing kernel , @xmath34 , that approximates a local integral . the smoothing operation provides a basis from which to obtain derivatives . thus , estimates of density related physical quantities and gradients are generated . the summation aspect led to sph being described as a monte carlo type method ( with @xmath35 errors ) however it was shown by that the method is more closely related to interpolation theory with errors @xmath36 , where @xmath37 is the number of dimensions . a general smoothed estimate for some quantity @xmath38 at particle @xmath39 given particles @xmath40 at positions @xmath41 takes the form : @xmath42 where @xmath43 is a kernel function and @xmath44 is a smoothing length indicative of the range of interaction of particle @xmath40 . it is common to convert this particle - weighted sum to volume weighting using @xmath45 in place of @xmath46 where the @xmath47 and @xmath48 are the particle masses and densities , respectively . for momentum and energy conservation in the force terms , a symmetric @xmath49 is required . we use the kernel - average first suggested by , @xmath50 for @xmath51 we use the standard spline form with compact support where @xmath52 if @xmath53 @xcite . we employ a fairly standard implementation of the the hydrodynamics equations of motion for sph @xcite . density is calculated from a sum over particle masses @xmath47 , @xmath54 the momentum equation is expressed , @xmath55 where @xmath56 is pressure , @xmath57 velocity and the artificial viscosity term @xmath58 is given by , @xmath59 where @xmath60 , @xmath61 and @xmath62 is the sound speed . @xmath63 and @xmath64 are coefficients we use for the terms representing shear and von neumann - richtmyer ( high mach number ) viscosities respectively . when simulating strongly rotating systems we use the multiplicative switch @xmath65 to suppress the viscosity in non - shocking , shearing environments . the pressure averaged energy equation ( analogous to equation [ sphmom ] ) conserves energy exactly in the limit of infinitesimal time steps but may produce negative energies due to the @xmath56 term if significant local variations in pressure occur . we employ the following equation which also conserves energy exactly in each pairwise exchange but is dependent only on the local particle pressure , @xmath66 where @xmath67 is the internal energy of particle @xmath39 , which is equal to @xmath68 for an ideal gas . in this formulation entropy is closely conserved making it similar to alternative entropy integration approaches such as that proposed by . sph derivative estimates , such as the rate of change of thermal energy , vary sufficiently from the exact answer so that over a full cosmological simulation the cooling due to universal expansion will be noticeably incorrect . in this case we use comoving divergence estimates and add the cosmological expansion term explicitly . finding neighbours of particles is a useful operation . a neighbour list is essential to sph , but it can also be a basis for other local estimates , such as a dark matter density , as a first step in finding potential colliders or interactors via a short range force . stadel developed an efficient search algorithm using priority queues and a k - d tree ball - search to locate the @xmath69-nearest neighbours for each particle ( freely available as the smooth utility at http://www-hpcc.astro.washington.edu ) . for gasoline we use a parallel version of the algorithm that caches non - local neighbours via mdl . the sph interaction distance @xmath70 is set equal to the @xmath69-th neighbour distance from particle @xmath39 . we use an exact number of neighbours . the number is set to a value such as 32 or 64 at start - up . we have also implemented a minimum smoothing length which is usually set to be comparable to the gravitational softening . to calculate the fluid accelerations using sph we perform two smoothing operations . first we sum densities and then forces using the density estimates . to get a kernel - averaged sum for every particle ( equations [ sum],[kernelavg ] ) it is sufficient to perform a gather operation over all particles within @xmath70 of every particle @xmath39 . if only a subset of the particles are active and require new forces , all particles for which the active set are neighbours must also perform a gather so that they can scatter their contribution to the active set . finding these scatter neighbours requires solving the @xmath69-inverse nearest neighbour problem , an active research area in computer science . fortunately , during a simulation the change per step in @xmath71 for each particle typically less than 2 - 3 percent , so it is sufficient to find scatter neighbours , @xmath40 for which some active particles is within @xmath72 . we use a tree search where the nodes contain sph interaction bounds for their particles estimated with @xmath73 . a similar scheme has been employed by . for the forces sum the inactive neighbours need density values which can be estimated using the continuity equation or calculated exactly with a second inverse neighbour search . in astrophysical systems the cooling timescale is usually short compared to dynamical timescales which often results in temperatures that are close to an equilibrium set by competing heating and cooling processes . we have implemented a range of cases including : adiabatic ( no cooling ) , isothermal ( instant cooling ) , and implicit energy integration . hydrogen and helium cooling processes have been incorporated . ionization fractions are calculated assuming equilibrium for efficiency . gasoline optionally adds heating due to feedback from star formation , an uniform uv background or using user defined functions . the implicit integration uses a stiff equation solver assuming that the hydrodynamic work and density are constant across the step . the second order nature of the particle integration is maintained using an implicit predictor step when needed . internal energy is required on every step to calculate pressure forces on the active particles . the energy integration is @xmath3 but reasonably expensive . to avoid integrating energy for every particle on the smallest timestep we extrapolate each particle forward on its individual dynamical timestep and use linear interpolation to estimate internal energy at intermediate times as required . the range of physical scales in astrophysical systems is large . for example current galaxy formation simulations contain 9 orders of magnitude variation in density . the dynamical timescale for gravity scales as @xmath74 and for gas it scales as @xmath75 . for an adiabatic gas the local dynamical time scales as @xmath76 . with gas cooling ( or the isothermal assumption ) simulations can achieve very high gas densities . in most cases gas sets the shortest dynamical timescales , and thus gas simulations are much more demanding ( many more steps to completion ) than corresponding gravity only simulations . time adaptivity can be very effective in this regime . = 5.5 in gasoline incorporates the timestep scheme described as kick - drift - kick ( kdk ) in . the scheme uses a fixed overall timestep . starting with all quantities synchronised , velocities and energies are updated to the half - step ( half - kick ) , followed by a full step position update ( drift ) . the positions alone are sufficient to calculate gravity at the end of the step ; however , for sph , velocities and thermal energies are also required and obtained with a predictor step using the old accelerations . finally another half - kick is applied synchronising all the variables . without gas forces this is a symplectic leap - frog integration . the leap - frog scheme requires only one force evaluation and minimum storage . while symplectic integration is not required in cosmology it is critical in solar system integrations or any system where natural instabilities occur over very many dynamical times . an arbitrary number of sub - stepping rungs factors of two smaller may be used as shown in figure [ timestep ] . the scheme is no longer strictly symplectic if particles change rungs during the integration which they generally must do to satisfy their individual timestep criteria . after overheads , tree - based force calculation scales approximately with the number of active particles so large speed - ups may be realised in comparison to single stepping ( see section [ performance ] ) . figure [ timestep ] compares kdk with drift - kick - drift ( dkd ) , also discussed in . for kdk the force calculations for different rungs are synchronised . in the limit of many rungs this results in half as many force calculation times with their associated tree builds compared to dkd . kdk also gives significantly better momentum and energy conservation for the same choice of timestep criteria . we use standard timestep criteria based on the particle acceleration , and for gas particles , the courant condition and the expansion cooling rate . @xmath77 @xmath78 is the maximum value of @xmath79 ( from equation [ artifvisc ] ) over interactions between pairs of sph particles . for cosmology in place of the comoving velocity @xmath80 , we use the momentum @xmath81 which is canonical to the comoving position , @xmath82 . as described in detail in appendix a of @xcite , this results in a separable hamiltonian which may be integrated straightforwardly using the drift and kick operators , @xmath83 where @xmath4 is the perturbed potential given by @xmath84 , @xmath85 is the cosmological expansion factor and @xmath86 is the mean density . thus no hubble drag term is required in the equations of motion , and the integration is perfectly symplectic in the single stepping case . = 5.5 in gasoline was built on the pkdgrav @xmath0-body code which achieves excellent performance on pure gravity in serial and parallel . performance can be measured in floating point calculations per second but the measure of most interest to researchers is the science rate . we define this in terms of resolution elements updated per unit wallclock time . in the case of gasoline this is particles updated per second . this measure directly determines how long it takes to finish the simulation . figure [ fig : particlespersecond ] shows the scaling of particles per second with numbers of alpha processors for a single update for velocities and positions for all particles . this requires gravitational forces for all particles and sph for the gas particles ( half the total ) . the simulation used is a @xmath87 gas and @xmath87 dark matter particle , purely adiabatic cosmological simulation ( @xmath88cdm ) in a 200 mpc box at the final time ( redshift , @xmath89 ) . at this time the simulation is highly clustered locally but material is still well distributed throughout the volume . thus , it is still possible to partition the volume to achieve fairly even work balancing among a fair number of processors . as a rule of thumb we aim to have around 100,000 particles per processor . as seen in the figure , the gravity scales very well . for pure gravity around @xmath90 scaling efficiency can be achieved in this case . the cache based design has small memory overheads and uses the large amounts of gravity work to hide the communication overheads associated with filling the cache with remote particle and cell data . with gas the efficiency is lowered because the data structures to be passed are larger and there is less work per data element to hide the communication costs . thus the computation to communication ratio is substantially lowered with more processors . fixed costs such as the treebuilds for gravity and sph scale well , but the gas - only costs peel away from the more efficient gravity scaling . when the overall rate is considered , gasoline is still making substantial improvements in the time to solution going from 32 to 64 processors ( lowest curve in the figure ) . the overall rate includes costs for domain decomposition and @xmath3 updates of the particle information such as updating the positions from the velocities . the net science rate compares well with other parallel tree and sph codes . = 5.5 in figure [ fig : particlespersecond ] only treats single timestepping . multiple timesteps provide an additional speed - up in the science rate of a factor typically in the range of 2 - 5 that is quite problem dependent . the value of individual particle timesteps is illustrated separately in figure [ fig : multistep ] . for this example we analysed the time spent by gasoline doing a single major step ( around 13 million years ) of a million particle galaxy formation simulation at redshift @xmath91 . the simulation used a range of 128 in substeps per major step . the top curve in the figure is the cost for single stepping . it rises exponentially , doubling with every new bin , and drops off because the last bin was not always occupied . sample numbers of particles in the bins ( going from 0 to 7 ) were 473806 , 80464 , 63708 , 62977 , 85187 , 129931 , 1801 and 20 respectively . using individual timesteps gasoline was running 4 times faster than an equivalent single stepping code . the test was performed in parallel on 8 processors , and it is heartening that despite the added difficulties of load balancing operations on subsets of the particles the factor of 4 benefit was realized . tree building and other fixed costs that do not scale with the number of active particles can be substantially reduced using tree - repair and related methods which would bring the speed - up to around 5 . in the limit of uniform costs per particle independent of the number of active particles the speed - up would approach 10 . in this example the timestep range was limited partly due to restrictions imposed on the sph smoothing length . in future work we anticipate a wider range of timestep bins . with a few particles consistently on a timestep 1/256th of the major step the theoretical speed - up factor is over 30 . in practice , overheads will always limit the achievable speed - up . in the limit of very large numbers of rungs the current code would asymptotically approach a factor of 10 speedup . with the fixed costs of tree builds and domain decomposition made to scale with the number of active particles ( e.g. tree - repair ) , the asymptotic speed up achievable with gasoline would be 24 times . the remaining overheads are increasingly difficult to address . for small runs , such as the galaxy formation example used here , the low numbers of the particles on the shortest timesteps make load balancing difficult . if more than 16 processors are used with the current code on this run there will be idle processors for a noticeable fraction of the time . though the time to solution is reduced with more processors , it is an inefficient use of computing resources . the ongoing challenge is to see if better load balance through more complex work division offsets increases in communication and other parallel overheads . = 5.5 in there are three key points to keep in mind when evaluating the performance of sph on highly symmetric tests . the first is that the natural particle configuration is a semi - regular three dimensional glass rather than a regular mesh . the second is that individual particle velocities are smoothed before they affect the dynamics so that the low level noise in individual particle velocities is not particularly important . the dispersion in individual velocities is related to continuous settling of the irregular particle distribution . this is particularly evident after large density changes . thirdly , the sph density is a smoothed estimate . any sharp step in the number density of particles translates into a density estimate that is smooth on the scale of @xmath92 particle separations . when relaxed irregular particle distributions are used , sph resolves density discontinuities close to this limit . as a lagrangian method sph can also resolve contact discontinuities just as tightly without the advective spreading of some eulerian methods . we have performed standard shock tube tests used for the original treesph @xcite . we find the best results with the pairwise viscosity of equation [ artifvisc ] which is marginally better than the bulk viscosity formulation for this test . the one dimensional tests often shown do not properly represent the ability of sph to model shocks on more realistic problems . the results of figure [ tube ] demonstrate that sph can resolve discontinuities in a shock tube very well when the problem is setup to be comparable to the environment in a typical three dimensional simulation . = 5.5 in the shocks of the previous example have a fairly low mach number compared to astrophysical shocks found in collapse problems . @xcite first introduced a spherical adiabatic collapse as a test of gasdynamics with gravity . this test is nearly equivalent to the self - similar collapse of and has comparable shock strengths . we compare gasoline results on this problem with very high resolution 1d lagrangian mesh solutions in figure [ adcollfig ] . we used a three - dimensional glass initial condition . the solution is excellent with 2 particle spacings required to model the shock . the deviation at the inner radii is a combination of the minimum smoothing length ( 0.01 ) and earlier slight over - production of entropy at a less resolved stage of the collapse . the pre - shock entropy generation ( bottom left panel of fig . [ adcollfig ] occurs in any strongly collapsing flow and is present for both the pairwise ( equation [ artifvisc ] ) and divergence based artificial viscosity formulations . the post - shock entropy values are correct . the rotating isothermal cloud test examines the angular momentum transport in an analogue of a typical astrophysical collapse with cooling . grid methods must explicitly advect material across cell boundaries which leads to small but systematic angular momentum non - conservation and transport errors . sph conserves angular momentum very well , limited only by the accuracy of the time integration of the particle trajectories . however , the sph artificial viscosity that is required to handle shocks has an unwanted side - effect in the form of viscous transport away from shocks . the magnitude of this effect scaling with the typical particle spacing , and it can be combatted effectively with increased resolution . the switch detects shearing regions so that the viscosity can be reduced where strong compression is not also present . we modelled the collapse of a rotating , isothermal gas cloud . this test is similar to a tests performed by @xcite and @xcite except that we have simplified the problem in the manner of @xcite . we use a fixed @xcite ( concentration @xmath93=11 , mass=@xmath94 ) potential without self - gravity to avoid coarse gravitational clumping with associated angular momentum transport . this results in a disk with a circular velocity of 220 km / s at 10 kpc . the @xmath95 , @xmath96 kpc gas cloud was constructed with particle locations evolved into a uniform density glass initial condition and set in solid body rotation ( @xmath97 ) corresponding to a rotation parameter @xmath98 . the gas was kept isothermal at @xmath99 k rather than using a cooling function to make the test more general . the corresponding sound speed of @xmath100 km / s implies substantial shocking during the collapse . = 5.5 in we ran simulations using 64 , 125 , 500 and 4000 particle clouds for @xmath101 gyr ( 43 rotation periods at @xmath100 kpc ) . we present results for monaghan viscosity , bulk viscosity ( hernquist - katz ) and our default set - up : monaghan viscosity with the balsara switch . results showing the effect of resolution are shown in figure [ figrotsummary ] . both monaghan viscosity with the balsara switch and bulk viscosity result in minimal angular momentum transport . the monaghan viscosity without the switch results in steady transport of angular momentum outwards with a corresponding mass inflow . the monaghan viscosity case was run with and without multiple timesteps . with multiple timesteps the particle momentum exchanges are no longer explicitly synchronized and the integrator is no longer perfectly symplectic . despite this the evolution is very similar , particularly at high resolution . the resolution dependence of the artificial viscosity is immediately apparent as the viscous transport drastically falls with increasing particle numbers . it is worth noting that in hierarchical collapse processes the first collapses always involve small particle numbers . our results are consistent with the results of @xcite . in that paper self - gravity of the gas was also included , providing an additional mechanism to transport angular momentum due to mass clumping . as a result , their disks show a gradual transport of angular momentum even with the balsara switch ; however , this transport was readily reduced with increased particle numbers . = 5.5 in in figure [ figrotzoom ] we focus on the bounce that occurs during the collapse using bulk viscosity . bulk viscosity converges very slowly towards the correct solution and displays unwanted numerical behaviours even at high particle numbers . in cosmological adiabatic tests it tends to generate widespread heating during the early ( pre - shock ) phases of collapse . thus we favour monaghan artificial viscosity with the balsara switch . = 5.5 in the cluster comparison involved simulating the formation of an extremely massive galaxy cluster with adiabatic gas ( no cooling ) . profiles and bulk properties of the cluster were compared for many widely used codes . we ran this cluster with gasoline at two resolutions : @xmath102 and @xmath103 . these were the two resolutions which other sph codes used in the original paper . profiles of the simulated cluster have previously been shown in . as shown in figure [ figcluscomp ] , the gasoline cluster bulk properties are within the range of values for the other codes at @xmath89 . the gasoline x - ray luminosity is near the high end ; however , the results are consistent with those of the other codes . the large variation in the bulk properties among codes was a notable result of the original comparison paper . we investigated the source of this variation and found that a large merger was occurring at @xmath89 with a strong shock at the cluster centre . the timing of this merger is significantly changed from code - to - code or even within the same code at different resolutions as noted in the original paper . running the cluster with gasoline with @xmath102 ( solid ) instead @xmath103 ( dotted ) particles changes the timing noticeably . the different resolutions have modified initial waves which affect overall timing and levels of substructure . these differences are highlighted by the merger event near @xmath89 . as shown in the figure , the bulk properties are changing very rapidly in a short time . this appears to explain a large part of the code - to - code variation in quantities such as mean gas temperature and the kinetic to thermal energy ratio . timing issues are likely to be a feature of future cosmological code comparisons . the authors have applied gasoline to provide new insight into clusters of galaxies , dwarf galaxies , galaxy formation , gas - giant planet formation and large scale structure . these applications are a testament to the robustness , flexibility and performance of the code . used gasoline at high resolution to show convincingly that giant planets can form in marginally unstable disks around protostars in just a few orbital periods . these simulation used 1 million gas particles and were integrated for around thousand years until dense proto - planets formed . current work focusses on improving the treatment of the heating and cooling processes in the disk . borgani ( 2001,2002 ) used gasoline simulations of galaxy clusters in the standard @xmath88cdm cosmology to demonstrate that extreme levels of heating in excess of current estimates from supernovae associated with the current stellar content of the universe are required to make cluster x - ray properties match observations . the simulations did not use gas cooling , and this is a next step . used gasoline with star formation algorithms to produce a realistic disk galaxy in the standard @xmath88cdm cosmology . current work aims to improve the resolution and the treatment of the star formation processes . mayer ( 2001a , b ) simulated the tidal disruption of dwarf galaxies around typical large spiral galaxy hosts ( analogues of the milky way ) , showing the expected morphology of the gas and stellar components . used a 270 million particle cosmological gasoline simulation to provide detailed estimates of the sunyaev - zeldovich effect on the cosmic microwave background at small angular scales . this simulated 400 mpc cube contains thousands of bright x - ray clusters and provides a statistical sample that is currently being matched to observational catalogues @xcite . a common theme in these simulations is large dynamic ranges in space and time . these represent the cutting edge for gasdynamical simulations of self - gravitating astrophysical systems , and were enabled by the combination of the key features of gasoline described above . the first feature is simply one of software engineering : a modular framework had been constructed in order to ensure that pkdgrav could both scale well , but also be portable to many parallel architectures . using this framework , it was relatively straightforward to include the necessary physics for gasdynamics . of course , the fast , adaptable , parallel gravity solver itself is also a necessary ingredient . thirdly , the gasdynamical implementation is state - of - the - art and tested on a number of standard problems . lastly , significant effort has gone into making the code adaptable to a large range in timescales . increasing the efficiency with which such problems can be handled will continue be a area of development with contributions from computer science and applied mathematics . the authors would like to thank carlos frenk for providing the cluster comparison results and matthias steinmetz for providing high resolution results for the spherical collapse . development of pkdgrav and gasoline was supported by grants from the nasa hpcc / ess program . bond , j. , contaldi , c. , pen , u - l . , pogosyan , d. , prunet , s. , ruetalo , m. , wadsley , j. , zhang , p. , mason , b. , myers , s. , pearson , t. , readhead , a. , sievers , j. , udomprasert , p. 2002 , apj ( submitted ) , astro - ph/0205386 frenk , c. , white , s. , bryan , g. , norman , m. , cen , r. , ostriker , g. , couchman , h. , evrard , g. , gnedin , n. , jenkins , a. , pearce , f. , thomas , p. , navarro , h. , owen , j. , villumsen , j. , pen , u - l . , steinmetz , m. , warren , j. , zurek , w. , yepes , g. & klypin , a. , 1999 , ap . j. , 525 , 554 anderson , r. & tjaden , b. 2001 , `` the inverse nearest neighbor problem with astrophysical applications '' , proceedings of the twelfth annual acm - siam symposium on discrete algorithms ( soda ) , january 2001 .

relativistic heavy ion collision experiments at the bnl - ags ( au(10.7 agev)+au ) and at the cern - sps ( pb(160 agev)+pb ) have yielded a large variety of fascinating data . various observables like the strong @xmath2 suppression , enhanced yield of intermediate mass dilepton pairs , enhanced ( anti-)hyperon yields , the creation of antimatter clusters and strong transverse flow seem to indicate the formation of very dense and highly excited matter @xcite . since these observables are connected in a non - trivial way it is a tempting task for theoreticians to model high energy heavy ion collisions in a consistent way and simultaneously predict this wide range of observables from a few hundred mev up to several thousand gev per nucleon at lhc . bear in mind , however , that up to now there is no unique theoretical description of the underlying hadron@xmath3hadron interactions , with their vastly different characteristics at different incident energies and in different kinematic intervals . perturbative quantum chromodynamics ( pqcd ) can be applied to describe hard processes , i.e. processes with large four - momentum , @xmath4 , transfer . but pqcd is formally inappropriate for the description of the soft interactions because of the absence of the large @xmath5scale . therefore , low@xmath6 collisions are described in terms of phenomenological models . early on , multiple production of secondaries in relativistic hadronic collisions has been described within the hydrodynamic approach @xcite . then regge theory @xcite and multiperipheral models have been developed to understand the phenomenology of the soft interactions . they avoid the difficulties attributed to the statistical models . an inconvenient point of this approach is the large number of free parameters , which have to be fixed by comparison to experiment . subsequently , various qcd - motivated quark@xmath3parton models have been introduced . consequently , a vast variety of models for hadronic- and nuclear collisions have been developed . they may be subdivided into macroscopic ( statistical and hydrodynamical ) models @xcite and microscopic ( string- , transport- , cascade- , etc . ) models like , e.g. urqmd @xcite , which is applied in the present paper , fritiof @xcite , venus@xcite , qgsm @xcite , rqmd @xcite and others @xcite including the parton cascade approach @xcite . in the hydrodynamical ( thermal ) model one assumes local ( global ) equilibrium - the dynamics is characterized by the equation of state employed . the microscopic models describe subsequent individual hadron@xmath3hadron collisions . for low and intermediate energies hadron@xmath3hadron and nucleus@xmath3nucleus collisions are described in terms of the interactions between hadrons and their excited states , resonances , i.e. on the ( quasi-)particle level . at high energies the quark and gluon degrees of freedom can not be neglected . then the concept of color string excitations is introduced with their subsequent fragmentation into hadrons . in lead@xmath3on@xmath3lead collisions at the full sps energy one finds in the urqmd model that the ten most frequent hadron@xmath3hadron collision types namely , @xmath7 ( in decreasing order of frequency ) describe only 50% of the total h - h collisions . the inclusion of an additional 120 h - h collision types allows modeling up to 90% of the collisions predicted in the urqmd model , while several thousand different h - h combinations are needed to cover more than 99% of the total number of h - h collisions . since only a few of these cross sections are measured , one relies heavily on extrapolations ( and transformation , e.g. via the detailed balance principle ) of known processes . therefore , here we want to present and analyze the detailed elementary h - h input used in the transport model urqmd . the h - h predictions are a necessary basis for our understanding of the dynamics of the complex heavy ion reactions . this paper is structured as follows : a brief description of the basic principles of the urqmd model is given in sec . section [ sec3 ] presents urqmd - results of the different h - h cross - sections for different reactions and a comparison with the available experimental data . the additive quark model ( aqm ) is used to calculate unknown cross - sections . urqmd is the first microscopic model which attempts to include the color coherent phenomena . the implications of the effects of color opacity and color transparency in the model are discussed . the treatment of the formation and decay of resonances and strings is described in detail in sec . the importance of finite size effects in the fragmentation of strings is demonstrated . section [ sec5 ] discusses the generation of the transverse momentum of particles in the model . in sec . [ sec6 ] several predictions of observables for elementary channels are presented , which are especially interesting for the upcoming proton - proton run of the na49 collaboration . finally , a summary and conclusions are given . the urqmd - model @xcite is a microscopic transport theory based on the covariant propagation of all hadrons on classical trajectories in combination with stochastic binary scatterings , color string formation and resonance decay . it represents a monte carlo solution of a large set of coupled partial integro - differential equations for the time evolution of the various phase space densities @xmath8 of particle species @xmath9 etc . , which non - relativistically assumes the boltzmann form : + + = st f_i(x , p ) , where @xmath10 and @xmath11 are the position and momentum of the particle , respectively , and @xmath12 denotes the collision ( or rather source- ) term of these particle species , which are connected to any other particle species @xmath13 . the exchange of electric and baryonic charge , strangeness and four momentum in the @xmath14-channel is considered for baryon - baryon ( bb ) collisions at low energies , while meson - baryon ( mb ) and meson - meson ( mm ) interactions are treated via the formation and decay of resonances , i.e. the @xmath15-channel reactions . @xmath14-channel reactions for mb and mm collisions are taken into account from @xmath16 gev on increasing to the only mb , mm interaction type above @xmath17 gev . for nucleus - nucleus collisions the soft binary and ternary interactions between nucleons can be described by the real part of the in - medium g - matrix , which is approximated by a non - relativistic density - dependent skyrme potential of the form v^sk = t_1 _ i j ( x_i - x_j ) + t_2 _ i j k ( x_i - x_j ) ( x_j - x_k ) , where @xmath18 denotes the coordinate variable in the quantum phase space . the first term simulates the attractive potential of the nn - interaction , and the second one yields the saturation . in addition , yukawa and coulomb potentials are implemented in the model . the potentials allow to calculate the equation of state of the interacting many body system , as long as it is dominated by nucleons . note that these potential interactions are only used in the model for baryons / nucleons with relative momenta @xmath19 of less than 2@xmath20gev/@xmath21 . for the hadronic collisions discussed here , the potential interactions are omitted . further details of the application of the urqmd model to heavy - ion reactions may be found in @xcite . this framework allows to bridge with one concise model the entire available range of energies from the sis energy region ( @xmath22gev ) to the rhic energy ( @xmath23gev ) . at the highest energies , a huge number of different particle species can be produced . the model should allow for subsequent rescatterings . the collision term in the urqmd model includes more than fifty baryon species and five meson nonetts ( 45 mesons ) . the baryons and baryon resonances included in the urqmd are listed in table [ tab1 ] . in addition , their antiparticles have been implemented using charge - conjugation to assure full baryon - antibaryon symmetry . figures [ fig1 ] and [ fig2 ] depict the implemented meson multiplets : pseudo - scalar , vector , scalar , pseudo - vector and ( not shown in the figs . ) the tensor mesons as well as the heavy vector meson resonances @xmath24 , @xmath25 , @xmath26 , and @xmath27 . extremely heavy meson resonances ( @xmath28 gev ) are not explicitly implemented , however they may be important when investigating , e.g. the dynamics of @xmath29 correlations in future experiments . all particles can be produced in hadron - hadron collisions and can interact further with each other . the different decay channels all nucleon- , @xmath30- and hyperon - resonances up to 2.25@xmath20gev/@xmath31 mass as well as the meson ( e.g. k@xmath32 ) decays etc . are implemented . at higher energies we take advantage of the hadron universality and use a string model for the decay of intermediate states . the cross - sections of various hadronic processes as well as the formation and fragmentation of the strings are discussed in the subsequent chapters . a basic input into the microscopic transport models are the particle species and -energy dependent cross - sections of hadron - hadron interactions . the total cross - sections are interpreted geometrically . a collision between two hadrons will occur if @xmath33 , where @xmath34 and @xmath35 are the impact parameter of the hadrons and the total cross - section of the two hadrons , respectively . in the urqmd model the total cross - section @xmath35 depends on the isospins of colliding particles , their flavor and the c.m . however , partial cross - sections are then used to calculate the relative weights for the different channels . only a small fraction of all possible hadronic cross - sections has been measured . in the following sections , we compare the urqmd cross - sections with experimental data . if no data are available , the additive quark model and detailed balance arguments are used to extrapolate such unknown observables . the total bb cross - section of the reaction @xmath36 has the general form _ tot^bb ( ) ( 2s_d + 1 ) ( 2s_e + 1 ) |m|^2 , with the spins of the particles , @xmath37 , momenta of the pairs of particles , @xmath38 , in the two - particle rest frame , and the matrix element @xmath39 . the matrix element @xmath39 , however , can take on a very complicated form and may be in general a function of of all the particle s quantum numbers as well as it s momenta and the c.m . energy . if high quality experimental data on the respective cross section exists , a phenomenological fit to the respective data is by far the most accurate approach for implementing the cross section . otherwise , we have to rely on simplified assumption for the matrix element and employ general symmetries , like the principle of detailed balance ( see section [ res_sec ] ) . let us start by investigating the total cross - section of proton - proton collisions from a beam momentum of 0.1@xmath20gev/@xmath21 up to @xmath40gev/@xmath21 as shown in fig . [ fig3 ] . the total and inelastic cross - sections of the @xmath41 reaction are well measured in this energy region @xcite . one finds a complex structure in this cross - section : local minima at 700@xmath20mev/@xmath21 ( @xmath42gev ) and 100@xmath20gev/@xmath21 ( @xmath43gev ) , the maximum at 2@xmath20gev/@xmath21 ( @xmath44gev ) and a rise above 100@xmath20gev/@xmath21 . note that the steep rise in the data below 300@xmath20mev/@xmath21 is due to soft coulomb interaction of the protons and is taken care of via the potential interaction . the structure in the @xmath41 cross - section is mainly due to the inelastic channels which are shown in fig . one clearly sees the @xmath30 excitation with its increasing cross - section at low energies . the different partial cross sections depicted in fig . [ fig4 ] are discussed in section [ res_sec ] . a detailed comparison of prominent outgoing channels is depicted in fig . [ ppc ] . here we show calculations of exclusive ( @xmath45 ) and inclusive ( @xmath46 ) cross sections for the production of neutral mesons @xmath47 as a function of the excess energies @xmath48 . here @xmath49 is the energy of the production threshold calculated as @xmath50 with the proton mass @xmath51 and the pole mass of the meson @xmath52 . in the case of @xmath53 mesons we count only those with masses within @xmath54 mev around the pole mass to compare with data . the exclusive @xmath55 production just above threshold @xcite is overestimated by a factor of about two . note that above 3.5 gev the exclusive cross sections become less important because the string and multiple decay channels open and allow for multiple resonance production . for upcoming gsi - sis experiments relevant @xmath56 values are below 4 gev , where so far no data on the inclusive channels are available and therefore rely heavily on extrapolations . in fig . [ pbar ] the cross section of @xmath57 reactions is shown . good agreement with the data @xcite is found over a large energy range . at higher energies , the contributions of the different nucleonic resonances decrease and give way to the excitation of color strings , which is the dominant process at high energies in our model . the total cross section above the resonance region is given by the cern - hera parameterization @xcite as shown in table [ tab2 ] . partonic pqcd scattering is not included into the urqmd model in the present version . the difference between the total and the elastic cross - section is taken as the inelastic cross - section . the mb cross - sections are dominated by the formation of @xmath15-channel resonances , i.e. the formation of a transient state of mass @xmath58 , containing the total c.m . energy of the two incoming hadrons . on the quark level such a process implies that a quark from the baryon annihilates an antiquark from the incoming meson . below 2.2 gev c.m . energy intermediate resonance states get excited . the total cross - section of these reactions are given by the expression : @xmath59 which depends on the total decay width @xmath60 , on the partial decay width @xmath61 and on the c.m . energy @xmath56 . at higher energies the quark - antiquark annihilation processes become less important . there , @xmath14-channel excitations of the hadrons dominate , where the exchange of mesons and pomeron exchange determines the total cross - section of the mb interaction @xcite . figures [ fig5 ] and [ fig6 ] show the cross - section of pion - proton reactions at different energies . in fig . [ fig5 ] ( @xmath62 ) one probes predominantly the creation of the @xmath63 ( @xmath64 ) resonance . note that the low energy s - wave @xmath65 scattering is not included into the urqmd fit . the resonance peak at @xmath66 gev / c is from the @xmath67 resonances . in comparison fig . [ fig6 ] ( @xmath68 ) depicts many strong uncharged non - strange baryon resonances , e.g. the @xmath69 etc . the total cross - section in the intermediate energy regime is therefore the sum of the individual excitation modes of baryon resonances , the s - wave at lower energies is left out . let us now investigate collisions of strange mesons with baryons . for @xmath70 mesons strange @xmath15-channel resonances can be formed on non - strange baryons due to the annihilation of the @xmath71-quark . a comparison of these processes from urqmd with the experimental data is shown in fig . the formation of hyperon resonances is clearly visible at lower energies , while the universal @xmath14-channel reaction dominates the high energy tail . [ fig8 ] shows the cross - section of @xmath72-mesons ( @xmath73 ) on protons . in this case , the formation of resonances is forbidden , since the @xmath74-quark can not be annihilated by non - strange baryons ( for strange baryons the formation of resonances is still possible ) . here we use only the elastic channel and the @xmath14-channel excitation of both particles . the cross section at very high energies is given by the cern - hera parametrization as shown in table [ tab2 ] . due to the fact that the experimental preparation of meson beams and targets is restricted to @xmath75 s and k s , only very little is known about mm collisions in general . for the description of heavy ion collisions the importance of this channel increases with energy : at 1 a@xmath20gev beam energy we find that the production of new hadrons ( mostly pions ) is only a ten percent effect . at ags energies ( 10 a@xmath20gev ) the amount of mesons roughly equals the number of incoming nucleons . going on to the sps ( 160 a@xmath20gev ) the picture changes drastically : the produced particles dominate the reactions , while the incoming nucleons have dropped to a 15% admixture in particle density and multiplicity @xcite . to describe the total meson - meson reaction cross - sections , we make use of the additive quark model ( see below ) and the principle of detailed balance , which assumes the reversibility of the particle interactions . fig . [ fig9 ] compares the calculated cross - section of @xmath76 scattering to experimental data @xcite . the spectrum is dominated by the formation of the @xmath77 with a mass of 770 mev , the other two small peaks belong to the @xmath78 and @xmath79 resonances . the @xmath78 resonance is not visible in the data since the experimental analysis of meson - meson scattering is model dependent . figure [ fig10 ] shows the implemented elastic @xmath80 cross - section , which remains constant in the whole energy range where data are available @xcite . strangeness production in the meson - meson channel is possible , e.g. via the reaction @xmath81 as shown in fig . [ fig11 ] . rescattering of strange mesons is implemented via the resonance formation ( cf . [ fig12 ] dominated by the @xmath82 resonance @xcite ) , or elastically as depicted in fig . [ fig13 ] @xcite . finally , we predict the cross - sections of @xmath83 , @xmath84 and @xmath85-reactions ( fig . [ fig14 ] ) which are of utmost importance for the production of thermal photons and dileptons . at higher energies other meson resonances can be formed . to model mm interactions above the resonance region @xmath86 gev we use the rescaled total @xmath65 cross section : _ tot^mm(s > 1.7)=_tot^p(s ) . this is justified , since at high energies the total cross section is given by quark counting . in the energy region from @xmath87 gev to @xmath88 gev @xmath15-channel interactions are taken into account , while from @xmath89 gev on @xmath14-channel excitation of both mesons becomes the mm interaction process of increasing importance in the model . the cross section for high energetic reactions are taken from the aqm - rescaled @xmath90 paramatrization by the cern - hera group ( see table [ tab2 ] ) . the physics of baryon@xmath3antibaryon interactions has been an area of much theoretical and experimental activity for a rather long period . it is well - known that at energies @xmath91gev/@xmath21 an important contribution to the total interaction cross - section comes from the process of annihilation , where only mesons are left in the final state . though the earlier experiments on @xmath92annihilation revealed a number of differences from the non - annihilation channels , it is not clearly understood whether these differences arose simply due to the kinematic restrictions on the available phase space , or whether they are related to dynamical differences between the non - annihilation and annihilation mechanisms . the experimental results obtained in @xcite by comparison of @xmath41 with non - annihilation @xmath93 interactions at 32 gev/@xmath21 support the conclusion of equivalence of @xmath41 and non - annihilation @xmath93 interaction processes . still , the nature of the baryon annihilation is subject to theoretical discussions . in the framework of the quark models based on topological @xmath94 expansion , the annihilation process is associated with the annihilation of string junctions , i.e. the point where strings are connected , such that the baryons have a @xmath95-form . in this case three @xmath96 strings are formed . the theory also allows for diagrams where the string junctions and one or two of the valence quarks can annihilate , corresponding to the formation of two strings or one string . other theories consider an annihilation mechanism without invoking the string junction hypothesis . this intriguing question has yet to be clarified ( for review see , e.g. @xcite , and references herein ) . to avoid the difficulties attributed to these theoretical approaches , urqmd is adjusted to known experimental data . the total @xmath97 cross - section is shown in fig . [ fig15 ] , as well as the annihilation and the elastic cross - sections . the urqmd parameterizations depicted by lines are taken from koch and dover @xcite : _ ann^|p p = _ 0^n , with @xmath98 mb , @xmath99 , a = 50 mev and b = 0.6 . the @xmath100 cross - section does not differ significantly from the @xmath97 cross - section @xcite , hence they are set equal in the urqmd model . at higher energies , cern - hera parameterizations @xcite are used for the total and elastic channel : ^|p p_tot , el(p)=a + b p^n + cln^2(p ) + dln(p ) , with the laboratory momentum @xmath11 in gev/@xmath21 and the cross - section @xmath101 in mb . the parameters of the fit are listed in table [ tab2 ] . below @xmath102 gev / c the following parameterization is used : _ tot(p ) = \ { ll 75.0 + 43.1 p^-1 + 2.6 p^-2 -3.9 p & ; 0.3<p<5 + 271.6 & ; p < 0.3 . _ el(p ) = \ { ll 31.6 + 18.3 p^-1 -1.1 p^-2 - 3.8 p & ; 0.3<p < 5 + 78.6 & ; p < 0.3 . the sum of annihilation and elastic cross - sections do not yield the total cross - section : = _ tot-_el-_ann in urqmd this difference @xmath103 is interpreted as the diffractive cross - section which describes the excitation at least one of the collision particles to a resonance or to a string via pomeron exchange . the annihilation of baryon@xmath3antibaryon pairs proceeds in the urqmd model according to _ rearrangement _ diagrams . here the formation of two @xmath104-strings of equal energies in the c.m . system is assumed while the remaining constituent quarks are rearranged into newly produced hadrons . the generalization of the @xmath93 cross - section towards all possible antibaryon@xmath3baryon collisions can be done in different ways : 1 . the anti - baryon baryon cross - section at a given c.m . energy @xmath56 is equal to the @xmath93 annihilation cross - section at the same @xmath56 : _ |bb|_=_|pp| _ 2 . the anti - baryon baryon cross - section at a given relative momentum @xmath105 is equal to the @xmath93 annihilation cross - section at the same @xmath105 ( fig . [ fig16 ] ) : _ |bb|_p_rel=_|pp|_p_rel in the urqmd we have chosen the first parameterization . since @xmath106 , the annihilation cross - section drops rapidly with rising particle mass . the different treatment this cross - section can lead to systematic shifts in the antibaryon distributions for massive systems . therefore , anti - baryon production off nuclei may be used to solve this ambiguity . unknown cross - sections are calculated on the basis of the additive quark model ( aqm ) @xcite , which assumes the existence of dressed valence quarks , interacting very weakly inside of the hadron . at the phenomenological level , the aqm gives a correct quantitative and qualitative description of , e.g. , the asymmetry of c.m . spectra of secondaries in meson@xmath3nucleon and photon@xmath3nucleon reactions . the aqm has predicted also the important role of resonances for the multiple production of particles in hadronic interactions . to apply the aqm to calculations of heavy ion reactions , one needs to know the cross - sections of the quark interactions , which can be evaluated from the quark masses . then , the unknown total cross - section of the high energy reaction can be calculated assuming a 40% reduced @xmath15-quark cross - section ( compared to that of @xmath107- and @xmath34-quark ) . the elastic cross - section is derived from regge theory @xcite : _ elastic = 0.039 _ total^ [ mb ] , where _ total = 40()^m_1+m_2 ( 1 - 0.4 ) ( 1 - 0.4 ) [ mb ] , where @xmath108 for particle @xmath109 being a meson ( a baryon ) and @xmath110 is the number of strange quarks in particle @xmath109 . this formula results from the high energy reactions , therefore there is no difference between antiparticles and particles . for bb - reactions no additional energy dependence is employed in collisions involving strange baryons . non - strange baryon cross - sections are not treated via the additive quark model , they have an explicit energy dependence in line with experimental data . the mb and mm cross - sections are rescaled via : _ the hyperon - nucleon cross - section , which is taken in the urqmd model from the additive quark model , is in good agreement with the data above @xmath111 mev ( cf . [ fig17 ] ) . the total cross - sections calculated for baryon - baryon , meson - baryon and meson - meson interactions are listed in tables [ tab3]@xmath3[tab5 ] . quantum chromodynamics has important applications of the dynamical role of color degrees of freedom to the strong interactions at ultrarelativistic energies ( for a review , see @xcite and references therein ) . the theory is presented in ref . @xcite in detail , here we just sketch the main ideas of color optics and coherent phenomena in high energy physics . hadrons are dynamical objects which come in fock space of configurations of very different spatial sizes . at high energies , incident hadronic quark@xmath3gluon configurations can be considered frozen as a result of lorentz time dilation . due to the long coherence length at such high energies one can apply geometrical color optics . small objects produced in hard processes with high momentum transfer @xmath4 then have reduced interaction cross sections . in processes with moderate @xmath4 such compact objects , which are a coherent superposition of eigenstates of the qcd hamiltonian , should gain size . when the quark@xmath3gluon configuration is large , it will lead to an increased interaction cross - section of the hadron with the nuclear medium . therefore , the fluctuations of the hadron s spatial extent give rise to the color transparency and color opacity phenomena : when a small object is produced , it interacts only very weakly with other hadrons due to color screening . moreover , since - at sufficiently high energies - the small - sized configuration of this object is frozen , the nuclear medium appears to be transparent for such hadrons ( color transparency ) . in contrast , hadronic configurations which are larger than average interact with larger cross - section , giving rise to color opacity . nucleus - nucleus collisions provide a tool to investigate the effect of color transparency , for instance , in the production of leading nucleons . on the other hand , the complementary color opacity effect , i.e. large - sized configurations , can cause stronger stopping and significant fluctuations in the transverse energy of secondaries in central reactions . a first step to investigate these qcd effects within a microscopic transport model is made by incorporating the color fluctuations in the elementary hadron@xmath3hadron reactions in the urqmd model . thus , one needs to know the probability @xmath112 that a given configuration interacts with a nucleon with a total cross - section @xmath101 . it is convenient to consider moments of the distribution : ^0 & = & |^0 p ( ) = 1 , + ^1 & = & |^1 p ( ) =| , etc . , where @xmath113 denotes the average cross - section . the second moment @xmath114 can be determined from the diffractive dissociation experiments . in addition , further information can be obtained from qcd , which implies @xcite : p()^n_q-2 , for @xmath115 , where @xmath116 is the number of valence quarks . thus , for the nucleon and the pion distributions it follows for @xmath115 : p_n ( ) & & , + p _ ( ) & & constant . from these arguments , @xmath112 can be construct . [ fig18 ] shows the resulting broad @xmath112 distribution for proton projectiles and the even broader one for the pions @xcite . the effect of the color fluctuations on proton - proton interactions at different impact parameters , @xmath117 , is shown in fig . [ fig19 ] for the urqmd model . the charged pion multiplicity distribution decreases monotonically with rising @xmath117 . @xmath118 has a non - vanishing tail for @xmath119fm , in contrast to the abrupt geometrical edge of the distribution as calculated in the static geometric models without color fluctuations . the production and decay of resonances is the most important h - h reaction channel below @xmath120 gev for bb and 3 gev for mm and mb reactions . baryon resonances are produced in two different ways , namely * _ hard production _ : n+n@xmath121n , @xmath122 , n@xmath32n , etc . * _ soft production _ : @xmath123p@xmath124 , k@xmath125+p@xmath126 ... the formation of @xmath15-channel resonances is fitted to measured data , e.g. in the reaction @xmath36 we use the general form _ tot^bb ( ) ( 2s_d + 1 ) ( 2s_e + 1 ) |m(m_d , m_e)|^2 , with the spins of the particles , @xmath37 , momenta of the pairs of particles , @xmath38 , in the two - particle rest frame , and the matrix element @xmath127 , which here depends only on the masses of the outgoing hadrons , @xmath128 . there are six channels of the excitation of non - strange resonances in the urqmd model , namely @xmath129 . the @xmath130 is explicitly listed , whereas higher excitations of the @xmath30 resonance have been denoted as @xmath131 . for each of these 6 channels specific assumptions are made with respect to the form of the matrix element @xmath132 , and the free parameters are adjusted to the available experimental data , when available : 1 . @xmath133 excitation : @xmath134 with @xmath135 gev , @xmath136 gev and @xmath137 . note that this form of the matrix element has been adjusted to fit the data shown in figure [ pp - nd1232 ] . 2 . @xmath138 , @xmath139 , @xmath140 and @xmath141 excitation : @xmath142 with @xmath143 for @xmath138 , @xmath144 for @xmath139 and @xmath145 for @xmath140 . in the case of @xmath141 there are insufficient data available , therefore we use the same matrix element and parameters as in the case of @xmath140 . since @xmath146 is valid for all above cases , the matrix element can not diverge . @xmath147 excitation : @xmath148 with @xmath149 . figure [ pp - nd1232 ] shows the fit of the urqmd @xmath150 cross section to experimental measurements @xcite . the measurements refer to the @xmath151 exit channel and have been rescaled to match the full isospin - summed cross section . in the case of the exclusive @xmath130 cross section the quality of the data and thus also the quality of the resulting fit is very good . for higher resonance excitations this is unfortunately no longer the case and additional measurements are needed to clarify the situation . one has to keep in mind , however , that the experimental extraction of exclusive resonance production cross sections is only possible via two- or three - particle correlations ( e.g. a pion - nucleon correlation in the case of the @xmath30 ) which introduces large systematic errors , especially for broad resonances . in figure [ pp - ppstar ] the urqmd cross sections for the processes @xmath152 , @xmath153 @xmath154 and @xmath155 are compared to data @xcite . one single parameter has been used to fit all four cross sections . the data situation is not as good as in the case of the @xmath130 resonance , some ambiguities are visible which results in the quality of the fit being not as good as in the previous case . the parameters for the other classes are fitted in the same fashion . the cross section for exclusive @xmath156 can be seen in figure [ pp-2d1232 ] . the data points @xcite hint at a resonance like structure which can not be reproduced with the urqmd ansatz for resonance - excitation cross sections . however , the data deviates considerably from other cross sections for resonance excitation ( e.g. @xmath157 ) . considering hadron - universality and the similarities in all other resonance excitation cross sections this casts a certain doubt on the accuracy of the measurement of the resonance - like peak . figure [ pp - nd1920 ] shows the urqmd fit for the exclusive @xmath158 production . the same matrix element is used for the entire class of @xmath159 reactions . in the case of exclusive @xmath160 production the matrix element has been fitted to the @xmath161 exit channel ( see figure [ pp - d1232n1680 ] ) . the extrapolation to the case of @xmath162 production can be seen in figure [ pp - d1232n1520 ] . for the exclusive @xmath163 reaction class the data situation is unsatisfactory , therefore we used the same matrix - element as in the @xmath164 case . the decay of the resonances proceeds according to the branching ratios compiled by the particle data group @xcite . the resonance decay products have isotropical distributions in the rest frame of the resonance . if the resonance decays into @xmath165 particles , then the corresponding @xmath166body phase space is used to calculate their @xmath167 momenta stochastically . it is necessary to note that a consistent description of angular momentum distributions points to a rather intricate problem of transport theory itself : if one considers the whole scattering interaction to be described by one single quantum mechanical process there are correlations between the final and the initial stage . for instance the angular distribution of the final particles with respect to the axes of the incoming momenta in the cms system . however , a fitting of the angular distributions to experimental data may conflict with the basic assumption of transport theories that the multiple scattering processes can be considered to be of markovian type , i.e. after each scattering process or resonance formation the outgoing particles completely forget about their entrance channels . in the case of a spin 0 resonance there is no preferred direction for the emission of the final particles , while for spin 1 ( and other ) the different magnetic quantum numbers are statistically occupied , so that also in these cases there is no preferred angle of emission . all produced particles are able to rescatter within the nuclear medium , therefore the excitation of resonances by the annihilation of mesons on baryons included as depicted in fig . [ fig5 ] for the reaction @xmath168 . also the @xmath169 channel ( fig . [ fig6 ] ) shows a rich structure of baryon resonances . the total meson - baryon cross section is given by formula [ mbbreitwig ] . there , the total and partial decay widths also define the inverse reaction , i.e. the different decay - channels of the respective resonance . thus , the principle of detailed balance is applied . based on this principle we calculate all resonance formation cross sections from the measured decay properties of the possible resonances up to c.m . energies of 2.25 gev/@xmath31 for baryon resonance and 1.7 gev/@xmath31 in the case of mm and mb reactions . above this energy collisions are modeled by the formation of @xmath15-channel string or , at higher energies ( beginning at @xmath170 gev ) , by one / two @xmath14-channel strings . in the strangeness channel elastic collisions are possible for those meson - baryon combinations which are not able to form a resonance , while the creation of @xmath14-channel strings is always possible at sufficiently large energies ( c.f . fig . [ fig7 ] for the formation of hyperon resonances and fig . [ fig8 ] for the non - resonant channel ) . at high collision energies both cross section become equal due to quark counting rules . in more general terms , the principle of detailed balance can be derived by assuming time - reversal invariance of the interaction hamiltonian and can be formulated in the following way : [ detbal ] ( yx ) p_y^2 g_y = ( xy ) p_x^2 g_x , with @xmath171 denoting the c.m .- momenta of the particles and @xmath172 being the spin - isospin degeneracy factors . thus , if the cross section of the reaction @xmath173 is known , the back reaction @xmath174 can be easily obtained . this principle is in urqmd widely applied for the calculation of baryon - resonance absorption cross sections , such as @xmath175 . for a detailed discussion of the application of the principle of detailed balance to resonance absorption and @xmath56-dependent decay widths we refer to @xcite . as was mentioned above , not only baryon - baryon and meson - baryon collisions have to be included in the proposed scheme . at high energies and in massive aa systems meson - meson collisions may dominate the multiple production of secondaries . unfortunately , there are only few channels for which the experimental information exists , like the process of @xmath176 scattering ( fig . [ fig9 ] ) , which is fairly described by the urqmd model . both , for the high energy regime and for baryon@xmath3antibaryon annihilation we apply a string model ( similar to the lund model @xcite ) to describe the inelastic reactions . the constituents , quark and diquark ( or anti - quarks ) , of the incoming hadron also define the predominant emission patterns of the events . the amount of stopping in nucleus - nucleus is strongly correlated to the detailed dynamics of the diquark in the hadronic medium @xcite . recently different additional mechanisms of baryon number transport in nucleus - nucleus reactions have been investigated : baryonic junctions as suggested by @xcite which yield an enhanced hyperon and proton production cross section at central rapidities @xcite ; di - quark breaking due to interaction with the hadronic medium as predicted by @xcite . the di - quark breaking component is also taken into account in the urqmd model ( di - quark breaking probability@xmath177 ) . however , the above mechanisms are of minor importance in the urqmd approach since rescattering of the leading di - quark with hadrons is explicitly taken into account . since gluons are massless particles with spin @xmath178 , the static strong interaction between quarks at small distances @xmath179 may be described by a potential @xmath180 . at sufficiently large distances the color field between two quarks or anti - quarks transforms into the color string . denoting the string tension as @xmath181 one defines the linear string potential @xmath182 between the quarks / diquarks located at @xmath183 and @xmath184 , respectively this form of the potential is chosen from heuristic considerations , based on the quark confinement hypothesis , and is supported by lattice qcd calculations @xcite . the transverse directions have not to be taken into account , since they are negligibly small compared to the longitudinal excitation of the hadron string . hence , we get the dynamics of the quark system ( with quark momenta @xmath185 and @xmath186 ) from the hamiltonian @xmath187 @xmath188 which leads to the following equations of motion for the massless endpoints of the string : @xmath189 a change in momentum is directly related to the sign of @xmath190 , while the direction of propagation is defined by the sign of the momentum @xmath191 of the quark . this results in a typical yo - yo type evolution of the quark system . if the momentum transfer is large enough , the excitation of the string may exceed some critical limit . after that it will be energetically favorable to break the string into pieces by producing @xmath104-pairs from the vacuum . each of the produced @xmath104-pairs will have small relative momenta in their rest frame . owing to the fact that the color string is uniformly stretched , the hadrons produced as a result of the string fragmentation will be uniformly distributed within the kinematically allowed interval between @xmath192 and @xmath193 . the probability of the pair production process has been calculated by schwinger for the case of an infinite homogeneous electrical field . his result can be used to motivate the decay of qcd color field between the quarks . note that qcd is a non - abelian theory , therefore the color field need not be homogeneous and it is definitely not infinite . this leads to the modifications of string decays which will be discussed later . the probability @xmath194 for the creation of a quark - antiquark pair with mass @xmath195 in a color field with a string tension @xmath181 is : @xmath196 where a typical value for @xmath181 is 1 gev / fm . this relation is motivated by schwinger s qed - based result for particle - antiparticle creation in a strong electric field ( see the discussion of formula [ schw ] in section [ finite ] ) . the relative production probabilities of the different quark flavors are adjusted to @xmath197-data : @xmath198 the production of charmed ( and heavier ) quarks is strongly suppressed in the string picture , hence they are exclusively produced in hard qcd processes . the elementary diquark is introduced to allow for baryon - antibaryon production in the string . a schematic view of the decaying string is shown in fig . [ fig20 ] @xmath3 a non - strange baryon decays into a hyperon , a kaon and a pion . to decide which type of hadron is produced from the quark configuration that is created in the color field we choose in the case of a produced ( * i * ) _ baryon _ @xmath3 the octet and decuplet with equal probabilities ; ( * ii * ) _ meson _ @xmath3 the meson nonet with a probability proportional to the spin degeneracy and inverse mean mass @xmath195 , @xmath199 the singlet states are mixtures of @xmath200 , @xmath201 and @xmath202 . they are projected onto su@xmath203 hadrons with the flavor mixing angles from the quadratic gell - mann - okubo mass formula @xcite . for the scalar mesons this formula is not applicable , here an ideal mixing angle ( @xmath204 ) is assumed ( the mixing angles are depicted in table [ mixang ] ) . the field@xmath3feynman fragmentation mechanism @xcite , which allows the independent string decay from both ends of the string is used in the urqmd model . the string break - up is treated iteratively : string @xmath205 hadron + smaller string . the conservation laws are fulfilled . the diquark is permitted to convert into mesons via the breaking of the diquark link , thus transporting the baryon number into central rapidities . on both sides of the fragmenting string the new particles are formed randomly . if a resonance is produced , its mass is determined by a breit - wigner mass distribution . the transverse momentum is assigned to this particle according to a thermal momentum distribution , resulting in a temperature of 170 mev . after that the fragmentation function determines the fraction of the longitudinal momentum of the string transferred to the hadron . this procedure can be described in a covariant manner by the light cone variables defined as : @xmath206 the light cone momentum @xmath207 given to the newly produced hadron is : @xmath208 the fragmentation of a baryonic string reads : @xmath209 the main input is the fragmentation function which yields the probability distribution @xmath210 . this function regulates the fraction of energy and momentum given to the produced hadron in the stochastic fragmentation of the color string . for newly produced particles the field - feynman function @xcite : @xmath211 is used . @xmath212 drops rapidly with increasing @xmath213 ( fig . [ fig21 ] ) . therefore , the longitudinal momenta of e.g. produced antibaryons ( fig . [ fig22 ] ) and pions ( fig . [ fig23 ] ) are small ( they stick to central rapidities ) , in line with the experimental data . the rapidity spectra of these particles have a characteristic gaussian - like shape , in contrast to the baryon spectra in pp , as it is clearly seen in figure [ fig22 ] . the proton is on average less stopped , since it is build up from the leading diquark in the string ( leading particle effect ) . [ fig24 ] compares the @xmath214 distribution of protons and @xmath215 s for the feynman scaling variable @xmath216 measured in pp reactions at 205 gev/@xmath21 . the data on leading baryons can only be reproduced when a modified fragmentation function is used for the leading baryons ( cf . [ fig21 ] , dashed curve ) . this leading baryon fragmentation function is of gaussian form : @xmath217 } \quad,\ ] ] with parameters @xmath218 and @xmath219 . it is obvious that modeling the momentum loss in elementary collisions has a strong influence on the rapidity spectra of particles produced in heavy - ion collisions . this can be seen if one compares the spectra of p , @xmath215 s ( created from the leading baryon ) and @xmath220 s or mesons ( created from newly produced quarks ) in lead - lead collisions at the sps energies @xcite . let us now discuss finite size effects in the process of string fragmentation . the string is essentially a color field which connects two color charges , the @xmath221 $ ] quark and the @xmath222 $ ] di - quark or anti - quark , at the ends of the string . quantum electrodynamics ( qed ) predicts spontaneous particle@xmath3antiparticle creation in strong electric fields @xcite . this effect should hold for particle creation in a strong color fields . schwinger s qed result w_=^_n=1 [ schw ] is often adopted to the case of color fields by equating @xmath223 with the string tension @xmath181 . however , for strings , several important assumptions , which lead to schwinger s result are not fulfilled : firstly , the color field is not infinitely extended . it is bound radially by the interaction length of the gluons and longitudinally roughly by the @xmath221 $ ] and @xmath222 $ ] endpoints . secondly , the two endpoints of the string move ( with close to the speed of light ) in opposite directions . finally , the assumption of a constant field strength @xmath181 seems to be fulfilled along the string , but may become invalid in hot and dense matter @xcite . the influence of the finite radial size of the string can be studied by solving the dirac equation for the ( newly produced ) pairs in a finite volume . the field is restricted to a cylindrical volume of length @xmath224 and and area @xmath225 , where @xmath226 is the cylinder radius . along the cylinder axis a homogeneous color field is assumed . the boundary condition on the surface of the cylinder leads to enclosure of the color charges . taking @xmath227 neglects the longitudinal direction . the constraint on the cylinder surface is given by linear mit - boundary conditions @xcite , thus we have to solve the equations : @xmath228 the mit model allows for an analytic solution@xcite . the occupation numbers in the limits @xmath229 yield the pair creation probability as @xcite w_r=^_n=1 \ { _ n , } . thus the string radius is directly related to the pair creation rate @xmath230 , with @xmath231 being the n@xmath232 momentum eigenvector in the solution of eqs . ( [ dirac1 ] and [ dirac2 ] ) for a given projection @xmath233 of the corresponding bessel functions . the expression in curly brackets gives the deviation from the schwinger formula . for @xmath234 this second factor vanishes and one is back to the infinite case . this result allows to calculate the strangeness suppression @xmath235 for different string radii as shown in fig . [ fig26 ] . the overall pair creation rate has been calculated as a function of the longitudinal size of the color field for the corresponding dirac equation @xcite in terms of confluent hypergeometric functions . the resulting particle production rate per volume d@xmath236 and time interval d@xmath14 , @xmath237 at @xmath238 is shown in fig . [ fig27 ] as a function of @xmath213 , the distance from the string center , for different string lengths @xmath224 . a clear depletion of particle production near the endpoints of the string endings is visible ( eq.[schw ] ) . for short strings the particle production gets enhanced toward the string center . the above discussed finite size effects have a strong influence on the results , especially on the production of heavy quarks and particles like anti - baryons , -hyperons , etc . up to now these corrections are included only in the non - strange antibaryon sector . this effect is very important especially in the case of the @xmath215-particle , since it is mostly created at the string ends in the fragmentation of a leading proton . indeed , it has been reported that most of the transport models which use a string fragmentation scheme based on the schwinger formalism tend to overestimate the @xmath215 s @xcite . therefore , this finite size effects shall be included for strange baryons , in order to correct the overestimated number of @xmath215 s and @xmath239 s . in hot and dense nuclear matter , most hadrons suffer interactions from the many surrounding particles . as a result , the effective mass may change with density . many dynamical properties of hadrons are modified in the medium . in - medium two - body scattering cross - sections may differ from the free space values . those effects can be studied in the framework of relativistic transport theory , i.e. the relativistic boltzmann - uehling - uhlenbeck ( rbuu ) equation . this type of transport equation has been used extensively to the study of relativistic heavy - ion collisions and turned out to be very successful . the urqmd approach uses an analytical expression for the differential cross - section of in - medium @xmath240 elastic scattering derived from the collision term of the rbuu equation @xcite to determine the scattering angles between the outgoing particles in elementary hadron - hadron collisions . it is assumed that the angular distributions for all relevant two - body processes are similar modified in an analogous manner . they are approximated by the differential in - medium @xmath240 elastic scattering cross - section : @xmath241 with the direct term @xmath242 and the exchange term @xmath243 the ( pseudo-)scalar and vector coupling constants are @xmath244 , @xmath245 , and @xmath246 and @xmath247 is the in - medium mass , @xmath248 are the mandelstam variables . the in - medium single - particle energy is given by @xmath249 the formula for the differential cross section of in - medium @xmath240 elastic scattering can be used for elementary hadron - hadron collision if it is scaled by @xmath250 where @xmath251 and @xmath252 denote the effective masses of the incoming hadrons . furthermore , we take into account finite size effects of the hadrons and part of the short range correlation by introducing a phenomenological form factor at each vertex . for the baryon - baryon - meson vertex the common form @xmath253 is used , where @xmath254 is the cut - off mass for meson @xmath255 . + the total energy and the masses of the incoming hadrons serve as input for calculating the angular distribution . it is worth to stress again that eq.([angeq ] ) is used to calculate only the angular distributions for all elementary elastic two - body processes but not the corresponding total cross sections . the inverse slopes ( temperatures ) as calculated in the urqmd from the transverse momentum spectra of pions - by fitting the @xmath256 distribution with an exponential - compare well with values extracted from thermal fits to data , @xmath257 and @xmath197 . ] @xcite ( fig . [ fig28 ] ) . the urqmd model reproduces nicely the total , elastic and inelastic cross - sections of numerous hadronic reactions . the model also predicts the particle multiplicities ( i.e. the inclusive cross - sections ) as well as the ( lorentz - invariant ) cross - sections , which may come in the form of @xmath214- , rapidity- , or transverse momentum distributions . the abundances of the most important particle species produced in @xmath41 collisions at 12 gev/@xmath21 @xcite are listed in table [ tab12gev ] - the model predictions are in line with the data within 15% . the yields of various particle species produced in @xmath41 collisions at 205 gev/@xmath21 ( @xmath258 gev ) @xcite are listed in table [ tab6 ] . it is easy to see that the model predictions lie generally well within the 10% of the data except for the strange baryons . table [ tab7 ] presents the results on particle production in @xmath41 interactions at @xmath259 gev @xcite . again , the agreement between the experimental and theoretical results is good . as discussed above , the production of @xmath215(@xmath220 ) s is overestimated by a factor of 2 - 3 due to the neglect of finite size effects in the strings . the rapidity spectra of pions and @xmath214-distributions of baryons , as obtained from the urqmd model for @xmath41 interactions at 205 gev/@xmath21 ( fig . [ fig22 ] and [ fig23 ] , respectively ) have been discussed already in the previous section . correlations between the transverse and longitudinal momenta of charged pions , produced in the same reaction , can be studied @xcite . the transverse momenta of both positively and negatively charged pions have evident minima at @xmath260 as shown in fig . [ fig29 ] ( the sea - gull effect ) . then the transverse momentum increase nearly linearly with the rising longitudinal momenta . the agreement with the experimental data is fair . the single event correlation between the transverse momenta of the produced pions and the multiplicity of the negatively charged hadrons in the event is presented in fig . [ fig30 ] . the measured distribution is reproduced nicely for @xmath261 s , while for the @xmath262 s , 20% deviations of the calculated spectrum from the data are evident at high ( @xmath263 ) event multiplicities . to probe the ability of the urqmd model to describe hadron - hadron collision even at the today highest available bombarding energies for nuclei , we compare the calculated he+he collision at isr with data @xcite as shown in fig . [ hehe ] . it is not surprising , that the light helium system is transparent . a baryon free area of 3 units in rapidity is produced . the urqmd model prediction describes the data fairly well . the calculated particle yields can be increased by 15% if one includes multiple jet creation into the model description @xcite . the urqmd model reproduces the cross - sections and spectra of particles in hadronic collisions fairly well . since hadronic interactions build up the basic input to simulate the hadron - nucleus and nucleus - nucleus interactions in the model , it is interesting to examine the applicability of the urqmd model to these reactions . the full comparison with the experimental @xmath264 and @xmath265 data is an ongoing program , which is not completed yet @xcite . the ability of the model to reproduce e.g. dilepton yields in @xmath41 collisions , which is of interest in high energy physics is shown here as an example . [ fig31 ] shows the urqmd calculations of the dilepton spectrum for @xmath11+be ( which serves as a substitute for pp reactions ) at 450 gev/@xmath21 . dilepton sources considered here are dalitz decays ( @xmath266 , @xmath55 and @xmath267 ) and direct vector meson decays ( @xmath77 , @xmath267 and @xmath268 ) . dalitz decays of heavier meson and baryon resonances are included explicitly via their emission of @xmath77 mesons ( assuming vector meson dominance ) . to avoid double counting , the @xmath77 mesons from @xmath55 s , and @xmath267 s are excluded from the @xmath77 contribution . pion annihilation is included dynamically into the contribution of decaying @xmath77 mesons via the channel @xmath269 . the calculation of dilepton yields without modifications of the @xmath77 mass pole is compatible with the ceres data @xcite . the basic hadronic interaction processes incorporated into the urqmd approach are given . the implemented cross - sections of various hadron - hadron collisions , as well as their extrapolations into the high energy region are presented . the model treatment of excitation and decay of intermediate objects like resonances and strings is described . the importance of the finite size effects for the process of string fragmentation is discussed . the urqmd model is a microscopic transport model which allows to study optionally color fluctuations , i.e. color coherence phenomena , as color transparency and color opacity , as well as the expansion of small wave packets from the point of the production . the model predictions are compared with the available experimental data on particle yields in hadronic interactions for a wide range of c.m . the urqmd model treats the elementary processes reasonably well . more accurate data on proton - proton , proton - neutron , as well as meson - baryon collisions are needed to improve the extrapolation to nucleus - nucleus interactions at high energies , where already in the present model - about @xmath270 elementary hadron - hadron reactions are possible . this work was supported by the graduiertenkolleg fr theoretische und experimentelle schwerionenphysik , frankfurt@xmath3giessen , the bundesministerium fr bildung und forschung , the gesellschaft fr schwerionenforschung , darmstadt , deutsche forschungsgemeinschaft , the alexander von humboldt@xmath3stiftung , the josef buchmann - stiftung , daad and in part by doe grant de - fg02 - 96er40945 . quark matter96 , heidelberg , germany , nucl . phys . a * 610 * ( 1996 ) ; quark matter97 , tsukuba , japan , nucl . a * 638 * ( 1998 ) ; s. a. bass , m. gyulassy , h. stcker , w. greiner , j. phys . g ( 1998 ) in print ; strange quark matter 97 , j. phys . g * 23 * ( 1997 ) ; strange quark matter 98 , j. phys . g ( 1998 ) . landau , e.m . lifshitz , fluid mechanics ( pergamon , oxford , 1959 ) ; h. stcker , w. greiner , phys . rep . * 137 * , 277 ( 1986 ) ; r.b . clare , d. strottman , phys . rep . * 141 * , 177 ( 1986 ) ; e. schnedermann , j. sollfrank , u. heinz , phys . c * 48 * , 2462 ( 1993 ) ; j. brachmann , a. dumitru , j. maruhn , h. stcker , w. greiner , d. rischke , nucl . a * 619 * , 391 ( 1997 ) . bass , m. belkacem , m. bleicher , m. brandstetter , l. bravina , c. ernst , l. gerland , m. hofmann , s. hofmann , j. konopka , g. mao , l. neise , s. soff , c. spieles , h. weber , l.a . winckelmann , h. stcker , w. greiner , ch . hartnack , j. aichelin , n. amelin , prog . . phys . * 41 * , 225 ( 1998 ) . zabrodin , l.v . bravina , o.l . kodolova , n.a . kruglov , a.s . proskuryakov , l.i . sarycheva , m.y . bogolyubsky , m.s . levitsky , v.v . maximov , a.a . minaenko , a.m. moiseev , s.v . chekulaev , phys . d * 52 * , 1316 ( 1995 ) . t. kafka , r. engelmann , r. godbole , j. hanlon , m. pratap , h. wahl , r. singer , y. cho , t. fields , l.g . hyman , l. voyvodic , r. walker , j. whitmore , phys . d * 16 * , 1261 ( 1977 ) ; g. charlton et al . . lett . * 30 * , 574 ( 1973 ) . ( 75,40)(0,0 ) ( 3,16)@xmath262 ( 17,16)@xmath266 ( 26,16)@xmath261 ( 17,13)@xmath271 ( 11,13)@xmath55 ( 8,27)@xmath272 ( 22,27)@xmath72 ( 8,2)@xmath273 ( 22,2)@xmath274 ( 0,15)(1,0)38 ( 15,0)(0,1)38 ( 5,15)(1,2)5 ( 10,25)(1,0)10 ( 20,25)(1,-2)5 ( 25,15)(-1,-2)5 ( 20,5)(-1,0)10 ( 10,5)(-1,2)5 ( 10,16)(0,-1)2 ( 20,16)(0,-1)2 ( 15,15 ) ( 15,15 ) ( 15,15 ) ( 5,15 ) ( 10,25 ) ( 20,25 ) ( 25,15 ) ( 20,5 ) ( 10,5 ) ( 34,13)@xmath275 ( 18,32)@xmath95 ( 43,16)@xmath276 ( 57,16)@xmath277 ( 66,16)@xmath278 ( 57,13)@xmath279 ( 51,13)@xmath267 ( 48,27)@xmath280 ( 62,27)@xmath281 ( 48,2)@xmath282 ( 62,2)@xmath283 ( 40,15)(1,0)38 ( 55,0)(0,1)38 ( 45,15)(1,2)5 ( 50,25)(1,0)10 ( 60,25)(1,-2)5 ( 65,15)(-1,-2)5 ( 60,5)(-1,0)10 ( 50,5)(-1,2)5 ( 50,16)(0,-1)2 ( 60,16)(0,-1)2 ( 55,15 ) ( 55,15 ) ( 55,15 ) ( 45,15 ) ( 50,25 ) ( 60,25 ) ( 65,15 ) ( 60,5 ) ( 50,5 ) ( 74,13)@xmath275 ( 58,32)@xmath95 ( 75,40)(0,0 ) ( 3,16)@xmath284 ( 17,16)@xmath285 ( 26,16)@xmath286 ( 17,13)@xmath287 ( 11,13)@xmath288 ( 8,27)@xmath289 ( 22,27)@xmath290 ( 8,2)@xmath291 ( 22,2)@xmath292 ( 0,15)(1,0)38 ( 15,0)(0,1)38 ( 5,15)(1,2)5 ( 10,25)(1,0)10 ( 20,25)(1,-2)5 ( 25,15)(-1,-2)5 ( 20,5)(-1,0)10 ( 10,5)(-1,2)5 ( 10,16)(0,-1)2 ( 20,16)(0,-1)2 ( 15,15 ) ( 15,15 ) ( 15,15 ) ( 5,15 ) ( 10,25 ) ( 20,25 ) ( 25,15 ) ( 20,5 ) ( 10,5 ) ( 34,13)@xmath275 ( 18,32)@xmath95 ( 43,16)@xmath293 ( 57,16)@xmath294 ( 66,16)@xmath295 ( 57,13)@xmath296 ( 51,13)@xmath297 ( 48,27)@xmath298 ( 62,27)@xmath299 ( 48,2)@xmath300 ( 62,2)@xmath301 ( 40,15)(1,0)38 ( 55,0)(0,1)38 ( 45,15)(1,2)5 ( 50,25)(1,0)10 ( 60,25)(1,-2)5 ( 65,15)(-1,-2)5 ( 60,5)(-1,0)10 ( 50,5)(-1,2)5 ( 50,16)(0,-1)2 ( 60,16)(0,-1)2 ( 55,15 ) ( 55,15 ) ( 55,15 ) ( 45,15 ) ( 50,25 ) ( 60,25 ) ( 65,15 ) ( 60,5 ) ( 50,5 ) ( 74,13)@xmath275 ( 58,32)@xmath95

the rare decays @xmath1 are among the most important probes of flavour physics . they are potentially sensitive to dynamics beyond the standard model ( sm ) and have been intensely studied in the literature @xcite . measurements have been performed at the @xmath0-meson factories and at the fermilab tevatron @xcite . excellent future prospects for detailed measurements are provided by the lhc experiments atlas , cms , and lhcb at cern @xcite , and , in the longer run , by super flavour factories based on @xmath7 colliders . the calculability of @xmath1 decay rates and distributions benefits from the fact that these processes are , to first approximation , semileptonic modes . correspondingly , the hadronic physics is described by @xmath8 form factors , which multiply a perturbatively calculable amplitude . this simple picture is not exact because also the nonleptonic weak hamiltonian at scale @xmath9 has @xmath1 matrix elements . the prominent example is given by hadronic interactions of the form @xmath10 , where the charm quarks annihilate into @xmath11 through a virtual photon . such charm - loop contributions are more complicated theoretically than the form - factor terms . even though the charm loops are subdominant numerically in the kinematical regions of interest , they can not be completely neglected . in particular , the related uncertainty needs to be properly estimated in order to obtain accurate predictions . we need to distinguish three regions in the dilepton invariant mass @xmath2 , for which the properties of charm loops are markedly different . for @xmath12 the presence of very narrow @xmath13 resonances leads to huge violations of quark hadron duality @xcite and the hadronic backgrounds from @xmath14 , followed by @xmath15 , dominate the short distance rate by two orders of magnitude . this region in @xmath2 can be removed by experimental cuts . for @xmath16 the kaon is very energetic and the charm loops can be computed systematically in the heavy - quark limit using qcd factorization for @xmath0 decays into light - like mesons @xcite . this approach was first employed for @xmath1 in @xcite . the results have many applications . a summary with detailed references can be found in @xcite ( see also @xcite for a recent analysis ) . the high-@xmath2 region , @xmath17 has received comparatively little attention . in this case the kaon energy is around a @xmath18 or below , and ( soft - collinear ) qcd factorization is less justified , becoming invalid close to the endpoint of the spectrum at @xmath19 . on the other hand , the large value of @xmath2 defines a hard scale for the hadronic contribution to @xmath1 . consequently an operator product expansion ( ope ) can be constructed , which generates an expansion of the amplitude in powers of @xmath20 ( or @xmath21 ) . charm loops , and other hadronic contributions , are thus approximated as effective interactions that are local on the soft scales set by @xmath22 and @xmath23 . this simplifies the computation substantially . in fact , to leading order in the ope the hadronic contribution reduces to a standard form - factor term . this picture has been first discussed at lowest order in the ope in @xcite , where it was applied to the endpoint region of @xmath1 and @xmath24 . in @xcite the ope was considered in some detail , including a discussion of power corrections . in the present paper we formulate the ope for the high-@xmath2 region of @xmath1 from the outset . although our approach is similar in spirit to the analysis of @xcite , the concrete implementation is different . we will also go beyond the estimates presented in @xcite in several ways . an important difference is that @xcite combines the ope with heavy - quark effective theory ( hqet ) , whereas we prefer to work with @xmath25-quark fields in full qcd . the latter formulation has the advantage of a simplified operator basis , which makes the structure of power corrections and their evaluation considerably more transparent . we also retain the kinematical dependence on @xmath2 in the coefficient functions , rather than expanding it around @xmath26 . we further discuss the issue of quark - hadron duality , which appears relevant because of the existence of @xmath13 resonance structure in the @xmath2 region of interest . violations of duality are effects beyond any finite order in the ope . using a resonance model based on a proposal by shifman , we quantify for the first time the size of duality violations in the high-@xmath2 region of @xmath1 . further aspects and new results of our analysis will be summarized in sections 8 and 9 . the main conclusion is that @xmath1 is under very good theoretical control also for @xmath17 . precise predictions can be obtained in terms of the standard form factors , with essentially negligible effects from the additional hadronic parameters related to power corrections and duality violation . the paper is organized as follows . section 2 collects basic expressions for later reference . in section 3 our formulation of the ope for @xmath1 at high @xmath2 is described and the power expansion is constructed explicitly , complete to second order in @xmath27 and with a discussion of weak annihilation as an example of a ( small ) third - order correction . in section 4 we present an estimate of the matrix elements of higher - dimensional operators and quantify their impact on the decay amplitudes for both @xmath28 and @xmath29 transitions . section 5 discusses the connection between the ope for large @xmath2 and qcd factorization for energetic kaons , which are shown to give consistent results at intermediate @xmath30 . in section 6 we address the subject of duality violation in the context of a toy model analysis . the estimate is then adapted to the case of @xmath31 in section 7 . in this section we also address conceptual aspects relevant for the existence of the ope and the notion of quark - hadron duality . a comparison of our approach with the literature is given in section 8 before we conclude in section 9 . details on the basis of operators in the ope are described in appendix a and some numerical input is collected in appendix b. the effective hamiltonian for @xmath32 transitions reads @xcite @xmath33\ ] ] where @xmath34 the operators are given by @xmath35 note that the numbering of @xmath36 is reversed with respect to the convention of @xcite . our coefficients @xmath37 correspond to @xmath38 in @xcite and we include the factor of @xmath39 in the definition of @xmath40 . the sign conventions for the electromagnetic and strong couplings correspond to the covariant derivative @xmath41 . with these definitions the coefficients @xmath42 are negative in the standard model . we define the kinematic quantities @xmath43 ( where @xmath2 is the dilepton invariant mass squared ) , @xmath44 , and @xmath45 the differential branching fractions for @xmath46 can then be written as @xcite @xmath47 the coefficient @xmath48 contains the wilson coefficient @xmath49 combined with the short - distance parts of the @xmath46 matrix elements of operators @xmath50 . the coefficient @xmath48 multiplies the matrix element of the local operator @xmath51 in the decay amplitude . the coefficient @xmath52 of the operator @xmath53 is determined by very short distances @xmath54 and is precisely known . the corresponding formulas for @xmath55 can for instance be found in @xcite . the amplitudes for the exclusive decays @xmath56 , where @xmath57 , @xmath58 , or a similar meson , are given by the matrix element of the effective hamiltonian in ( [ heff ] ) between the initial @xmath0 meson and the @xmath59 final state . the dominant contribution comes from the semileptonic operators @xmath40 . their matrix elements are simple in the sense that all hadronic physics is described by a set of @xmath60 transition form factors . this is also true for the electromagnetic operator @xmath61 . the matrix elements of the hadronic operators @xmath62 are more complicated . they are induced by photon exchange and can be expressed through the matrix element of a correlator between the hadronic part of the effective hamiltonian @xmath63 and the electromagnetic current of the quarks @xmath64 where @xmath65 is the electric charge quantum number of quark flavour @xmath66 and a summation over @xmath66 is understood . the decay amplitude may thus be written as @xmath67\ ] ] where @xmath68 and @xmath69 are the lepton spinors and @xmath70 for @xmath71 transitions the contribution from @xmath72 is suppressed by the prefactor @xmath73 and can be neglected . exploiting the presence of the large scale @xmath74 , an operator product expansion ( ope ) can be performed for the non - local term @xmath75 which describes the contribution of 4-quark operators to the @xmath32 amplitude . such an ope corresponds to integrating out the hard quark loop , leading to a series of local effective interactions for the high-@xmath2 region . to leading order in the large-@xmath2 expansion this has been presented in @xcite . a discussion of the ope including higher - order contributions has been given in @xcite . before going into more detail we discuss the basic structure of the ope for @xmath76 . the expansion may be written as @xmath77 the operators @xmath78 are composed of quark and gluon fields and have the flavour quantum numbers of @xmath79 . they are ordered according to their dimension @xmath80 and carry an index @xmath81 labeling different operators with the same dimension . the @xmath82 are the corresponding wilson coefficients , which can be computed in perturbation theory . the large scales justifying the expansion are @xmath83 and @xmath2 . they are counted as quantities of the same order . the coefficients then scale as @xmath84 in the heavy - quark limit . since the matrix elements @xmath85 scale as @xmath86 , the matrix element of each term in ( [ khco ] ) behaves as @xmath87 . current conservation implies that all operators are transverse in @xmath66 , @xmath88 it is convenient to work with the @xmath25-quark field in full qcd . this field could be further expanded within heavy - quark effective theory ( hqet ) , in order to make the @xmath89-dependence fully explicit . in such an approach many additional operators would arise whose hadronic matrix elements are not readily known . in contrast , the advantage of using the @xmath25-field in full qcd is that fewer operators appear and that the matrix elements of the leading ones are given by common form factors . in this method the ope becomes particularly transparent and we will adopt it here . at leading order in the ope ( @xmath90 ) , illustrated in fig . [ fig : kh3 ] , and in the chiral limit ( @xmath91 ) there are two operators @xmath92 using the equations of motion for the external quarks it can be shown that all possible bilinears @xmath93 and @xmath94 arising from the correlator @xmath76 can be expressed in terms of ( [ o31 ] ) and ( [ o32 ] ) . consequently , no independent dimension-4 operators of the form @xmath94 can appear in the ope . the complete proof is given in appendix a. as an example , the operator @xmath95 satisfies the equations - of - motion identity ( for @xmath91 ) @xmath96 for any @xmath97 matrix element with momentum transfer @xmath66 this is equivalent to @xmath98 because of current conservation only the transverse part @xmath99 of such an operator @xmath100 can appear in the ope . from ( [ sdbq ] ) we see that this part can be reduced to a linear combination of ( [ o31 ] ) and ( [ o32 ] ) . if we keep @xmath101 , two additional operators have to be considered @xmath102 since @xmath103 is small , and numerically similar to @xmath104 , we may formally count these as operators of dimension 4 . because they are absent at order @xmath105 , their impact will be suppressed to the level of @xmath106 , which is negligible . note that these operators do in any case not introduce new hadronic form factors . at @xmath107 ( fig . [ fig : kh5 ] ) we encounter operators with a factor of the gluon field strength @xmath108 , which have the form @xmath109 where the @xmath110 denote dirac and lorentz structures . we will treat the ope explicitly to the level of @xmath107 , that is including power corrections up to second order in @xmath104 . although we will not give a full treatment of dimension-6 corrections , we consider as an example the effect of weak annihilation ( fig . [ fig : kh6 ] ) . this contribution is characterized by the annihilation of the two valence quarks in the @xmath111 meson in the @xmath112 transition through the weak hamiltonian . it is described by 4-quark operators , which read schematically @xmath113 with lorentz and dirac structures indicated by @xmath114 , and the light quark field @xmath115 , @xmath80 in the case of non - strange @xmath111 mesons . weak annihilation provides a mechanism to break isospin symmetry , directly at the level of the transition operator . in @xmath76 weak annihilation , in addition to being a third order power correction , comes only from qcd penguin operators , which have small coefficients . the contribution to isospin breaking from this source will therefore be strongly suppressed . in this section we give explicitly the first few terms in the ope to leading order in renormalization - group improved perturbation theory , that is neglecting relative corrections of @xmath117 . this order for @xmath118 is relevant in the next - to - leading logarithmic approximation to the @xmath119 amplitude . we may then write @xmath120 the lower indices of the terms on the r.h.s . denote the dimension @xmath80 of the corresponding local operators , which come with a coefficient of order @xmath121 . the first term reads @xmath122\end{aligned}\ ] ] the coefficient in ( [ kh3 ] ) requires a uv renormalization , which has to be consistent with the definition of @xmath123 . the expression given here corresponds to the ndr scheme used in @xcite . the function @xmath124 is ( @xmath125 , @xmath126 , @xmath127 ) @xmath128 next we have @xmath129\ , \bar s\gamma_\lambda(1-\gamma_5)gg_{\alpha\rho}b\ , \frac{c_1 q_c}{q^2 } f(x ) \nonumber \\ & & -\frac{q_\lambda}{m_b } \bar s gg_{\alpha\beta } ( g^{\alpha\lambda}\sigma^{\beta\mu } - g^{\alpha\mu}\sigma^{\beta\lambda } ) ( 1+\gamma_5)b\ , \frac{4 c_8 q_b}{q^2}\end{aligned}\ ] ] here @xmath130 with @xmath131 . the charm - loop contribution in ( [ kh5 ] ) , proportional to @xmath132 , can be inferred from @xcite . note that here we use the convention @xmath133 . in writing ( [ kh5 ] ) we have neglected terms with the small qcd penguin coefficients @xmath134 . finally , weak - annihilation diagrams give the dimension-6 term @xmath135 \ , ( \delta_{ij}\delta_{kl } c_4 + \delta_{il}\delta_{kj } c_3 ) \nonumber\\ & + & \frac{16\pi^2i}{q^4}q_\lambda\sum_{r = u , d } \bigg [ q_r ( \bar r_i(1-\gamma_5)b_j\ , \bar s_k\sigma^{\mu\lambda}(1+\gamma_5)r_l + \bar r_i\sigma^{\mu\lambda}(1-\gamma_5)b_j\ , \bar s_k(1+\gamma_5)r_l ) \nonumber\\ & & -\frac{1}{3 } ( \bar r_i(1-\gamma_5)b_j\ , \bar s_k\sigma^{\mu\lambda}(1+\gamma_5)r_l -\bar r_i\sigma^{\mu\lambda}(1-\gamma_5)b_j\ , \bar s_k(1+\gamma_5)r_l)\bigg ] \nonumber\\ & & ( \delta_{ij}\delta_{kl } c_6 + \delta_{il}\delta_{kj } c_5)\end{aligned}\ ] ] the terms in ( [ kh6 ] ) only arise from qcd penguin operators , which have small coefficients . we remark that all operators in ( [ kh3 ] ) , ( [ kh5 ] ) and ( [ kh6 ] ) vanish identically when contracted with @xmath136 , as required by gauge invariance . the non - factorizable @xmath117 corrections to the charm loop arise from diagrams like the one shown on the r.h.s . of fig . [ fig : kh3 ] . the @xmath2-dependence has been recently calculated in analytic form as a taylor expansion in the small parameter @xmath137 @xcite . analytic results for @xmath138 had been presented in @xcite . in the kinematical range relevant to our considerations , it has been shown that the convergence of the series is very good . we therefore use the mathematica input files provided by the authors of @xcite in the online preprint publication for a numerical estimate . we find that the non - factorizable @xmath117 corrections to the charm loop lead to a 10 - 15% reduction of the real part of @xmath139 and contribute a negative imaginary part of again 10 - 15% relative to the short - distance contribution from @xmath123 ( the precise value is scheme - dependent ) . this is in agreement with the effect found for the _ inclusive _ @xmath140 rate in the high-@xmath2 region , as discussed in @xcite , and is similar to the effect observed for the low-@xmath2 region in the exclusive decay modes , see table 5 in @xcite . it is to be stressed that these corrections almost compensate the factorizable charm - loop contribution ( diagram on the l.h.s . in fig . [ fig : kh3 ] ) . the reason why the @xmath117 corrections are not suppressed stems from the different colour structure of the diagrams . whereas the factorizable charm loop comes with a colour - suppressed combination of wilson coefficients , the additional gluon exchange allows the @xmath13-pair to be in a colour - octet state with no such suppression . at even higher orders in perturbation theory , @xmath141 with @xmath142 , on the other hand , the numerical effect on @xmath139 should really be small , as no new additionally enhanced colour structures will arise . the computation of the amplitude from the ope requires the evaluation of the matrix elements of the local operators . we estimate in particular the matrix elements of the higher - dimensional contributions . this will allow us to quantify power corrections to the @xmath1 amplitude at high @xmath2 . the cases of @xmath28 and @xmath29 transitions will be considered in turn . the matrix element of the leading dimension-3 operator is given in terms of the familiar form factors @xmath143 , defined by ( @xmath144 ) @xmath145\ , q^\mu\ ] ] at the level of the dimension-5 correction in ( [ kh5 ] ) one encounters operators of the form @xmath146 . their matrix elements introduce , in general , new nonperturbative form factors . using lorentz invariance and the antisymmetry of @xmath147 and @xmath148 one can show that @xmath149 in order to estimate the remaining term we assume @xmath150 for the kaon energy @xmath22 . in this limit the matrix element can be computed in qcd factorization . to leading order we then find @xmath151\ ] ] where @xmath152 , @xmath153 is the number of colours , and @xmath154 is the first inverse moment of the @xmath0-meson light - cone distribution amplitude . this matrix element scales as @xmath155 relative to ( [ bkffpm ] ) in the heavy quark limit , and as @xmath156 for large @xmath153 . in a similar way we can estimate the weak annihilation term @xmath157\ ] ] this contribution is power - suppressed as @xmath158 relative to ( [ bkffpm ] ) . the suppression by the small wilson coefficients @xmath159 is partly compensated by a large numerical factor of @xmath160 . to relative order @xmath158 there is no contribution from the term with @xmath161 and @xmath162 . note that the result in ( [ bkkh6 ] ) also corresponds to the matrix element obtained when naively factorizing the four - quark operators . normalized to the amplitude coefficient @xmath163 , the power corrections from @xmath164 and @xmath165 read @xmath166 @xmath167 where @xmath115 , @xmath80 refers to the spectator quark in the @xmath0 meson . numerically , we find @xmath168 at @xmath169 for central values of the parameters . this number comes with a substantial uncertainty , in particular from @xmath170 and @xmath171 . nevertheless , the correction to @xmath172 is very small , most likely below @xmath173 in magnitude . the correction in ( [ dela9h5 ] ) diminishes further for larger @xmath2 , reaching @xmath174 at the endpoint . towards the endpoint the kaon becomes soft and the result in ( [ dela9h5 ] ) , based on @xmath175 , can only be viewed as a rough model calculation . the conclusion that @xmath176 remains negligibly small should however still hold . the weak annihilation correction is @xmath177 at @xmath169 for @xmath115 and therefore entirely negligible , a consequence also of the small wilson coefficients . in the case of the decay into a vector meson the relevant form factors are defined as ( @xmath178 ) a_1(q^2 ) + & - a_2(q^2 ) it is convenient to treat the decay into longitudinally and transversely polarized vector mesons separately . omitting terms proportional to @xmath179 @xmath180\end{aligned}\ ] ] @xmath181 in the large energy limit @xmath182 , which we may use in the normalization of the power corrections , ( [ bkparff ] ) and ( [ bkperpff ] ) simplify to @xcite @xmath183 the case of a longitudinally polarized @xmath58 is very similar to the case of a pseudoscalar @xmath184 , discussed in section [ subsec : btok ] , and we find @xmath185 where @xmath186 is the decay constant of @xmath187 . for a @xmath58 with transverse polarization we obtain @xmath188 where @xmath189 is the decay constant of @xmath190 . numerically we have @xmath191 and @xmath192 at @xmath193 for our standard set of parameters . these corrections are of similar size as for the pseudoscalar kaon and they are likewise negligible . finally , we quote the corrections from weak annihilation a_9,h6a(k^ * _ ) & = -(c_4 + ) + a_9,h6a(k^ * _ ) & = -(c_6 + ) ( q_r- ) where @xmath115 , @xmath80 refers to the spectator quark in the b meson . the corrections are tiny , at @xmath193 we have @xmath194 for @xmath115 , and @xmath195 , @xmath196 for @xmath197 . this is again a consequence of the @xmath198 suppression at large values of @xmath2 , see also @xcite . the ope for the correlator ( [ khdef ] ) , applied to @xmath1 , is valid as long as the energy of the kaon in the @xmath0-meson rest frame @xmath199 is small compared to @xmath200 . this condition is certainly fulfilled in the vicinity of the endpoint , but even for @xmath2 as low as @xmath201 , just above the narrow - resonance region , @xmath202 is still fairly small in comparison to the hard scale . on the other hand , such a value of @xmath22 is already larger than the qcd scale @xmath23 and one could consider using the factorization methods applicable to the case of energetic kaons . for @xmath2 at @xmath201 , or somewhat above , we have a transition region for the applicability of factorization methods for large @xmath22 and the ope for small @xmath22 . it is this transition we wish to explore in the present section . qcd factorization ( qcdf ) at large kaon recoil requires @xmath175 for arbitrary @xmath2 . the ope method requires @xmath203 , @xmath23 . both scenarios are consistent with the case @xmath204 which can be realized , at least approximately , as we have seen above . in both , the qcdf and the ope scenario , the leading contributions to the @xmath1 amplitudes are given by short - distance quantities multiplying the standard @xmath8 form factors . in qcdf , the first corrections to these form - factor terms come from hard - spectator interactions , which have been computed in @xcite . normalized to @xmath139 ( called @xmath205 in @xcite ) these corrections may be written as @xmath206 , where the indices refer to the notation of @xcite . for definiteness we will consider the case of a pseudoscalar kaon in the following . neglecting the small penguin coefficients @xmath134 and adapting the expressions in @xcite to our notation , the second contribution reads ( @xmath207 ) @xmath208\end{aligned}\ ] ] in the ope limit , defined by treating @xmath209 and taking ( [ q2eklambda ] ) , we have @xmath210 . this term is therefore subleading with respect to ( [ dela9h5 ] ) , consistent with the absence of a @xmath211 term in @xmath176 . the remaining term is given by @xmath212 where ( @xmath213 ) @xmath214 @xmath215 @xmath216 @xmath217 @xmath218 the formulas from @xcite have been slightly rewritten here to make them more convenient for application in the high-@xmath2 region where @xmath219 . to make contact with the ope result we eliminate @xmath2 from ( [ tpardef ] ) using @xmath220 and expand in @xmath221 . the terms proportional to @xmath222 and @xmath223 cancel and we find @xmath224 with the function @xmath225 from ( [ fxdef ] ) . in the limit @xmath226 the qcdf formula ( [ da9p ] ) then reduces to the ope result ( [ dela9h5 ] ) . we note that the hard - spectator correction ( [ da9p ] ) , which is a leading - power contribution in qcdf , becomes a power - correction @xmath227 in the ope regime . this behaviour is related to the form factor @xmath228 , which scales as @xmath229 at small @xmath2 and as @xmath230 at large @xmath2 . we may use the preceding comparison to check the validity of the ope at the relatively low values of @xmath2 around @xmath201 . assuming that the large - energy limit for the kaon is a reasonable approximation , the qcdf expression ( [ da9p ] ) and the ope result ( [ dela9h5 ] ) differ only by the replacement @xmath231 where the numerical values are obtained at @xmath169 for our standard set of parameters , neglecting the kaon mass for consistency . the real part of the ope approximation is @xmath232 larger in magnitude than the qcdf result , for the imaginary part the discrepancy is about @xmath233 . such differences are to be expected since the expansion leading from qcdf to the ope is governed by @xmath234 . within this accuracy , the ope formula still gives a very good estimate of the more complete qcdf result at @xmath169 . for larger values of @xmath2 the numerical difference between the two sides in ( [ da9ph5 ] ) becomes smaller and towards the endpoint the ope is the more appropriate description . the above exercise suggests that even at @xmath2 as low as @xmath201 the ope is a valid method to obtain the second - order correction ( [ dela9h5 ] ) . the difference with the qcdf estimate is immaterial in view of the very small overall size of the effect . the transition from large to small recoil energy has already been considered in @xcite for the @xmath28 form factor ratio @xmath235 . also in this and in analogous cases , hard - spectator corrections , which are leading - power effects at large recoil , become power suppressed for soft kaons . similar observations hold for weak annihilation @xcite . the ope defines a systematic framework to compute the correlator ( [ khdef ] ) at high @xmath2 in qcd . in the minkowski region @xmath236 , there will be uncertainties in the ope - based predictions that go beyond those due to neglected orders in @xmath116 or in @xmath104 . such effects are refered to as violations of quark - hadron duality . we investigate their importance first within a toy model for the charm loops in @xmath31 . in section 7 the model is generalized to be closer to the realistic case . it is illuminating to consider the systematics of duality for the charm - loop contribution to @xmath31 in the simplified context of the toy model introduced in @xcite . this model assumes the existence of two leptons , @xmath237 with a large mass @xmath238 and @xmath239 with mass @xmath240 , and the effective weak hamiltonian @xmath241\ ] ] all particles have standard strong and electromagnetic interactions . then @xmath242 gives rise to a loop - induced process @xmath243 via charm- and top - quark penguin diagrams with a gim - like cancellation between them . the hadronic physics is fully contained in current correlators of the form @xmath244 with @xmath245 . the decay amplitude can then be written as @xmath246 where @xmath247 is the difference of the charm and top contributions . we take @xmath248 , and thus @xmath249 comes only from the charm sector . the correlator @xmath250 fulfills the dispersion relation @xmath251 where the subtraction constant @xmath252 is fixed within the model and can be computed in perturbation theory . to leading order it reads @xmath253 the form of ( [ al12ee ] ) for the amplitude holds to lowest order in @xmath254 and @xmath255 , but to all orders in the strong coupling . the decay @xmath256 in this model shares important similarities with @xmath257 , but the hadronic dynamics is simplified to the physics of quark - current correlators . the role of resonances and quark - hadron duality can thus be illustrated in a transparent way . from ( [ al12ee ] ) we obtain the differential decay rate ( with @xmath258 ) @xmath259 where we defined @xmath260 and @xmath261 to lowest order we have @xmath262 and the partonic expression for @xmath263 in one - loop approximation is @xmath264 \ln\frac{1+\sqrt{1-x}}{1-\sqrt{1-x } } - i\pi \,;~ & x<1 \end{array } \right.\ ] ] where @xmath131 . for typical values of the parameters ( e.g. @xmath265gev , @xmath266gev , @xmath267gev ) we have @xmath268 , whereas @xmath269 first rises from 0 at @xmath270 to 4/3 at @xmath271 and then drops again monotonically to the small negative value @xmath272 at @xmath26 . the coefficient @xmath273 represents the short - distance contribution of the amplitude . it is real and larger ( parametrically as well as numerically ) than the quark - level charm contribution @xmath274 . the decay rate ( [ gaml12ee ] ) is proportional to @xmath275 is small compared to @xmath273 , there is a clear hierarchy among the three terms on the r.h.s . of ( [ ccdel ] ) : the first , short - distance term @xmath276 dominates and the next term gives the correction to first - order in @xmath277 , whereas the final term enters at second order . these features are qualitatively similar in the case of @xmath31 . in order to investigate the systematics of duality violation in @xmath256 , we find it convenient to consider first a simple model for the quark - current correlator , which has been proposed in @xcite . in this model the correlator is represented as an infinite sum over resonances , which include finite width effects . in its original form it applies to massless quarks and we will correspondingly neglect the charm - quark mass in the present section . a detailed discussion of the model and its use in illustrating duality violation in the @xmath278 ratio and similar quantities has been given in @xcite . the model has also been used to study duality violation in @xmath279 decays in @xcite . in shifman s model the correlator @xmath277 in ( [ deldef ] ) reads @xmath280\ ] ] where @xmath281 is the digamma function and @xmath282 @xmath283 is the number of colours , @xmath284 is a scale corresponding to the string tension in qcd and @xmath285 is a ( small ) parameter related to the width - to - mass ratio of the resonances . the model expression for @xmath277 in ( [ delshm ] ) has the correct analytic behaviour ( a cut for positive @xmath2 but no other singularities on the physical sheet ) and it reproduces the asymptotic result of qcd in the limit of large @xmath2 , @xmath286 using the identity @xmath287_1 + \left[-\pi \cot \pi z + i\pi \right]_2\ ] ] the function in ( [ delshm ] ) can be decomposed into two parts , @xmath288 , corresponding to the two brackets in ( [ psiid ] ) . @xmath289 is an oscillating function of @xmath2 , exponentially suppressed for large @xmath2 . it represents the duality violating component of @xmath277 . the function @xmath290 is a monotonous function , which gives the ope approximation to @xmath277 . the real part of these two functions is shown in fig . [ fig : rddvope ] . a plot of the imaginary part of @xmath277 can be found in @xcite . the duality violating part of @xmath291 can be approximated as @xmath292\approx -\frac{n\pi}{3 } \exp(-2\pi b r ) \sin(2\pi r)\ ] ] if @xmath293 and @xmath294 . in the decay rate integrated over the high-@xmath2 part of the spectrum , the duality violating contribution enters proportional to @xmath295 where @xmath296 in ( [ dvrate ] ) we have used the approximation from ( [ deldv ] ) . typical values of the parameters are @xmath297 the value of @xmath298 corresponds for instance to @xmath299 and @xmath300 . the quantity in ( [ dvrate ] ) is shown in fig . [ fig : rddvint ] as a function of @xmath301 . we comment on several important aspects of these results . * the duality violating component of @xmath291 in ( [ deldv ] ) exhibits the characteristic oscillating behaviour in @xmath302 with an exponential suppression governed by @xmath303 . the analogous expression for the duality violating term in @xmath304 , which has a cosine instead of the sine , has been given in @xcite . ( [ dvrate ] ) displays the duality violating contribution from @xmath291 to the partially integrated decay rate . the integration over @xmath305 extends from a suitably chosen lower limit @xmath301 up to the end of the spectrum . the parameter @xmath301 should be large enough such that the ope still remains reasonable at the corresponding value of @xmath2 . using the approximation in ( [ deldv ] ) and expanding in the small quantities @xmath25 and @xmath306 , we find the explicit result written as the last expression in ( [ dvrate ] ) . this expression is an oscillating function of @xmath301 , with frequency @xmath307 , multiplied by an exponential suppression factor . the latter is active when the exponent is large , at least @xmath308 . in addition , the entire term is further suppressed by one power of @xmath309 . this effect remains even if the exponential suppression is not fully developped . the power suppression arises because the oscillating contributions average out in the integral , except for a remainder @xmath310 near the lower end of integration @xmath301 . * the duality violating term ( [ dvrate ] ) is plotted in fig . [ fig : rddvint ] for a semi - realistic choice of parameters , where @xmath311 and @xmath299 . the smallest values of @xmath301 shown , @xmath312 , correspond to @xmath313 and @xmath314 , which is already on the low side of the allowed range . from fig . [ fig : rddvint ] we observe , first , that in the scenario considered here the simple approximation to ( [ dvrate ] ) agrees very well with the full result . second , the numerical size of the duality violating term is about @xmath315 for the relatively low values of @xmath301 around @xmath316 . the effect diminishes quickly for larger @xmath301 due to the exponential suppression . the variation @xmath315 from ( [ dvrate ] ) amounts to @xmath317 when compared with the corresponding ope expression @xmath318 we next turn to a discussion of the @xmath319 term in ( [ ccdel ] ) , where the situation is systematically different from the case of @xmath291 . the @xmath319 term receives a contribution from the duality violating component given by @xmath320 where in the second step an approximation similar to the one in ( [ deldv ] ) has been used . there is still an exponential suppression , which makes the entire term negligible for sufficiently large @xmath321 . on the other hand , the power suppression with @xmath306 observed in ( [ dvrate ] ) is softened into a behaviour as @xmath322 . for small @xmath25 ( comparable to @xmath306 or smaller ) the violation of duality may become large . this is in qualitative agreement with the discussion of duality violation for the squared correlator @xmath323 given in @xcite . the enhanced impact of duality violation is related to the absence of oscillations with alternating sign in the integrand of ( [ delsqdv ] ) . we conclude that the @xmath319 term in the integrated rate is particularly susceptible to violations of quark - hadron duality , which may lead to substantial ( positive ) deviations from the ope result , unless @xmath2 is large enough for a strong exponential suppression of the effect . however , the uncertainties in the @xmath319 term may be immaterial if this contribution is only a small correction to the dominant @xmath276 part in ( [ ccdel ] ) . let us finally illustrate the relative importance of the various contributions to the rate of @xmath256 using the shifman model for the charm loop with the numerical input defined above . in the limit @xmath324 considered here the function @xmath273 from ( [ cclnmtmc ] ) is modified to @xmath325 the relative contributions to the decay rate ( [ gaml12ee ] ) , ( [ ccdel ] ) then read @xmath326\label{cdnum}\\ \int^1_{0.1 } ds\ , g(s)\ , |\delta_1|^2 & = & 1.25 \qquad [ + 0.10]\label{ddnum}\end{aligned}\ ] ] with @xmath327 . the central values in ( [ cdnum ] ) and ( [ ddnum ] ) are based on the ope result for @xmath277 , the square brackets indicate the impact of duality violation . the uncertainty in ( [ cdnum ] ) gives the variation due to @xmath328 . it has the relative size of @xmath317 , which is reduced to @xmath329 in the sum of all contributions . the shift from duality violation in ( [ ddnum ] ) is positive and essentially negligible in the present example . in order to obtain a more realistic picture of the @xmath13 spectrum , we are going to fit the bes data @xcite for the @xmath278ratio in the @xmath13region to a modified version of shifman s model . the spectra of @xmath13 mesons can be accounted for by linear relations for the squared masses , @xmath330 , @xmath331 , similarly to the case of the light mesons @xcite . the trajectory of the @xmath332 charmonia , the @xmath333 states @xmath334 , @xmath335 , @xmath336 , @xmath337 , @xmath338 , for instance , follows this pattern . starting form the third resonance ( @xmath339 ) , these states can decay into open charm and have widths of the order of @xmath23 . the first two are extremely narrow and may be described separately , but their properties are unimportant for duality violation , which is related to the infinite tower of high-@xmath81 resonances . we therefore choose an ansatz where the sum over resonances begins at @xmath339 rather than @xmath340 . a finite width is included in analogy to ( [ zdef ] ) and the variable @xmath2 is shifted by a constant into @xmath341 . this leads to the following expression for the imaginary part of the correlator or , equivalently , the @xmath278 ratio @xmath342 the individual resonances are located at @xmath343 ( @xmath344 ) in the limit @xmath345 . we observe that a rough description of the bes data @xcite can already be obtained with this formula , where we find @xmath346 from the measured r - ratio below charm threshold , @xmath347 , @xmath348 and @xmath349 . this yields the result shown in fig . [ fig : fit0 ] , corresponding to a @xmath350 . we remark that our values for the fit parameters @xmath351 and @xmath352 are in agreement with the results of @xcite . there is a second trajectory of @xmath353 charmonia , the @xmath354 states . of these the first two resonances @xmath355 and @xmath356 are known . the first one is barely above threshold and still rather narrow . it may be considered separately , similar to @xmath334 and @xmath335 . note also that @xmath355 is still below our default choice for the lower limit of the high-@xmath2 region , @xmath357 . the remaining resonances @xmath354 are rather close to the resonances @xmath358 for @xmath142 . for an approximate treatment it appears justified to subsume such a pair of close resonances under a single peak and keep the ansatz given in ( [ step1 ] ) . the accuracy of this description can be gauged by inspecting fig . [ fig : fit0 ] . in any case , the normalization of the second term of @xmath278 in ( [ step1 ] ) is fixed in the large-@xmath2 limit by the free - quark result . simple fit to bes data for @xmath278 vs. @xmath359 . right : detailed view.,title="fig:",scaledwidth=45.0% ] simple fit to bes data for @xmath278 vs. @xmath359 . right : detailed view.,title="fig:",scaledwidth=45.0% ] the fit could be refined by including the low - lying , narrow resonances in the fit ansatz . any finite number of resonances does not change the asymptotic behaviour responsible for duality violation . also the fit parameters are not much affected by such modifications . for example , including the @xmath355 reduces the @xmath360 , while the parameters of the continuum ansatz remain almost unchanged , @xmath361 , @xmath347 , @xmath362 and @xmath349 . we will therefore be content with the simple representation given in ( [ step1 ] ) above . the analytic structure of the matrix element of the operator product in ( [ khdef ] ) can be infered from @xmath363 in order to discuss the properties of the matrix element in ( [ khas ] ) we make the following simplifications . first , we neglect the small penguin contributions in the weak hamiltonian , that is we take @xmath364 second , in the electromagnetic current we retain only the charm - quark component , @xmath365 , and consider only contributions where the charm fields in @xmath366 are contracted with those in @xmath367 . this neglects terms where the charm - anticharm pair from @xmath366 annihilates into gluons before connecting to the electromagnetic current . such contributions are of higher order in @xmath116 and not essential for the problem we want to address . we consider the matrix element in ( [ khas ] ) as a function of @xmath368 , keeping the kaon energy @xmath369 fixed at a value of order @xmath370 . this can be achieved by injecting a spurion 4-momentum @xcite @xmath371 into the @xmath366 vertex . then @xmath372 and for @xmath373 and @xmath369 fixed at their physical values the variable @xmath374 grows with @xmath368 . the physical kinematics is recovered for @xmath375 . under the simplifying assumptions specified above , the intermediate state @xmath376 in the first sum in ( [ khas ] ) always contains a @xmath13 pair and a strange quark , in general together with other hadronic states , and thus @xmath377 . the state @xmath376 in the second sum contains a @xmath13 pair and a @xmath25 quark , and @xmath378 . the matrix element in ( [ khas ] ) is then seen to be an analytic function of @xmath368 in the entire @xmath368-plane , except for two branch cuts at @xmath379 and at @xmath380 . if we would relax the simplifications above and allowed for intermediate states @xmath376 without @xmath13 pairs , the cuts would extend down to lower values of @xmath381 on the real axis . the ope of the matrix element in ( [ khas ] ) can be justified for @xmath368 on the imaginary axis , sufficiently far from the origin , that is at @xmath382 , for @xmath383 . the ope defined in this way in the euclidean can then be analytically continued , term by term , from imaginary @xmath368 onto the positive real axis , corresponding to the minkowskian domain . terms that are exponentially suppressed in @xmath384 for large positive @xmath385 become oscillating functions of @xmath368 in the minkowskian case , that is for large positive @xmath368 @xcite . these oscillating terms are invisible at any finite order in the ope and represent the duality violating contribution . for a quantitative estimate of duality violation we have to resort to a model of the hadronic correlator in ( [ khas ] ) . to this end we write the hamiltonian in ( [ hcnopeng ] ) as @xmath386 and assume a factorization of the currents , that is we neglect interactions between @xmath387 and the @xmath388 system . the coefficient @xmath389 is then treated as a phenomenological parameter . with these simplifications the correlator in ( [ khas ] ) reduces to @xmath390 where we omitted the longitudinal component @xmath391 . @xmath392 is the current correlator defined in ( [ pimunu ] ) . the charm loop in ( [ khfact ] ) contributes to the coefficient @xmath139 in the amplitude of @xmath46 a term @xmath393 in the model of section [ sec : charmbes ] we have @xmath394\ ] ] where @xmath395 and @xmath396 . for the parameters we use the following values @xmath397 we will not employ the model ( [ defd ] ) to describe the entire charm - loop contribution , but only to estimate its duality violating component . the remainder is more reliably obtained by the ope itself . in order to extract the term that represents duality violation within the model , we decompose @xmath398_1 + \left[-\pi\cot \pi z + i\pi\right]_2\ ] ] this decomposition is useful for @xmath17 , when the second term starts being exponentially suppressed and gives the duality violating contribution . in correspondence with ( [ psiz312 ] ) we have @xmath399 and the duality violating term is @xmath400_2 \approx-\frac{8\pi}{3}\exp(-2\pi br)\left(\sin 2\pi r - i\cos 2\pi r\right)\ ] ] when integrating the @xmath401 part of the @xmath46 rate over the high-@xmath2 region , from a lower limit @xmath402 to the end of the spectrum , the relative size of the duality violating effect is then given by @xmath403 to first order in the charm - loop contribution . here we have assumed @xmath389 to be real . using the approximation in ( [ defd2 ] ) and proceeding as in ( [ dvrate ] ) we find @xmath404 this formula gives an excellent approximation of the full result based on ( [ rdv1def ] ) . we note the ( mild ) exponential suppression and the power suppression by @xmath405 . for @xmath389 we take the value @xmath406 , which is large enough to reproduce the measured @xmath407 branching fraction within the factorization ansatz . we recall that @xmath172 . then , for @xmath408 , that is for @xmath409 in the vicinity of @xmath201 , the relative correction @xmath410 is below @xmath411 . in the rate for @xmath46 , @xmath412 is added to @xmath401 , which roughly doubles the result . the net effect of the uncertainty from ( [ rdv1approx ] ) for the rate integrated over the high-@xmath2 region is therefore only about @xmath413 . as discussed in section [ sec : dualitytoy ] , the second order effect in @xmath414 is qualitatively different . it has no cancellations due to oscillating terms , giving a positive correction , and its impact increases with decreasing @xmath25 . if local duality is at least roughly fulfilled , as is the case for high enough @xmath17 , the duality violation from @xmath415 is still suppressed , being of second order in the small quantity @xmath416 . the relative size of this component is @xmath417 where the second term uses the approximation in ( [ defd2 ] ) . the approximate form and the full result in ( [ rdv2 ] ) agree reasonably well . with our set of parameters we find @xmath418 at @xmath419 , about half the size of @xmath410 . the phenomenological value @xmath406 used above is larger than the perturbative result for the coefficient @xmath389 . thus it effectively absorbs the ( partly unknown ) effects from factorizable and non - factorizable corrections . we should emphasize here that the analytic structure of the latter is , in general , more complicated than it is implied by the approximation to @xmath420 in ( [ khfact ] ) . globally using the larger value of @xmath389 in our numerical estimate of duality violation thus corresponds to the pessimistic scenario where the oscillating terms from the non - factorizable corrections to the charm - loop are added coherently . in reality , we expect that at least some destructive interference between the various contributions appears , and the amount of duality violation should even be smaller than our estimate . our model for the charm - loop in ( [ khfact ] ) is similar to the ansatz originally proposed in @xcite and used since then in many phenomenological studies . however , our motivation for considering this model is essentially different . in contrast to @xcite it is not our goal to model the hadronic effects on the spectrum point by point in the @xmath2 distribution . more relevant than the detailed shape of the spectrum is the rate integrated over the entire high-@xmath2 region , which is best described in a model - independent way by the ope , as mentioned above . we rather employ the model to get an indication of the duality violating effects , which are not captured in an ope calculation . for this purpose the shifman model for the current correlator is adapted to the charm - quark case @xmath392 , with a choice of parameters consistent with the basic features of the most recent experimental data from bes . note that the krger - sehgal ansatz for @xmath392 @xcite , consisting of a spectral function with a finite number of resonances and a flat continuum for large @xmath2 , contains no information on duality violation . the model we use to estimate duality violation is very simple and involves many assumptions . still we expect it to indicate the systematics and the typical size of the effect , which presumably is closely connected with the @xmath13 resonance structure in the charm - loop contribution . the high-@xmath2 region of @xmath421 has also been analyzed in the framework of an ope in @xcite and a recent application was presented in @xcite . our approach differs from the analysis of @xcite in several respects . we go beyond the work of @xcite by addressing in detail the issue of duality violation and the basis of the ope formalism . in addition , we extend the ope to include second - order power corrections , which we estimate quantitatively . a list of new results is given in the conclusions . here we would like to comment further on two conceptual differences between @xcite and our formulation . this concerns , first , the construction of operators in the ope and , second , the treatment of charm quarks . the operators in our approach are built from @xmath25-quark fields in full qcd rather than using hqet . this is convenient because the operator basis is simpler and the ope becomes particularly transparent . another advantage is that the matrix elements of the leading operators are given by the usual form factors in full qcd . unlike in a hqet framework , not all dependence on @xmath89 is made explicit , but this is not essential from a practical point of view and still allows the consistent inclusion of power corrections to a given order in the expansion . this approach has been used for instance in the ope for inclusive @xmath0 decays applied to the computation of the lifetime difference of @xmath422 mesons @xcite . these investigations included power corrections @xcite as well as corrections of @xmath117 @xcite . by contrast , in @xcite a matching onto hqet operators is performed from the start . this leads to a proliferation of operators , whose matrix elements are not given by the usual form factors . in fact , to reach a simplification of the resulting hqet expressions , the authors of @xcite partly undo the matching , from hqet back to full qcd expressions . we prefer to employ the full - qcd formulation throughout , in view of the advantages mentioned above . we next turn to the second point , the treatment of charm . the authors of @xcite perform an explicit expansion in @xmath423 , which corresponds to assuming the hierarchy @xmath424 . this implies that the charm quark is not integrated out at the scale @xmath425 and continues to be an active field below this scale . operators with charm - quark fields are then present in the ope . we will argue that it is conceptually simpler to integrate out charm immediately at the scale @xmath89 , assuming a hierarchy @xmath426 , and that this can be done without any loss in accuracy . charm - quark effects are then entirely contained in the coefficients of the operators , as is apparent from the formulation of the ope given in section 3 . to illustrate the point we consider the following example . let us take the scenario where @xmath424 . in this case the operator coefficients should be evaluated with @xmath138 . on the other hand , additional operators involving charm would have to be included in the ope . with explicit @xmath13 fields , such operators arise at dimension 6 or higher . for instance , radiating the virtual photon from the two charm lines in the effective hamiltonian ( [ hcnopeng ] ) and leaving all four quark lines open , gives the dimension-6 operator @xmath427 \ , ( \delta_{ij}\delta_{kl } c_1 + \delta_{il}\delta_{kj } c_2)\ ] ] in analogy to ( [ kh6 ] ) . this operator has @xmath428 matrix elements contributing to the decay amplitude . assuming next that @xmath429 , the contribution ( [ kh6c ] ) can be simplified by integrating out the charm fields in a further step . contracting the charm lines into a loop , to which a gluon field is attached , induces below @xmath351 the dimension-6 operator @xmath430 the coefficient of this operator reproduces the logarithmic @xmath351-dependence at this order in the ope . a similar discussion has been presented in @xcite , illustrating how the logarithmic term @xmath431 in the coefficient of the leading dimension-3 operator ( [ kh3 ] ) is recovered in an effective theory with four - quark operators of the type @xmath10 . whereas the latter effect vanishes for @xmath324 , we note that the coefficient in ( [ kh6d ] ) has a logarithmic divergence in this limit . this is of no consequence if @xmath429 , where @xmath351 still represents a hard scale . in fact , for higher - dimensional operators generated by ( [ kh6c ] ) , that is operators with more factors of the gluon field and its derivatives , the coefficients will scale as inverse powers of @xmath351 . a similar situation exists for inclusive semi - leptonic @xmath432 decays , where four - quark operators with charm , analogous to ( [ kh6c ] ) , also appear at third order in the ope for @xmath424 . the effect of the corresponding charm loops has been refered to as intrinsic charm in @xcite , where the issue was discussed in great detail . the above consideration indicates how the nonperturbative @xmath433 matrix element of operators such as ( [ kh6c ] ) can be treated in an expansion in @xmath434 . however , as demonstrated in @xcite for semi - leptonic @xmath432 decays , the same effects are also described in a framework where charm is integrated out at the @xmath89 scale . no active charm fields need to be considered in this case and ( [ kh6c ] ) is absent from the ope . a difference between the ope with or without active charm is that the framework with charm fields and with a strong hierachy @xmath424 assumed , would offer the possibility of efficiently resumming logarithmic terms @xmath435 . since such logarithms are not very large and in view of the additional power suppression of such terms , such resummations appear not to be necessary in practice . we also stress that effects such as ( [ kh6c ] ) , irrespective of their detailed treatment , are suppressed at least as @xmath436 . small differences in the method of their calculation are therefore hardly relevant . we thus conclude that integrating out charm at the scale @xmath89 is entirely justified . the ope is then constructed in a single step and with a simpler operator basis . for these reasons the approach appears preferable and we have adopted it here . the amplitude for @xmath437 , @xmath438 , has the general form given in ( [ abmll ] ) . it contains the component @xmath439 @xmath440 which receives a nonlocal , hadronic contribution @xmath441 from the matrix element of the nonleptonic weak hamiltonian @xmath366 in addition to the semileptonic form factor terms ( @xmath442 ) . although the hadronic part @xmath443 is relatively small numerically outside the narrow - resonance region ( @xmath444 of @xmath445 ) , it needs to be reliably computed in order to achieve very accurate predictions . we have presented a detailed study of the hadronic contribution in the region of large dilepton invariant mass @xmath17 , based on an operator product expansion in inverse powers of the hard scale @xmath200 . working with @xmath25-quark fields in full qcd and factorizing the dependence on @xmath89 , @xmath200 and @xmath351 into the coefficient functions , we obtain the following results : * to leading order in the ope and to all orders in @xmath116 , @xmath446 is expressed in terms of the standard form factors parametrizing the matrix elements @xmath447 and @xmath448 of dimension-3 operators , up to coefficients calculable in perturbation theory . to lowest order in @xmath116 only @xmath447 is present . * in the chiral limit ( @xmath91 ) , the first power corrections appear only at second order ( @xmath449 ) and are governed by dimension-5 operators with a gluon field of the form @xmath450 . the corrections are computed explicitly , using the limit @xmath175 for the hadronic matrix elements , and are shown to be smaller than @xmath173 for @xmath445 . * for @xmath101 the dimension-4 operators @xmath451 and @xmath452 , with right - handed strange quarks , can arise in the ope . because they are absent at order @xmath105 , their contribution will be suppressed to the negligible level of @xmath106 . besides , no new form factors will be introduced by these operators . * within the ope framework the effect of weak annihilation is a natural ingredient , which we have briefly discussed , mainly for illustration . since in addition to being a third - order power correction it comes with small wilson coefficients , its numerical impact of a few permille is entirely negligible . * we have clarified the relationship between the ope at high @xmath2 and qcd factorization at low @xmath2 by showing that both descriptions yield consistent results at intermediate values of @xmath30 . * a relevant topic at high @xmath2 is the issue of quark - hadron duality , which is closely related to the existence of an ope . we have defined the ope with the help of a spurion momentum that allows for an independent scaling of @xmath2 at fixed @xmath453 and kaon energy . this allows one to clarify the analytic structure of the matrix elements of the correlator ( [ khconcl ] ) . we then employed a model based on an infinite series of charm resonances to estimate quantitatively the amount of duality violation , resulting in about @xmath329 for the rate integrated over the high-@xmath2 region . an important aspect is the different sensitivity to duality violation of the contributions to the rate linear or quadratic in @xmath443 . the quadratic term is more vulnerable to duality violation , but numerically suppressed as a term of second order in the small charm - loop contribution . the systematics of duality have further been studied in a toy model for the rare decays , in which the factorization of the charm loop is exact . the main conclusion is that the high-@xmath2 region of @xmath1 is theoretically under excellent control . decay rates and distributions are perturbatively calculable up to the nonperturbative effects accounted for by the standard form factors in full qcd . further nonperturbative corrections are strongly suppressed and negligible within the accuracy of a few percent . the existence of the ope implies that at high @xmath2 the theory of @xmath1 has an even more solid basis than at low @xmath2 . an example for applications is the combined analysis of @xmath31 and @xmath454 @xcite . here the form factor dependence can be essentially eliminated , which leads to precision observables sensitive to new physics effects @xcite . beyond @xmath1 , the results of our analysis apply to the high-@xmath2 region of exclusive rare decays with similar final states such as @xmath24 , @xmath455 and , with appropriate modifications , to @xmath456 or @xmath457 . we show that the ope of @xmath76 in ( [ khdef ] ) can be expressed in terms of @xmath458 at the level of dimension-3 and dimension-4 operators in full qcd , in the chiral limit @xmath91 . for @xmath101 , @xmath459 , right - handed strange quarks can also appear , together with a suppression factor of @xmath103 . this leads to the two additional operators @xmath460 and @xmath461 in ( [ o41 ] ) and ( [ o42 ] ) , which can be counted as operators of dimension 4 . the matrix elements of @xmath462 , @xmath463 , @xmath460 and @xmath461 between @xmath111 and @xmath464 are all given by the standard @xmath428 form factors . together this implies that at leading ( dimension 3 ) and next - to - leading order ( dimension 4 ) in the ope , and to all orders in @xmath116 , standard form factors are the only hadronic matrix elements required . corrections to these terms arise only at second order ( @xmath465 ) in the ope through operators of dimension 5 . in other words , there are no first - order power corrections ( @xmath466 ) in the ope of @xmath76 , except for the purely kinematical dependence on @xmath2 and @xmath83 in the coefficient functions . these functions contain first - order terms such as @xmath467 , which however are calculable and do not introduce unknown hadronic matrix elements . we demonstrate the completeness of the basis @xmath462 , @xmath463 , @xmath460 and @xmath461 for operators of dimension 3 and 4 by enumerating all possibilities consistent with the relevant symmetries , and using the equations of motion . we first assume @xmath91 . operators of dimension 3 built from @xmath468 and @xmath25 have the form @xmath469 where lorentz indices can be contracted with the metric @xmath470 , the @xmath471-tensor and factors of @xmath66 to yield a 4-vector with index @xmath472 . useful relations are @xmath473 which holds for @xmath474 matrix elements , and @xmath475 or similar identities . exactly three structures can be formed : @xmath476 this can be seen as follows . for @xmath477 in ( [ opdim3 ] ) the only possibility is the third operator in ( [ opdim3b ] ) . for @xmath478 we obtain the first operator in ( [ opdim3b ] ) and @xmath479 which is equivalent to the third term in ( [ opdim3b ] ) . for @xmath480 we can write the second operator in ( [ opdim3b ] ) . in addition , we could form @xmath481 which leads back to the same structure . this exhausts the possibilities and proves the completeness of the basis in ( [ opdim3b ] ) at the dimension-3 level . we next consider operators of dimension 4 . these have , in addition to the fields in ( [ opdim3 ] ) , one covariant derivative acting on the strange quark . their general form is @xmath482 because of @xmath483 terms with extra covariant derivatives acting on @xmath25 do not lead to independent dimension-4 operators and may be disregarded . for @xmath477 in ( [ opdim4 ] ) we can write the operator @xmath484 . using the identity in ( [ sdbq ] ) this operator can be expressed in terms of the basis in ( [ opdim3b ] ) : @xmath485 the other possible structure can be reduced as @xmath486 for @xmath478 the dimension-4 operator has the form @xmath487 we list the possible contractions of ( [ sdgamb ] ) into a 4-vector together with the reduction to the basis in ( [ opdim3b ] ) , neglecting operators of dimension 5 . @xmath488 the r.h.s . of ( [ gammac ] ) is reduced to the basic operators via ( [ sdbop3 ] ) . the terms on the r.h.s . of ( [ gammad ] ) lead back to the expressions ( [ gammac ] ) and ( [ gammab ] ) . for @xmath480 the dimension-4 operator becomes @xmath489 we list again the possible contractions and their reduction to the operators in ( [ opdim3b ] ) . @xmath490 the r.h.s . of ( [ sigmaa ] ) is given by ( [ sdbop3 ] ) . the r.h.s . of ( [ sigmab ] ) is equivalent to ( [ gammab ] ) and ( [ oneqq ] ) , and ( [ sigmac ] ) reduces to ( [ sigmaa ] ) . this completes the proof that all operators of dimension 3 and 4 , ( [ opdim3 ] ) and ( [ opdim4 ] ) , can be expressed in terms of ( [ opdim3b ] ) for @xmath91 . taking current conservation into account leaves us with @xmath462 and @xmath463 in ( [ o31o32 ] ) . similar arguments hold if @xmath468 is replaced by @xmath491 . since right - handed strange quarks come with a factor @xmath492 , the two further operators @xmath460 and @xmath461 in ( [ o41 ] ) and ( [ o42 ] ) are obtained for @xmath101 . similarly to the derivations above it can be shown that dimension-5 operators of the form @xmath493 , contracted to a lorentz vector with index @xmath472 , can always be reduced to linear combinations of ( [ opdim3b ] ) and operators containing a factor of the gluon field strength @xmath147 . the former terms correspond to dimension-3 operators with a purely kinematic power suppression . genuine operators of dimension 5 therefore have the form @xmath494 , as quoted in ( [ o5n ] ) . in this appendix we collect input we have used in numerical calculations . the numbers quoted are our central values . the form factors have an uncertainty of roughly @xmath495 . the @xmath29 form factors are parametrized as @xcite @xmath496 @xmath497 the @xmath28 form factor is parametrized as @xcite @xmath498 with the default choice @xmath499 , @xmath500=1.6 and @xmath501 . further parameters are summarized in table [ tab : parinput ] . .[tab : parinput ] input parameters in @xmath18 . @xmath502 . [ cols="^,^,^,^",options="header " , ] we thank the galileo galilei institute for theoretical physics for hospitality and the infn for partial support during the initial phase of this project . g.b . thanks the cern theory division for support and hospitality while this paper was being completed . we thank alexander khodjamirian and thomas mannel for discussions and critical remarks . this work was supported in part by the dfg cluster of excellence ` origin and structure of the universe ' and by the dfg graduiertenkolleg gk 1054 . m. artuso _ et al . _ , eur . j. c * 57 * ( 2008 ) 309 [ arxiv:0801.1833 [ hep - ph ] ] . a. ishikawa _ et al . _ [ belle collaboration ] , phys . rev . * 91 * ( 2003 ) 261601 [ arxiv : hep - ex/0308044 ] . b. aubert _ et al . _ [ babar collaboration ] , phys . d * 73 * ( 2006 ) 092001 [ arxiv : hep - ex/0604007 ] . b. aubert _ et al . _ [ babar collaboration ] , phys . rev . d * 79 * ( 2009 ) 031102 [ arxiv:0804.4412 [ hep - ex ] ] . b. aubert _ et al . _ [ babar collaboration ] , phys . * 102 * ( 2009 ) 091803 [ arxiv:0807.4119 [ hep - ex ] ] . j. t. wei _ et al . _ [ belle collaboration ] , phys . * 103 * ( 2009 ) 171801 [ arxiv:0904.0770 [ hep - ex ] ] . e. pueschel [ cdf collaboration ] , arxiv:1005.5572 [ hep - ex ] ; t. aaltonen _ et al . _ [ cdf collaboration ] , arxiv:1101.1028 [ hep - ex ] . m. bona _ et al . _ , arxiv:0709.0451 [ hep - ex ] . t. kageyama , aip conf . * 842 * ( 2006 ) 1064 . t. browder _ et al . _ , jhep * 0802 * ( 2008 ) 110 [ arxiv:0710.3799 [ hep - ph ] ] . t. e. browder , t. gershon , d. pirjol , a. soni and j. zupan , arxiv:0802.3201 [ hep - ph ] . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , eur . j. c * 61 * ( 2009 ) 439 [ arxiv:0902.4446 [ hep - ph ] ] . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , phys . * 83 * ( 1999 ) 1914 [ arxiv : hep - ph/9905312 ] . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , nucl . phys . b * 591 * ( 2000 ) 313 [ arxiv : hep - ph/0006124 ] . m. beneke , t. feldmann and d. seidel , nucl . b * 612 * ( 2001 ) 25 [ arxiv : hep - ph/0106067 ] . a. khodjamirian , t. mannel , a. a. pivovarov and y. m. wang , jhep * 1009 * ( 2010 ) 089 [ arxiv:1006.4945 [ hep - ph ] ] . g. buchalla and g. isidori , nucl . b * 525 * ( 1998 ) 333 [ arxiv : hep - ph/9801456 ] . b. grinstein and d. pirjol , phys . d * 70 * ( 2004 ) 114005 [ arxiv : hep - ph/0404250 ] . g. buchalla , a. j. buras and m. e. lautenbacher , rev . * 68 * ( 1996 ) 1125 [ arxiv : hep - ph/9512380 ] . a. j. buras and m. mnz , phys . d * 52 * ( 1995 ) 186 [ arxiv : hep - ph/9501281 ] . m. misiak , nucl . b * 393 * ( 1993 ) 23 [ erratum - ibid . b * 439 * ( 1995 ) 461 ] . m. bartsch , m. beylich , g. buchalla and d. n. gao , jhep * 0911 * ( 2009 ) 011 [ arxiv:0909.1512 [ hep - ph ] ] ; g. buchalla , arxiv:1010.2674 [ hep - ph ] . g. buchalla , g. hiller and g. isidori , phys . d * 63 * ( 2001 ) 014015 [ arxiv : hep - ph/0006136 ] . g. buchalla , g. isidori and s. j. rey , nucl . b * 511 * ( 1998 ) 594 [ arxiv : hep - ph/9705253 ] . c. greub , v. pilipp and c. schpbach , jhep * 0812 * ( 2008 ) 040 [ arxiv:0810.4077 [ hep - ph ] ] . d. seidel , phys . d * 70 * ( 2004 ) 094038 [ arxiv : hep - ph/0403185 ] . m. beneke , t. feldmann and d. seidel , eur . j. c * 41 * ( 2005 ) 173 [ arxiv : hep - ph/0412400 ] . j. charles _ et al . _ , phys . d * 60 * ( 1999 ) 014001 [ arxiv : hep - ph/9812358 ] . m. beneke and t. feldmann , nucl . b * 592 * ( 2001 ) 3 [ arxiv : hep - ph/0008255 ] . t. feldmann and j. matias , jhep * 0301 * ( 2003 ) 074 [ arxiv : hep - ph/0212158 ] . b. blok , m. a. shifman and d. x. zhang , phys . d * 57 * ( 1998 ) 2691 [ erratum - ibid . d * 59 * ( 1999 ) 019901 ] [ arxiv : hep - ph/9709333 ] . m. a. shifman , arxiv : hep - ph/0009131 . m. shifman , _ in the proceedings of qcd @ work 2003 : 2nd international workshop on quantum chromodynamics : theory and experiment , conversano , italy , 14 - 18 june 2003 , pp 001_. o. cata , m. golterman and s. peris , phys . rev . d * 77 * ( 2008 ) 093006 [ arxiv:0803.0246 [ hep - ph ] ] . j. z. bai _ et al . _ [ bes collaboration ] , phys . * 88 * ( 2002 ) 101802 [ arxiv : hep - ex/0102003 ] . m. ablikim _ et al . _ , phys . * 97 * ( 2006 ) 262001 [ arxiv : hep - ex/0612054 ] . j. h. kuhn , m. steinhauser and c. sturm , nucl . b * 778 * ( 2007 ) 192 [ arxiv : hep - ph/0702103 ] . s. s. gershtein , a. k. likhoded and a. v. luchinsky , phys . d * 74 * ( 2006 ) 016002 [ arxiv : hep - ph/0602048 ] . b. chibisov , r. d. dikeman , m. a. shifman and n. uraltsev , int . . phys . a * 12 * ( 1997 ) 2075 [ arxiv : hep - ph/9605465 ] . f. krger and l. m. sehgal , phys . b * 380 * ( 1996 ) 199 [ arxiv : hep - ph/9603237 ] . c. bobeth , g. hiller and d. van dyk , jhep * 1007 * ( 2010 ) 098 [ arxiv:1006.5013 [ hep - ph ] ] . m. beneke , g. buchalla and i. dunietz , phys . d * 54 * ( 1996 ) 4419 [ arxiv : hep - ph/9605259 ] . m. beneke , g. buchalla , c. greub , a. lenz and u. nierste , phys . lett . b * 459 * ( 1999 ) 631 [ arxiv : hep - ph/9808385 ] . i. i. bigi , n. uraltsev and r. zwicky , eur . j. c * 50 * ( 2007 ) 539 [ arxiv : hep - ph/0511158 ] . c. breidenbach , t. feldmann , t. mannel and s. turczyk , phys . d * 78 * ( 2008 ) 014022 [ arxiv:0805.0971 [ hep - ph ] ] . i. bigi , t. mannel , s. turczyk and n. uraltsev , jhep * 1004 * ( 2010 ) 073 [ arxiv:0911.3322 [ hep - ph ] ] . t. hurth and d. wyler _ in _ j. l. . hewett _ et al . _ , arxiv : hep - ph/0503261 . p. ball and r. zwicky , phys . d * 71 * ( 2005 ) 014029 [ arxiv : hep - ph/0412079 ] . m. beneke and m. neubert , nucl . b * 675 * ( 2003 ) 333 [ arxiv : hep - ph/0308039 ] . p. ball , g. w. jones and r. zwicky , phys . d * 75 * ( 2007 ) 054004 [ arxiv : hep - ph/0612081 ] .

recently , there has been a lot of interest in cosmological solutions in the presence of a cosmological constant , when the latter is significant compared to the present total energy density of the universe . indeed , the hubble diagram based on observations of type ia supernovae up to a redshift @xmath1 seems to imply that our universe is presently accelerating @xcite . these data , when combined with the observed location of the first acoustic peak of the cmb temperature fluctuations , favor a spatially flat universe whose energy density is dominated by a `` cosmological constant''-like term . the flatness of the universe is corroborated by the latest boomerang and maxima data @xcite , in accordance with the inflationary paradigm , though a marginally closed universe is still allowed by the position of the first acoustic ( doppler ) peak at @xmath2 . a significant cosmological constant may help in resolving the dark matter problem for dustlike matter alone observations seem to imply @xmath3 and in reconciling flat cold dark matter ( cdm ) models with observations in the framework of @xmath4cdm models . finally , a cosmological constant is an elegant way to allow a high hubble constant @xmath5 with @xmath6 and a sufficiently old universe @xmath7gyr @xcite ( see also , e.g. , @xcite for a recent comprehensive review and references therein ) . therefore , this interpretation , if confirmed by future observations , constitutes a fundamental progress towards the solution of the dark matter problem and the formation of large - scale structure in the universe out of primordial fluctuations generated by some inflationary model . that is certainly what makes it so appealing and gives it , maybe somehow prematurely , the status of new paradigm . a striking consequence for our universe is then its present acceleration , for a large range of equations of state @xcite . of course , from the point of view of particle physics , a pure cosmological constant of the order of magnitude @xmath8 , interpreted as the vacuum energy , is extremely problematic . this is why attempts were made to find some alternative explanation to the origin of the acceleration under the form of some scalar field @xmath9 ( sometimes called quintessence @xcite , `` @xmath4''-field , etc . ) whose slowly varying energy density would mimic an effective cosmological constant . this is very reminiscent of the mechanism producing the inflationary phase itself with the fundamental difference that this scalar field , which does not have to be _ a priori _ the inflaton , is accelerating the expansion today , therefore at a much lower energy scale . this of course has problems of its own as this effective cosmological constant term started dominating the universe expansion only in the very recent past ( the so - called `` cosmic coincidence '' problem ) . indeed , the energy density of the field @xmath9 must remain subdominant at very early stages and come to dominate in the recent past only . hence , specific evolution properties are required to meet these constraints and were indeed shown to hold for particular potentials , partly alleviating the problem of the initial conditions . for inverse power - law potentials the energy density of the scalar field was shown to decrease less rapidly than the background energy density so that it can be negligible in the early universe and still come to dominate in the recent past @xcite . for exponential potentials , @xcite the scalar field energy density has the very interesting behavior that it tends to a fixed fraction of the total energy density , these are the so - called `` tracker solutions '' . hence a pure exponential potential is excluded if data confirm that the energy density of the scalar field is dominating today , as this fraction had to be small at the time of nucleosynthesis . a slightly different potential is proposed in @xcite and a classification of the scaling behavior of the scalar field for various potentials has been given in @xcite . hence , though a minimally coupled scalar field is an attractive possibility , some degree of fine tuning still remains in the parameters of the potential @xcite . if one admits that it is some minimally coupled scalar field which plays the role of an effective cosmological constant while gravity is described by general relativity , the question immediately arises : what is the `` right '' potential @xmath10 of this scalar field ? in a recent work by starobinsky @xcite , the following `` phenomenological '' point of view was adopted : instead of looking for more or less well motivated models , like the interesting possibilities discussed above , it is perhaps more desirable to extract as much information as possible from the observations ( a similar approach can also be adopted to reconstruct the inflaton potential ) in order to reconstruct the scalar field potential , if the latter exists at all . cosmological observations could then be used to constrain the particle physics model in which this scalar field is supposed to originate . in the context of general relativity plus a minimally coupled scalar field , it was shown that the reconstruction of @xmath10 can be implemented once the quantity @xmath11 , the luminosity distance as a function of redshift , is extracted from the observations @xcite , something that is expected in the near future . can already be reconstructed from present experimental data , although not yet very accurately . ] the snap ( supernovae acceleration probe ) satellite will notably make measurements with an accuracy at the percent level up to @xmath12 . of course , in this way only the recent past of our universe , up to redshifts @xmath13 ( for reference , we will push some of our simulations up to @xmath14 ) , is probed and so the reconstruction is made only for the corresponding part of the potential . crucial information is therefore gained on the microscopic lagrangian of the theory through relatively `` low '' redshift cosmological observations . a further step is to generalize the same mechanism in the framework of scalar - tensor theories of gravity , sometimes called `` generalized quintessence '' . the usual minimally coupled models are certainly ruled out if , for example , it turns out that this component of the energy density obeys an equation of state @xmath15 with @xmath16 ( @xmath17 ) . strangely enough , such an unexpected equation of state which in itself implies new physics , is in fair agreement with the observations @xcite . also the inequality @xmath18 must hold for a minimally coupled scalar field , hence its violation would force us to consider more complicated theories , possibly scalar - tensor theories . there are also strong theoretical motivations . these theories , in which the scalar field participates in the gravitational interaction , are the most natural alternatives to general relativity ( gr ) . indeed , scalar partners to the graviton generically arise in theoretical attempts at quantizing gravity or at unifying it with other interactions . for instance , in superstrings theory , a dilaton is already present in the supermultiplet of the 10-dimensional graviton , and several other scalar fields ( called the moduli ) also appear when performing a kaluza - klein dimensional reduction to our usual spacetime . moreover , contrary to other alternative theories of gravity , scalar - tensor theories respect most of gr s symmetries : conservation laws , constancy of ( non - gravitational ) constants , local lorentz invariance ( even if a subsystem is influenced by external masses ) , and they also have the capability of satisfying the weak equivalence principle ( universality of free fall of laboratory - size objects ) even for a strictly massless scalar field . nevertheless , they can describe many possible deviations from gr , and their predictions have been thoroughly studied in various situations : solar - system experiments @xcite , binary - pulsar tests @xcite , gravitational - wave detection @xcite . finally these scalar - tensor theories could play a crucial role in the very early universe , for example in the pre big bang inflationary model ( see e.g. @xcite ) . thus , in this work we are investigating the possibility to have an accelerating universe in the context of scalar - tensor theories of gravity instead of pure gr . this has indeed attracted a lot of interest recently and such cosmological models have been studied and possibly confronted with observations like cmb anisotropies , or the growth of energy density perturbations ( see for instance @xcite ) . however , we emphasize once more that the central point of view adopted here , in analogy with starobinsky @xcite , is to constrain the model with the _ experimental _ knowledge of the hubble diagram up to @xmath13 . this is precisely why use of the redshift @xmath19 as basic variable is crucial for our purpose : quantities like @xmath20 are directly observable , in contrast to , say , can be obtained from the knowledge of @xmath20 thanks to the relation @xmath21 $ ] , but the directly observable quantity is @xmath20 . ] @xmath22 or @xmath23 . for instance , we have access to @xmath20 through the direct measurement of the luminosity distance in function of redshift @xmath11 . in a recent letter @xcite , it was shown that the knowledge of both @xmath20 and @xmath24 is sufficient to reconstruct the full theory ( again , in the range probed by the data ) . this means that we do _ not _ choose any specific theory _ a priori _ , but instead we reconstruct whatever theory possibly realized in nature . as we will see , the knowledge of @xmath20 on its own , though insufficient in order to fully reconstruct a scalar - tensor theory unless one makes additional assumptions , turns out to be already very constraining when subclasses of models are considered . this is particularly interesting because it means that cosmological observations at low redshifts implying an accelerated expansion might well give new constraints on scalar - tensor theories . we will show that this is indeed the case . throughout the paper , we use natural units for which @xmath25 , and the signature @xmath26 , together with the sign conventions of @xcite . in section ii , we introduce the general formalism of scalar - tensor theories of gravity and their different parametrizations . in section iii , we briefly review the severe experimental restrictions imposed on these theories _ today_. in section iv , we consider frw universes in the framework of scalar - tensor gravity and we give the equations for the different parameterizations . in section v , we review the full reconstruction problem . in section vi , we give a detailed study of subclasses of models , which are investigated using the background equations . finally , in section vii , our results are summarized and discussed . we are interested in a universe where gravity is described by a scalar - tensor theory , and we consider the action @xcite @xmath27\ . \label{s_jf}\ ] ] here , @xmath28 denotes the bare gravitational coupling constant ( which differs from the measured one , see eq . ( [ geff ] ) below ) , @xmath29 is the scalar curvature of @xmath30 , and @xmath31 its determinant . in ref . @xcite , we used different conventions ( corresponding to the choice @xmath32 in the above action ) ; here , the quantity @xmath33 is dimensionless . this factor @xmath33 needs to be positive for the gravitons to carry positive energy . the action of matter @xmath34 is a functional of some matter fields @xmath35 and of the metric @xmath30 , but it does not involve the scalar field @xmath9 . this ensures that the weak equivalence principle is exactly satisfied . the dynamics of the real scalar field @xmath9 depends _ a priori _ on three functions : @xmath33 , @xmath36 , and the potential @xmath10 . however , one can always simplify @xmath36 by a redefinition of the scalar field , so that @xmath33 and @xmath36 can be reduced to only one unknown function . two natural parametrizations are used in the literature : ( i ) the brans - dicke one , corresponding to @xmath37 and @xmath38 ; and ( ii ) the simple choice @xmath39 and @xmath33 arbitrary . this second parametrization is however sometimes pathological . [ the derivatives of @xmath9 can become imaginary in perfectly regular situations ; see the discussion about eq . ( [ positiveenergy ] ) below . ] in the following , we will write the field equations in terms of the two functions @xmath33 and @xmath36 , so that any particular choice can be recovered easily . the variation of action ( [ s_jf ] ) gives straightforwardly @xmath40 [ 2.2 ] where @xmath41 is the trace of the matter energy - momentum tensor @xmath42 . the scalar - field equation ( [ boxphi ] ) can of course be rewritten differently if one uses the trace of eq . ( [ einstein ] ) to replace the curvature scalar @xmath29 by its source , and one gets the brans - dicke - like equation @xmath43 where @xmath44 . [ in the brans - dicke representation where @xmath45 and @xmath46 , this factor @xmath47 reduces to the well - known expression @xmath48 . ] in the following , we will however use the form ( [ boxphi ] ) , which will simplify considerably our calculations . the above equations are written in the so - called jordan frame ( jf ) . since in action ( [ s_jf ] ) , matter is universally coupled to @xmath30 , this `` jordan metric '' defines the lengths and times actually measured by laboratory rods and clocks ( which are made of matter ) . all experimental data will thus have their usual interpretation in this frame . in particular , the observed hubble parameter @xmath49 and the measured redshifts @xmath19 of distant objects are jordan - frame quantities . however , it is usually much clearer to analyze the equations and the mathematical consistency of the solutions in the so - called einstein frame ( ef ) , defined by diagonalizing the kinetic terms of the graviton and the scalar field . this is achieved thanks to a conformal transformation of the metric and a redefinition of the scalar field . let us call @xmath50 and @xmath51 the new variables , and define @xmath52 [ 2.4 ] action ( [ s_jf ] ) then takes the form @xmath53\ , \label{s_ef}\ ] ] where @xmath54 is the determinant of @xmath50 , @xmath55 its inverse , and @xmath56 its scalar curvature . note that the first term looks like the action of general relativity , but that matter is now explicitly coupled to the scalar field @xmath51 through the conformal factor @xmath57 . quantities referring to the einstein frame will always have an asterisk ( either in superscript or in subscript ) , e.g. @xmath58 and @xmath59 for the covariant derivative and the dalembertian with respect to the einstein metric . the indices of einstein - frame tensors will also be lowered and raised with the einstein metric @xmath50 and its inverse @xmath55 . the field equations deriving from action ( [ s_ef ] ) take the simple form @xmath60 [ 2.6 ] where @xmath61 is the coupling strength of the scalar field to matter sources @xcite , and @xmath62 is the trace of the matter energy - momentum tensor @xmath63 in einstein - frame units . from its definition , one can deduce the relation @xmath64 with its jordan - frame counterpart . let us underline that the cauchy problem is well posed in the einstein frame @xcite , because all the second - order derivatives of the fields are separated in the left - hand sides of eqs . ( [ 2.6 ] ) , whereas they are mixed in the jf equations ( [ 2.2 ] ) . action ( [ s_ef ] ) also shows that the helicity-2 degree of freedom is described by the fluctuations of the einstein metric @xmath50 ( whose kinetic term is the standard einstein - hilbert one ) , and that the ef scalar @xmath51 is the true helicity-0 degree of freedom of the theory ( since its kinetic term has the standard form ) . on the other hand , the fluctuations of the jordan metric @xmath30 actually describe a _ mixing _ of helicity-2 and helicity-0 excitations , and the jf scalar @xmath9 is related to the helicity-0 degree of freedom via the complicated relation ( [ varphi ] ) , because its kinetic term in action ( [ s_jf ] ) comes not only from the naive contribution @xmath65 but also from the cross term @xmath66 . in conclusion , the theory can be mathematically well defined only if it is possible to write the ef action ( [ s_ef ] ) , notably with its negative sign for the scalar - field kinetic term ( so that @xmath51 carries positive energy ) . if it happens that the transformation ( [ 2.4 ] ) is singular for particular values of @xmath9 , the consistency of the theory should be analyzed in the ef . some singularities may be artifacts of the parametrization which is chosen to write action ( [ s_jf ] ) , and may not have any physical significance . on the other hand , jordan - frame quantities may look sometimes regular while there is an actual singularity in the einstein frame ( a typical example is provided when @xmath33 vanishes ) . in this case , the solution should be considered as mathematically inconsistent . in the following , we will see that the jf is better suited than the ef for our cosmological study , but we will always check the consistency of our results by finally translating them in terms of einstein - frame quantities . the predictions of general relativity in weak - field conditions , and at present , are confirmed by solar - system experiments at the @xmath67 level @xcite . one should therefore verify that the scalar - tensor models we are considering are presently close enough to einstein s theory . if the scalar field is very massive ( say , if @xmath68 is large with respect to the inverse of the astronomical unit ) , its influence is exponentially small in solar - system experiments , even if it is strongly coupled to matter . this situation corresponds to the particular scalar - tensor model considered in ref . @xcite ( namely @xmath37 and @xmath69 in action ( [ s_jf ] ) , but assuming a large enough value for @xmath70 ) . although this situation is phenomenologically acceptable , it remains somewhat problematic from a field theoretical viewpoint , since the massive scalar would _ a priori _ desintegrate into lighter ( matter ) particles . on the contrary , if the scalar mass is small with respect to the inverse solar - system distances , it must be presently very weakly coupled to matter for the theory to be consistent with experimental data . at the first post - newtonian order ( @xmath71 with respect to the newtonian interaction ) , the deviations from general relativity can be parametrized by two real numbers , that eddington @xcite denoted as @xmath72 and @xmath73 . in the present framework , they take the form @xcite @xmath74 [ 3.1 ] where the first expressions are given in terms of the einstein - frame notation ( [ s_ef])-([alpha ] ) , whereas the last ones correspond to the jordan - frame general representation ( [ s_jf ] ) . to simplify , the second expression of eq . ( [ betappn ] ) has been written in terms of the derivative of ( [ gammappn ] ) with respect to @xmath9 . using the upper bounds on @xmath73 from solar - system measurements @xcite , we thus get the constraint @xmath75 where an index @xmath76 means the present value of the corresponding quantity . on the other hand , the experimental bounds on @xmath72 can not be used to constrain the derivative @xmath77 appearing in eq . ( [ betappn ] ) , since it is multiplied by a factor @xmath78 consistent with 0 . because of nonperturbative strong - field effects , binary - pulsar tests are however directly sensitive to this derivative , _ i.e. _ , to the ratio @xmath79 . in a generic class of scalar - tensor models , refs . @xcite have obtained the bound @xmath80 from action eq . ( [ s_jf ] ) , one can naively define newton s gravitational constant as the inverse factor of the curvature scalar @xmath29 : @xmath81 however , @xmath82 does not have the same physical meaning as newton s gravitational constant in gr . indeed , the actual newtonian force measured ( in cavendish - type experiments ) between two close test masses @xmath83 and @xmath84 is of the form @xmath85 , where the effective gravitational constant reads @xcite @xmath86 the contribution @xmath87 is due to the exchange of a graviton between the two bodies , whereas @xmath88 comes from the exchange of a scalar particle between them . of course , when the distance between the bodies becomes larger than the inverse mass of the scalar field , its influence becomes negligible and one gets @xmath89 . note that as usual , the last expression in eq . ( [ geff ] ) , in terms of jordan - frame notation , is much more complicated than its einstein - frame counterpart . in the particular brans - dicke representation , @xmath45 and @xmath46 , it however reduces to the simpler ( and well - known ) form @xmath90 . the experimental bound ( [ df0 ] ) shows that the present values of @xmath91 and @xmath82 differ by less than @xmath92 . however , they can _ a priori _ differ significantly in the past . it should be noted that the experimental limit on the time variation of the gravitational constant , @xmath93 @xcite , does _ not _ imply any constraint on @xmath94 . indeed , @xmath91 can be almost constant even if @xmath95 ( or @xmath96 ) varies significantly . a simple example is provided by barker s theory @xcite , in which @xmath97 : one gets @xmath98 , which is strictly constant independently of the time variations of @xmath99 . nevertheless , as pointed out in @xcite , under reasonable cosmological assumptions , one can derive @xmath100 with @xmath101 accuracy up to redshifts @xmath102 . the equations derived in this section generalize those of our previous paper @xcite in several ways . first , we use the most general representation ( [ s_jf ] ) of the theory , instead of the simpler choice @xmath103 that was made in @xcite . second , we take into account a possible spatial curvature of the universe , which will be an interesting possibility in our studies of sec . third , we write the equations for an arbitrary pressure of the perfect fluid describing matter in the universe . this will not be useful for our reconstruction program of the following sections , as matter can be assumed to be simply dustlike for the redshifts @xmath104 that we will consider , but these general equations may be interesting for further cosmological studies of earlier epochs of the universe . finally , we comment on the einstein - frame version of these equations , which are mathematically simpler , but actually more difficult to use for our purpose . we consider a friedmann - robertson - walker ( frw ) universe whose background metric in the jordan frame is given by @xmath105 [ 4.1 ] where @xmath106 , @xmath76 , or @xmath107 for spatially open , flat , or closed universes respectively . the scalar field @xmath9 ( or @xmath51 , in the ef ) is also assumed to depend only on time . since the relation between the ef and jf is given by @xmath108 , see eqs . ( [ 2.4 ] ) , our universe is still of the frw type in the ef , with @xmath109 and @xmath110 in the following , matter will be described by a perfect fluid , and we will write its energy - momentum tensor as @xmath111 where @xmath112 and @xmath113 are the spacetime components of the four - dimensional unit velocity of matter , in jf and ef units respectively . as we are interested in a frw background , the spatial components @xmath114 and @xmath115 ( @xmath116 ) all vanish . from ( [ t ] ) , we deduce the relation between the matter density and pressure in both frames : @xmath117 the background equations in the jf follow from ( [ einstein])([matter ] ) , and read @xmath118 [ 4.5 ] where @xmath119 , and a dot denotes differentiation with respect to the jordan - frame time @xmath120 . as usual , if @xmath121 , eq . ( [ dmatter ] ) is trivially integrated as @xmath122 ( and in particular @xmath123 for dustlike matter ) . equation ( [ ddotphi ] ) is actually a consequence of the other three , and we will not need it in the following . since these equations correspond to the most general parametrization ( [ s_jf ] ) of scalar - tensor theories , many particular cases are easily recovered . for instance , the case of a minimally coupled scalar field @xcite is obtained for constant values of @xmath96 and @xmath124 ( say , @xmath125 and @xmath126 ) , and the particular model considered in @xcite is recovered immediately for @xmath45 and @xmath127 . the corresponding background equations in the ef are very similar to those in general relativity . they follow from eq . ( [ einstein * ] ) , and read @xmath128 [ 4.6 ] where @xmath129 is the einstein - frame hubble parameter . it is obvious from ( [ acc ] ) that a vanishing potential @xmath130 implies @xmath131 , so that the universe is decelerating in the einstein frame . however , because of the relation @xmath132 , see eq . ( [ avsa * ] ) , the _ observed _ ( jordan - frame ) expansion rate @xmath133 may be positive even in this case , and we will see concrete examples in sec . vi.a below . this is an important point to remember : although we are looking for cosmological frw backgrounds whose expansion is accelerating , the sign of @xmath134 is _ a priori _ not fixed . the scalar - field equation of motion in the ef follows from eq . ( [ boxvarphi ] ) , and reads @xmath135 it is also similar to the usual klein - gordon equation , with the notable difference of a source term on the right - hand side , with the coupling strength @xmath136 defined in eq . ( [ alpha ] ) above . it is tempting to tackle our problem in the ef as the equations are simpler and we can rely on experience gained in general relativity . however , a crucial difficulty that we encounter is that all physical quantities which appear in the ef background equations are not those that come from observations . moreover , the behavior of matter in the ef is complicated by the relations ( [ rho*vsrho ] ) : instead of the simple power law @xmath137 for dustlike matter in the jf , one gets @xmath138 in the ef , where @xmath139 can have _ a priori _ any shape . to avoid these problems , we will thus work in the jf , and show that the `` reconstruction '' program can equally well be implemented , like in general relativity , although it is mathematically very different . we will nevertheless check at the end the consistency of the solutions obtained by translating them in terms of ef quantities . we now consider the perturbations in the longitudinal gauge . for this problem , we will restrict our discussion to the case of a spatially flat frw universe ( @xmath140 ) , and write the jf and ef metrics as @xmath141 [ 4.8 ] in the ef , the perturbation equations deriving from eq . ( [ einstein * ] ) are strictly the same as in general relativity plus a minimally coupled scalar field . one thus finds notably @xmath142 . on the other hand , the equations for scalar - field and matter perturbations are modified by the matter - scalar coupling , proportional to @xmath136 in eqs . ( [ boxvarphi ] ) and ( [ matter * ] ) . for our purpose , it will be more useful to write the perturbation equations in the ( physical ) jf . let us define the gauge invariant quantity introduced in @xcite : @xmath143 . ] @xmath144 where @xmath145 is the matter peculiar velocity potential ( such that @xmath146 is the perturbation of the four - dimensional unit velocity @xmath147 ) . we now work in fourier space , and assume a spatial dependence @xmath148 , with @xmath149 . the conservation equations of matter ( [ matter ] ) give @xmath150 [ 4.10 ] on the other hand , the einstein equations ( [ einstein ] ) give @xmath151 [ 4.11 ] note that @xmath152 in the jf , in contrast to the corresponding problem in general relativity or in the ef . equation ( [ psi ] ) is actually an obvious consequence of the relation between @xmath50 and @xmath30 , eq . ( [ g * ] ) , and of the fact that @xmath153 . finally , eq . ( [ boxphi ] ) yields the equation for the dilaton fluctuations @xmath154 : @xmath155\delta\phi = \nonumber\\ & & = \left[{k^2 \over a^2}(\phi-2\psi ) - 3(\ddot\psi+4h\dot\psi+h\dot\phi)\right]{1\over z}\,{df\over d\phi } + ( 3\dot\psi+\dot\phi)\dot\phi - 2\,{\phi\over z}\,{du\over d\phi}\ . \label{fluctphi}\end{aligned}\ ] ] in the particular representation @xmath103 used in ref . @xcite , this equation reduces to the simpler form @xmath156\delta\phi = \nonumber\\ & = & \left[{k^2 \over a^2}(\phi-2\psi ) - 3(\ddot\psi+4h\dot\psi+h\dot\phi)\right]{df\over d\phi } + ( 3\dot\psi+\dot\phi)\dot\phi - 2\phi\,{du\over d\phi}\ . \label{fluctphiz1}\end{aligned}\ ] ] the reconstruction of the potential @xmath10 was shown in @xcite to be possible in the framework of general relativity plus a minimally coupled scalar field , the @xmath4-field or quintessence , provided the hubble diagram ( and thus also @xmath20 ) can be extracted from the observations . an essential difference arises when one deals with scalar - tensor theories : we have to reconstruct two unknown functions instead of one , hence we need to extract two quantities ( as functions of the redshift @xmath157 ) from the observations . actually , in the minimally coupled case , the knowledge of the luminosity distance @xmath158 and of the clustering of matter @xmath159 , both in function of @xmath19 , provides two independent ways to reconstruct the scalar field potential @xcite . without any ambiguity in the minimally coupled case , one needs to know both @xmath11 and the present energy density of dustlike matter @xmath160 , or both @xmath24 and the present value of the hubble constant @xmath5 . in our general scalar - tensor case , we need to know the two functions @xmath11 and @xmath24 , but no independent measurement of @xmath160 or @xmath5 is necessary . ] in our case , both quantities are necessary and the reconstruction itself is significantly more complicated . the present section generalizes our previous results of ref . @xcite not only by considering the most general parametrization ( [ s_jf ] ) of scalar - tensor theories and by taking into account the possible spatial curvature of the universe , but also by discussing particular cases that were excluded in this reference . from now on , we will restrict our discussion to the case of a pressureless perfect fluid ( @xmath161 ) , because all matter in the universe will be assumed to be simply dustlike , of course besides that part needed to account for the present accelerated expansion ( _ i.e. _ , the scalar field in the present framework ) . the first step of the reconstruction program is the same as in general relativity , since it is purely kinematical and does not depend on the field content of the theory : if the luminosity distance @xmath158 is experimentally determined as a function of the redshift @xmath19 , one can deduce the quantity @xmath20 from the relation @xmath162^{-1/2 } , \label{hvsdl}\ ] ] where the prime denotes the derivative with respect to @xmath19 . the large square brackets contain a corrective factor involving the present energy contribution @xmath163 of the spatial curvature of the universe . it was not written explicitly in refs . @xcite , which focused their discussions on the flat - space case ( @xmath164 ) , but it is a straightforward consequence of eqs . ( 23)(25 ) of ref . since present experimental data suggest that @xmath165 is small , the flat - space expression for @xmath166'$ ] is _ a priori _ a good approximation anyway . note that even if one uses the exact equation ( [ hvsdl ] ) , it reduces to the flat - space expression for @xmath167 ( because @xmath168 ) , and therefore @xmath5 is always known without any ambiguity . to determine @xmath20 precisely at higher @xmath19 , one then needs to know both @xmath11 and @xmath169 . by eliminating @xmath170 from the background equations ( [ h2 ] ) and ( [ doth ] ) , we then obtain the equation @xmath171 which , when rewritten in terms of the redshift @xmath19 , gives the fundamental equation @xmath172~f ' + \left[\frac{6}{(1+z)^2 } - \frac{2}{1+z}(\ln h)'-4\left({h_0\over h}\right)^2\omega_{\kappa,0}\right]~f = & & \nonumber\\ = \frac{2 u}{(1+z)^2 h^2 } + 3 ( 1+z ) \left({h_0\over h}\right)^2 f_0 \omega_{m,0}\ . & & \label{f}\end{aligned}\ ] ] as before , an index @xmath76 means the present value of the corresponding quantity , and we use again the notation @xmath173 . in this equation , @xmath174 stands for the present energy density of dustlike matter relative to the critical density @xmath175 . to simplify , this critical density is defined in terms of the present value of newton s gravitational constant ( [ g_n ] ) , @xmath176 , instead of the effective gravitational constant ( [ geff ] ) actually measured in cavendish - type experiments . indeed , solar - system experiments tell us that their present values differ by less than @xmath92 , as discussed in sec . [ note in passing that by changing the value of @xmath28 , one can always set @xmath177 without loss of generality . ] in conclusion , we are left with a non - homogeneous second order differential equation for the function @xmath178 , a situation very different from that prevailing in general relativity . however , the right - hand side also depends on the unknown potential @xmath179 , so that this equation does not suffice to fully reconstruct the microscopic lagrangian of the theory . as we will show in sec . vi below , it can nevertheless be used for a systematic study of several scalar - tensor models , provided one of the two unknown functions is given ( or a functional dependence between them is assumed ) . this can be useful as we do not expect a simultaneous release of data yielding @xmath20 and @xmath24 . we will see that such a study already yields powerful constraints on the family of theories which are viable . on the other hand , if @xmath24 is also experimentally determined , and if we assume a spatially flat frw universe ( @xmath180 ) , we will see in the next subsection ( v.b ) that the value of @xmath160 as well as the function @xmath178 can be obtained independently of @xmath179 . equation ( [ f ] ) then gives @xmath179 in an _ algebraic _ way from our knowledge of @xmath20 , @xmath178 and @xmath160 . let us now assume that both @xmath178 and @xmath179 are known , either because one of them was given from theoretical naturalness assumptions , or because @xmath24 has been experimentally determined with sufficient accuracy . we will also assume that both @xmath160 and @xmath169 are known . it is then straightforward to reconstruct the various functions of @xmath9 entering the microscopic lagrangian ( [ s_jf ] ) . in the brans - dicke representation , one has @xmath45 , therefore the knowledge of @xmath178 and @xmath179 suffices to reconstruct the potential @xmath10 in a parametric way . however , to fully determine the theory , one also needs to know @xmath181 , or equivalently an equation giving the @xmath19-dependence of @xmath124 . on the other hand , in the simpler representation @xmath182 and @xmath33 unknown , we need an equation giving the @xmath19-dependence of @xmath9 to reconstruct @xmath33 and @xmath10 parametrically . these two cases , as well as any other possible parametrization of the theory , are solved thanks to eq . ( [ doth ] ) above , which reads in function of the redshift @xmath183~f ' + 2\left[\frac{(\ln h)'}{1+z } - \left({h_0\over h}\right)^2\omega_{\kappa,0}\right]~f \nonumber\\ & & - 3(1+z)\left({h_0\over h}\right)^2 f_0 \omega_{m,0}\ , \label{phi}\end{aligned}\ ] ] or equivalently @xmath184 in the @xmath182 representation , @xmath185 is thus obtained by a simple integration . in the brans - dicke representation , on the other hand , @xmath186 is given by an _ algebraic _ equation in terms of @xmath20 , @xmath187 , and their derivatives . it is rather obvious but anyway important to note that if the microscopic lagrangian ( [ s_jf ] ) can be reconstructed in the jf , it can also be directly obtained in the ef , eq . ( [ s_ef ] ) . this allows us to check the mathematical consistency of the theory , and notably if the helicity-0 degree of freedom @xmath51 always carries positive energy . one can also verify that the function @xmath188 defining the coupling of matter to the scalar field is well defined , and notably single valued . finally , the second derivative of the potential @xmath130 also gives us the sign of the square of the scalar mass , and negative values would strongly indicate an instability of the model . these important features can not easily be checked in the jf , because the sign of @xmath36 in eq . ( [ s_jf ] ) is not directly related to the positivity of the scalar - field energy ( see below ) , and also because the second derivative of @xmath10 does not give the precise value of its squared mass . [ as shown by eq . ( [ v ] ) , the helicity-0 degree of freedom @xmath51 may have a mass , @xmath189 , even if @xmath10 is strictly constant , provided @xmath33 varies . ] let us thus assume that @xmath20 , @xmath160 and @xmath169 are known , and that @xmath178 and @xmath179 were reconstructed as above . equation ( [ a ] ) then gives @xmath190 , _ i.e. _ , the einstein - frame coupling factor @xmath95 as a function of the _ jordan - frame _ redshift @xmath19 ( which is the redshift we observe ) . combining now eq . ( [ varphi ] ) with ( [ phi ] ) , we get @xmath191{f'\over f } + { ( \ln h ) ' \over 1+z}\nonumber\\ & & -\left({h_0\over h}\right)^2\omega_{\kappa,0 } -{3\over 2}(1+z)\left({h_0\over h}\right)^2 { f_0\over f}\ , \omega_{m,0}\ , \label{varphiprime}\end{aligned}\ ] ] [ 5.5 ] or also , not eliminating the potential @xmath192 @xmath193 the ef scalar @xmath51 is thus also known as a function of the _ jordan - frame _ redshift @xmath19 ( up to an additive constant @xmath194 which can be chosen to vanish without loss of generality ) , and one can reconstruct @xmath188 in a parametric way . similarly , the ef potential @xmath130 , eq . ( [ v ] ) , can be reconstructed from our knowledge of @xmath178 , @xmath179 and @xmath195 . since @xmath51 describes the actual helicity-0 degree of freedom of the theory , this field must carry only positive energy excitations , and @xmath196 must be positive . on the other hand , the tensor and scalar degrees of freedom are mixed in the jf , and the positivity of energy does not imply that @xmath197 should always be positive . actually , eq . ( [ positiveenergy ] ) shows that it can become negative when @xmath198 happens to be larger than @xmath199 , which can occur in perfectly regular situations . [ we will see an explicit example in sec . vi.a below . ] this underlines that the parametrization @xmath182 can sometimes be singular : the derivatives of @xmath9 may become purely imaginary although the scalar degree of freedom @xmath51 is well defined . on the other hand , the brans - dicke representation is well behaved ( @xmath200 remains always positive ) , and the positivity of energy simply implies the well - known inequality @xmath201 . actually , the particular value @xmath202 is also singular , as it corresponds to an infinite coupling strength @xmath203 between matter and the helicity-0 degree of freedom @xmath51 . the domain for which the @xmath103 parametrization is pathological although the theory remains consistent simply corresponds to @xmath204 , or @xmath205 . although the perturbations will not be used in sec . vi below , we emphasize that the phenomenological reconstruction of the full microscopic lagrangian can be implemented without any ambiguity if fluctuations are taken into account . for completeness , we review now this part of our program . we assume that both @xmath20 and the matter density perturbation @xmath24 are experimentally determined with enough accuracy , and as in sec . iv.b above , we focus our discussion on the case of a spatially flat frw universe ( @xmath180 ) . we also assume that matter is dustlike ( @xmath206 ) , and the perturbation equations of sec . iv.b are thus simplified . in particular , eq . ( [ conserv_z ] ) reduces to the mere identity @xmath207 . we consider comoving wavelengths @xmath208 much shorter ( for recent times ) than the hubble radius @xmath209 , and also shorter than the inverse mass of the scalar field : @xmath210 two different reasonings can now be used to reach the same conclusions . the first one , explained in ref . @xcite , consists in taking the formal limit @xmath211 in the various perturbation equations . then , the leading terms are either those containing @xmath159 or those multiplied by the large factor @xmath212 . one also needs to consider only the growing adiabatic mode of eq . ( [ fluctphi ] ) , for which @xmath213 . the other reasoning needs a simpler ( but _ a priori _ stronger ) hypothesis . one assumes that the logarithmic time derivative of any quantity , say @xmath214 , is at most of order @xmath49 : @xmath215 suffices to derive straightforwardly all the following approximations . note that both reasonings correspond in fact to the same physical situation of a weakly - coupled light scalar field . in the case of a strongly - coupled but very massive scalar ( see the second paragraph of sec . iii ) , the equations can not be approximated as shown below , and the time evolution of density fluctuations does not follow the same law . for instance , in the particular model considered in ref . @xcite , one always finds a strong clustering of the scalar field at small scales . indeed , this model corresponds to the choice @xmath45 and @xmath127 in action ( [ s_jf ] ) , and eq . ( [ fluctphi ] ) can then be rewritten as @xmath216 . therefore , even if the scalar field is very massive ( @xmath70 large ) , one finds that it is anyway strongly clustered for comoving wavelengths @xmath217 shorter than the inverse mass , _ i.e. _ , in the formal limit @xmath211 . although this is _ a priori _ not forbidden by observations of gravitational clustering , since the inverse mass must be much smaller than the astronomical unit in this model , this is anyway an indication of its probable instability . we will not consider such heavy scalar fields any longer in this paper , and we now come back to the class of weakly - coupled light - scalar models , which are the most natural alternatives to general relativity . setting @xmath218 and making use of ( [ conserv_z ] ) , one can write ( [ conserv_0 ] ) as @xmath219 where the right - hand side is negligible with respect to each separate term of the left - hand side because of the above hypotheses . note that ( [ ddotdeltam ] ) just reproduces the standard evolution equation for matter perturbations . using ( [ fluctphiz1 ] ) , we also arrive at @xmath220 where the second equality is a consequence of eq . ( [ psi ] ) . in the case of gr plus a minimally coupled scalar field , one finds that @xmath221 in the limit @xmath211 , so that the scalar field is not gravitationally clustered at small scales @xcite . this is in agreement with the observational fact that the dark matter described by the @xmath4-term should remain unclustered up to comoving scales @xmath222 mpc ( where we recall that @xmath223 ) . on the other hand , in our scalar - tensor framework , ( [ dphi ] ) shows that the scalar field _ is _ clustered at arbitrarily small scales , but only weakly because the derivative @xmath224 is experimentally known to be small [ see the solar - system constraint ( [ df0 ] ) , and the limit @xmath225 justified in @xcite for redshifts @xmath226 . the class of models we are considering , involving a light scalar field weakly coupled to matter , is thus also in agreement with observations of gravitational clustering . finally , still under the above hypotheses , eq . ( [ fluctmetr ] ) implies @xmath227 remembering the definition ( [ geff ] ) for @xmath91 , and using ( [ dphi ] ) above , eq . ( [ poi ] ) can be recast in a form which exhibits its physical content : @xmath228 poisson s equation is thus simply modified by the substitution of newton s constant @xmath229 by @xmath91 , the effective gravitational constant between two close test masses ! this conclusion was also reached in @xcite , but only for brans - dicke theory with a constant parameter @xmath230 , while we have derived it for an arbitrary ( light ) scalar - tensor theory . as discussed in sec . iii above , expression ( [ geff ] ) is valid only if the distance between the test masses is negligible with respect to the inverse scalar mass . the physical reason why this expression appears in poisson s equation ( [ poi2 ] ) is just that we are working in the short wavelength limit ( [ largek ] ) : the frequency of the waves we are considering is so large that the scalar field behaves as if it were massless . combining ( [ ddotdeltam ] ) with ( [ poi2 ] ) , we now arrive at our final evolution equation for @xmath159 : @xmath231 in terms of the redshift @xmath19 , this reads @xmath232 provided we can extract from observation both physical quantities @xmath20 and @xmath24 with sufficient accuracy , the explicit reconstruction of the microscopic lagrangian is obtained in the following way . starting from ( [ delz ] ) and using the fact that _ today _ @xmath233 and @xmath234 differ by less than @xmath92 , eq . ( [ delz ] ) evaluated at present gives us the cosmological parameter @xmath160 with the same accuracy . then , returning to eq . ( [ delz ] ) for arbitrary @xmath19 , we get @xmath235 , where @xmath236 is a known function of the observables @xmath20 , @xmath24 , and their derivatives . using now eq . ( [ phi ] ) and expression ( [ geff ] ) for @xmath237 , we get a nonlinear second order differential equation for @xmath178 , which can be solved for given @xmath238 and @xmath239 [ one can always set @xmath177 without loss of generality , while @xmath239 is constrained by eq . ( [ df0 ] ) ] . after we have found @xmath178 , we can plug it into ( [ f ] ) to determine @xmath179 in an algebraic way . the final step is explained in the previous subsection , above eq . ( [ phi ] ) , for the various possible parametrizations of action ( [ s_jf ] ) : in the @xmath103 parametrization , @xmath185 is obtained by a simple integration of eq . ( [ phi ] ) , while in the brans - dicke parametrization ( @xmath37 ) , @xmath186 is given algebraically by the same eq . ( [ phi ] ) . this enables us to reconstruct @xmath33 ( or @xmath240 ) and @xmath10 as functions of @xmath241 for that range corresponding to the data . actually , for sufficiently low redshifts @xmath242 , eq . ( [ delz ] ) can be simplified without losing too much accuracy . indeed , as shown in ref . @xcite , the square of the matter - scalar coupling strength @xmath243 , eq . ( [ alpha ] ) , is at most of order @xmath244 for such redshifts . moreover , under natural assumptions , much smaller values of @xmath245 are generically predicted in scalar - tensor theories @xcite . therefore , @xmath91 and @xmath82 differ by less than @xmath101 for redshifts @xmath242 , and eq . ( [ delz ] ) can be used to obtain @xmath246 with the same accuracy . the interest of this simplification is that @xmath178 is now given by an _ algebraic _ equation . in the brans - dicke representation , _ all _ the steps of the reconstruction program are thus algebraic , eq . ( [ f ] ) giving @xmath179 , and eq . ( [ phi ] ) giving @xmath186 . the only non - algebraic step is the final parametric reconstruction of @xmath10 and @xmath240 . let us end this section by a few comments on the observational accuracy which will be needed for this reconstruction program to be implemented . first , eq . ( [ delz ] ) allows to reconstruct @xmath178 only if @xmath247 and @xmath248 are both determined with enough accuracy . moreover , the second derivative of this reconstructed @xmath178 is needed in eq . ( [ f ] ) to obtain @xmath179 . therefore , the actual reconstruction of the potential depends _ a priori _ on the fourth derivative of @xmath24 , so that extremely clean data seem to be necessary . however , the situation is better than this naive derivative counting suggests . indeed , the above estimates for @xmath245 show that @xmath178 does not vary much on the redshift interval @xmath249 . therefore , the first two terms of eq . ( [ f ] ) , involving @xmath250 and @xmath251 , are expected to be negligible with respect to the third one involving @xmath96 . a noisy experimental determination of @xmath252 and @xmath253 is thus not a serious difficulty for our reconstruction program . on the other hand , clean enough data are still needed to determine @xmath178 from eq . ( [ delz ] ) , using @xmath24 and its first two derivatives . before such clean data are available , it will be sufficient to verify that eq . ( [ delz ] ) is consistent with a slowly varying @xmath178 . in the next section , we will show that interesting theoretical constraints can anyway be obtained without knowing at all the density fluctuation @xmath24 , but using only the luminosity distance @xmath11 and consistency arguments within particular subclasses of scalar - tensor models . @xcite , a fit of presently known supernovae events has been performed to obtain the luminosity distance @xmath11 up to redshifts @xmath1 , of course still with large uncertainties . although this is not yet sufficient to constrain seriously scalar - tensor models , we can expect clean data on @xmath11 in the near future from additional supernovae events , and anyway earlier than for the density perturbations @xmath24 . the snap satellite will in particular observe thousands supernovae events up to @xmath12 . in this section , we will concentrate on the theoretical constraints that can be extracted from the knowledge of @xmath11 alone , and therefore of @xmath20 using eq . ( [ hvsdl ] ) . we will thus only use the results of subsection v.a above . since the knowledge of this function does not suffice to fully reconstruct the microscopic lagrangian ( [ s_jf ] ) , we will need additional assumptions on one of the functions it involves , either @xmath96 ( or @xmath124 , depending on the parametrization ) or the potential @xmath192 . one may also assume a functional relation between @xmath96 and @xmath192 ( for instance @xmath254 as in ref . @xcite ) . to emphasize as clearly as possible what kind of constraints can be imposed on scalar - tensor theories , we shall consider the worst situation for them . let us assume that the observed function @xmath20 will be exactly given by eq . ( [ h2 ] ) for @xmath140 , @xmath255 , and @xmath256 : @xmath257 of course , such an observation would _ a priori _ call for the following standard interpretation : gravity is correctly described by general relativity , and we live in a flat universe filled with dustlike matter and a cosmological constant , with corresponding present energy densities ( relative to the critical density ) @xmath160 and @xmath258 . however , for our purpose , eq . ( [ h2gr ] ) should just be considered as kinematical . it tells us how the universe expands with redshift @xmath19 , but we are free to assume that the dynamics of the expansion is governed by a scalar - tensor theory . therefore , @xmath160 and @xmath258 are here mere parameters , whose names refer to their physical significance in the framework of gr . of course , one should not forget that they do not have the same interpretation within scalar - tensor theories . for our numerical applications , we will further take the present estimates based on combined cmb fluctuations and supernovae observations ( they will be determined more accurately by future experiments ) : @xmath259 for these numerical values , ( [ h2gr ] ) is consistent with the presently available luminosity distance @xmath11 up to @xmath1 . actually the best - fit universe , if we assume flatness , gives @xmath260 and @xmath261 . we have chosen to work directly with the exact form ( [ h2gr ] ) , instead of the @xmath11 extracted from observation , in order to clarify the physical content of our results . indeed , the present observational estimates for @xmath11 are still too imprecise to constrain strongly the class of scalar - tensor theories we are considering . moreover , some of our results below depend crucially on the fact that @xmath20 keeps the form ( [ h2gr ] ) up to redshifts @xmath262 , which have not yet been reached experimentally . to relate our results to those obtained in @xcite using fitting functions or an expansion in powers of @xmath19 , one just needs to use eq . ( [ hvsdl ] ) : our exact expression ( [ h2gr ] ) for @xmath20 corresponds to some exact expression for @xmath11 . to summarize , we are assuming in this section that future observations of the luminosity distance @xmath11 will provide a @xmath20 of the form ( [ h2gr ] ) with the numerical values ( [ omegavalues ] ) . this implies notably that our universe is presently accelerating . on the other hand , we are _ not _ assuming that the correct theory of gravity is necessarily gr plus a cosmological constant . the main question that we will address is therefore the following : would such an `` observed '' @xmath20 necessarily rule out the existence of a scalar partner to the graviton ? if not , would it be possible to reproduce ( [ h2gr ] ) within a more natural scalar - tensor theory , in which @xmath258 could be explained by a `` generalized quintessence '' mechanism ? we will first analyze in subsection a the simplest subclass of scalar - tensor theories that we can consider , namely when @xmath263 in action ( [ s_jf ] ) . since this is _ a priori _ the subclass which differs the most from gr plus a cosmological constant , this study will be rather detailed , and it will allow us to underline the mathematical and physical meaning of the constraints that are obtained . subsection b will be again devoted to the case of a massless scalar field , but combined with a cosmological constant . as its conclusions basically confirm those of subsection a , we will present them more concisely . finally , we will briefly discuss in subsection c the cases where one imposes particular forms for the coupling function @xmath96 in action ( [ s_jf ] ) , and one reconstructs the potential @xmath192 from the background equations ( [ f])([phi ] ) . the case of a given functional dependence between @xmath96 and @xmath192 will also be addressed . since a cosmological constant can be interpreted as a particular case of scalar - field potential , it is instructive to analyze whether an observed expansion like ( [ h2gr ] ) could be reproduced in a theory _ without _ any potential , and we now study eqs . ( [ f])([5.5 ] ) for @xmath264 . this case can be analyzed using the second order differential equation ( [ f ] ) for @xmath96 , which simplifies significantly if one introduces a function @xmath214 such that @xmath265 [ as mentioned in sec . v.a above , one can also set @xmath177 without loss of generality . ] then , using the assumed `` experimental '' expression ( [ h2gr ] ) for @xmath20 , and writing ( [ f ] ) in terms of @xmath266 , we get @xmath267 to avoid any confusion , let us recall that @xmath258 ( and the two occurrences of @xmath160 in the left - hand side ) comes from the `` observed '' cosmological function ( [ h2gr ] ) , notwithstanding the fact that there is _ no _ cosmological constant in the model we are considering . the value @xmath160 appearing in the right - hand side stands for the present relative energy density of dustlike matter . we assume that it takes the same numerical value ( [ omegavalues ] ) as in the `` observed '' @xmath20 ( [ h2gr ] ) . equation ( [ eqf ] ) tells us how we should choose @xmath268 to mimic exactly this @xmath20 in the present potential - free theory . in other words , @xmath258 and @xmath160 are two numbers assumed to be given by experiment , and we wish to fit @xmath268 and @xmath169 to satisfy eq . ( [ eqf ] ) . to integrate this second - order differential equation , we need two initial conditions for @xmath214 and its derivative . the first one is an obvious consequence of eq . ( [ smallf ] ) taken at @xmath269 , and we simply get @xmath270 . the second one should be such that the solar - system bound ( [ df0 ] ) is satisfied . for instance , if @xmath271 does not vanish , it is _ sufficient _ to impose @xmath272 , _ i.e. _ , @xmath273 using eq . ( [ smallf ] ) . this corresponds to a scalar - tensor theory which has been attracted towards an extremum of @xmath96 during the cosmological expansion of the universe ( _ cf . _ @xcite ) , so that it is presently strictly indistinguishable from general relativity in solar - system experiments . [ the full allowed domain for @xmath274 will be explored below in a numerical way . ] we consider first our potential - free model in a spatially flat frw universe ( @xmath164 ) . then eq . ( [ eqf ] ) becomes a first - order differential equation for @xmath275 , and its integration yields @xmath276\ , , \label{fprime}\ ] ] where we have set @xmath277 , and where the final constant inside the square brackets has been chosen to impose @xmath273 ( _ i.e. _ , @xmath278 ) . the function @xmath279 can be explicitly written in terms of generalized hypergeometric functions , but its complicated expression will not be useful for our purpose . let us just quote the first order of its expansion in powers of @xmath280 : @xmath281 in conclusion , eq . ( [ eqf ] ) could be integrated analytically , in the particular case of a spatially flat universe . this means that at least in the vicinity of @xmath269 , there _ a priori _ exists a potential - free scalar - tensor theory which exactly mimics general relativity plus a cosmological constant . however , the theory is mathematically consistent only if @xmath178 remains strictly positive . [ if @xmath96 vanishes , then the coupling function @xmath188 , eq . ( [ a ] ) , between matter and the helicity-0 degree of freedom @xmath51 diverges , and if @xmath96 becomes negative , the graviton carries negative energy . ] let us thus compute the value @xmath282 for which @xmath283 , or @xmath284 , vanishes for the first time . because of the complexity of the solution @xmath214 , we did not find a close analytical expression for @xmath282 , but its expansion in powers of @xmath280 can be obtained straightforwardly : @xmath285 { \omega_{m,0}^2\over\omega_{\lambda,0}^2 } + o\left({\omega_{m,0}^3\over\omega_{\lambda,0}^3}\right ) . \label{zmax}\ ] ] numerically , for the values ( [ omegavalues ] ) of @xmath258 and @xmath160 , we find @xmath286 . in conclusion , this scalar - tensor model is able to mimic general relativity plus a cosmological constant , but only on the small interval @xmath287 . if future observations of type ia supernovae give a behavior of @xmath20 of the form ( [ h2gr ] ) on a larger interval , say up to @xmath1 , then the present scalar - tensor theory will be ruled out . this example of a vanishing potential illustrates a conclusion that we will reobtain below for more general theories : the determination of the form of @xmath20 over some ( even rather small ) redshift interval is in fact more constraining than the precise value of the parameters @xmath288 themselves . indeed , eq . ( [ zmax ] ) clearly shows that @xmath282 can not exceed 1 even in the presumably unrealistic case of @xmath289 . [ a calculation using the exact expression for @xmath268 shows that @xmath290 would exceed 1 only for @xmath291 . ] note that all the results obtained are independent of the parameter @xmath5 . one could try to increase @xmath282 by considering a spatially curved frw universe . we did not solve eq . ( [ eqf ] ) in the most general case , but since we wish to compute the corrections to eq . ( [ zmax ] ) due to a small value of @xmath292 , it is sufficient to work at zeroth order in @xmath280 . let us thus set @xmath293 in eq . ( [ eqf ] ) , which reduces to @xmath294 its solution is obviously a sine if @xmath295 ( _ i.e. _ , @xmath296 , closed universe ) , or a hyperbolic sine for @xmath297 ( _ i.e. _ , @xmath106 , open universe ) . taking into account the initial conditions @xmath270 and @xmath298 , we thus get @xmath299 [ fok0 ] the first zero of @xmath300 is then reached either at @xmath301 or at @xmath302 . in both cases , the expansion in powers of @xmath303 gives @xmath304 . working perturbatively , one can also compute the correction to this expression due to the nonzero value of @xmath160 , and one finds that @xmath282 is given by eq . ( [ zmax ] ) above plus the following correction : @xmath305 { \omega_{\kappa,0}\over\omega_{\lambda,0 } } + o\left({\omega_{\kappa,0}^2\over\omega_{\lambda,0}^2}\right)\ , . \label{deltazmax}\ ] ] in conclusion , @xmath282 can be slightly enlarged if we consider our potential - free scalar - tensor theory in an open frw universe ( @xmath297 ) . numerically , for the values ( [ omegavalues ] ) of @xmath258 and @xmath160 , we find @xmath306 . since the latest experimental data on cmb temperature fluctuations already constrain @xmath165 to be small ( see the latest boomerang and maxima data ) , and actually an open universe is unlikely while a marginally closed universe is still acceptable , we thus recover the same qualitative conclusion as in the spatially flat case : it is possible to mimic general relativity plus a cosmological constant within a potential - free scalar - tensor theory only on a small redshift interval @xmath307 . the above conclusions have been confirmed by numerical integrations of eqs . ( [ f])([5.5 ] ) , still assuming a hubble diagram consistent with ( [ h2gr ] ) . instead of considering only theories which are presently indistinguishable from general relativity ( @xmath278 ) , we imposed arbitrary initial conditions for @xmath250 , and computed the corresponding value of the present scalar - matter coupling strength @xmath308 , eq . ( [ alpha ] ) . in the case of a spatially flat frw universe , we recovered that the solar - system bound ( [ df0 ] ) imposes the limit @xmath309 , consistently with the above analytical estimate ( [ zmax ] ) . in other words , the constraint ( [ df0 ] ) is so tight that even taking the largest allowed value for @xmath310 does not change significantly @xmath282 . figure 1 displays the reconstructed @xmath178 for this maximal @xmath310 , and one can note that its slope at @xmath167 is visually indistinguishable from the horizontal . this figure also plots the einstein - frame scalar @xmath51 , eq . ( [ varphi ] ) , which is the actual helicity-0 degree of freedom of the theory . notice that it diverges at @xmath282 , so that the theory loses its consistency beyond this value of the redshift . curiously , we found that even if no experimental constraint like ( [ df0 ] ) is imposed on @xmath310 ( _ i.e. _ , even if we forget that solar - system experiments confirm very well general relativity ) , then the mathematical consistency of the theory anyway imposes @xmath311 . in fact , eq . ( [ f ] ) alone can be solved for arbitrary large values of @xmath19 , _ i.e. _ , there exist initial values of @xmath239 such that @xmath178 remains positive for any @xmath19 . however , the values of @xmath239 needed to integrate eq . ( [ f ] ) beyond @xmath312 correspond to negative values of @xmath313 ( where @xmath314 denotes the present value of the brans - dicke parameter ) . in other words , the expression of @xmath196 given by eq . ( [ 5.5 ] ) would become negative around @xmath167 , and the helicity-0 degree of freedom would thus need to carry negative energy at least on a finite interval of @xmath19 , if one wished to integrate eqs . ( [ f])([5.5 ] ) beyond @xmath315 . figure 2 displays the maximum redshift @xmath282 consistent with the positivity of energy of both the graviton and the scalar field , but for any value of the present matter - scalar coupling strength @xmath310 . as underlined above , one finds that @xmath290 can never be larger than @xmath316 . this figure also indicates the present solar system bound on @xmath310 , corresponding to @xmath309 as in fig . 1 . the limiting case of a vanishing @xmath310 , _ i.e. _ , of a scalar - tensor theory which is presently _ strictly _ indistinguishable from gr in the solar system , corresponds to @xmath286 , as was derived analytically in eq . ( [ zmax ] ) . figure 2 also indicates the range of values for @xmath310 that are generically obtained in refs . @xcite while studying the cosmological evolution of scalar - tensor theories at earlier epochs in the matter - dominated era : the theory is attracted towards a maximum of @xmath96 ( _ i.e. _ , a minimum of @xmath317 ) so that the present value of @xmath310 is expected to be extremely small . finally , this figure also displays the maximum value of @xmath310 for which the parametrization @xmath318 of action ( [ s_jf ] ) has a meaning . beyond @xmath319 ( _ i.e. _ , for a brans - dicke parameter @xmath320 ) , one would get @xmath321 in this parametrization . in other words , eqs . ( [ f])([phi ] ) can not be integrated consistently beyond @xmath322 if one sets @xmath39 , whereas the brans - dicke or the einstein - frame representations show that the theory can be mathematically consistent up to @xmath323 ( @xmath199 remains positive ) . this underlines that the @xmath103 parametrization may be sometimes pathological . our numerical integration of eq . ( [ varphiprime ] ) not only allowed us to check the positivity of the scalar field energy , but also to reconstruct parametrically the matter - scalar coupling function @xmath188 . since @xmath324 , eq . ( [ a ] ) , we know that @xmath325 is finite and strictly positive over the interval @xmath326 , but we also checked that it is single valued over this interval . this means that if @xmath195 can take several times the same value for different @xmath19 , they must correspond also to the same value of @xmath325 . actually , since eq . ( [ varphiprime ] ) does not fix the sign of @xmath327 , one should keep in mind that @xmath51 can oscillate around a constant value @xmath328 . if the numerical integration confuses the two points @xmath329 , but if @xmath188 happens not to be symmetrical around @xmath328 , it may look like a bi - valued function . when such a situation occurred in our programs , we always verified that a single - valued @xmath188 could be defined consistently by unfolding it around the oscillation points of @xmath51 . figure 3 illustrates such a situation , for an intentionally unrealistic value of @xmath310 in order to clarify the plots . [ the value @xmath330 is inconsistent with the solar - system bound ( [ df0 ] ) , but it corresponds anyway to a mathematically consistent theory , although the @xmath103 parametrization can not be used in this case . ] all the functions @xmath317 that we reconstructed have similar convex parabolic shapes . this is consistent with the results of refs . @xcite , showing that the scalar field is generically attracted towards a minimum of @xmath317 during the expansion of the universe . if we had found models such that the present epoch ( @xmath167 ) is close to a _ maximum _ of @xmath331 , this would have meant that the theory is unstable , and that we have extremely fine tuned it to be consistent with solar - system constraints . on the contrary , the convex functions @xmath317 that we obtained show that these scalar - tensor models are cosmologically stable , _ i.e. _ , that the tight bounds ( [ df0 ] ) are in fact natural consequences of the attractor mechanism described in @xcite . we have checked that reducing the parameter @xmath258 allows us to extend the integration region in the past , consistently with the above analytical results . for instance , when we vary @xmath258 , still satisfying @xmath332 and setting @xmath164 , we find that @xmath333 is required in order to integrate the equations up to a redshift @xmath334 . this would correspond to @xmath335 , _ i.e. _ , 100 times larger than present estimates . we also added random noise to our assumed @xmath20 , eq . ( [ h2gr ] ) , and verified that the conclusions are not changed qualitatively provided @xmath20 is known over a wide enough redshift interval . this means that the experimental determination of the luminosity distance @xmath11 needs not be very precise to be quite constraining , provided redshifts of order @xmath262 are probed . as an illustration , let us take the exact expression @xmath20 of eqs . ( [ h2gr])([omegavalues ] ) for discrete values of the redshift , say @xmath336 , and let us add or subtract randomly between @xmath76 and @xmath337 to the corresponding @xmath20 . then , we may fit a polynomial through these `` noisy '' values of @xmath20 , and use our numerical programs to integrate the background equations ( [ f])([5.5 ] ) and reconstruct @xmath96 . we found that there always exists a maximum redshift beyond which @xmath96 is negative ( and the theory thus inconsistent ) . figure 4 displays the two extreme values of @xmath282 that we obtained with hundreds of such `` deformed '' @xmath20 : it is sometimes even smaller than for the `` exact '' @xmath20 of eq . ( [ h2gr ] ) , and sometimes larger but never greater that @xmath338 . it should be noted that for the @xmath337 noise we chose , the @xmath20 of pure gr with a vanishing cosmological constant could have been obtained . in that case , a potential - free scalar - tensor model with @xmath339 would of course have fitted perfectly this @xmath20 up to @xmath340 . the reason why we never managed to go beyond @xmath341 is that we considered _ random noise _ , instead of such a precise _ bias _ of our assumed function @xmath20 , eq . ( [ h2gr ] ) . we are aware that our deformed functions of fig . 4 do not reproduce a realistic experimental noise . however , they illustrate in a well defined way that an inaccurate determination of @xmath20 over a wide redshift interval is actually more constraining than a precise measurement over a small redshift interval only . the conclusion of the present subsection is therefore that a scalar - tensor theory without potential can accommodate a hubble diagram consistent with ( [ h2gr ] ) , but only on a small redshift interval if @xmath258 is significant . the experimental determination of the luminosity distance @xmath11 , either accurately for @xmath242 or even with large ( tens of percents ) uncertainties up to redshifts @xmath262 , severely constrains this subclass of theories . future observations should thus be able to distinguish them from general relativity , and to confirm or rule them out without any ambiguity . it is worth noting that such future determinations of @xmath11 would _ a priori _ be much more constraining than solar - system experiments and binary pulsars tests . indeed , although the precision of the latter is quite impressive ( see e.g. @xcite ) , they anyway probe only the first two derivatives of @xmath317 , eqs . ( [ df0])-([beta0 ] ) , whereas cosmological observations should give access to the full shape of this function . let us also recall that the constraints we found crucially depend on the fact that the theory should contain only positive - energy excitations to be consistent , and notably that the function @xmath96 should remain always strictly positive . we did not use any other cosmological observation , but obviously , once the microscopic lagrangian of a scalar - tensor theory has been reconstructed using @xmath11 , all its other cosmological predictions should also be checked . for instance , a bound @xmath342 is given in ref . @xcite for the value of the function @xmath96 at nucleosynthesis time . are shown to be consistent with the observed abundances of light elements , provided @xmath343 is large enough , where @xmath188 is the matter - scalar coupling function ( [ a ] ) . ] if one assumes that @xmath178 is monotonic , the reconstructed function of fig . 1 would not be consistent with this nucleosynthesis bound beyond @xmath344 . this would be even more constraining than the bound @xmath345 we obtained just from mathematical consistency requirements . alternatively , a reconstructed function @xmath178 like the one of fig . 1 would be consistent with the above nucleosynthesis bound only if it were non - monotonic beyond @xmath346 . although this would not be forbidden from a purely phenomenological point of view , this would be anyway unnatural , and more difficult to justify theoretically . to confirm the results of the previous subsection , let us now consider the case of a massless scalar field together with a cosmological constant whose value _ differs _ from the one entering our assumed @xmath20 , eq . ( [ h2gr])-([omegavalues ] ) . the question that we wish to address is the following : can _ part of _ the observed @xmath258 be due to the presence of a massless scalar field ? to impose a cosmological constant in a scalar - tensor theory , one would naively choose a constant potential @xmath10 in action ( [ s_jf ] ) . however , as shown by eq . ( [ v ] ) , the corresponding potential @xmath130 of the helicity-0 degree of freedom @xmath51 would not be constant in this case ( because @xmath33 is _ a priori _ varying ) , and its second derivative would give generically a nonvanishing scalar mass . to avoid any scalar self - interaction , and in particular to set its mass to @xmath76 , one needs in fact to impose @xmath347 in the einstein - frame action ( [ s_ef ] ) . this defines a consistent cosmological `` constant '' in a massless scalar - tensor theory . note that the corresponding jordan - frame potential @xmath10 is then proportional to @xmath348 , and therefore that it does not correspond to the usual notion of cosmological _ constant _ in action ( [ s_jf ] ) . since our assumed `` observed '' @xmath20 involves a parameter denoted @xmath258 , eqs . ( [ h2gr])-([omegavalues ] ) , let us introduce a different notation for the contribution due to the constant potential @xmath349 : @xmath350 it is easily checked that for @xmath351 , the solution @xmath352 ( or @xmath353 ) is recovered , _ i.e. _ , a scalar field minimally coupled to gravity with a constant potential acting like a cosmological constant . indeed , in terms of the function @xmath268 defined in ( [ smallf ] ) , eq . ( [ f ] ) reads @xmath354 note that this is now a _ non_-linear equation in @xmath96 , contrary to eq . ( [ eqf ] ) above for the case of a vanishing potential . if @xmath351 , one finds that @xmath355 is an obvious solution , _ i.e. _ , @xmath356 a constant scalar field @xmath9 ( or @xmath51 ) then satisfies eqs . ( [ phi])([varphiprimebis ] ) . if we now consider a scalar - tensor theory for which @xmath357 differs from the `` observed '' @xmath358 , we find that like in the previous subsection , there exists a maximum redshift @xmath282 beyond which @xmath178 becomes negative , and therefore beyond which the theory loses its mathematical consistency . figure 5 displays this maximum redshift as a function of @xmath357 . we plot this figure for the initial condition @xmath278 ( _ i.e. _ , for a theory which is presently indistinguishable from gr in the solar system ) , but as before , we verified that the curve is almost identical if one takes the maximum value of @xmath359 consistent with the solar - system bound ( [ df0 ] ) . we also assume @xmath164 ( spatially flat universe ) for this figure , as we know from the previous discussion that a value even as large as @xmath360 does not change qualitatively the results . for @xmath361 , we recover the result @xmath362 derived above for a vanishing potential . when @xmath363 , @xmath282 becomes even smaller . as expected , this is worse than in the potential - free case . on the contrary , when @xmath357 is positive ( _ i.e. _ , when it contributes positively to part of the `` observed '' @xmath258 ) , the maximum redshift @xmath282 increases . this is just due to the fact that our massless scalar field needs to mimic a smaller fraction of the `` observed '' @xmath258 , so that the theory can remain consistent over a wider redshift interval . however , we find that @xmath282 is still smaller than @xmath364 for @xmath365 , and a @xmath20 of the form ( [ h2gr])-([omegavalues ] ) observed up to @xmath262 would thus suffice to rule out the model . if such a @xmath20 could be confirmed up to @xmath14 , one would need @xmath366 for our massless scalar - tensor theory to fit it ! even so , the theory would anyway become pathological at slightly higher redshifts . in conclusion , a massless scalar _ can not _ account for a significant part of the observed cosmological constant if @xmath20 is experimentally found to be of the form ( [ h2gr ] ) over a wide redshift interval . let us note finally that for @xmath367 , eq . ( [ eqfvconst ] ) does admit strictly positive solutions for @xmath214 ( or @xmath96 ) up to arbitrarily large redshifts . this ensures that the graviton energy is always positive . however , it is now the scalar field which needs to carry negative energy . indeed , eq . ( [ varphiprimebis ] ) gives a negative value for @xmath199 , basically because of the presence of the large negative number @xmath368 in this equation . [ in the @xmath318 parametrization , @xmath200 is obviously also negative , because of eq . ( [ positiveenergy ] ) , or directly from eq . ( [ phibis ] ) which also involves a @xmath368 term . ] therefore , there is only one possibility for a consistent massless scalar - tensor theory to reproduce ( [ h2gr ] ) over a wide redshift interval : it must involve a cosmological constant , whose contribution @xmath357 is equal to ( or very slightly smaller than ) the parameter @xmath258 entering ( [ h2gr ] ) . in other words , the theory should be extremely close to gr plus a cosmological constant , and the massless scalar field must have a negligible contribution . this illustrates again the main conclusion of our paper : the experimental determination of the luminosity distance @xmath11 over a wide redshift interval , up to @xmath262 , will suffice to rule out ( or confirm ) the existence of a massless scalar partner to the graviton . in the previous two subsections , the matter - scalar coupling function @xmath33 ( or @xmath188 ) was reconstructed from the assumed knowledge of @xmath20 , for theories whose potential @xmath10 ( or @xmath130 ) had a given form . we now consider the inverse problem . we still assume that future observations will provide a hubble diagram consistent with ( [ h2gr])-([omegavalues ] ) , but we wish now to reconstruct the scalar - field potential @xmath192 for given forms of the coupling function @xmath96 . we first consider a generic two - parameter family of scalar - tensor theories , which has already been studied in great detail for solar - system , binary - pulsar and gravity - wave experiments @xcite , as well as for cosmology starting with the matter - dominated era @xcite and even back to nucleosynthesis @xcite . its definition is simplified if we work in the einstein frame ( [ 2.4])-([s_ef ] ) . the matter - scalar coupling function is simply given by @xmath369 in which the present value of the scalar field , @xmath194 , may be chosen to vanish without loss of generality . any analytical function @xmath317 may be expanded in such a way , but we here assume that no higher power of @xmath51 appears , _ i.e. _ , that @xmath370 is strictly parabolic : it depends only on the two parameters @xmath308 and @xmath371 . the latter is a simplified notation for @xmath77 , and should not be confused with the post - newtonian parameter @xmath372 defined in ( [ betappn ] ) . [ actually , this equation shows that @xmath373 . ] solar - system experiments impose @xmath374 , eq . ( [ df0 ] ) , while binary pulsars give @xmath375 , eq . ( [ beta0 ] ) , for this class of theories . we first study these models for the case of a spatially flat universe ( @xmath164 ) . as shown by eq . ( [ kg ] ) , a constant scalar field @xmath376 may be a solution if @xmath377 ( so that @xmath378 vanishes too ) and if the potential @xmath130 is also constant . our assumed @xmath20 , eqs . ( [ h2gr])-([omegavalues ] ) , can thus always be reproduced if the parameter @xmath308 vanishes identically , and the reconstructed potential merely reduces to the constant @xmath379 . this corresponds simply to gr plus a cosmological constant , and the massless scalar degree of freedom @xmath51 remains unexcited , frozen at an extremum of the parabola ( [ generica ] ) . actually , eq . ( [ kg ] ) shows that this extremum corresponds to a stable situation only if it is a minimum , _ i.e. _ , if @xmath380 in ( [ generica ] ) . this is consistent with the results of refs . @xcite : if the theory involves a cosmological constant whose value equals the `` observed '' one in eqs . ( [ h2gr])-([omegavalues ] ) , a massless scalar field is cosmologically attracted towards a minimum of the coupling function @xmath317 , and the present value of its slope , @xmath308 , is expected to be generically very small . on the other hand , if @xmath308 is _ not _ assumed to vanish , say if its value is comparable to the solar - system bound ( [ df0 ] ) , then our reconstruction of the potential @xmath130 from eqs . ( [ f])([5.5 ] ) leads to serious difficulties . their nature depends on the magnitude of the curvature parameter @xmath371 of parabola ( [ generica ] ) . string - inspired models @xcite suggest that @xmath371 may be as large as 10 , or even 40 . with such large values ( and assuming non - vanishing @xmath308 ) , our numerical integrations of eqs . ( [ f])([5.5 ] ) give concave potentials @xmath130 , unbounded from below . this corresponds to unstable theories , and thereby to extremely fine - tuned initial conditions : changing slightly the derivative of the scalar field , @xmath327 , at high redshifts would _ a priori _ yield a totally different universe at present . this result tells us that this kind of models can not be consistent over a wide redshift interval with the exact form of @xmath20 we chose in ( [ h2gr ] ) , unless the parameter @xmath308 is extremely small . actually , this is just another way to present the results of refs . @xcite : since they predict that @xmath308 should be almost vanishing at present , assuming a significant non - zero value implies that the theory is unnatural . to obtain convex - shaped potentials @xmath130 ( _ i.e. _ , stable theories ) while still assuming a non - vanishing @xmath308 , we typically need values of @xmath381 . however , the reconstructed potentials always exhibit sudden changes of their slope . basically , they reproduce a cosmological constant over a finite interval around @xmath194 ( _ i.e. _ , around @xmath269 ) , and become rapidly divergent beyond a critical value of the scalar field ( depending on @xmath371 ) . therefore , as in subsection vi.b above , we find that such scalar - tensor models can reproduce ( [ h2gr ] ) only if they involve a cosmological constant , whose energy contribution is close to the parameter @xmath258 entering @xmath20 , and if the scalar field has a negligible enough influence . in other words , such models would not explain the small but nonzero value of the observed cosmological constant by a `` quintessence '' mechanism , and would not be more natural than merely assuming the existence of @xmath4 . the above results are significantly changed if we take into account the possible spatial curvature of the universe . indeed , smoother potentials @xmath130 are obtained for closed universes ( @xmath295 ) , and the present value of the cosmological constant thus becomes more `` natural '' . to illustrate this feature , let us consider the case of a minimally coupled scalar field ( as in @xcite ) , corresponding to @xmath382 in eq . ( [ generica ] ) . for an open universe ( @xmath297 ) , we find from eq . ( [ 5.5 ] ) that the scalar field would need to carry negative energy to reproduce ( [ h2gr ] ) . on the other hand , for a closed universe ( @xmath295 ) , one can derive analytically the parametric form of the potential @xmath130 . it can be expressed in terms of the hypergeometric function @xmath383 ( solution of the differential equation @xmath384{\cal f } ' - ab { \cal f } = 0 $ ] ) : @xmath385 [ parametricv ] where as before @xmath386 . if @xmath165 is very small , we recover that @xmath130 exhibits a sudden change of slope , as was obtained above in the flat case . this is illustrated by the left panel of fig . 6 . on the contrary , if @xmath165 is large enough , the same analytical expression ( [ parametricv ] ) gives nice regular potentials , like the one displayed in the right panel of fig . this reconstructed @xmath130 , as well as those obtained numerically for weakly varying @xmath317 , eq . ( [ generica ] ) , are natural in the sense that they can be approximated by the exponential of simple polynomials in @xmath51 . in that case , the observed value of the cosmological constant does not appear as a mere parameter introduced by hand in the lagrangian , but corresponds basically to the present value of @xmath387 . it should be noted that a value as large as @xmath388 is not excluded by the latest boomerang data , though it would be problematic in the framework of the inflationary paradigm . in conclusion , the existence of non - singular solutions over a long period of time is again the constraining input . a non - minimally coupled scalar field is essentially incompatible with ( [ h2gr ] ) over a wide redshift interval , unless the scalar field is frozen at a minimum of @xmath317 ( consistently with @xcite ) . if future experiments provide a hubble diagram in accordance with ( [ h2gr ] ) and also give a very small value for @xmath169 , it will be possible to conclude that scalar - tensor theories ( either non - minimally or minimally coupled ) _ can not _ explain in a natural way the existence of a cosmological constant . on the other hand , if the universe is closed and @xmath165 large enough , a `` quintessence '' mechanism in a scalar - tensor theory seems more natural than a mere cosmological constant . the above conclusions can be confirmed by starting from a given @xmath178 ( or @xmath325 ) , rather than @xmath33 ( or @xmath188 ) . we consider here `` scaling solutions '' , _ i.e. _ , we assume that these functions behave as some power of the scale factor @xmath389 . one may for instance write @xmath390 , with @xmath391 . as before , our aim is to reconstruct a regular potential @xmath130 from the knowledge of @xmath20 , assumed to be of the form ( [ h2gr])-([omegavalues ] ) . the strongest constraint on this class of theories is imposed by the solar - system bound ( [ df0 ] ) . indeed , using the definition ( [ alpha ] ) for @xmath136 , one can also write it as @xmath392 , and eq . ( [ 5.5 ] ) evaluated at @xmath167 then yields the following second - order equation for @xmath393 : @xmath394 note that this equation does not depend on the full form of eq . ( [ h2gr ] ) , but only on its first derivative at @xmath269 , _ i.e. _ , on the deceleration parameter @xmath395 . the constraints on @xmath393 derived below are thus valid as soon as @xmath396 is of order @xmath397 , consistently with the estimated value ( 6.2 ) for @xmath160 . in the case of a spatially flat universe ( @xmath164 ) , ( [ eq.p ] ) gives immediately @xmath398 , so that the solar - system bound ( [ df0 ] ) imposes @xmath399 . therefore , the scalar field needs to be almost minimally coupled . if @xmath393 vanishes identically , we recover as before the trivial solution of gr plus a cosmological constant , together with an unexcited minimally - coupled scalar field . on the other hand , if @xmath393 does not vanish , one finds that the scalar field needs to carry negative energy beyond @xmath400 . even without trying to reconstruct the potential @xmath130 , one can thus conclude that such scaling solutions would be ruled out by the observation of a @xmath20 of the form ( [ h2gr ] ) up to @xmath262 . paradoxically , this result is valid even for an infinitesimal ( but nonzero ) value of @xmath393 . indeed , there exists a discontinuity between the case of a strictly constant @xmath96 and that of a scaling solution @xmath401 . at first order in @xmath393 , and still assuming @xmath164 , one can write eq . ( [ 5.5 ] ) as @xmath402 this equation confirms that @xmath403 when @xmath404 , and therefore that the scalar field tends towards a constant in this limit . however , it carries positive energy ( @xmath405 ) only if @xmath406 since the right - hand side is estimated to be @xmath407 , the maximum value of @xmath19 is obtained for @xmath408 , so that the large numerical factor coming from @xmath409 in the left - hand side is compensated by the small term inside the second parentheses . working iteratively , this maximum redshift can be better approximated by @xmath410 , and the actual numerical resolution of equality ( [ psmallpositivee ] ) for the values ( [ omegavalues ] ) gives @xmath411 . therefore , even if @xmath393 is vanishingly small , a scaling solution @xmath401 can not be consistent with ( [ h2gr ] ) beyond this maximum redshift . this illustrates once more that the experimental determination of @xmath20 up to @xmath262 would be more constraining that solar - system experiments for this class of theories , provided one takes into account the requirement of positive energy . let us underline that the above value for @xmath282 is valid for a monomial @xmath412 but not for more complicated polynomial expressions . indeed , as shown for instance in sec . vi.b above , there do exist scalar - tensor theories consistent with ( [ h2gr ] ) up to arbitrarily large redshifts , and they do not need to be _ strictly _ equivalent to gr plus a cosmological constant ( although they must be close enough to it ) . moreover , the above maximum redshift is a consequence of the _ exact _ form for @xmath20 we chose in eq . ( [ h2gr ] ) . a slightly different function may of course allow a positive - energy scalar field up to much higher redshifts . it suffices that the right - hand side of eq . ( [ varphiprime ] ) be strictly positive for @xmath413 , and the case of a closed universe discussed below provides an example , since the contribution @xmath414 is then positive in eq . ( [ varphiprime ] ) . the case of a spatially open universe ( @xmath297 ) is forbidden by eq . ( [ eq.p ] ) , unless @xmath169 is smaller than @xmath415 . such a situation would be indistinguishable from the spatially flat case . in a spatially closed universe ( @xmath295 ) , @xmath393 is given by the positive root of the second - order equation eq . ( [ eq.p ] ) . remembering the solar - system bound ( [ df0 ] ) , one may consider the case @xmath416 , and one gets @xmath417 . even if one considered values of @xmath169 as large as @xmath418 , this would limit @xmath393 to @xmath419 . therefore , in this case again , solar - system constraints impose that the scalar field should be almost minimally coupled , if one looks for such scaling solutions . the difference with the spatially flat case is that eqs . ( [ f])([varphiprimebis ] ) can now be integrated for any redshift @xmath19 ( from future infinity , @xmath420 , to arbitrarily large @xmath19 ) . since @xmath178 needs to be almost constant , we recover solution ( [ parametricv ] ) for the potential @xmath130 . as in sec . vi.c.1 above , we can thus conclude that such models would be consistent with ( [ h2gr ] ) over a wide redshift interval only if they are ( almost ) minimally coupled , and they would provide a natural `` quintessence '' mechanism to explain the presently observed cosmological constant only if the universe is ( marginally ) closed . let us end this paragraph by a remark concerning scaling solutions , for which the scalar - field energy density scales like a power of @xmath389 . as mentioned in the introduction , they have attracted a lot of attention recently . for a minimally coupled field , the possible scaling behaviors and the corresponding potentials can be classified @xcite . as for a non - minimally coupled field , a subclass of theories was considered in @xcite , for which @xmath421 where @xmath422 and @xmath423 are constants . since , besides these two constants , there is only one unknown function of @xmath9 , the knowledge of @xmath20 suffices to reconstruct the full microscopic lagrangian from eqs . ( [ f])([5.5 ] ) above . however , the main conclusion of ref . @xcite can be recovered from a simple argument , without any numerical integration . indeed , it was shown in this reference that there exists a universal behavior of these theories , depending on @xmath423 but not on the precise shape of @xmath33 . as emphasized in @xcite , this result was obtained in the strong coupling limit , corresponding formally to @xmath424 in action ( [ s_jf ] ) . taking into account the assumed relation ( [ ufm ] ) , the class of theories under consideration is thus defined by @xmath425\ . \label{s_jfz0}\ ] ] if we now introduce a new scalar variable @xmath426 , we notice that @xmath9 disappears totally from the action . no physical result can thus depend on the precise form of @xmath33 , and we recover the conclusion of @xcite . the constant @xmath422 may also be set to 1 by a change of length units , and this class of theories is thus parametrized by the single real number @xmath423 . any physical prediction must therefore depend only on @xmath423 . in this work we have investigated the constraints that arise from the _ experimental _ knowledge of the luminosity distance in function of the redshift up to @xmath262 , corresponding to @xmath20 given by ( [ h2gr])-([omegavalues ] ) . in particular , our universe is then presently accelerating and we have studied the viability of subclasses of scalar - tensor theories of gravity . we have shown that the subclass of models in which the scalar partner @xmath9 of the graviton has no potential at all , and which satisfy the present - day existing constraints , are inevitably ruled out if an expansion of the form of eq . ( [ h2gr ] ) holds even for a redshift interval as tiny as @xmath427 ( see the precise numbers in section vi ) . we see that these theories become pathological in the form of a vanishing @xmath96 , already at such low redshifts for which @xmath20 will be experimentally accessible in the near future @xcite . hence we show that a cosmological observation of the background evolution according to eq . ( [ h2gr ] ) in the `` recent '' epoch will be enough to rule out such models . [ on the other hand , future observations might provide a @xmath20 which confirms the existence of a scalar partner to the graviton and rule out pure gr ! ] the main reason why we obtained so constraining results is that we took into account the mathematical consistency of the theory , _ i.e. _ , the fact that it should contain only positive - energy excitations to be well behaved . this requirement severely restricts the class of viable models . a non - flat universe can alleviate in some cases the tight constraints we found . however , the latest cmb data released by boomerang and maxima @xcite favor a flat universe ( in accordance with the inflationary paradigm ) , and only a marginally closed universe is still allowed by the location of the first acoustic ( doppler ) peak at @xmath428 , while an open universe is more unlikely . the most impressive conclusion is that future cosmological observations may prove to be more constraining for massless scalar - tensor theories than solar - system and binary - pulsar tests . indeed , even if the determination of the luminosity distance @xmath11 will not reach very quickly the impressive accuracy obtained in the solar system or with binary pulsars , it will nevertheless give access to the full coupling function @xmath33 in action ( [ s_jf ] ) , or @xmath188 in the einstein - frame rewriting ( [ s_ef ] ) , whereas only its first two derivatives are presently probed . we thank our collaborators a. starobinsky and b. boisseau for discussions . one of us ( gef ) also wishes to thank a. riazuelo for informative discussions . centre de physique thorique is unit propre de recherche 7061 . s. perlmutter , g. aldering , m. della valle _ _ , nature * 391 * , 51 ( 1998 ) ; s. perlmutter , g. aldering , g. goldhaber _ et al . _ , astroph . j. , * 517 * , 565 ( 1999 ) . p.m. garnavich , r.p . kirshner , p. challis _ et al . _ , astrophys . j. lett . * 493 * , l53 ( 1998 ) ; a.g . riess , a.v . philipenko , p. challis _ et al . . j. * 116 * , 1009 ( 1998 ) . bernardis _ et al . _ , nature * 404 * , 955 ( 2000 ) . s. hanany _ et al . _ , astro - ph/0005123 ; a. balbi _ et al . _ , astro - ph/0005124 . bahcall , j.p . ostriker , s. perlmutter , and p.j . steinhardt , science , * 284 * , 1481 ( 1999 ) . v. sahni and a.a . starobinsky , ijmpd , to appear ( 2000 ) ; astro - ph/9904398 . m. chevallier and d. polarski , submitted , gr - qc/0009008 r.r . caldwell , r. dave , and p.j . steinhardt , phys . rev . lett . * 80 * , 1582 ( 1998 ) . b. ratra and p.j.e . peebles , phys . d * 37 * , 3406 ( 1988 ) . ferreira and m. joyce , phys . d * 58 * , 023503 ( 1998 ) ; phys . * 79 * , 4740 ( 1998 ) . i. zlatev , l. wang and p.j . steinhardt , phys . lett . * 82 * , 896 ( 1999 ) . a.d . liddle and r.j . scherrer , phys . rev . d * 59 * , 0223509 , ( 1998 ) . c. kolda and d.h . lyth , phys . b * 458 * , 197 ( 1999 ) . starobinsky , jetp lett . * 68 * , 757 ( 1998 ) ; gravit . & cosmol . * 4 * ( suppl . ) , 88 ( 1998 ) . d. huterer and m.s . turner , phys . d * 60 * , 081301 ( 1999 ) ; t. nakamura and t. chiba , mon . not . ast . soc . * 306 * , 696 ( 1999 ) . saini , s. raychaudhury , v. sahni , and a.a . starobinsky , astro - ph/9910231 . caldwell , astro - ph/9908168 . will , _ theory and experiment in gravitational physics _ , ( cambridge university press , cambridge , england , 1993 ) . t. damour and g. esposito - farse , class . * 9 * , 2093 ( 1992 ) . t. damour and g. esposito - farse , phys . d * 53 * , 5541 ( 1996 ) . t. damour and g. esposito - farse , phys . d * 54 * , 1474 ( 1996 ) . . will , phys . d * 50 * , 6058 ( 1994 ) . t. damour and g. esposito - farse , phys . d * 58 * , 042001 ( 1998 ) . m. gasperini and g. veneziano , astropart . phys . * 1 * ( 1993 ) 317 ; mod . phys . lett . a*8 * ( 1993 ) 3701 . t. damour and k. nordtvedt , phys . lett . * 70 * , 2217 ( 1993 ) ; phys . d * 48 * , 3436 ( 1993 ) . d.i . santiago , d. kalligas , and r.v . wagoner , phys . d * 58 * , 124005 ( 1998 ) . t. damour and b. pichon , phys . d * 59 * , 123502 ( 1999 ) . a. navarro , a. serna , and j.m . alimi , phys . d * 59 * , 124015 ( 1999 ) . uzan , phys . d * 59 * , 123510 ( 1999 ) . t. chiba , phys . d * 60 * , 083508 ( 1999 ) . x. chen and m. kamionkowski , phys . d * 60 * , 104036 ( 1999 ) . f. perrotta , c. baccigalupi , and s. matarrese , astro - ph/9906066 . holden and d. wands , gr - qc/9908026 . n. bartolo and m. pietroni , hep - ph/9908521 . e. gaztanaga and j.a . lobo , astro - ph/0003129 . b. boisseau , g. esposito - farse , d. polarski , and a.a . starobinsky , phys . * 85 * ( 2000 ) 2236 . misner , k.s . thorne , j.a . wheeler , _ gravitation _ ( san francisco , freeman , 1973 ) . bergmann , int . * 1 * , 25 ( 1968 ) ; k. nordtvedt , astrophys . j. * 161 * , 1059 ( 1970 ) ; r. wagoner , phys . rev . d * 1 * , 3209 ( 1970 ) . _ , bull . phys . soc . , abstract # k 11.05 ( 1997 ) . dickey _ et al . _ , science * 265 * , 482 ( 1994 ) ; j.g . williams , x.x . newhall , and j.o . dickey , phys . d * 53 * , 6730 ( 1996 ) . p. fiziev , gr - qc/9911037 . eddington , _ the mathematical theory of relativity _ , ( cambridge university press , cambridge , 1923 ) . b.m . barker , astrophys . j. * 219 * , 5 ( 1978 ) . bardeen , phys . d * 22 * , 1882 ( 1980 ) . l. amendola , phys . d * 60 * , 043501 ( 1999 ) . j. weller and a. albrecht , astro - ph/0008314 ; i. maor , r. brustein and p.j . steinhardt , astro - ph/0007297 ; t. chiba and t. nakamura , astro - ph/0008175 . t. damour and a. polyakov , nucl . * b423 * , 532 ( 1994 ) ; gen . * 26 * , 1171 ( 1994 ) .

bg geminorum was discovered by hoffmeister ( 1933 ) and jensch ( 1938 ) as a possible rv tau star with an uncertain period of @xmath0 60 days . with a photographic magnitude of @xmath0 14 , the star languished in the general catalog of variable stars ( @xcite ) until 1992 , when we began photometric observations to improve the period estimate and to verify the rv tau classification . early data revealed a repeatable ellipsoidal variation with a long period , instead of the more irregular variation expected from an rv tau star . additional photometry allowed us to refine the period and plan spectroscopic observations . together with the photometry , high quality spectra acquired around the orbit establish bg gem as a rare eclipsing binary system with an unseen primary star and a lobe - filling k supergiant secondary star . this paper describes our data and analysis . we begin with a summary of the observations in 2 , continue with a detailed analysis in 3 , and conclude with the discussion in 4 . various student observers acquired optical photometry of bg gem with standard vr@xmath1i@xmath1 filters and a photometrics pm512 camera mounted on the wellesley college 0.6-m sawyer telescope . observations beginning 23 july 1997 used a tk1024 back - illuminated ccd . we processed the data using standard tasks within noao iraf . each image was bias - subtracted , dark - subtracted , and flat - fielded using twilight flats . we extracted photometry of bg gem and several comparison stars using phot with a 3 aperture and a 3 sky annulus 17 from the star . wellesley is not a photometric site ; we thus derived photometry of bg gem relative to comparisons verified to be non - variable . the relative photometry has 1@xmath5 probable errors of @xmath60.011 mag at v , @xmath60.014 mag at r@xmath1 , and @xmath60.011 mag at i@xmath1 based on repeat observations of comparison stars in the field of bg gem . table 1 lists the relative photometry of bg gem and a comparison star as a function of time ( the heliocentric julian date , jd ) and photometric phase @xmath7 defined below . p. berlind , m. calkins , and several other observers acquired low resolution optical spectra of bg gem with fast , a high throughput , slit spectrograph mounted at the fred l. whipple observatory 1.5-m telescope on mount hopkins , arizona ( @xcite ) . we used a 300 g mm@xmath8 grating blazed at 4750 , a 3 slit , and a thinned loral 512 @xmath9 2688 ccd . these spectra cover 38007500 at a resolution of @xmath0 6 . we flux- and wavelength - calibrated the spectra in iraf . after trimming the ccd frames at each end of the slit , we corrected for the bias level , flat - fielded each frame , applied an illumination correction , and derived a full wavelength solution from calibration lamps acquired immediately after each exposure . the wavelength solution for each frame has a probable error of @xmath60.51.0 . we extracted object and sky spectra using the optimal extraction algorithm within apextract . the absolute flux - calibration for each night relies on observations of 2 - 5 standard stars ( barnes & hayes 1982 ) and has an uncertainty of @xmath65%10% . nights with clouds or poor seeing have poor absolute calibrations ( @xmath6 20% or worse ) but good relative calibrations . figure 1 shows a typical spectrum of bg gem . the spectrum resembles a late g or early k star , with a red continuum and strong absorption from the g band , mg i , na i , and the ba i blend at @xmath106495 . strong h i balmer emission lines are also visible , with h10 and h11 present on some spectra . the h@xmath11 and h@xmath12 lines are often double peaked . some spectra show a weak blue continuum and weak he i emission at 5876 and 6678 . there is no evidence for absorption lines from a hot , early - type star . to analyze the spectra , we measured continuum magnitudes and indices of strong absorption and emission lines using narrow passbands ( oconnell 1973 ; worthey 1994 ) . table 2 lists the central wavelength @xmath10 and width @xmath13 for each . the continuum magnitudes are normalized to the zero - point of the v magnitude scale , @xmath14 = @xmath15 @xmath1621.1 , where @xmath17 is the flux in the passband . the absorption and emission indices , derived using sbands within iraf , are @xmath18 = @xmath162.5 log(@xmath17/@xmath19 ) , where @xmath17 is the average flux in the passband @xmath10 and @xmath19 is the continuum flux interpolated between the fluxes in the neighboring blue band centered at @xmath20 and the red band centered at @xmath21 . table 3 lists the measured indices along with emission line equivalent widths as a function of jd and @xmath7 . table 4 lists continuum magnitudes measured on spectrophotometric nights . we searched various databases for previous observations of bg gem . bg gem is not a known point source in any of the online _ iras _ , radio , or x - ray catalogs ( e.g. , _ asca _ , _ rosat _ ) . it lies close to several triggers in the cgro / batse 4b catalog ( e.g. , trigger # 4157 ) , but large positional offsets from the nominal trigger positions makes association with these events unlikely . the average of 100 days of data from the all - sky monitor ( asm ) of the _ rossi x - ray timing explorer _ is @xmath22 mcrab at 212 kev ( remillard 1999 ; 1 mcrab is 0.0755 counts s@xmath8 with the asm ( levine 1996 ) , which corresponds roughly to @xmath23 erg @xmath24 s@xmath8 at 212 kev assuming a spectrum similar to that other x - ray binaries [ see @xcite ] ) . bg gem is a 2mass point source , # 0603308 + 274150 in the 1999 spring release point source catalog , as summarized below . figure 2 shows light curves for bg gem . the data have been folded on the best period derived from fourier analysis and periodograms . the primary minimum has the ephemeris @xmath25 we define @xmath7 = 0.0 at primary minima . the curves at 4050 and 4400 are derived from fast spectra . the zero points for the v and r data are set from fast light curves at 5550 and 6500 using offsets derived from the jacoby , hunter & christian ( 1984 ) spectral atlas and published broadband magnitudes , v = @xmath26 + 0.16 and r@xmath1 = @xmath27 . the zero point for the i@xmath1 light curve assumes the secondary is a normal k0 i star reddened by @xmath28 1.65 mag ( see below ) , with an intrinsic r@xmath1i@xmath1 = 0.52 ( bessell 1990 ) . the light curves show several clear features . ellipsoidal variations of the secondary star are prominent at vr@xmath1i@xmath1 . the large amplitudes of @xmath0 0.5 mag indicate that the secondary probably fills its roche lobe and that the orbital inclination is close to 90 . the deep primary minimum at short wavelengths shows that the primary star is much hotter than the secondary . the eclipse of the primary is long and flat - bottomed , which implies that the primary is extended . this extended primary is probably responsible for the eclipse of the secondary star visible in the v , r@xmath1 and i@xmath1 filters as a deep secondary minimum at @xmath7 = 0.5 . the absorption line indices shown in figure 3 provide additional support for a hot , extended , primary star in bg gem . due to a decrease in the continuum flux , all of the lines are strongest during the eclipse of the primary . the blue lines strengthen much more than the red lines , indicating that the primary has a blue continuum . the eclipse of the primary covers a significant fraction , @xmath29 , of the orbital period . we can measure the spectral type of the secondary star by examining the absorption lines at @xmath7 = 0.0 , when the primary is in eclipse . based on a comparison of the spectral indices with indices derived from the jacoby ( 1984 ) and worthey ( 1994 ) spectral atlases , all of the absorption lines imply a late g or early k spectral type . our best estimate is k0 i , with an uncertainty of 12 subclasses . the supergiant classification is required for the combination of strong mg i , na i , and fe i absorption features . this result agrees with the low gravity implied for a lobe - filling secondary . the balmer emission line equivalent widths also vary with phase ( figure 4 ) . the h@xmath11 and h@xmath12 lines are eclipsed at primary minimum . the eclipses are long and never reach totality . both lines strengthen relative to the continuum during secondary eclipses , because the flux from the k0 i secondary weakens considerably . similar variations may be visible in he i @xmath106678 , but the line is weak with a typical equivalent width of 0.5 or less . other he i lines are too difficult to measure accurately : @xmath105876 is blended with na i and @xmath107065 is very weak . the light curves constrain the sizes of the h ii emission region and the primary and secondary stars . for an edge - on system with a circular orbit , the eclipse duration yields the sum of the radii of the two components ; the duration of totality establishes the difference of the two radii . in bg gem , the length of the eclipses at 4400 and 4050 results in @xmath30 0.37 @xmath31 , where @xmath31 is the orbital semi - major axis , @xmath32 is the radius of the primary star , and @xmath33 is the radius of the k0 i secondary star . from the length of totality , @xmath34 0.12 @xmath31 . these two constraints set @xmath35 0.12 @xmath31 and @xmath36 0.25 @xmath31 . the apparent length of totality at @xmath7 = 0.5 in the r@xmath1 and i@xmath1 filters suggests a somewhat larger radius for the primary at long wavelengths . the length of the h@xmath11 eclipse implies @xmath37 0.55 @xmath31 , which results in @xmath38 0.3 @xmath31 . the lack of a total eclipse for h@xmath11 confirms that the h ii region is larger than the k0 i secondary . to derive absolute constraints on the binary components , we need an independent measure of @xmath31 . we now consider radial velocity data to establish the orbital parameters of bg gem . we derived radial velocities from the strong absorption and emission lines on fast spectra . for absorption line velocities , we cross - correlated the fast spectra against the best - exposed spectrum , where the velocity is set by cross - correlation against standard stars with known velocities ( see tonry & davis 1979 ; kurtz & mink 1998 ) . to avoid contamination from the hot primary , we restricted the cross - correlation to @xmath3950006800 . we measured emission line velocities from cross - correlations with an emission - line template , as described by kurtz & mink ( 1998 ) . we adopted the velocity of h@xmath12 as the emission line velocity , because h@xmath11 may be blended with [ n ii ] emission on our low resolution spectra . we estimate errors of @xmath615 for absorption lines and @xmath620 for emission lines . table 5 lists the measured velocities as a function of jd and @xmath7 . figure 5 shows the absorption line radial velocities as a function of photometric phase . we analyzed these observations using monet s ( 1979 ) fourier transform algorithm ( see also kenyon & garcia 1986 ) . the best spectroscopic period , @xmath40 , agrees with the photometric period . a circular orbit with @xmath41 from equation ( 1 ) fits the orbit well . this solution has an orbital semi - amplitude , @xmath42 , and a fractional semi - major axis , @xmath43 au . spectroscopic conjunctions occur 1.234 days prior to photometric minima , @xmath44 this phase difference has a significance of only 1.3@xmath5 . the mass function is @xmath45 . if we adopt sin @xmath46 = 1 , this result becomes @xmath47 , where @xmath48 is the mass ratio . solutions with eccentric orbits do not improve the fit to the data . iterations in fourier and configuration space yield @xmath49 . the lucy & sweeney ( 1971 ) test confirms that the non - zero @xmath50 has only 1.5@xmath5 significance . we thus prefer the circular solution with the parameters quoted above . these results place a lower limit on the mass of the primary , @xmath51 4 , if the k0 i secondary has negligible mass and sin @xmath46 = 1 . the emission line velocities and double - peaked profiles indicate that the emission lines form in a circumstellar disk surrounding the primary star ( figure 6 ) . the h@xmath12 line has a nearly constant velocity outside primary eclipse , with a mean velocity of + 33.8 @xmath6 16.3 . this velocity is identical to the systemic velocity @xmath52 = 28.0 @xmath6 3.2 . the line velocity increases as the eclipse of the blue continuum begins . the velocity reaches @xmath53 at @xmath7 = @xmath160.05 to @xmath160.04 , decreases to the systemic velocity at @xmath7 = 0.00 , falls to @xmath54 at @xmath7 = 0.04 to 0.05 , and returns to the systemic velocity as the eclipse of the blue continuum ends . this classical disturbance the rossiter effect is consistent with the eclipse of a rapidly rotating disk . the blue - shifted half of the disk is eclipsed first during ingress and is revealed first during egress . both h@xmath11 and h@xmath12 decrease in full - width at half maximum during eclipse , supporting this interpretation . if we associate the apparent rotational velocity of @xmath0 125150 with disk rotation , the radius of the disk is @xmath38 3550 @xmath55 . together with our limit on @xmath56 from the eclipse duration , this result leads to a rough estimate for the mass ratio , @xmath57 0.1 , and the semi - major axis , @xmath58 150 . the small mass ratio supports our conclusion that the k0 supergiant fills its roche lobe . the effective radius of the roche lobe for @xmath59 is @xmath60 , indistinguishable from our estimate for the radius of the secondary , @xmath61 . the effective radius of the roche lobe for the primary star is @xmath62 @xmath63 @xmath64 . the radius of the h@xmath11 emission region is thus roughly half of the tidal radius . the radius of the primary at blue wavelengths is only @xmath0 20% of the tidal radius , but large compared to the radius of a normal main sequence star . these results suggest a simple dynamical model for bg gem . the system consists of a lobe - filling k0 supergiant which transfers material into a large luminous accretion disk surrounding a primary star with @xmath65 45 . the disk must have a blue continuous spectrum to explain the deep primary eclipses at @xmath66 4500 . this disk is also responsible for the secondary eclipses at longer wavelengths . to refine this model , we now consider the light curve and spectra in more detail . we begin with an analysis of the light curve using the wilson - devinney ( 1971 ) code . improved constraints on the orbital parameters and on the physical characteristics of the binary components lead to better limits on the nature of the accretion disk . we then use the reddening derived from the balmer lines and the optical continuum to set the distance to the system . wilson & devinney ( 1971 ) developed a modern light curve synthesis program to derive physical characteristics of binary stars from multi - color light curves . wilson ( 1990 , 1994 , and references therein ) describes recent improvements to the code , which is now available by anonymous ftp . the code consists of two programs : lc , used to compute model light curves or radial velocity curves as a function of various input parameters ; and dc , which can be iterated to derive the best set of binary parameters from a least squares fit of the model to actual observations . the analysis of 3.1 and 3.2 demonstrates that the primary component of bg gem is an extended disk surrounding a massive star . we view the system at close to 90 . few light curve synthesis programs include radiation from a disk ; none reliably compute fluxes for an edge - on disk . therefore , for this study , we used lc to derive reliable ranges for binary parameters that produce light curves similar to those in figures 24 , but omitted the iteration of these parameters with dc . figure 7 shows light curves from figure 2 along with model light curves . to determine the best fit binary system ( dashed curves ) , we varied several parameters to derive a ` best ' approximation to the observations , ( i ) the mass ratio , @xmath67 ; ( ii ) the inclination , @xmath46 ; and ( iii ) the brightness of the primary star relative to a lobe - filling secondary with an effective temperature , 4500 k , appropriate for a k0 supergiant star . the mass ratio and inclination set the amplitude of the ellipsoidal variations ; the ellipsoidal amplitude grows with sin @xmath46 and inversely with @xmath67 . the relative brightness of the primary sets the depth of the primary eclipse and is also constrained by the variation of absorption line indices with wavelength . the dashed light curve models in figure 7 have several successes and failures . the models match the depth of primary eclipse at each wavelength and the overall amplitude and shape of the light curves at @xmath68 0.050.40 and @xmath68 0.600.95 . the length of the model eclipse at @xmath7 = 0 is too short , because the model primary is a star instead of a disk . the secondary minimum at @xmath7 = 0.5 is too shallow , because the small primary star can not eclipse very much of the lobe - filling k0 i secondary . the primary star must be hot and small to match the depths of primary eclipse at 40009000 . to make a more realistic light curve synthesis , we considered a crude disk model for the primary star ( solid curves in figure 7 ) . we assumed a disk with a rectangular cross - section as viewed in the orbital plane . this disk has height @xmath69 above the orbital plane and radius @xmath70 in the orbital plane . we assumed @xmath70 = @xmath71 and @xmath69 = 0.050.1 @xmath70 , where @xmath72 is the effective radius of the roche lobe of the primary . we then took the output of lc the fluxes of the primary and secondary as a function of @xmath7 and constructed new model light curves . we computed ( i ) the fraction of the disk eclipsed by the k0 secondary star at @xmath73 , and ( ii ) the fraction of the k0 secondary eclipsed by the disk at @xmath74 . we accurately accounted for roche geometry and assumed that both light sources are completely opaque and uniformly illuminated . this approach ignored limb and gravity darkening of the secondary , but should yield light curves closer to those actually observed . the simple disk - shaped primary can explain the depths and shape of both primary and secondary eclipses , as indicated by the solid lines in figure 7 . models with @xmath69 = 0.07 @xmath70 , @xmath75 , and @xmath76 0.80.9 yield the ` best ' fit to the data . less extended disks do not eclipse enough of the secondary at @xmath77 . more extended disks occult too much of the secondary . the large variation in disk radius with orbital phase suggests that the opaque , outer part of the disk is much cooler than the inner part of the disk . our model is too simple to constrain either the size of the disk as a function of wavelength or the brightness temperature distribution within the disk , as in vrielmann , horne , & hessman ( 1999 ) for example . we plan more detailed analyses in future studies . the models shown in figure 7 assume @xmath78 , @xmath79 , and a primary star with an effective temperature of 10,000 k. primary stars with different effective temperatures do not change the model light curves as long as the ratio of the luminosity of the primary star to the luminosity of the secondary star at each wavelength remains fixed . these ratios are set by the eclipse depth and can not vary by much more than 10% . for fixed inclination , smaller mass ratios produce shallower secondary mimina . the disk radius also shrinks as @xmath67 increases , which leads to a smaller `` disk eclipse '' at secondary minimum . we estimate @xmath59 for sin @xmath46 = 1 . any geometry with @xmath80 80 does not produce a primary eclipse and is thus ruled out by the data . improved constraints on these values using our data requires a wd - type code that includes a reliable model for an edge - on accretion disk . to derive better limits on the nature of the primary star , we need good estimates for the reddening and distance in addition to the parameters derived above . with @xmath81 and @xmath82 , bg gem lies close to the galactic plane in the direction of the galactic anti - center . previous extinction surveys suggest modest visual extinctions , @xmath83 12 mag , for distances of 15 kpc ( e.g. , neckel & klare 1980 ; hakkila 1997 ) . adopting @xmath84 = 1.5 @xmath6 0.5 mag yields @xmath85 = 11.5 @xmath6 0.5 for the system at maximum and @xmath86 kpc for a k0 i star with @xmath87 = @xmath161.9 from the lc program and a bolometric correction of 1.5 mag . we can place better limits on the extinction from near - ir photometry and the emission line data . a k0 supergiant should have v j = 1.51.6 and v k = 2.12.3 . data from 2mass j = 10.32 @xmath6 0.03 , h = 9.61 @xmath6 0.03 , and k = 9.34 @xmath6 0.03 and v = 13.0 at maximum imply @xmath88 for a normal extinction law ( e.g. , rieke & lebofsky 1985 ; bessell & brett 1988 ) . the mean intensity ratio for h@xmath11 and h@xmath12 is @xmath89(h@xmath11)/@xmath89(h@xmath12 ) = 5.81 @xmath6 1.01 ; this ratio suggests @xmath90 for a standard extinction law ( mathis 1990 ) . we thus favor @xmath88 for the extinction and @xmath91 kpc . this combination agrees with the distance - dependent extinction measurements summarized in hakkila ( 1997 ) . figure 8 shows the observed spectral energy distribution for bg gem . filled points indicate observations dereddened by @xmath92 . the k0 supergiant produces most of the optical and near - infrared radiation , as indicated by the solid and dot - dashed curves . the primary source contributes a modest fraction of the v light . its fraction of the total light grows substantially towards shorter wavelengths , as shown by the dashed curve . the slope of the hot continuum , @xmath11 = ( d log @xmath17/d log @xmath10 ) = @xmath93 , is best - fit by the spectrum of an accretion disk but is marginally consistent with the spectrum of an early type star . the measured extent of the primary , @xmath94 , is much larger than a normal early - type main sequence star . the predicted v magnitude of a spherical star with @xmath95 k and @xmath96 is roughly 23 mag _ brighter _ than the k0 secondary . our observed limit on the v brightness of the primary , @xmath0 10% of the k0 secondary , demonstrates that the primary is not a star . however , a flattened disk with @xmath97 0.050.10 satisfies the limits on @xmath32 and the total brightness derived from the light curves . constructing a ` best ' model for the disk in bg gem requires ultraviolet or x - ray data to provide better constraints on the spectrum and on the radius of the primary star . in this paper , we simply develop a model consistent with the observed spectrum and masses without considering whether it is ` best ' or unique . we first examine the case where the disk surrounds a hot primary star with parameters defined in 3.3 . the primary star either has a low luminosity compared to the disk or is occulted by the disk . otherwise , our model can not account for the primary eclipse as outlined above . requiring a self - luminous disk to produce all of the optical flux yields a large accretion rate , @xmath98 , which is ruled out by the weak h i emission and lack of high ionization lines , such as he ii @xmath104686 , on the optical spectrum ( see kenyon & webbink 1984 ) . a ` reprocessing ' disk which absorbs and reradiates light from the primary can have a spectrum similar to a viscous accretion disk , because the temperature distributions are similar ( kenyon & hartmann 1987 ) . tests of several purely reprocessing disk models surrounding b - type stars indicate , however , that the half - light radius of the disk is @xmath0 10 , instead of the @xmath0 20 derived from the optical eclipses . we thus consider models where accretion and reprocessing contribute comparable amounts to the observed spectrum . the light solid line in figure 8 plots the spectrum from an accretion disk that satisfies several observational constraints . the disk surrounds a b3 main sequence star . a large inclination , @xmath99 85 , allows the central star to hide behind the outer parts of the disk . the disk may also shield the k0 star from the radiation field of the b star . an accretion rate of a few @xmath100 provides roughly half of the observed continuum radiation . the model predicts a disk temperature @xmath101 k at a disk radius @xmath0 20 and a temperature @xmath0 3000 k at @xmath0 8090 . these results are consistent with eclipse data which require a small hot primary at short wavelengths and a larger , cooler primary at long wavelengths ( 3.3 ) . cool material in the outer disk may also produce the extra mg i absorption implied by the apparent rise in the mg i index at secondary minimum ( fig . the uncertainties in this model are modest if we require the primary to lie on the main sequence . stars much cooler than @xmath0 15,000 k ( b6 v ) produce disks that are too cold to match the optical spectrum . stars much hotter than @xmath0 30,000 k ( b0 ) should produce substantial he i emission ( from ionization of the disk ) that is not observed . this model has one main failure . it is difficult for a b3 star to produce enough photons to ionize h to the observed level . we estimate an extinction - corrected h@xmath12 luminosity of @xmath102 , which requires @xmath103 h - ionizing photons s@xmath8 if the b star ionizes material in the disk . this limit is 12 orders of magnitude larger than expected from a b3 v star . a boundary layer at the inner edge of the disk probably also produces too few ionizing photons , unless we have underestimated the accretion rate ( see , for example , @xcite ) . olson ( 1991 ) considered h@xmath11 line formation within the disks of several algol systems , and produced line profiles similar to those observed . it may be possible to account for the line flux in a similar way , but such a calculation is beyond the scope of this paper . it is unlikely that any b - type star in bg gem could be underluminous compared to a normal main sequence star . main sequence stars accreting material from a disk are at least as luminous as non - accreting main sequence stars of the same mass ( e.g. , kippenhahn & meyer - hofmeister 1977 ) . the best alternative to this model is a black hole primary . although less likely than a b - type primary star , a black hole accreting at a small fraction of the eddington limit might account for the observed radiation at 20 ( depending on the geometry ) . to evaluate the plausibility of this model , we considered a disk surrounding a 4.5 black hole . the thick , inner disk produces a luminosity of @xmath104 if the accretion rate is 1% of the eddington limit . we naively assumed that the inner disk emits x - rays isotropically ; the outer disk absorbs and re - radiates this emission at the local blackbody temperature . for a steady - state disk where @xmath105 , the disk temperature at 420 is @xmath0 50008000 k. the predicted luminosity of the outer disk in this model , @xmath106 , is close to our estimated disk luminosity of 510 if the inclination is @xmath0 85 . our simple example suggests that a black hole accretion disk can plausibly produce the disk emission observed in bg gem for accretion rates much smaller than the eddington limit . the model optimistically assumes that the inner disk radiates efficiently , @xmath0 10% , compared to advection - dominated accretion flows ( e.g. , @xcite ) and that the outer disk radiates as a blackbody . despite this optimism , we note that esin ( 1997 ) derive optical luminosities , @xmath107 , of @xmath0 @xmath108@xmath109 erg s@xmath8 for quiescent disks and @xmath109@xmath110 erg s@xmath8 for low state disks in black hole binaries with inclinations of 6090 . the compact disks assumed in the esin models are more than factor of 100 smaller than the disk in bg gem , so it may be possible to achieve the observed optical flux from a larger disk surrounding a black hole in a wider binary system . the black hole model has one main advantage and one main disadvantage compared to models where the primary is a b - type main sequence star . a black hole accreting at 1% of the eddington limit easily produces enough high energy photons to satisfy our constraints on the h@xmath11 and h@xmath12 fluxes unless the accretion efficiency is much less than 1% ( esin 1997 ) . but , hiding an accreting black hole in an edge - on binary may be more difficult than hiding a b - type main sequence star . the x - ray luminosity , @xmath0 1300 , needed to power the observed optical luminosity in bg gem is roughly 3 orders of magnitude larger than the upper limit derived from the asm observations , @xmath0 2 . the x - ray luminosities of quiescent black hole binary systems are also low , @xmath111 ( e.g. , @xcite ) . however , the region that emits x - rays in a black hole binary is very small , @xmath112 to @xmath113 cm , and easily occulted when the binary is viewed at high inclination . for example , esin ( 1997 ) show that the x - ray luminosity of the small disk in an edge - on black hole binary system is several orders of magnitude smaller than the same disk viewed pole - on . we expect that the larger disk in bg gem can occult a larger fraction of the x - ray flux than an edge - on disk in a smaller binary system . future calculations of black hole binaries should address this issue . our results indicate that bg gem is a textbook example of a very evolved , semi - detached binary system . the system consists of a low mass , lobe - filling k0 supergiant which transfers material into a disk surrounding a massive primary star . the large ellipsoidal light variations of the k0 secondary indicate a large mass ratio , @xmath114 0.1 , and large orbital inclination , @xmath115 . the radial velocities of h@xmath11 emission in the disk support the large mass ratio . we prefer @xmath116 to allow the disk to occult the primary star , which is not visible in the optical . the system is remarkably stable for an interacting binary . there is little scatter in either the optical brightness ( for 25 orbital cycles ) or the emission line equivalent widths ( for 34 orbital cycles ) . studies of archival plate material would place useful constraints on long - term variability and outburst frequency . our data are insufficient to choose between the two possible alternatives for the primary star . the optical spectra resemble spectra of several algol systems e.g. , uu cnc ( p = 96.7 days , @xcite ) and rz oph ( p = 262 days , @xcite ; @xcite ; @xcite ) as well as quiescent spectra of several black hole binaries e.g. , v616 mon ( p = 0.32 days , @xcite and references therein ) and v404 cyg ( p = 6.5 days , @xcite and references therein ) . none of the long - period algol binaries have as extreme mass ratios as in bg gem ( see @xcite ) , although rz oph has @xmath117 . the structure of the binary in rz oph has several other features in common with bg gem , including a lobe - filling k - type star , an invisible primary , and an extended accretion disk that occults the secondary star at @xmath7 = 0.5 ( see olson 1987 ) . however , most of the energy from the disk in rz oph is produced by accretion , because the disk in rz oph is brighter and less stable than the disk in bg gem . producing a large h@xmath11 flux from the intrinsically less luminous disk in bg gem is also a challenge , as noted above . bg gem has many features in common with quiescent black hole binaries , despite the 12 order of magnitude difference in orbital periods . many black hole binaries have ellipsoidal light curves and extreme mass ratios similar to those in bg gem ( e.g. , @xcite ; @xcite ; @xcite ) . quiescent black hole binaries often have modest emission line spectra , with strong , double - peaked h i lines and weaker , double - peaked he i lines ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ) . higher ionization emission lines , such as he ii @xmath104686 , appear only in outburst , when the accretion rate through the disk increases dramatically . quiescent black hole binaries often escape detection as very intense x - ray sources until they undergo major eruptions , and they often remain in quiescence for decades ( e.g. , @xcite ; @xcite ) . all of the currently known black hole binary systems have small orbital inclinations @xmath118 ( see @xcite ; @xcite ; @xcite ) . the @xmath116 derived here for bg gem may allow the disk to occult even an intense x - ray source ( see @xcite , 1999b , and references therein for typical outburst fluxes ) . despite the lack of x - rays , bg gem may still contain a black hole primary surrounded by a quiescent accretion disk . if confirmed , bg gem would be the black hole binary with the longest known orbital period . other black hole binaries have periods of a few days or less . any black hole binary is a challenge to modern theories of the evolution of binary systems ( e.g. , kalogera 1999 ) . bg gem would provide a severe test of these theories . bg gem would also be the only eclipsing system known among black hole binaries . the black hole mass is unambiguous in an eclipsing binary . our limit for the primary mass is close to the maximum possible neutron star mass ( see , for example , burrows 1998 ) . bg gem thus may place new constraints on the minimum black hole mass . with these considerations in mind , it is clear that additional observations are needed to understand the nature of the primary in bg gem . ultraviolet spectra at 10003000 would probe the continuum and emission lines produced in the inner portions of the accretion disk and might reveal stellar absorption lines if the primary is a b - type star . better limits on the x - ray flux from instruments on board _ chandra _ and the _ x - ray multi - mirror _ would also provide better constraints on the nature of the primary . detection of an x - ray source that is eclipsed at primary minima would favor a black hole primary , because b - type main sequence stars are weak x - ray sources . whatever the nature of the primary , future studies of the secondary star and the accretion disk surrounding the primary should yield a better understanding of accretion in wide binary systems . eclipses of the blue continuum and the emission lines provide useful information about disk structure , through maximum entropy reconstruction among other techniques . higher resolution observations of mg i and other absorption features probe conditions in the secondary star as well as the cool , outer portions of the disk . near - infrared observations covering the orbit would improve mass estimates for the secondary star derived from ellipsoidal variations and might reveal emission from an optically thick `` bright spot , '' where material lost by the k0 supergiant impacts the disk . emission from an optically thin spot might be visible in h@xmath11 or h@xmath12 . finally , high resolution observations might yield a radial velocity curve for the primary and improved estimates for the mass ratio . the small distance and reddening make bg gem a prime target for these and other observations of a binary system at an interesting phase of its evolution . we thank p. berlind , m. calkins , and other queue observers at the flwo 60 telescope for acquiring the fast spectra used in this project . susan tokarz made the preliminary reductions of the fast spectra . we are grateful to dan fabricant and his excellent group for designing and building a wonderful optical spectrograph . comments from g. sobczak , j. mcclintock , r. webbink , and an anonymous referee improved our presentation . r. remillard graciously derived the asm upper limit . we also thank the wellesley college undergraduates who obtained the images of bg gem for photometry . we acknowledge summer support for a. b. ( nsf grant ast-9417359 ) and a. d. ( w. m. keck foundation grant to the keck northeast astronomy consortium ) . p. j. b. acknowledges support from nsf grant ast-9417359 and the brachman hoffman grants from wellesley college . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center , funded by the national aeronautics and space administration and the national science foundation . this research also used data through from the high energy astrophysics science archive research center online service , provided by the nasa / goddard space flight center . l c c c c c c c 48694.5086 & 0.205 & @xmath1190.187 & 1.154 & @xmath1190.023 & 1.125 & @xmath160.161 & 1.122 + 48698.5333 & 0.249 & @xmath1190.151 & 1.147 & @xmath1190.059 & 1.115 & @xmath160.172 & 1.122 + 48984.5759 & 0.370 & @xmath1190.354 & 1.133 & @xmath1190.167 & 1.135 & @xmath160.037 & 1.118 + 48990.7176 & 0.437 & @xmath1190.492 & 1.158 & @xmath1190.274 & 1.154 & @xmath1190.036 & 1.141 + 49006.5533 & 0.610 & @xmath1190.373 & 1.165 & @xmath1190.200 & 1.121 & @xmath160.012 & 1.111 + 49044.5032 & 0.024 & @xmath1190.835 & 1.161 & @xmath1190.635 & 1.123 & @xmath1190.435 & 1.130 + 49084.5480 & 0.461 & @xmath1190.566 & 1.145 & @xmath1190.317 & 1.116 & @xmath1190.070 & 1.147 + 49106.5242 & 0.701 & @xmath1190.195 & 1.136 & @xmath1190.008 & 1.142 & @xmath160.180 & 1.134 + 49368.7090 & 0.562 & @xmath1190.477 & 1.169 & @xmath1190.277 & 1.142 & @xmath1190.081 & 1.123 + 49375.6981 & 0.638 & @xmath1190.313 & 1.151 & @xmath1190.110 & 1.132 & @xmath160.069 & 1.125 + 49382.7523 & 0.715 & & & @xmath160.007 & 1.185 & @xmath160.157 & 1.126 + 49390.6297 & 0.801 & @xmath1190.193 & 1.157 & @xmath1190.025 & 1.141 & @xmath160.132 & 1.112 + 49391.5136 & 0.811 & @xmath1190.190 & 1.137 & @xmath1190.002 & 1.143 & @xmath160.171 & 1.127 + 49398.5720 & 0.888 & @xmath1190.373 & 1.160 & @xmath1190.203 & 1.143 & @xmath1190.027 & 1.119 + 49402.7024 & 0.933 & @xmath1190.513 & 1.149 & @xmath1190.371 & 1.111 & @xmath1190.152 & 1.131 + 49406.6050 & 0.975 & @xmath1190.784 & 1.156 & @xmath1190.603 & 1.173 & @xmath1190.411 & 1.125 + 49409.4841 & 0.007 & @xmath1190.879 & 1.162 & @xmath1190.637 & 1.143 & @xmath1190.430 & 1.136 + 49410.6664 & 0.020 & @xmath1190.899 & 1.146 & @xmath1190.649 & 1.139 & @xmath1190.484 & 1.119 + 49411.5224 & 0.029 & @xmath1190.838 & 1.156 & @xmath1190.625 & 1.139 & @xmath1190.396 & 1.155 + 49412.5232 & 0.040 & @xmath1190.810 & 1.154 & @xmath1190.588 & 1.138 & @xmath1190.348 & 1.143 + 49413.5668 & 0.051 & @xmath1190.702 & 1.150 & @xmath1190.462 & 1.122 & @xmath1190.302 & 1.118 + 49416.5775 & 0.084 & @xmath1190.533 & 1.168 & @xmath1190.326 & 1.134 & @xmath1190.116 & 1.118 + 49423.6071 & 0.161 & @xmath1190.283 & 1.147 & @xmath1190.110 & 1.133 & @xmath160.055 & 1.123 + 49435.6265 & 0.292 & & & @xmath1190.034 & 1.164 & @xmath160.115 & 1.120 + 49438.5772 & 0.324 & @xmath1190.271 & 1.138 & @xmath1190.048 & 1.155 & @xmath160.107 & 1.118 + 49442.5804 & 0.368 & @xmath1190.335 & 1.167 & @xmath1190.166 & 1.150 & @xmath1190.009 & 1.116 + 49443.6201 & 0.379 & @xmath1190.377 & 1.156 & @xmath1190.170 & 1.134 & @xmath1190.035 & 1.103 + 49444.5683 & 0.390 & @xmath1190.400 & 1.158 & @xmath1190.213 & 1.147 & @xmath1190.041 & 1.132 + 49447.5587 & 0.422 & @xmath1190.438 & 1.151 & @xmath1190.253 & 1.113 & @xmath1190.047 & 1.132 + 49461.5167 & 0.575 & @xmath1190.408 & 1.171 & @xmath1190.230 & 1.136 & @xmath1190.045 & 1.131 + 49744.6085 & 0.664 & @xmath1190.229 & 1.155 & @xmath1190.049 & 1.134 & @xmath160.099 & 1.114 + 49746.6087 & 0.685 & @xmath1190.182 & 1.169 & @xmath1190.028 & 1.143 & @xmath160.127 & 1.116 + 49747.6180 & 0.697 & @xmath1190.171 & 1.176 & @xmath160.021 & 1.147 & @xmath160.164 & 1.120 + 49748.5698 & 0.707 & @xmath1190.174 & 1.161 & @xmath160.003 & 1.143 & @xmath160.153 & 1.116 + 49751.6107 & 0.740 & @xmath1190.076 & 1.166 & @xmath160.088 & 1.147 & @xmath160.202 & 1.102 + 49752.5963 & 0.751 & @xmath1190.115 & 1.145 & & & @xmath160.197 & 1.114 + 49755.5755 & 0.783 & @xmath1190.134 & 1.163 & @xmath160.021 & 1.135 & @xmath160.186 & 1.111 + 49757.5714 & 0.805 & @xmath1190.165 & 1.158 & @xmath1190.007 & 1.126 & @xmath160.151 & 1.110 + 49761.5820 & 0.849 & @xmath1190.235 & 1.174 & @xmath1190.061 & 1.144 & @xmath160.082 & 1.122 + 49762.5920 & 0.860 & @xmath1190.251 & 1.165 & @xmath1190.080 & 1.126 & @xmath160.056 & 1.112 + 49763.5168 & 0.870 & @xmath1190.255 & 1.169 & @xmath1190.089 & 1.143 & @xmath160.062 & 1.121 + 49766.5277 & 0.903 & @xmath1190.381 & 1.150 & @xmath1190.213 & 1.124 & @xmath1190.040 & 1.125 + 49771.5679 & 0.958 & @xmath1190.703 & 1.170 & @xmath1190.504 & 1.159 & @xmath1190.338 & 1.118 + 49780.6224 & 0.057 & @xmath1190.621 & 1.170 & @xmath1190.410 & 1.160 & @xmath1190.246 & 1.111 + 49781.6495 & 0.068 & @xmath1190.581 & 1.150 & @xmath1190.347 & 1.136 & @xmath1190.160 & 1.124 + 49804.5517 & 0.318 & @xmath1190.256 & 1.159 & @xmath1190.053 & 1.142 & @xmath160.112 & 1.125 + 49810.5121 & 0.383 & @xmath1190.360 & 1.171 & @xmath1190.144 & 1.154 & @xmath160.013 & 1.117 + 49815.5164 & 0.437 & @xmath1190.455 & 1.169 & @xmath1190.223 & 1.149 & @xmath1190.056 & 1.109 + l c c c c c c c 49831.5523 & 0.612 & @xmath1190.381 & 1.162 & @xmath1190.158 & 1.144 & @xmath1190.001 & 1.117 + 50114.6766 & 0.702 & @xmath1190.141 & 1.155 & @xmath160.018 & 1.152 & @xmath160.162 & 1.119 + 50118.6449 & 0.745 & @xmath1190.111 & 1.172 & @xmath160.045 & 1.154 & @xmath160.199 & 1.131 + 50120.6499 & 0.767 & @xmath1190.133 & 1.157 & @xmath160.023 & 1.142 & & + 50131.5678 & 0.886 & @xmath1190.310 & 1.162 & @xmath1190.145 & 1.141 & @xmath160.022 & 1.120 + 50132.5850 & 0.897 & @xmath1190.342 & 1.165 & @xmath1190.186 & 1.140 & @xmath1190.016 & 1.127 + 50139.6648 & 0.974 & @xmath1190.702 & 1.172 & @xmath1190.523 & 1.148 & @xmath1190.353 & 1.126 + 50140.5501 & 0.984 & @xmath1190.736 & 1.157 & @xmath1190.548 & 1.140 & @xmath1190.383 & 1.116 + 50143.5953 & 0.017 & @xmath1190.777 & 1.171 & @xmath1190.587 & 1.149 & @xmath1190.403 & 1.137 + 50146.6049 & 0.050 & @xmath1190.646 & 1.161 & @xmath1190.456 & 1.140 & @xmath1190.266 & 1.124 + 50147.5501 & 0.060 & @xmath1190.602 & 1.174 & @xmath1190.394 & 1.139 & @xmath1190.207 & 1.121 + 50153.5759 & 0.126 & @xmath1190.317 & 1.161 & @xmath1190.136 & 1.139 & @xmath160.046 & 1.116 + 50160.5511 & 0.202 & @xmath1190.128 & 1.179 & @xmath160.034 & 1.156 & @xmath160.198 & 1.132 + 50161.5227 & 0.213 & @xmath1190.142 & 1.170 & @xmath160.025 & 1.157 & @xmath160.186 & 1.126 + 50462.6408 & 0.499 & @xmath1190.633 & 1.166 & @xmath1190.392 & 1.153 & @xmath1190.185 & 1.121 + 50463.6463 & 0.509 & @xmath1190.620 & 1.167 & @xmath1190.398 & 1.142 & @xmath1190.178 & 1.125 + 50472.6658 & 0.608 & @xmath1190.369 & 1.155 & @xmath1190.160 & 1.140 & @xmath160.015 & 1.109 + 50489.5630 & 0.792 & @xmath1190.137 & 1.147 & @xmath160.013 & 1.138 & @xmath160.175 & 1.121 + 50490.6027 & 0.804 & @xmath1190.136 & 1.161 & & & & + 50504.5811 & 0.956 & @xmath1190.679 & 1.167 & @xmath1190.494 & 1.120 & @xmath1190.288 & 1.162 + 50523.5677 & 0.163 & @xmath1190.241 & 1.124 & @xmath1190.107 & 1.101 & @xmath160.117 & 1.116 + 50782.8269 & 0.992 & @xmath1190.792 & 1.158 & @xmath1190.603 & 1.138 & @xmath1190.431 & 1.122 + 50828.6566 & 0.492 & @xmath1190.644 & 1.153 & @xmath1190.398 & 1.134 & @xmath1190.170 & 1.125 + 50836.5599 & 0.579 & @xmath1190.374 & 1.159 & @xmath1190.177 & 1.143 & @xmath1190.015 & 1.132 + 50839.5264 & 0.611 & @xmath1190.268 & 1.144 & @xmath1190.140 & 1.118 & @xmath160.045 & 1.107 + 50840.5498 & 0.622 & @xmath1190.329 & 1.164 & @xmath1190.125 & 1.124 & @xmath160.069 & 1.122 + 50858.6164 & 0.819 & @xmath1190.163 & 1.164 & @xmath1190.022 & 1.154 & @xmath160.163 & 1.131 + 50860.6880 & 0.842 & @xmath1190.224 & 1.158 & @xmath1190.064 & 1.140 & @xmath160.085 & 1.099 + 50883.6207 & 0.092 & @xmath1190.407 & 1.144 & @xmath1190.218 & 1.131 & @xmath1190.052 & 1.132 + 50897.5577 & 0.244 & @xmath1190.156 & 1.163 & @xmath160.008 & 1.131 & @xmath160.165 & 1.124 + 50898.5383 & 0.255 & & & @xmath160.016 & 1.134 & @xmath160.168 & 1.119 + 50903.5061 & 0.309 & @xmath1190.173 & 1.151 & @xmath1190.005 & 1.149 & @xmath160.149 & 1.134 + 50914.5891 & 0.430 & @xmath1190.432 & 1.161 & @xmath1190.221 & 1.147 & @xmath1190.036 & 1.148 + 50915.5573 & 0.441 & @xmath1190.453 & 1.168 & @xmath1190.247 & 1.145 & @xmath1190.057 & 1.126 + 50931.5498 & 0.615 & & & @xmath1190.077 & 1.175 & @xmath160.077 & 1.142 + 50932.5137 & 0.626 & @xmath1190.251 & 1.168 & @xmath1190.068 & 1.158 & @xmath160.095 & 1.122 + 51210.6585 & 0.661 & @xmath1190.262 & 1.142 & @xmath1190.059 & 1.137 & @xmath160.116 & 1.136 + 51219.5829 & 0.758 & @xmath1190.145 & 1.159 & @xmath160.017 & 1.143 & @xmath160.179 & 1.121 + 51225.5590 & 0.823 & @xmath1190.192 & 1.157 & @xmath1190.024 & 1.147 & @xmath160.136 & 1.123 + 51226.6063 & 0.835 & @xmath1190.213 & 1.166 & @xmath1190.050 & 1.147 & @xmath160.097 & 1.121 + 51233.5798 & 0.911 & @xmath1190.426 & 1.167 & @xmath1190.265 & 1.133 & @xmath1190.089 & 1.122 + 51240.5426 & 0.987 & @xmath1190.844 & 1.151 & @xmath1190.626 & 1.140 & @xmath1190.444 & 1.107 + 51248.5614 & 0.074 & @xmath1190.487 & 1.135 & @xmath1190.245 & 1.145 & @xmath1190.019 & 1.111 + 51255.5686 & 0.151 & @xmath1190.381 & 1.140 & @xmath1190.195 & 1.154 & @xmath160.028 & 1.128 + 51263.4974 & 0.237 & @xmath1190.206 & 1.165 & @xmath1190.001 & 1.144 & @xmath160.174 & 1.128 + 51269.5027 & 0.303 & @xmath1190.250 & 1.150 & @xmath1190.064 & 1.129 & @xmath160.139 & 1.125 + 51273.5037 & 0.346 & @xmath1190.309 & 1.158 & @xmath1190.113 & 1.148 & @xmath160.059 & 1.131 + 51277.6396 & 0.392 & @xmath1190.309 & 1.192 & @xmath1190.107 & 1.171 & @xmath160.033 & 1.133 + l c c c c c he i & i@xmath120 & 6678 & 6600 & 6750 & 30 + h@xmath11 & i@xmath121 & 6563 & 6525 & 6600 & 30 + ba i & i@xmath122 & 6495 & 6475 & 6525 & 25 + na i & i@xmath123 & 5893 & 5825 & 5965 & 30 + mg i & i@xmath124 & 5175 & 5050 & 5300 & 30 + h@xmath12 & i@xmath125 & 4861 & 4825 & 4900 & 30 + fe i & i@xmath126 & 4392 & 4364 & 4448 & 35 + + 7025 cont & @xmath127 & 7025 & & & 30 + 6370 cont & @xmath128 & 6370 & & & 30 + 5550 cont & @xmath26 & 5550 & & & 30 + 4400 cont & @xmath129 & 4400 & & & 30 + 4050 cont & @xmath130 & 4050 & & & 30 + l c c c c c c c 51070.0128 & 0.626 & 0.083 & 0.144 & 0.176 & 0.045 & @xmath168.5 & @xmath1627.7 + 51070.0150 & 0.626 & 0.103 & 0.150 & 0.182 & 0.041 & @xmath169.1 & @xmath1628.3 + 51086.0142 & 0.801 & 0.107 & 0.154 & 0.179 & 0.040 & @xmath167.9 & @xmath1625.8 + 51101.0046 & 0.964 & 0.130 & 0.179 & 0.177 & 0.052 & @xmath165.8 & @xmath1620.1 + 51102.9545 & 0.985 & 0.148 & 0.205 & 0.177 & 0.047 & @xmath162.6 & @xmath1616.4 + 51109.9734 & 0.062 & 0.115 & 0.158 & 0.174 & 0.040 & @xmath169.5 & @xmath1627.4 + 51115.9804 & 0.128 & 0.092 & 0.151 & 0.174 & 0.045 & @xmath168.6 & @xmath1628.1 + 51129.9315 & 0.280 & 0.107 & 0.141 & 0.149 & 0.035 & @xmath166.6 & @xmath1623.8 + 51132.9985 & 0.313 & 0.124 & 0.143 & 0.164 & 0.037 & @xmath167.0 & @xmath1625.8 + 51135.9320 & 0.345 & 0.110 & 0.147 & 0.155 & 0.038 & @xmath167.0 & @xmath1626.4 + 51138.9114 & 0.378 & 0.115 & 0.151 & 0.154 & 0.041 & @xmath167.6 & @xmath1627.9 + 51141.9962 & 0.411 & 0.112 & 0.151 & 0.169 & 0.047 & @xmath169.5 & @xmath1629.9 + 51144.8588 & 0.443 & 0.104 & 0.186 & 0.184 & 0.040 & @xmath1610.5 & @xmath1635.5 + 51147.9806 & 0.477 & 0.111 & 0.170 & 0.197 & 0.042 & @xmath1613.0 & @xmath1640.7 + 51157.7598 & 0.583 & 0.097 & 0.158 & 0.177 & 0.049 & @xmath1610.7 & @xmath1632.0 + 51160.8444 & 0.617 & 0.106 & 0.144 & 0.160 & 0.042 & @xmath169.2 & @xmath1629.1 + 51163.9600 & 0.651 & 0.096 & 0.139 & 0.156 & 0.029 & @xmath168.7 & @xmath1628.2 + 51166.8621 & 0.683 & 0.102 & 0.154 & 0.159 & 0.048 & @xmath167.8 & @xmath1627.0 + 51169.8996 & 0.716 & 0.103 & 0.150 & 0.160 & 0.045 & @xmath167.0 & @xmath1625.2 + 51172.8716 & 0.748 & 0.109 & 0.151 & 0.163 & 0.047 & @xmath167.6 & @xmath1626.0 + 51175.9043 & 0.781 & 0.092 & 0.143 & 0.154 & 0.043 & @xmath168.3 & @xmath1627.9 + 51187.7566 & 0.911 & 0.094 & 0.164 & 0.157 & 0.048 & @xmath1610.7 & @xmath1629.4 + 51188.7424 & 0.922 & 0.121 & 0.161 & 0.165 & 0.044 & @xmath169.4 & @xmath1627.9 + 51191.7695 & 0.955 & 0.140 & 0.176 & 0.160 & 0.051 & @xmath166.4 & @xmath1621.1 + 51194.8215 & 0.988 & 0.135 & 0.201 & 0.183 & 0.060 & @xmath162.0 & @xmath1615.3 + 51196.7938 & 0.009 & 0.143 & 0.187 & 0.187 & 0.049 & @xmath162.9 & @xmath1615.9 + 51198.7117 & 0.030 & 0.129 & 0.176 & 0.174 & 0.052 & @xmath165.1 & @xmath1618.9 + 51201.7889 & 0.064 & 0.110 & 0.160 & 0.151 & 0.046 & @xmath169.8 & @xmath1627.6 + 51216.6831 & 0.226 & 0.108 & 0.129 & 0.148 & 0.049 & @xmath167.3 & @xmath1624.6 + 51218.6966 & 0.248 & 0.108 & 0.125 & 0.153 & 0.046 & @xmath166.9 & @xmath1624.3 + 51223.5895 & 0.302 & 0.100 & 0.141 & 0.157 & 0.043 & @xmath167.1 & @xmath1625.8 + 51224.6268 & 0.313 & 0.106 & 0.129 & 0.151 & 0.042 & @xmath167.0 & @xmath1625.7 + 51226.7300 & 0.336 & 0.105 & 0.122 & 0.145 & 0.035 & @xmath166.8 & @xmath1629.7 + 51228.6119 & 0.357 & 0.083 & 0.144 & 0.127 & 0.043 & @xmath167.2 & @xmath1628.1 + 51230.7042 & 0.379 & 0.114 & 0.130 & 0.155 & 0.041 & @xmath168.3 & @xmath1629.1 + 51232.7299 & 0.402 & 0.101 & 0.131 & 0.154 & 0.048 & @xmath168.8 & @xmath1630.6 + 51247.7040 & 0.565 & 0.100 & 0.160 & 0.170 & 0.044 & @xmath1612.8 & @xmath1636.0 + 51249.5973 & 0.586 & 0.100 & 0.150 & 0.160 & 0.043 & @xmath1611.0 & @xmath1633.2 + 51257.6133 & 0.673 & 0.105 & 0.134 & 0.156 & 0.046 & @xmath168.6 & @xmath1627.5 + 51259.6068 & 0.695 & 0.102 & 0.153 & 0.159 & 0.043 & @xmath168.0 & @xmath1627.2 + 51278.6080 & 0.902 & 0.122 & 0.143 & 0.146 & 0.051 & @xmath169.1 & @xmath1628.2 + 51280.6209 & 0.924 & 0.116 & 0.151 & 0.163 & 0.052 & @xmath168.7 & @xmath1625.7 + 51284.6181 & 0.968 & 0.139 & 0.169 & 0.168 & 0.040 & @xmath163.3 & @xmath1618.1 + 51287.6168 & 0.000 & 0.143 & 0.198 & 0.176 & 0.055 & @xmath161.7 & @xmath1614.0 + 51289.6136 & 0.022 & 0.126 & 0.188 & 0.187 & 0.049 & @xmath164.1 & @xmath1617.6 + 51291.6213 & 0.044 & 0.122 & 0.173 & 0.173 & 0.049 & @xmath168.3 & @xmath1622.8 + l c c c c c c 51101.0046 & 0.964 & 15.19 & 14.40 & 13.33 & 13.15 & 13.07 + 51102.9545 & 0.985 & 15.34 & 14.59 & 13.47 & 13.35 & 13.20 + 51109.9734 & 0.062 & 14.40 & 14.05 & 13.05 & 12.87 & 12.82 + 51115.9804 & 0.128 & 14.35 & 14.00 & 13.06 & 12.89 & 12.80 + 51135.9320 & 0.345 & 14.31 & 13.99 & 13.06 & 12.89 & 12.87 + 51141.9962 & 0.411 & 14.29 & 14.06 & 13.18 & 13.05 & 13.07 + 51157.7598 & 0.583 & 14.57 & 14.17 & 13.13 & 12.94 & 12.90 + 51160.8444 & 0.617 & 14.43 & 14.07 & 13.09 & 12.94 & 12.93 + 51172.8716 & 0.748 & 14.10 & 13.75 & 12.83 & 12.69 & 12.65 + 51175.9043 & 0.781 & 14.21 & 13.86 & 12.91 & 12.76 & 12.74 + 51188.7424 & 0.922 & 14.44 & 14.05 & 13.15 & 13.08 & 13.07 + 51191.7695 & 0.955 & 15.02 & 14.55 & 13.45 & 13.26 & 13.20 + 51196.7938 & 0.009 & 15.33 & 14.65 & 13.48 & 13.29 & 13.13 + 51216.6831 & 0.226 & 14.32 & 13.91 & 13.07 & 12.91 & 12.77 + 51218.6966 & 0.248 & 14.23 & 13.80 & 13.06 & 12.92 & 12.84 + 51223.5895 & 0.302 & 14.31 & 13.90 & 13.04 & 12.90 & 12.85 + 51230.7042 & 0.379 & 14.22 & 13.92 & 12.97 & 12.84 & 12.81 + 51247.7040 & 0.565 & 14.70 & 14.32 & 13.33 & 13.15 & 13.15 + 51259.6068 & 0.695 & 14.26 & 13.86 & 12.91 & 12.81 & 12.72 + 51278.6080 & 0.902 & 14.45 & 14.07 & 13.16 & 13.06 & 13.01 + 51280.6209 & 0.924 & 14.41 & 14.04 & 13.10 & 13.00 & 13.01 + 51287.6168 & 0.000 & 15.36 & 14.65 & 13.47 & 13.34 & 13.22 + 51289.6136 & 0.922 & 14.44 & 14.05 & 13.05 & 12.98 & 12.95 + 51291.6213 & 0.944 & 14.77 & 14.28 & 13.24 & 13.06 & 13.01 + l c r r 2451070.0150 & 0.626 & @xmath1619.0 & 41.8 + 2451086.0142 & 0.801 & @xmath1636.9 & 41.2 + 2451101.0046 & 0.964 & 5.4 & 212.0 + 2451102.9545 & 0.985 & @xmath1611.1 & 171.5 + 2451109.9734 & 0.062 & 56.9 & 49.3 + 2451115.9804 & 0.128 & 77.4 & 41.6 + 2451129.9315 & 0.280 & 98.5 & 56.7 + 2451132.9985 & 0.313 & 80.0 & 21.9 + 2451135.9320 & 0.345 & 82.1 & 24.2 + 2451138.9114 & 0.378 & 77.1 & 17.5 + 2451141.9962 & 0.411 & 80.2 & 32.1 + 2451144.8588 & 0.443 & 56.8 & 5.4 + 2451147.9806 & 0.477 & 24.5 & 32.7 + 2451157.7598 & 0.583 & @xmath1623.2 & 18.0 + 2451160.8444 & 0.617 & @xmath1629.6 & 48.2 + 2451163.9600 & 0.651 & @xmath1630.6 & 33.9 + 2451166.8621 & 0.683 & @xmath1648.6 & 35.2 + 2451169.8996 & 0.716 & @xmath1631.2 & 55.1 + 2451172.8716 & 0.748 & @xmath1636.6 & 57.3 + 2451175.9043 & 0.781 & @xmath1635.0 & 25.7 + 2451187.7566 & 0.911 & @xmath165.2 & 61.9 + 2451188.7424 & 0.922 & 7.3 & 73.0 + 2451191.7695 & 0.955 & 12.7 & 207.9 + 2451194.8215 & 0.988 & 31.0 & 153.7 + 2451196.7938 & 0.009 & 42.6 & @xmath1646.3 + 2451198.7117 & 0.030 & 50.2 & @xmath1699.2 + 2451201.7889 & 0.064 & 87.1 & 23.8 + 2451216.6831 & 0.226 & 129.0 & 36.0 + 2451218.6966 & 0.248 & 118.0 & 66.0 + 2451223.5895 & 0.302 & 95.3 & 28.7 + 2451224.6268 & 0.313 & 125.0 & 31.7 + 2451226.7300 & 0.336 & 104.3 & 63.1 + 2451228.6119 & 0.357 & 41.8 & 27.6 + 2451230.7042 & 0.379 & 68.4 & 38.7 + 2451232.7299 & 0.402 & 43.8 & 16.8 + 2451247.7040 & 0.565 & @xmath1610.3 & 4.5 + 2451249.5973 & 0.586 & @xmath1621.4 & 16.0 + 2451257.6133 & 0.673 & @xmath1654.3 & 12.0 + 2451259.6068 & 0.695 & @xmath1640.9 & 43.0 + 2451278.6080 & 0.902 & @xmath163.2 & 74.9 + 2451280.6209 & 0.924 & 10.6 & 125.5 + 2451284.6181 & 0.968 & 18.3 & 178.2 + 2451287.6168 & 0.000 & 15.3 & 12.8 + 2451289.6136 & 0.022 & 33.4 & @xmath1671.5 + 2451291.6213 & 0.044 & 55.2 & @xmath1646.5 +

a long time ago a rather unorthodox physics of crystals possessing three - dimensional ( 3d ) pseudo - relativistic quasiparticles @xcite , exhibiting an electronic dispersion relation @xmath5 , where the velocity @xmath3 is of the order of the fermi velocity in regular condensed matter systems , was invoked to describe properties of @xmath6 . the ultra - relativistic linear dispersion relation describes two conical bands ( of opposite orientation ) sharing the same cone tip . recently several proposals ogata , vishwanath , kane , kane , bernevig revived an interest in materials with such excitations nowadays called weyl semi - metals . the fermi `` surface '' of such materials , typically with dominant spin - orbit interactions , consists just of a finite number of disconnected points ( called weyl or dirac points , defined below ) rather than forming a continuous fermi surface like electrons in usual metals . the revived interest emerged of course after years of intense experimental and theoretical study of graphene , a 2d weyl quasiparticle material @xcite . suspended graphene is just such a `` semimetal '' system and exhibits a number of remarkable properties . for example , despite having zero density of states at the fermi level and ideally no impurities , it still has a nonzero dc conductivity@xcite . while this band `` touching '' at a singular point was noticed in band structure calculations even before the seminal work of wallace on graphitewallace , the first explicit use of the dirac model in 3d was in the context of the two - band approximation model of bismuth@xcite . due to the strong spin - orbit interaction , the linear in @xmath7 terms in a low energy effective theory near the crystallographic point @xmath8 of the fcc brillouin zone , dominate over quadratic terms ( that dominate near the @xmath9 point , leading to a common `` effective mass '' description ) . the electronic excitations are described by an analog of the weyl equation of particle physics , which describes eight two - component chiral spinors @xmath10 ( two for each of the four @xmath8 points ) @xmath11 the sign describes the chirality of the mode . metallic bismuth is only approximately described by the ultrarelativistic `` massless '' dispersion relation since the quasiparticles of the opposite chirality are coupled and form four - component massive dirac bispinors . in @xmath6 therefore electrons are not weyl , but @xmath12 massive dirac electrons where , in addition , the fermi level is located away from the dirac point . a number of related suggestions for suitable realizations of weyl semimetals were recently put forward . kariyado and ogata@xcite calculated the band structure of cubic inverse perovskites like @xmath13 with significant spin - orbit coupling . they observed the appearance of six weyl points with a very small relativistic electron mass down to @xmath14 on the line connecting the @xmath9 - and the @xmath15 - points . in iridium - based pyrochlores such as @xmath16 , there are @xmath17 weyl points located near the four @xmath8 points of the fcc lattice . as noted invishwanath these materials `` in particular provide a unique opportunity to study the interplay of coulomb interactions , spin - orbit coupling , and the band topology of solids '' . also , strong spin - orbit interactions can lead to a novel phase of matter , the topological insulator@xcite and various possibilities to create weyl fermions combined into coincident opposite chirality points or separated in the brillouin zone in @xmath18kane , @xmath19 , @xmath20@xcite . these proposals generated a great deal of experimental efforts@xcite . the system with 3d weyl points was proposed to appear in optical lattices@xcite following the discovery of `` artificial graphene''@xcite . since the density of carriers in 3d weyl semimetals at zero temperature is zero ( as in suspended graphene in 2d ) , the coulomb interactions are unscreened and therefore are expected to be important to the understanding of the electrical and optical response of these materials . unsophisticatedly the dimensionless coupling,@xmath21is of order @xmath22 provided the dielectric constant @xmath23 is not large , since the analog of the light velocity , @xmath3 , is of the order typically of the fermi velocity . note that the same coulomb potential @xmath24 created by an electron influences many more electrons in 3d compared to 2d , so naively , in 3d its importance is expected to increase . while electric transport in noninteracting 3d weyl fermions was studied ( @xcite and references therein to earlier papers in the context of particle physics ) , the contributions of potentially very important coulomb interactions ( coulomb scattering corrections to transport as an example ) have not been studied theoretically , except basic renormalization effects vishwanath12 . this is in contrast to the situation in graphene . the effect of the coulomb interactions in undoped graphene turned out to be highly nontrivial , even within perturbation theory , and have evoked a scientific controversy@xcite due to the problem of the `` ultraviolet regularization '' , and was just recently settled@xcite by noting that the ambiguities are associated with the treatment of the separation of scales related to the chiral anomaly@xcite . some aspects of the weyl semimetal physics are _ not _ dominated by the weyl points of the brillouin zone at which the spectrum is gapless . for example , the ac conductivity of undoped graphene ( the weak logarithmic renormalization of the electron velocity @xcite does not influence the result ) , is given in terms of its value in the noninteracting theory , @xmath25 , by @xmath26 and is strictly dissipative ( real ) . this expression is independent of frequency ( provided corrections of order @xmath27 , @xmath28 being the hopping energy , are neglected ) . the value of the only numerical constant @xmath29 appearing here has been a matter of intense controversy . the first detailed calculation@xcite utilizing a sharp momentum cutoff regularization of the dirac model provided a value @xmath30 of order @xmath31 . the use of the sharp momentum cutoff was criticized by mishchenkomishchenko , who obtained an exceptionally small value of @xmath32 making a `` soft '' momentum cutoff regularization . he supported this choice by the consistency of the kubo and the kinetic equation calculations of conductivity with that of the polarization function . the consistency required a modification of the long - range interaction so that it becomes uv cutoff dependent . this apparently closed the issue . albeit such a small numerical value would have profound physical consequences even beyond the transport and dielectric properties . nevertheless the interaction strength @xmath29 was recalculated once again by vafek , juricic and herbut@xcite , who argued that the modification of the interaction requires simultaneously a pauli - villars regularization of massless fermions . they applied yet another regularization , making the space dimensionality fractional , @xmath33 that modified both the current operator and the interaction in such a way that they satisfy the ward identities and obtained @xmath34 . the dimensional regularization is questionable on physical grounds and in a comprehensive subsequent workmacdonald the authors reaffirmed the small value @xmath35 it seems that this value is the commonly accepted one . the tight binding calculation@xcite however demonstrated that @xmath36 is the correct one . to reveal the origin of the ambiguity exhibited by the various values of @xmath29 , the authors made use of a dynamical approach developed earlier@xcite ( to address the problem of separating the interband contributions from the intraband effects due to contactsziegler , lewkowicz11 ) directly in the dc case by `` switching on '' a uniform electric field in the tight binding model with coulomb interactions , and then considering the large - time limit . this approach ( known in field theory as the `` infinite hotel story '' ) is the best way to reveal physical effects of anomalies@xcite . one can directly separate the contributions from the neighborhood of dirac points and the `` anomalous '' contributions from the rest of the brillouin zone , so that one can decide what regularization of the effective weyl theory is the correct one . in this sense this is advantageous over the standard diagrammatic kubo formula calculation in continuummishchenko , herbut1,herbut2,macdonald that might encounter the so - called schwinger terms ( found in quantum electrodynamics , that is similar to the 3d weyl fermions ) . the purpose of the present paper is to study the effect of coulomb interactions in 3d weyl fermion systems with emphasis on dynamical aspects of electric transport and compare / contrast it with the corresponding results in 2d weyl fermions . to achieve this goal we define in section ii a tight binding description of weyl fermions on a hypercubic lattice of any dimensionality similar to a variant of the hamiltonian lattice model in field theory@xcite already used in its lagrangian version to simulate graphene@xcite . in this model the electron s spin is strongly coupled to momentum and therefore the model is very reminiscent , in this respect , of the wolff model of bismuth or lattice realizations of topological insulators . the lattice `` regularization '' is necessary to address the ultraviolet divergencies at the intermediated stages of calculations , a problem mentioned above . the universality of this description of weyl fermions ( of various origins ) is supported by the fact that such a 2d model gives the same result for the interaction corrections as the tight binding model on the honeycomb lattice with zero spin - orbit coupling . in section iii the correction to the self energy of a quasiparticle with momentum @xmath37 is considered . it is shown that the fermi velocity renormalization in 3d ( already noted in@xcite ) is logarithmic in the uv cutoff @xmath38 ( @xmath39 being the lattice spacing ) very much like in 2d@xcite:@xmath40 \text{. } \label{renv}\]]section iv is devoted to a general derivation and application of the dynamical approach to the electric response to the dc and ac electric field . in particular , we obtain the slow current decay in a dc field @xmath41 of the neutral , noninteracting 3d weyl plasma and show that the relaxation is powerwise , see fig.1 . the long time asymptotics is oscillating and depends on the microscopic details via cutoff @xmath39:@xmath42 \text{. } \label{cur_decay}\]]here @xmath43 is the number of weyl fermions . this explains how the pseudo - ohmic dc conductivity vanishes in 3d . the relaxation dynamics therefore is qualitatively different from 2d , where it is insensitive to the cutoff @xcite . for the ac electric field similar slow convergence to the ac conductivity occurs , see fig . 2 . even in the free weyl semimetal one gets , in addition to a finite real ( pseudo - dissipative ) partlinear in frequency@xcite , @xmath44an imaginary part , logarithmically divergent as function of the uv cutoff:@xmath45this is again different from graphene at zero doping and leads to important dielectric properties . relaxation to the asymptotic behavior is faster at higher frequencies . the leading interaction correction to the real and imaginary conductivities per weyl point are subject of section v , see fig.3 for comparison with a metal and a semiconductor . the result for the real part expressed via the renormalized fermi velocity defined still linear in frequency:@xmath46 \label{results } \\ c & = & -\frac{5}{3\pi } \approx -0.53\text{. } \notag\end{aligned}\]]the normalization frequency is @xmath47 . the imaginary part gets further logarithmically dependent on cutoff corrections,@xmath48 \text { , } \label{imsig}\]]despite an apparent `` renormalizability '' of the model to the two loop order at least . this is due to the fact that the complex conductivity is not simply related to basic green s functions like in the relativistic theory . the physical significance of the results including formulas including that for the complex dielectric constant , see fig.4 , are summarized in the concluding section vi . we speculate about obvious improvements like the random phase approximation , renormalization group and a possibility of stronger coupling effects like the exciton condensation . the noninteracting tight binding model is defined on the hypercubic lattice @xmath49 with lattice vectors @xmath50 of length @xmath39 by the hamiltonian : @xmath51 here @xmath52 are pauli matrices , operators @xmath53 create a two component spinor @xmath54 and @xmath55 is the hopping integral that in the presence of an external electromagnetic field , described by vector potential @xmath56 is @xmath57 , \label{wilson}\]]where the hopping energy @xmath58 is of order of the band width . it is important to derive the current density directly in tight binding model : @xmath59it defines the uv regularization of the current operator that obeys the ward identities . in linear response the current density operator is expanded up to the first order in @xmath60 as @xmath61 : @xmath62 where @xmath63 . space averages over volume @xmath64 for a homogeneous vector potential @xmath65 simplify:@xmath66expansion of the minimally coupled hamiltonian in electric field is @xmath67 defining fourier components by @xmath68 where the number of unit cells is @xmath69 one has:@xmath70diagonalization of @xmath71 is achieved adopting a reinterpretation of the absence of an electron in the valence band as a hole and using units @xmath72 by @xmath73 with spinors @xmath74 and @xmath75 given in appendix a. up to an additive constant it becomes @xmath76 where @xmath77and the notation @xmath78 was introduced . in 3d one observes 8 weyl points at which @xmath79 inside the brillouin zone ( bz ) . four are right handed : one in the center ( the @xmath9 point ) and three on the faces ( @xmath15 ) , while four are left handed : three on the edges ( @xmath80 ) and one in the corner ( @xmath81 ) . the chirality is determined by the expansion of the hamiltonian eq.([fourierk ] ) around a weyl point @xmath82 , @xmath83 , where @xmath84 . similarly , in 2d one has two right handed weyls at @xmath9 and @xmath81 and two left handed at @xmath80 . the paramagnetic and diamagnetic parts of the current due to an electric field ( along , let s say , the @xmath85 direction ) , @xmath86 @xmath87 , are : @xmath88 + hc ; \label{gpgddef } \\ j^{d } & = & \frac{e^{2}}{\mathcal{v}}a\left ( t\right ) \sum\limits_{\mathbf{k}}\frac{\widehat{k_{z}}}{\varepsilon _ { \mathbf{k}}}\left [ \frac{\widehat{k_{z}}}{2}\left ( a_{\mathbf{k}}^{+}a_{\mathbf{k}}+b_{-\mathbf{k}}^{+}b_{-\mathbf{k}}\right ) -\left ( \widehat{k_{x}}+i\widehat{k_{y}}\right ) b_{-\mathbf{k}}a_{\mathbf{k}}\right ] + hc , \notag\end{aligned}\ ] ] where the functions @xmath89 and @xmath90 are defined in appendix a. using the expression for the particle densities in momentum space ( einstein summation implied for spins only ) @xmath91 the coulomb interaction takes the form @xmath92 @xmath93 @xmath94 with fourier transform of the interaction being @xmath95 . it is important to note that the electric charge of the effective tight binding model should include the contributions to the screening due to the polarization constant @xmath23 caused by degrees of freedom not included in the model , so the `` bare '' charge includes this effect @xmath96 . we require charge neutrality that is achieved by leaving out all the contributions including @xmath97 . this prescription is always implied in what follows . it is convenient to normal order @xmath98 with respect to operators @xmath39 and @xmath99 diagonalizing the `` kinetic '' term @xmath71 via eq.([bogtrans]):@xmath100where the part @xmath101 contains @xmath102 creation operators @xmath103 or @xmath104 and @xmath105 annihilation operators @xmath39 or @xmath99 , all specified in appendix a. in principle the quadratic pair creation @xmath106 and the pair annihilation terms @xmath107 a could have appeared . the fact that they have not , explained in the appendix a , greatly simplifies the calculation and makes it competitive ( at least to the two - loop order ) with the diagrammatic approach . the `` time independent '' approach is however much more transparent , when one realizes that excitations can be created in fours rather than in pairs in this particular model . the energy of an electron above the fermi level with quasimomentum @xmath108 , @xmath109 , in the noninteracting model described by the hamiltonian eq.([k ] ) is : @xmath110 the interaction correction is@xmath111obviously only contributions with equal numbers of creation and annihilation operators , @xmath112 and @xmath113 can contribute . the first term contains ( see appendix a ) @xmath114 , @xmath115 and @xmath116 terms , of which only the last one could contribute to the expectation value eq.(qpenergy1 ) , yet it vanishes:@xmath117 we are left with the simple quadratic part @xmath113:@xmath118where@xmath119this results in:@xmath120 \text{. } \label{deleps_exact}\ ] ] it is instructive to estimate it using the expansion around any of the eight weyl points with the momenta restricted , say , by @xmath121 , thus omitting some contributions far from the weyl points that innocently could be thought to be small , but , as the graphene example taught us , might become perfidious if powerwise uv divergencies appear in the intermediate stages of the calculation . in the present case divergencies are just logarithmic though and one proceeds by expanding around one of the weyl points , @xmath122 where spherical coordinates , @xmath123 , were used . the sum is transformed into an integral:@xmath124the integrals over angles can be performed , see appendix b , resulting in ( for @xmath125 ) @xmath126 \text { , } \label{deps1}\]]where @xmath127 . changing variables in the remaining integral establishes the leading linear dependence on @xmath37 : @xmath128 \label{delepsfinal } \\ & = & \frac{e^{2}p}{3\pi } \left [ \log \left ( \frac{\lambda ^{2}}{p^{2}}\right ) + \frac{5}{3}\right ] \text{. } \notag\end{aligned}\]]comparing with the exact result for the `` universal '' model , eq.(deleps_exact ) , we can choose the value of the cutoff to be slightly outside of the continuum model applicability range : @xmath129 . this means that the continuum model can not determine correctly the constant term that is model dependent . to summarize , one indeed observes just a logarithmic divergence and therefore can renormalize the fermi velocity written in physical units in eq.([renv ] ) . in 2d the corresponding calculation gives a well known `` running '' of the graphene velocity towards higher velocities@xcite . therefore the running of the fermi velocity in weyl semimetal is very much like in ( undoped ) graphene . in this section we first calculate the time evolution of the current when the external electric field is switched on in order to obtain the dc and ac electric response . this is done however within a `` time independent '' formalism to carefully trace the states that contribute to the dynamics and investigate the emergence of the steady state . we use the dynamical approach@xcite to the semimetal response rather than the more customary diagrammatic method not just to investigate dynamics under dc or ac field , but also to clarify several fundamental issues in the following sections concerning the coulomb interaction corrections to the ac conductivity . first the nature of the dependence of the physical quantities on the uv `` cutoff '' @xmath130 ( renormalization ) is elucidated , next the time evolution is exploited to demonstrate the inductive response of the 3d semimetal even without electron - electron interaction and contrast it with the purely pseudo - ohmic response of graphene . the paramagnetic contribution in the heisenberg picture is given by the vacuum expectation value ( vev ) : @xmath131 the ground state @xmath132 is that of an interacting electron system without the external field . expanding to `` linear response '' in coupling to the external electric field ( considered homogeneous and oriented along the @xmath85 direction ) , eq.([ahomogen]),@xmath133 \text { , } \label{umc}\]]where @xmath134 , one gets @xmath135 the diamagnetic current , eq.([jpjd_def ] ) is already of the first order in electric field . in the next section the full model including the many body effects will be considered . in this section we neglect the interaction @xmath98 to solve exactly for the time evolution of the current density . in the absence of the coulomb interaction the vev of the average paramagnetic current density ( in the field direction @xmath85 , see eq.(jpjd_def ) ) takes a form : @xmath136 using definitions of the tight binding hamiltonian and the current density operator , eqs . ( [ ksecond ] ) and ( [ gpgddef ] ) , one obtains @xmath137 \text{. } \label{gpfree}\ ] ] similarly the diamagnetic contribution , using eq.([gpgddef ] ) , is @xmath138 to continue one has to specify the time dependence of the applied electric field . we start in the next subsection with the constant electric field and then continue to the ac case . in the homogeneous dc electric field described by the vector potential @xmath139 , one obtains from eq.([gpfree ] ) the average current . the integral over @xmath140 results in : @xmath141 \right\ } \text{. } \label{sigmadc}\ ] ] one expects that the neutral plasma system that does not possess electric charges on the fermi level will not have the `` acceleration '' terms linear in time @xmath142 that appear in the above equation . indeed the sum of the first two terms vanishes:@xmath143 = 0 . \label{sigmaacc}\]]this is seen as follows . the integral over the bz can be represented as an integral over the full @xmath144derivative like in graphenelewkowicz . since the bz can be taken periodic in the quasimomentum component @xmath145 , the integral over the derivative vanishes . the physical arguments put forward in the framework of graphene@xcite in order to comprehend this apply equally well here and are not repeated . the remaining non - accelerating part,@xmath146is presented in fig.1 . it decays as given ( in physical units ) in eq.(cur_decay ) , exhibiting the zero conductivity rather than a universal finite value as in graphene . this asymptotic value of zero is approached therefore powerwise and oscillating . note the dependence on the ultraviolet cutoff , that did not appear in graphene@xcite . is shown as a function of time . the current s asymptotic value of zero is approached powerwise.,width=226 ] the ac electric field , @xmath147 , is represented by the oscillating vector potential , @xmath148 . the `` universal '' nearest neighbours tight binding model is relevant only for frequencies @xmath149 smaller than the hopping integral @xmath150 . however , as we noticed in the context of graphene@xcite , the use of the expansion of the dispersion relation near the weyl points should be done with some care . performing the integral over @xmath140 in eq.([gpfree ] ) , one now obtains the current : @xmath151 as in graphene , it is very difficult to approach frequencies of interest @xmath152 , when the conductivity is presented in this form . however it becomes substantially simpler numerically when one subtracts the vanishing acceleration term eq.([sigmaacc ] ) , that we have encountered in the dc calculation , with @xmath142 replaced by @xmath153 : @xmath154 this is shown ( together with the electric field ) in fig.2 for several values of frequency . one observes that beyond certain relaxation time the response becomes periodic exhibiting a phase difference between the current ( points ) and the electric field ( solid lines ) . to obtain the steady state value of the complex conductivity we average over time with a damping factor @xmath155 using@xmath156the time integration in eq.([sigma0 ] ) with the current density of eq.([sigma0 ] ) results in @xmath157where @xmath158 . the exact integral is presented ( in physical units ) in eq.([sig0 ] ) and eq.([sig0im ] ) with nonuniversal cutoff related to the lattice spacing via @xmath159 . is typically of order @xmath160,width=226 ] now we proceed to try to use the continuum model near the weyl points . after the subtraction of `` anomalous '' acceleration terms most of the contributions for @xmath152 come from the immediate neighbourhoods of the weyl points@xcite . due to symmetries it suffices to consider one of them . the neighbourhood of the origin will be defined by @xmath161 . here two differences emanate compared to the transport in graphene . in graphene the lattice spacing is irrelevant , so that the conductivity can be calculated within the continuum weyl model and the ac conductivity is real for all frequencies @xmath152 . in the present case an inductive part appears and moreover depends logarithmically on the lattice spacing . to see this explicitly , let us calculate the ac conductivity approximately by using the weyl approximation ( linearization of the dispersion relation ) for the contribution of the weyl points . within the continuum approximation in spherical coordinates one writes the sum in eq.([sigma_eta ] ) as : @xmath162 where the factor @xmath163 is due to the multiplicity of the weyl points in our `` universal '' tight binding model . the integration over @xmath164 finally gives:@xmath165 in the last line small terms of the relative order @xmath166 were omitted . the finite result for the real part and regularized matsubara conductivity were first obtained in@xcite . let us emphasize that the electric properties in 3d differ from that in 2d ( graphene ) in that there appears an imaginary part of conductivity that is of the same size and sometimes even larger than the pseudo - dissipative one . now we turn to the study of the many - body effects in 3d weyl semimetal . the expressions for the diamagnetic and the paramagnetic components of the current density as a linear response to arbitrarily time dependent electric fields are given by eqs.([jpjd_def ] ) and ( [ jplinresp ] ) respectively . the evolution operator however should be modified as @xmath167 , \label{evolution}\]]and the ground state ( at initial time or schrdinger ) as@xmath168with the correction to energy being @xmath169 , since the constant term in the energy will be consistently omitted after the hamiltonian was normal ordered . only creation of four particles at once is possible within the `` universal '' model . as is shown in appendix b. the pair creating part @xmath170 vanishes . this simplifies the problem significantly . for example , the diamagnetic component is absent due to this . indeed the diamagnetic current is quadratic , eq([jpjd_def ] ) , and hence the correction@xmath171 the paramagnetic correction ( in homogeneous electric field along certain direction described by vector potential @xmath172 ) of eq.(jplinresp ) in the presence of coulomb interactions takes a form : @xmath173 expanding in interaction with help of eqs.([evolution ] ) and ( vaccumcor ) , the correction to the expectation value in the interacting ground state @xmath174 above in terms of the unperturbed vacuum expectation values is : @xmath175 @xmath176 the other contributions vanish , again due to absence of pair creations at this order . the first term corresponds to the fermion self energy correction , while the rest describe the vertex correction . these are readily calculated : @xmath177 @xmath178 \left\vert \chi _ { \mathbf{q}}\right\vert ^{2}g_{\mathbf{qp}}+g_{\mathbf{qp}}^{-}\frac{\sin \left [ 2\varepsilon _ { \mathbf{q}}\left ( t - t_{1}\right ) \right ] } { \varepsilon _ { \mathbf{q}}+\varepsilon _ { \mathbf{p}}}+\frac{1}{2}g_{\mathbf{qp}}^{+}\frac{\sin \left [ 2\varepsilon _ { \mathbf{p}}\left ( t - t_{1}\right ) \right ] -\sin \left [ 2\varepsilon _ { \mathbf{q}}\left ( t - t_{1}\right ) \right ] } { \varepsilon _ { \mathbf{q}}-\varepsilon _ { \mathbf{p}}}\right\ } \text { , } \notag\ ] ] where @xmath179 is defined in eq.([aqp ] ) and @xmath180 ; \text { \ } g_{\mathbf{qp}}^{+}=re\left [ \chi _ { \mathbf{q}}\chi _ { \mathbf{p}}^{\ast } \left ( v_{\mathbf{q}}^{\ast } \cdot v_{\mathbf{p}}\right ) ^{2}\right ] \text{.}\]]this can be calculated using the same methods as in the leading order in the previous section . the dc conductivity correction vanishes linearly . finally we calculate the correction to current for the ac electric field . in the homogeneous ac electric field , @xmath181 , the ac conductivity averaged over time as in the leading order , eq.([etatrick ] ) , is : @xmath182 \text { , } \label{sigmacorr1}\ ] ] where @xmath158 . subtracting the dc limit ( that vanishes after averaging over time ) like in the leading order , one obtains a much more converging expression proportional to the frequency : @xmath183 \text{. } \label{sigcorr2new}\ ] ] the numerical evaluation results in ( without subleading terms in the imaginary part)@xmath184 \text{. } \label{delsigexact}\]]it is however instructive to obtain this expression within the continuum ( weyl ) approximation valid within radius @xmath185 around each of the eight weyl points . near a weyl point one has the following expansion of the functions appearing in eq.([sigcorr2new ] ) in spherical coordinates around one of the points ( chosen to be the origin @xmath9):@xmath186 . \notag\end{aligned}\]]complementing eq.([aqpweyl ] ) for @xmath179 . the sum and the coulomb potential in eq.([sigcorr2new ] ) are written as @xmath187 } . \label{angleint}\ ] ] the integrals over all the angles can be performed , see appendix b , @xmath188 with @xmath127 . the integrals give @xmath189 \text{. } \label{weyl res}\ ] ] one observes that the subleading terms can not be given correctly by the weyl approximation as expected . we can use the renormalized velocity from eq.(renv ) to make the real part finite .. we have calculated the renormalization of the fermi velocity , eq.([renv ] ) and the ac conductivity of the clean weyl semimetal within the leading order in coulomb interaction . it is very tempting to try to use renormalization to improve the results of the `` bare '' perturbation theory presented to the `` two - loop '' order above by renormalizing the parameters of the theory . the expression for the ac conductivity both in the leading order and for the interaction correction contains dependence on the uv cutoff @xmath39 or @xmath185 and it is interesting to ask whether physically measurable quantities can be rewritten via `` renormalized '' parameters only or the microscopic details represented by the cutoff dependence are indeed unavoidable for certain measurable quantities like the ac conductivity . the possibility of `` renormalizability '' is expected for the 3d weyl model on the following two grounds . it has been claimed recently and shown to a very high order explicitly that the 2d version ( graphene ) is renormalizable@xcite . this is surprising due to the breaking of the relativistic invariance ( that ensures the renormalizability for a 2d model with a 3d electromagnetic coupling ) . the ac conductivity expression in 2d can be written via the renormalized fermi velocity , eq.([renv ] ) replacing the bare one in eq.([weyl res ] ) . the key point here is that the leading order conductivity is independent of both the velocity and the uv cutoff . the relativistic version , qed , is renormalizable . breaking of the relativistic invariance by taking just the static part of the interaction might not spoil this like in 2d . we perform the renormalization of parameters up to the two - loop order . replacing the bare fermi velocity with the renormalized one from eq.(renv ) in the real part of the conductivity from eq.([results ] ) , the two - loop result becomes finite and proportional to frequency : @xmath190 \text{. } \notag\end{aligned}\ ] ] where @xmath191 leading to the result ( in physical units ) given in eqs.([results ] ) . as in 2d the constant @xmath29 is positive and of order @xmath31 . note that the coefficient @xmath192 in the definition of the current is not directly related by relativistic invariance to the coulomb interaction in the present model , and thus is not renormalized . in the order considered there is no need to renormalize the static coulomb coupling . the imaginary part remains therefore `` divergent '' , namely logarithmically dependent on microscopic details . before discussing the applicability of the results to materials proposed as realizations of the weyl semimetal , let us summarize the electromagnetic properties of the clean 3d semimetals at zero temperature that can be extracted from the ac conductivity , eqs.([sig0],[sig0im],results,[imsig ] ) . extension of our formulas to finite temperature is trivial via the matsubara substitution @xmath193 . the properties of the clean semimetal are expected to be dissimilar from those of a band insulator and a metal . in particular , the dielectric and optical properties differ markedly from both of them . the complex conductivity of the neutral weyl plasma at zero temperature is very different from semiconductors and from metals . let us contrast it with the lorentz model of a band insulator and the drude model for metals with electron density @xmath194 and relaxation time @xmath195 . the ac conductivity of the lorentz model of a band insulator ( semiconductor ) , represented in fig . 3 by the red line , is @xmath196 where the central frequency of the band was taken to be rather small @xmath197 the relaxation time @xmath198 and the `` plasma '' frequency @xmath199 . the drude conductivity of a metal , represented by the blue line , is obtained from this formula by taking @xmath200 and the values of @xmath198 and @xmath201 . these are compared with the real part and the imaginary part , figs.3a and 3b respectively , of the weyl semimetal , represented by the magenta line . the number of weyl points is @xmath202 , the uv cutoff @xmath203 , @xmath204 , @xmath205 and the intrinsic dielectric constant @xmath206 ( due degrees of freedom not included in the model ) . at low frequencies the absorptive part is linear , so that at dc the weyl semimetal is insulating . however it becomes comparable with that of ( a poor ) metal at thz frequencies . the imaginary part is linear as in a metal but with the opposite sign ( capacitive like in insulator rather than inductive ) . the complex dielectric constant of the 3d semi - metal is @xmath207 @xmath208 ; \label{eps } \\ \varepsilon ^{\prime \prime } \left ( \omega \right ) & = & \frac{n_{w}e^{2}}{6v\hbar } \left [ 1+\,\alpha \left ( \frac{2}{3\pi } \log \frac{\omega } { \overline{\omega } } + c\right ) \right ] \text{. } \notag\end{aligned}\]]where @xmath209 with the renormalized ( measured ) value of the fermi velocity @xmath3 . the normalization frequency is @xmath2 and @xmath210 . the real part of the dielectric constant @xmath211 is like that of an ordinary _ dielectric _ material with a very weak frequency dependence , see fig . 4a , despite the nonzero ac conductivity which is linear in frequency . albeit note the logarithmic divergence for small frequencies : for example at @xmath212 , @xmath213 , see fig.4a . the imaginary part of the dielectric constant depends also only weakly on frequency and is universal in that it is of order @xmath214 for fermi velocity @xmath215 and intrinsic dielectric constant @xmath206 . let us compare the electric and optical properties of the pure 3d weyl semimetal at neutrality point with the corresponding ones in 2d . in the pure 2d weyl semimetal at the neutrality point ( suspended undoped graphene is considered to be a good realization of this model@xcite ) the real part of the conductivity is finite and frequency independent in infinite samples and with no contact work function@xcite , and there is no imaginary part in the ac conductivity , at least to leading order in interactions . the transport therefore is purely pseudo - dissipative . as noted before , the situation changes rather dramatically in 3d . we expect that problems of separation between the interband transitions ( between the valence and the conduction bands or electron - hole pairs effects ) considered in the present study and the intraband transitions ( including the klein scattering ) due to potential barriers and mesoscopic effects are less pronounced in 3d compared to 2d . let us now discuss the limitations of the model and point out some immediate extensions . a clean system was assumed , while disorder is expected to be present . the effects of disorder , neglecting the interactions , were studied in ref.vishwanath12 . there might be an interplay between the disorder and interaction effects , but the basic physics is expected to be unaltered unless interaction or disorder are strong . one also can hope that , like in graphene , the importance of disorder might be reduced compared to expectations based on physics of ordinary `` nonrelativistic '' quasiparticles . the use of an approximation like the tight binding model of section ii or even a continuum low energy weyl model is justified as long as the frequencies considered are much lower than the band width . as we argued in section v , the tight binding model is rather universal , but the use of the effective massless weyl theory on the condensed matter scale has to be dealt with care . it is well known in field theory@xcite and in graphene that massless fermions cause the absence of a perfect scale separation between high energies ( on atomic scale @xmath58 ) and low energies ( effective weyl theory on the condensed matter scale @xmath216 ) . it was demonstrated in the context of graphene@xcite that some aspects of the linear response physics , including the coulomb interactions corrections@xcite , are _ not _ dominated by the weyl points of the brillouin zone at which the effective low energy model is valid . for example , large contributions ( infinite , when the size of the brillouin zone is being considered infinite ) to the conductivity from the vicinity of the weyl points are cancelled by contributions from the region between them . another famous consequence of this scale nonseparation is the `` species doubling '' of lattice fermionssmit , which in the context of graphene means that there necessarily appears a pair of weyl points of opposite chirality . the uv regularization of the effective theory _ does matter _ and , if one were to use such a model , the only regularization known to date to be consistent with the tight binding is the space dimensional regularization developed in ref . herbut2 . there are a number of quite straightforward extensions of the leading order interaction calculation of the ac conductivity performed in the present work . converting it into the rpa - like approximation is simple , but in addition taking into account the finite momentum transfer is more involved than the analogous calculations in graphene@xcite . this would allow to study the plasmons effects and even the strong coupling phenomena like the exciton condensation@xcite . beyond the linear response , phenomena like the nonlinear i - v curves due to the particle - hole ( schwinger ) pair creation and relaxation due to their recombination studied in the context of graphene@xcite also can be extended to 3d in a straightforward fashion . we are indebted to t. maniv , h.c . kao and w.b . jian for valuable discussions . the noninteracting model is diagonalized in 3d with the following coefficients of the bogoliubov transformation eq.([bogtrans ] ) : @xmath217 where @xmath78 and @xmath218 . using matrix elements @xmath219 one obtains the coefficient of the electric current as@xmath220 the normal ordering is quite straightforward . @xmath221 where the part @xmath101 contains @xmath102 creation operators @xmath103 or @xmath104 and @xmath105 annihilation operators @xmath39 or @xmath99 . the quartic terms are : @xmath222 @xmath223 \label{o31 } \\ & & + \left ( v_{l^{\prime } } ^{\ast } \cdot u_{k^{\prime } } \right ) \left [ \left ( v_{l}^{\ast } \cdot v_{k}\right ) a_{l}^{+}a_{l^{\prime } } ^{+}b_{-k^{\prime } } ^{+}a_{k}-\left ( u_{l}^{\ast } \cdot u_{k}\right ) \right ] a_{l^{\prime } } ^{+}b_{-k}^{+}b_{-l}b_{-k^{\prime } } ^{+ } ; \notag\end{aligned}\ ] ] @xmath224 in principle there are three possible quadratic terms where @xmath225 @xmath226 relations between the scalar products for unequal momenta that appear here @xmath227 are useful in summation over momenta incorporating eq.([vdef ] ) , in particular for two equal momenta @xmath228most importantly , summing over momenta , the pair creation term vanishes : @xmath229 this leads to numerous simplifications . is:@xmath232 } \label{fi_int } \\ & = & \frac{1}{2r}\left [ \frac{1+r^{2}}{\sqrt{\left ( 1+r^{2}-2r\cos \theta _ { p}\cos \theta _ { q}\right ) ^{2}-\left ( 2r\sin \theta _ { p}\sin \theta _ { q}\right ) ^{2}}}-1\right ] . \notag\end{aligned}\ ] ] s.m . young , s. zaheer , j.c.y . teo , c.l.kane , e.j . mele and a.m. rappe , lett , 108 , 140405 ( 2012 ) ; z. wang , y. sun , x .- q . chen , . c. franchini , . g. xu , h. weng , x. dai , and z. fang , phys . b * 85 * , 195320 ( 2012 ) ; a. go , w. witczak - krempa , g. s. jeon , k. park , and y. b. kim , phys . * 109 , * 066401 ( 2012 ) ; w. witczak - krempa1 and y. b. kim , phys . b * * * * * 85 * , 045124 ( 2012 ) . k. wang , d. graf , h. lei , s. w. tozer , and c. petrovic , phys . b * * 84 * * , 220401 ( 2011 ) ; k. ueda , j. fujioka , y. takahashi , t. suzuki , s. ishiwata , y. taguchi , y. tokura , phys lett . * 109 * , 136402 ( 2012 ) ; f. f. tafti , j. j. ishikawa , a. mccollam , s. nakatsuji , s.r . julian , phys . b * * 85 * * , 205104 ( 2012 ) ; j. j. ishikawa , e. c. t. ofarrell , and s. nakatsuji , phys . b * * 85 * * , 245109 ( 2012 ) ; s. boseggia , r. springell , h. c. walker , h. m. ronnow , ch . ruegg , h. okabe , m. isobe , r. s. perry , s. p. collins , and d. f. mcmorrow , phys . . lett . * 110 * , 117207 ( 2013 ) . p. soltan - panahi , j. struck , a. bick , w. plenkers , g. meineke , c. becker , p. windpassinger , and k. sengstock , nature p. hauke and m. lewenstein ; z. lan , a. celi , w. lu , p. ohberg , and m. lewenstein , phys . lett . * 107 * , 253001 ( 2011 ) ; k. sun , . w. v. liu , a. hemmerich , and s. das sarma , nature physics * 8 * , 67 ( 2012 ) . v. n. kotov , b. uchoa , and a. h. castro neto , phys . b * * 78 * * , 035119 ( 2008 ) ; s. h. abedinpour , m. polini , w.k . tse , a. h. macdonald , phys . b * * 84 * * , 045429 ( 2011 ) ; i. sodemann , and m. m. fogler , phys . rev . b * * 86 * * , 115408 ( 2012 ) . j. e. drut and t. a. lahde , phys . rev . lett . * 102 * , 026802 ( 2009 ) ; phys . b * 79 * , 165425 ( 2009 ) ; p. v. buividovich , e. v. luschevskaya , o. v. pavlovsky , m. i. polikarpov , m. v. ulybyshev , phys . b * 86 * 045107 ( 2012 ) ; p.v . buividovich , m. i. polikarpov , phys . b * 86 * , 245117 ( 2012 ) ; m. v. ulybyshev , p. v. buividovich , m. i. katsnelson , m. i. polikarpov , monte - carlo study of the semimetal - insulator phase transition in monolayer graphene with realistic inter - electron interaction potential . arxiv:1304.3660 ( 2013 ) . a. barreiro , m. lazzeri , j. moser , f. mauri , a. bachtold , _ phys . lett . _ * 103 * , 076601 ( 2009 ) ; n. vandecasteele , a. barreiro , m. lazzeri , a. bachtold , f. mauri , _ phys . rev . b _ * 82 * , 045416 ( 2010 ) ; m. lewkowicz , h.c . kao , and b. rosenstein , phys . b * 86 * 245117 ( 2012 ) ; m. i. katsnelson , g. e. volovik , jetp let . * 95 * , 411 ( 2012 ) ; m. i. katsnelson , g. e. volovik , m. a. zubkov , ann . , * 331 * , 160 ( 2013 ) .

the effect of a magnetic field on an s=1/2 antiferromagnetic chain has been extensively investigated theoretically over many years . the hamiltonian is written : @xmath0.\label{ham}\ ] ] an important conclusion was that the groundstate remains gapless right up to the saturation field . the low - energy excitations can be described by bosonization which predicts gapless excitations at wave - vectors 0 and @xmath1 and also at the incommensurate wave - vectors @xmath2 where @xmath3 is the magnetization per site , @xmath4 . the first detailed experimental study of such systems at large fields with @xmath5 of @xmath6 were only performed very recently,@xcite on cu benzoate . this material has a relatively small exchange energy , @xmath7 mev so that @xmath8 for a field of 7 t. while these experiments verified , in detail , the expected field - dependent shift of the wave - vector at which a gap minimum occurs , they also discovered an unexpected result . a non - zero gap appeared which seemed to scale as approximately @xmath9 , with strong dependence on the field orientation . dender et al.@xcite suggested that this gap might arise from the staggered @xmath10 ( gyromagnetic)-tensor , associated with the low symmetry of the crystal structure and the presence of 2 crystallographically inequivalent cu sites on each chain . thus the last term in eq . ( [ ham ] ) must be replaced by : @xmath11s^b_j.\ ] ] this results in the presence of an effective staggered field , @xmath12 , upon the application of a uniform field . such a staggered field , which couples directly to the nel order parameter , is expected to produce an ordered antiferromagnetic moment and a gap which scale with field . this idea was developed in detail in ref . [ ] where it was found that a staggered dzyaloshinskii - moriya@xcite ( dm ) interaction also contributes a roughly equal amount to this effective staggered field . this corresponds to an additional term in the hamiltonian : @xmath13 it was found that several aspects of the experiments could be explained in detail by this model . these include the field orientation dependence of the gap and its scaling with field magnitude . much of this work used the bosonization technique which maps the problem onto the sine - gordon model , for which various exact results are available . the excitations observed in neutron scattering were identified with the soliton , antisoliton and `` breather '' ( soliton - antisoliton boundstate ) spectrum of the sine - gordon model . additional results , further supporting this approach , were obtained by essler and tsvelik.@xcite the purpose of this paper is to provide more details and some extensions of the results in [ ] . while cu benzoate is the only example of such a system that we discuss in the present paper , it should be possible to apply our theory to other quasi - one dimensional system with similar crystallographic structure . in the absence of a staggered field , the critical behaviour of the antiferromagnet is determined by 3 field dependent quantities : the magnetization , @xmath3 [ which determines the soft wave - vectors via eq . ( [ softk ] ) ] , the spin - wave velocity , @xmath14 and the boson compactification radius ( which determines the critical exponents ) , @xmath15 . all three of these quantities can be determined very accurately by numerical solution of bethe ansatz equations . furthermore , we derive an exact relationship between these three functions using field theory arguments . we also derive the logarithmic dependence of @xmath16 on @xmath17 , as @xmath18 , from the existence of a marginally irrelevant operator , using the renormalization group . we determine the scaling of gap with field , order parameter with field and susceptibility with temperature . the calculations are done including logarithmic corrections which arise from a marginally irrelevant operator and take into account both uniform and staggered fields . furthermore , the exact amplitudes of the scaling functions are determined using a recent result of lukyanov and zamalodchikov.@xcite ( after this calculation was finished we received the preprint [ ] which gives the same result for the gap , without a discussion of logarithmic corrections . ) we also give some further discussion of the structure factors @xmath19 , measured in neutron scattering . we discuss a hidden su(2 ) symmetry of the model . we prove that the longitudinal structure factor ( for @xmath20 corresponding to the uniform field direction ) gets contributions from only the soliton and anti - soliton intermediate states , in agreement with experiment . on the other hand , the transverse structure function gets contributions only from the breathers . using the approximate @xmath21 symmetry , we discuss the relative intensity of the various single - particle peaks in the neutron - scattering cross - section , taking into account the polarization dependent factors which arise from fourier transforming the dipole interaction between neutrons and spins which were omitted in ref . [ ] . a comparison is made with experimental results . in particular , the problem of determining a consistent value for the dm vector , @xmath22 is discussed . the susceptibility of the sine - gordon model is calculated , using the integrability of the model , giving essentially the field and temperature dependence of the staggered susceptibility . in section ii we discuss the dm interaction and the mapping of the system into a heisenberg model with orthogonal uniform and staggered fields . in section iii we treat this problem using conventional spin - wave theory . in section iv we discuss bosonization in the presence of a uniform magnetic field . in section v we extend the bosonization approach to the case with staggered field and analyze the induced gap . in section vi we discuss structure factors and compare with the observed neutron scattering cross - section . in section vii we present estimates of the dm interaction based on several experimental results . in section viii we discuss the magnetization and susceptibility . the crystal structure of cu benzoate is shown in fig . ( [ fig : crystal ] ) and fig . ( [ fig : crystallarge ] ) . the chain direction is the c - axis . note that each cu atom is surrounded by 6 ligands with a local symmetry which is almost tetragonal . however the principal axes for this tetragonal symmetry alternate along the chain , with the two inequivalent c - axes being rotated by @xmath23 relative to each other . these corresponding c - axes are labelled i and ii in fig . ( [ fig : crystal2 ] ) . the b - axes are the same for both cu sites . neither of these sets of principal axes correspond to the crystal axes . it is expected that the principal axes for the gyromagnetic tensor will also alternate , corresponding to the local tetragonal axes around each cu ion . the principal axis for the anisotropic exchange interaction is expected to be the @xmath24 axes , the perpendicular bisector of i and ii axes . on the other hand , the principal axis for the dipole interaction , which is of roughly the same order of magnitude is essentially the c - axis . combining these two types of contributions to the nearest neighbor spin - spin interaction , gives a principal axis which roughly bisects @xmath24 and @xmath25 , and is denoted @xmath26 in fig . ( [ fig : crystal3 ] ) . it is convenient to refer the g - tensor to this @xmath27-@xmath28-@xmath26 coordinate system . from electron spin resonance ( esr ) measurements,@xcite it takes the form : @xmath29 with the @xmath30 referring to the 2 inequivalent cu sites . @xmath31 and @xmath32 are the uniform and staggered parts of the g - tensor . this staggered g - tensor produces an effective staggered field , @xmath33 , while the uniform g - tensor produces an effective uniform field @xmath34 . in the special cases where the applied field is along the b axis or in the @xmath35 plane the effective staggered field is perpendicular to the applied field and also to the effective uniform field . for general directions of the applied field they are almost perpendicular ( to within a few % ) . as discussed by dzyaloshinskii@xcite and moriya@xcite , in magnetic crystals of low symmetry an additional antisymmetric exchange interaction occurs , the dm interaction : @xmath36 the possible values of the dm vector @xmath37 can be limited by considering crystal symmetries of cu benzoate . firstly , the compound is invariant under a translation along the c - axis by two sites . this means the dm vectors are the same among the even ( or odd ) links , but even and odd dm vectors can be different . secondly , there is a symmetry under rotation by angle @xmath1 about an axis parallel to the b - axis that passes through the mid - point of two neighboring sites ( @xmath38 and @xmath39 ) , along the chains ( c - axis ) . as noticed by moriya,@xcite this implies the dm vector for the interaction between @xmath38 and @xmath39 must be orthogonal to the b - axis . this can be shown as follows : assume we have the dm interaction @xmath40 . now apply the rotation described above . it acts on the spin operators as @xmath41 and @xmath42 . thus , for the @xmath28-component of @xmath43 , the dm interaction would be inverted while it is unchanged for the @xmath44-component of @xmath43 implying that @xmath45 . finally , the crystal structure is invariant under the combined operation of one site translation along the chain ( c ) direction _ and _ reflection in the ac - plane . considering the fact that the spin vector @xmath46 is an axial vector , the operation acts as @xmath47 and @xmath48 . since the dm vector is orthogonal to the @xmath28-axis ( and thus one factor of @xmath49 always appears in the outer product ) , the dm interaction term is inverted by the combined operation : @xmath50 . thus , the dm vector is alternating as in eq . ( [ hdm ] ) . there are apparently no other restrictions that can be placed on the dm vector using symmetry alone . by considering a tight - binding model for the exchange interactions it was estimated@xcite that @xmath51 is of o(@xmath52 where @xmath53 is the deviation of g from twice the identity matrix . apart from the antisymmetric dm interaction , the remaining exchange anisotropy is believed to be quite negligible ( about 1% of j ) and we will henceforth ignore it . taking @xmath54 , we may write the hamiltonian : @xmath55 \nonumber \\ & & + j \sum_j [ s^z_{2j-1 } s^z_{2j}+s^z_{2j}s^z_{2j+1}],\end{aligned}\ ] ] where @xmath56 . performing a rotation@xcite of the spins by an angle @xmath57 : @xmath58 where @xmath59 the hamiltonian is transformed to the standard xxz model : @xmath60.\ ] ] with some assumptions this anisotropic exchange may cancel the pre - existing one . in any event , it is small and we will ignore it . now consider an external magnetic field , approximating the @xmath10-tensor as 2 times the identity matrix . the spin redefinition of eq . ( [ redef ] ) introduces an effective staggered field . for example , for a uniform field in the x - direction : @xmath61.\ ] ] combining the actual form of the g - tensor with the dm interaction , we can obtain effective uniform and staggered fields corresponding to an arbitrary applied one . writing the rotation matrices by @xmath57 around @xmath22 as : @xmath62 the effective uniform and staggered fields are defined by : @xmath63\vec h \nonumber \\ h^s & \equiv & [ { \cal r}^s g^u + { \cal r}^ug^s]\vec h. \label{huaeff}\end{aligned}\ ] ] in general @xmath64 is nearly orthogonal to @xmath65 . for small @xmath32 and @xmath51 , the staggered field can be approximated as @xmath66 namely the sum of two contributions . henceforth , since we are ignoring the small residual exchange anisotropy and assuming that @xmath67 , we will take @xmath65 to be in the z - direction and @xmath64 to be in the x - direction and refer to them as simply @xmath17 and @xmath68 respectively . also setting @xmath69 we arrive at the simple effective hamiltonian : @xmath70 , \label{heff}\ ] ] with @xmath71 . [ note that we have switched the directions of the uniform and applied fields relative to our earlier paper@xcite which , unfortunately , contained some inconsistencies of notation . ] in this section we summarize the results of spin - wave theory ( leading order 1/s expansion ) for the effective hamiltonian of eq . ( [ heff ] ) . [ as far as we know , spin - wave theory results for this problem were first published in ref . although this misses certain features caused by quantum fluctuations in one dimension , it is still quite instructive . the classical groundstate is a canted antiferromagnetic structure , shown in figure ( [ fig : class ] ) . the spins on both sub - lattices lie in the @xmath72 plane canted towards the @xmath73-axis by an angle @xmath74 from the @xmath30 x - axis . the classical energy of this state is : @xmath75 this is minimized for @xmath76 the solution of : @xmath77 for @xmath78 , the solution is : @xmath79 in order to do a systematic 1/s expansion , it is convenient to regard @xmath17 and @xmath68 as being of o(s ) . we assume that @xmath17 is less than the saturation field , @xmath80 . the leading order spin - wave expansion for a spin pointing in the x - direction is : @xmath81\ ] ] here @xmath82 is a boson annihilation operator . to consider small fluctuations about the canted structure we simply write : @xmath83 where @xmath84 is a rotation about the y - axis by @xmath85 : @xmath86 and @xmath87 is a rotation by @xmath1 about the z - axis : @xmath88 using the facts that : @xmath89 the hamiltonian may be written in a manifestly translationally invariant way : @xmath90 . \label{hsw}\ ] ] it is the translational invariance of eq . ( [ hsw ] ) which motivated the somewhat peculiar looking choice of transformation matrices in eq . ( [ swt ] ) . substituting eq . ( [ swt ] ) into the hamiltonian of eq . ( [ hsw ] ) , we find that the term of @xmath91 is a c - number and the term of @xmath92 vanishes . the term of @xmath93 is : @xmath94.\end{aligned}\ ] ] fourier transforming and performing a bogoliubov transformation that mixes @xmath95 with @xmath96 , we obtain a single band of spin waves in the paramagnetic brillouin zone , @xmath97 with dispersion relation : @xmath98 ^ 2\nonumber \\ & & -[js(1+\cos 2\theta ) \cos k]^2\bigr\}^{1/2}.\end{aligned}\ ] ] the above transformation has allowed us to obtain a single band in the paramagnetic brillouin zone . we may equivalently fold the dispersion relation into the antiferromagnetic brillouin zone , @xmath99 . this gives us 2 branches of spin - waves with dispersion relations : @xmath100 ^ 2\nonumber \\ & & -[js(1+\cos 2\theta ) \cos k]^2\bigr\}^{1/2}.\end{aligned}\ ] ] while this is a fairly simple and explicit formula for the energies in terms of @xmath76 , it must be borne in mind that @xmath76 is determined in terms of @xmath101 , @xmath17 and @xmath68 by eq . ( [ theta ] ) . in the special case @xmath78 , using eq . ( [ thetah0 ] ) , we obtain : @xmath102^{1/2}.\ ] ] note that at @xmath103 , or equivalently @xmath104 , @xmath105 the @xmath106 mode is the goldstone mode corresponding to a uniform precession about the z - axis . a non - zero @xmath68 gives this mode a gap , pinning the spins along the x - axis . we may calculate this gap , to lowest order in @xmath68 using , from eq . ( [ theta ] ) @xmath107 where @xmath108 to linear order the gap is given by @xmath109 thus : @xmath110}[1-(h/4js)^2]^{1/4}+o(h^{3/2}).\ ] ] note that @xmath111 is a singular function of the staggered field , exhibiting a mean field exponent of 1/2 . on the other hand , it depends only weakly on the uniform field , being almost independent of @xmath17 up to @xmath17 of @xmath112 . while this mean field exponent changes when one - dimensional quantum fluctuations are taken into account , it is reasonable to expect this weak dependence on @xmath17 to remain true . the upper mode , @xmath113 depends strongly on h but only weakly on @xmath68 . for @xmath71 @xmath114 . in the one - dimensional case , an exact picture of the low - energy behavior can be obtained using bosonization and rg arguments . here we summarize the results for the case of a uniform field , but no staggered field . we begin with the case where the uniform field also vanishes . the low energy degrees of freedom of the quantum spin variables can be represented in terms of a free boson with lagrangian density : @xmath115 . \label{lag}\ ] ] [ here @xmath116 is the spin - wave velocity which we will generally set equal to 1 . ] the boson field , @xmath117 can be separated into left and right moving terms : @xmath118 their difference defines the dual field : @xmath119 for @xmath120 there are low energy degrees of freedom at wave - vectors 0 and @xmath1 . the spin operators can be approximated as : @xmath121 + c(-1)^je^{i2\pi r\tilde \phi}. \label{abbos}\end{aligned}\ ] ] ( the first constant above is universal . the next three ( real ) constants are not , but @xmath122 has been recently determined using the integrability of the model@xcite and will be discussed in the next section . ) for the h=0 heisenberg model , @xmath123 . for the xxz model @xmath16 ( and @xmath122 ) vary with the anisotropy parameter . writing the hamiltonian : @xmath124,\label{hxxz}\ ] ] exact bethe ansatz results determine : @xmath125 @xmath16 varies between @xmath126 and 0 along the xxz critical line , @xmath127 . in order to understand the vicinity of the isotropic antiferromagnetic point , @xmath128 , it is convenient to use non - abelian bosonization : @xmath129 here @xmath10 is the su(2 ) matrix field of the wess - zumino - witten , k=1 non - linear @xmath130-model ( wzw model ) . @xmath131 and @xmath132 are the left and right moving conserved currents associated with the su(2 ) symmetry of the spin chain . by comparing eqs . ( [ abbos ] ) and ( [ nonabbos ] ) we may read off the correspondences between the wzw fields and the free boson fields . these are : @xmath133 here @xmath134 . the bosonized spin - chain lagrangian contains , in addition to the free boson lagrangian , interaction terms : @xmath135 . \label{marg}\ ] ] for the isotropic heisenberg model , @xmath136 . including small anisotropy , @xmath137 . these obey the rg equations : @xmath138 the rg trajectories are hyperbolas as shown in fig . ( [ fig : kt1 ] ) . for @xmath139 , they end at the @xmath140 axis , corresponding to the xxz critical line . for @xmath141 they lead towards strong coupling , corresponding to the easy axis ordered phase . reverting to abelian bosonization , we see that the fixed point lagrangian contains the extra term : @xmath142.\ ] ] since this is proportional to the free lagrangian , we can eliminate it by a rescaling of @xmath117 . this corresponds to a rescaling of the parameter , @xmath16 , decreasing it by an amount of @xmath143 in agreement with the bethe ansatz result of eq . ( [ delta ] ) . we now consider the heisenberg model with a uniform external field ( but no staggered field ) . the extra term in the lagrangian becomes , upon bosonization , @xmath144 this term can be eliminated by a redefinition of the boson field : @xmath145 this leaves the free lagrangian unchanged . however , it does effect the interaction term and the bosonization formulae . the interaction term is changed due to the shift of the @xmath30 components of the currents : @xmath146 ( note that the phases add , rather than cancel in the interaction term @xmath147 . ) the effect of these phases on the rg equations can be determined from a consideration of the operator product expansion ( ope ) . one of the ope s gets shifted while the other one does not : @xmath148 the one - loop rg equations can be conveniently derived using an ultraviolet cut off on the distance between any pairs of insertions of the interaction lagrangian in perturbation theory . ( see for example [ ] . ) as the position space ultraviolet cut off is increased from @xmath20 to @xmath149 , we integrate over that range of separation of the two points , using the ope . the net effect is that , when the cut - off is small compared to @xmath150 the phase factor in the ope is nearly constant and can be ignored . however , when the cut - off is large compared to @xmath150 the phase factor produces rapid oscillations which tend to cancel out the term from the effective renormalization . reverting to an energy cut - off , this means that the rg equations of eq . ( [ rge ] ) are essentially correct for @xmath151 , but for @xmath152 the right hand side should be replaced by 0 in the first equation . that is , @xmath140 ceases renormalizing , at @xmath153 of o(h ) whereas @xmath154 continues to renormalize as before . thus the rg trajectories are essentially 2 straight lines . for @xmath151 the couplings renormalize along the isotropic separatrix , @xmath155 , but for @xmath152 @xmath140 is constant and @xmath154 renormalizes to 0 . see fig . ( [ fig : kt2 ] ) . thus , in order to determine @xmath156 we simply need to calculate @xmath157 using the isotropic rg equations of eq . ( [ rge ] ) . for @xmath158 , these give : @xmath159 this argument may seem rather naive and it certainly does not give the correct rg trajectory for e of o(h ) . however , due to the weak , logarithmic , dependence of @xmath160 on e we expect that this argument gives the correct behaviour of @xmath156 with h for @xmath158 . this argument then determines the dependence of r on h for @xmath158 : @xmath161 precisely this result was obtained from the bethe ansatz @xcite in the limit @xmath158 . ( [ fig : r(h ) ] ) . for larger values of h higher order terms in the rg equations would be needed and additional interactions would have to be considered , some of which break lorentz invariance . the net effect is that the h - dependence of r becomes more complicated at larger h and also the spin - wave velocity also change with h. we have so far set it equal to @xmath162 . it is known to have the value @xmath163 from the bethe ansatz , for the heisenberg model at @xmath120 . it can be determined numerically from the bethe ansatz integral equations @xcite , and is given in fig . ( [ fig : v ] ) . apart from the functions @xmath14 and @xmath15 we will also be interested in the behavior of the magnetization , @xmath164 from bosonization we obtain : @xmath165 as @xmath18 . one way of obtaining this result is from calculating the zero field susceptibility : @xmath166 bosonization leads to an exact relation between the three functions @xmath14 , @xmath15 and @xmath3 . this follows from calculating the susceptibility at arbitrary field , @xmath17 , using eq . ( [ abbos ] ) with the corresponding value of @xmath15 : @xmath167 in particular , in the limit of small @xmath17 this predicts a logarithmic correction to the magnetization : @xmath168 . \label{mh}\ ] ] eqs . ( [ rh ] ) , ( [ mrv ] ) and ( [ mh ] ) are universal . they should remain true for generic half - integer spin isotropic antiferromagnets in the gapless phase . for the particular case of the nearest neighbour s=1/2 heisenberg model we can set @xmath169 in eq . ( [ mh ] ) . ( [ mrv ] ) agrees very well with our numerical solution of the bethe ansatz equations . [ see figs . ( [ fig : r(h ) ] ) , ( [ fig : v ] ) , ( [ fig : mag ] ) . ] as far as we know the first ( numerical ) calculation of these quantities for @xmath170 were ref . [ ] for @xmath3 , refs . [ , ] for @xmath15 and ref . [ ] for @xmath14 . ( numerical calculations in refs . [ , ] were based on the bethe ansatz integral equations of ref . once @xmath171 , @xmath172 and @xmath16 are determined from the bethe ansatz , all low energy properties of the system are determined by bosonization . from eq . ( [ abbos ] ) the magnetization can be written : @xmath173 therefore the exact formula for the field - induced shift in @xmath117 is : @xmath174 all low energy green s functions are then determined from eq . ( [ abbos ] ) after shifting @xmath175 . in particular we see that @xmath176 has the soft wave - vectors @xmath177 and @xmath178 whereas @xmath179 has the soft wave - vectors @xmath180 and @xmath1 . now we consider the effective hamiltonian of eq . ( [ heff ] ) with both uniform and staggered fields non - zero and @xmath71 . we begin by using the results of the previous section to obtain the @xmath78 theory with the shifted and rescaled boson field characterized by @xmath15 , @xmath3 and @xmath14 . for @xmath78 , upon making these tranformations , the lagrangian density is simply the free boson one of eq . ( [ lag ] ) . from the bosonization formulae of eq . ( [ abbos ] ) the staggered field adds the interaction term : @xmath181 noting the duality transformation between @xmath117 and @xmath182 : @xmath183 we may also write the free lagrangian in terms of @xmath182 : @xmath184.\ ] ] hence we have the standard sine - gordon field theory . an impressive array of conjectured exact results are available on this model , which can be brought to bear on the spin - chain problem . the interaction term is sometimes written as @xmath185 , so we see that we have @xmath186 . in the next section we will discuss details of the excitation spectrum of this model . in this section we discuss the dependence of the gap on the uniform and staggered field . the renormalization group scaling dimension of this operator is @xmath187 from which it follows that the gap scales as : @xmath188 for some function @xmath189 . since , for small uniform fields , @xmath190 , it follows that the exponent is approximately 2/3 , as found in the experiment . note that this formula is valid for @xmath191 at fixed @xmath17 . this is a reasonable order of limits for describing the experiments since @xmath68 is only a few per cent of @xmath17 . the exponent is determined by @xmath15 , given in figure ( [ fig : r(h ) ] ) . we wish to improve on this result in two ways . first of all , if we consider the case where @xmath17 is strictly 0 , then this formula is modified to : @xmath192 we will calculate exactly the amplitude @xmath193 . this is not of direct relevance to the experiments however since they are in the opposite limit @xmath194 . more importantly , we can determine the amplitude function @xmath195 in eq . ( [ delta ] ) for @xmath158 : @xmath196 we will also determine exactly the numerical factor , @xmath197 , which is different than @xmath193 in eq . ( [ gaph0 ] ) . while the logarithmic factors in eqs . ( [ gaph0 ] ) and ( [ ah ] ) are universal and follow from an rg treatment of the marginal interaction , the exact numerical coefficients are specific to the ordinary nearest neighbour heisenberg s=1/2 model and are obtained by using a remarkable exact conjecture made recently by lukyanov and zamalodchikov@xcite extended to the heisenberg point following the method in ref . [ see also ref . our calculations follow the notation of refs . [ ] and [ ] . we consider the renormalization group equation obeyed by the effective coupling constant @xmath198 multiplying the @xmath199 interaction in eq . ( [ lsg ] ) , with bare value @xmath200 for a bare cut off @xmath101 , taking into acount , to linear order , the effect of the marginal interactions of eq . ( [ marg ] ) . this is : @xmath201 g . \label{rgg}\ ] ] the anomalous dimension is given , to low order , by : @xmath202 @xmath140 obeys the rg equation of eq . ( [ rge ] ) . by the usual scaling arguments , we determined the gap , @xmath111 by reducing the ultraviolet cut - off down to a scale @xmath111 such that @xmath203 is o(1 ) . taking into account the dependence of the effective coupling constant , @xmath198 on the bare coupling constant , @xmath204 , then determines the dependence of the gap on @xmath68 . in the case @xmath120 , the rg flow of @xmath205 is given by eq . ( [ rge ] ) for all @xmath153 . on the other hand , for finite @xmath17 , @xmath205 essentially stops renormalizing at a scale of order @xmath17 . integrating eq . ( [ rgg ] ) , gives : @xmath206 this integral can be conveniently evaluated by changing integration variables to @xmath207 using the second of eqs . ( [ rge ] ) . this gives : @xmath208^{1/4}.\ ] ] determining the gap by the condition @xmath209 and setting @xmath210 , gives : @xmath211^{1/4 } , \label{hlambda}\ ] ] for some non - universal constant @xmath212 , of o(1 ) . in the case @xmath120 , the solution of the rg equation eq . ( [ rge ] ) is : @xmath213 thus , @xmath214^{1/6}[1+o(1/\ln ( j / h ) ) ] . \label{deltah0}\ ] ] for non - zero @xmath17 , the first of eq . ( [ rge ] ) is only valid for @xmath151 . at lower energies , @xmath140 stops renormalizing . its fixed value at low e determines @xmath15 : @xmath215 we can extend somewhat the accuracy of our results to larger @xmath17 , by expressing the subsequent results in terms of @xmath15 , determined numerically from the bethe ansatz , rather than by using the above asymptotic small @xmath17 result for @xmath15 . for @xmath216 , we use the second of eq . ( [ rge ] ) with @xmath140 fixed at @xmath156 as given by eq . ( [ lambdaz0 ] ) and the initial condition @xmath217 to obtain : @xmath218(\frac{e}{h})^{2[1 - 2\pi r(h)^2]}. \label{lambdar}\ ] ] setting @xmath219 and substituting into eq . ( [ hlambda ] ) gives : @xmath220^{1/4}.\ ] ] thus we obtain eq . ( [ delta ] ) with @xmath221^{1/4}\right\}^{-1/(2-\pi r^2)}.\label{ah2}\ ] ] note that in eq . ( [ delta ] ) and ( [ ah2 ] ) @xmath16 is a function of @xmath17 , shown in fig . ( [ fig : r(h ) ] ) and given approximately by eq . ( [ lambdaz0 ] ) . as @xmath18 , we may evaluate @xmath222 explicitly , from eq . ( [ lambdaz0 ] ) . using : @xmath223 this gives : @xmath224 note that the same numerical constant , @xmath212 , occurs in the @xmath120 case , eq . ( [ deltah0 ] ) and the @xmath225 case , eq . ( [ delta ] ) and ( [ ah0 ] ) . however , in the latter case it gets multiplied by an extra factor of @xmath226 . finally we wish to determine the dimensionless amplitude , @xmath212 , appearing in eq . ( [ hlambda ] ) and below . this can be done using two remarkable results . one of them is the exact proportionality constant in the bosonization formula for the staggered part of @xmath227 , that is the constant c in eq . ( [ abbos ] ) . this fixes the coupling constant in the sine - gordon model , @xmath228 , in eq . ( [ lsg ] ) . the other recent result is the exact relationship between the sine - gordon coupling constant and the mass of the lightest particle in the spectrum of the sine - gordon model , which is the gap , @xmath111 . this determines the exact relationship between @xmath111 and @xmath68 . this calculation was done in zero uniform field for the xxz s=1/2 antiferromagnet of eq . ( [ hxxz ] ) . the calculation was performed for all @xmath229 along the critical line , @xmath230 . due to the logarithmic corrections at the isotropic point , @xmath128 , an additional calculation is needed at that point . this can be done using the rg . we essentially just need to apply eq . ( [ hlambda ] ) to the case of @xmath120 but @xmath229 slightly less than 1 . comparing to the exact result for all @xmath231 then determines the coefficient , @xmath212 . for @xmath231 and @xmath120 we use the rg equations of eq . ( [ rge ] ) at low energies . the rg flows are hyperbolas terminating on the positive @xmath140 axis . these flows are conveniently labelled by : @xmath232= 2 \frac{\cos^{-1}{\delta}}{\pi}. \label{epsilon}\ ] ] the solution of eq . ( [ rge ] ) with this initial condition is : @xmath233}. \label{lambdaep}\ ] ] substituting into eq . ( [ hlambda ] ) gives : @xmath234}\right\}^{1/4}.\ ] ] note that in the isotropic limit , @xmath235 , we recover our previous logarithmic result of eq . ( [ gaph0 ] ) . on the other hand , taking @xmath236 with @xmath237 held fixed it gives : @xmath238 by comparing this to the exact result for arbitrary @xmath229 ( and hence @xmath237 ) we may extract the value of the amplitude , @xmath212 . in ref . [ ] the spin correlation function in the xxz antiferromagnet is shown to have the asymptotic behavior : @xmath239 with an exact expression determined for the amplitude @xmath240 . in the isotropic limit , @xmath241 : @xmath242 in ref . [ ] the operator @xmath243 is normalized so : @xmath244\cos[2\pi r\tilde \phi ( r ) ] \rangle \to { 1\over 2}|r|^{-2\pi r^2},\ ] ] ( after accounting for a difference in normalization of the free boson lagrangian by a factor of @xmath245 ) . this determines the exact proportionality constant in the bosonization formula of eq . ( [ abbos ] ) : @xmath246 . \label{prop}\ ] ] hence the coupling constant in the sine - gordon lagrangian is precisely @xmath247 . the exact relationship between this coupling constant and the mass , @xmath111 , of the soliton of the sine - gordon model is : @xmath248^{2-\pi r^2}.\ ] ] here we have inserted , by dimensional analysis , the spin - wave velocity v. taking the isotropic limit on both sides of this equation , using @xmath249 , gives : @xmath250^{3/2}.\ ] ] we see that this is equivalent to the rg result of eq . ( [ hxxz ] ) with the amplitude determined to be : @xmath251^{3/2 } \label{b}\ ] ] from eq . ( [ deltah0 ] ) , for @xmath120 , the gap behaves as : @xmath252 with : @xmath253^{2/3 } { \gamma ( 1/6)\over \gamma ( 2/3)}\approx 1.77695.\label{a0ex}\ ] ] since we were not aware of the result [ ] at the time , in [ ] the behavior of the gap with staggered field ( for @xmath120 ) was estimated numerically by extrapolating lanczos results for lengths up to 22 sites . a very good fit to eq . ( [ deltah02 ] ) was obtained with : @xmath254 considering the numerical difficulties related to logarithmic corrections this is remarkably good agreement with the exact result of eq . ( [ a0ex ] ) . in the experimentally relevant case , @xmath216 , the gap behaves as in eq . ( [ delta ] ) and , for small @xmath255 , the amplitude is given by eq . ( [ ah ] ) with , from eq . ( [ ah0 ] ) , @xmath256 thus our expression for the gap becomes : @xmath257^{1/6}(h / j)^{1/[2-\pi r(h)^2)]}. \label{deltaf1}\ ] ] for larger @xmath255 greater accuracy might be obtained by using eq . ( [ ah2 ] ) with @xmath212 given in eq . ( [ b ] ) . that is : @xmath258/2}[2(1 - 2\pi r(h)^2)]^{1/4 } \right\}^{-1/[2-\pi r(h)^2 ] } ( h / j)^{1/[2-\pi r(h)^2]},\ ] ] where @xmath15 is given , from the bethe ansatz , in fig . ( [ fig : r(h ) ] ) . inserting its asymptotic value at low @xmath17 , @xmath259 gives back eq . ( [ deltaf1 ] ) . we note that , to actually fit the experimental data , we take @xmath260 for some constant of proportionality which depends on field direction but is generally of order a few % . thus the actual scaling of gap with field is not a pure power law . for @xmath120 it is convenient to use non - abelian bosonization , eq . ( [ nonabbos ] ) so that the interaction term is written : @xmath261 in this case the model has an su(2 ) symmetry . note that when both uniform and staggered fields vanish and ignoring the marginal operator , the symmetry is actually su(2)@xmath262su(2 ) : @xmath263 these two independent su(2 ) s act on left and right - movers separately . the ordinary su(2 ) symmetry of the spin chain is the diagonal subgroup with @xmath264 . this symmetry is broken by the staggered field . however , a different su(2 ) subgroup of the original su(2)@xmath262su(2 ) survives for which : @xmath265 we may redefine the field @xmath10 by @xmath266 in which case the interaction becomes @xmath267 , which has the diagonal su(2 ) symmetry . in fact , this continuum limit interaction arises from a staggered heisenberg exchange interaction , as occurs in the spin - peierls problem . the equivalence of the continuum limit of these two apparently very different problems , is a non - trivial consequence of the chiral symmetry which maps @xmath268 into @xmath269 . if we assume that the only effect of the uniform field is to add a term to the hamiltonian : @xmath270 then the uniform field can be removed by the gauge transformation : @xmath271 or equivalently : @xmath272 [ here we have set @xmath273 . ] this transforms the matrix field @xmath10 as : @xmath274 this gauge transformation leaves invariant the staggered field term tr@xmath275 . thus the exact su(2 ) symmetry remains in this approximation . however , additional irrelevant terms in the hamiltonian in the presence of a uniform field break the exact su(2 ) symmetry , as evidenced by the change in the parameter @xmath16 with field . the su(2 ) symmetry is only present for @xmath123 . the non - abelian bosonization formula for the staggered part of the spin operators ( for @xmath120 ) is : @xmath276 the magnetic field leads to the gauge transformation of eq . ( [ gauge ] ) . the parts of the spin operators with wave - vectors near @xmath1 thus become : @xmath277 now making the transformation @xmath278 , this becomes : @xmath279 in general the spectrum of the sine - gordon theory consists of the soliton , anti - soliton and breathers ( soliton - anti - soliton boundstates).@xcite in the su(2 ) symmetric case , the excitation spectrum of the sine - gordon model with @xmath280 consists of a triplet composed of soliton , anti - soliton and lowest breather and a second breather , heavier by a factor of @xmath281 . the degeneracy of the triplet is a result of the su(2 ) symmetry . due to the su(2 ) symmetry , the 3 elements of the triplet are produced by the 3 operators tr@xmath282 with equal intensity , whereas the singlet is produced by the operator tr@xmath10 . thus setting @xmath120 ( but not @xmath68 ) the structure functions @xmath283 and @xmath284 would be equal . note that , for @xmath120 , @xmath285 , creates the y - polarized member of the triplet . this can be regarded as a linear combination of the soliton ( created by tr@xmath286 ) and the anti - soliton ( created by tr@xmath287 ) . @xmath288 consists of 2 identical contributions from the @xmath289 and @xmath290 terms in eq . ( [ strans ] ) . each contributes exactly @xmath291 . the effect of @xmath17 is to split the soliton and anti - soliton contributions to @xmath284 into two separate contributions at different wave - vectors @xmath292 . thus , ignoring the small @xmath21 symmetry breaking : @xmath293 it is also interesting to note that the staggered part of the energy density is given by : @xmath294 this operator couples to lattice displacements ( phonons ) and is used to describe raman scattering experiments . upon making the gauge transformation of eq . ( [ gauge ] ) and the redefinition of eq . ( [ redefg ] ) this becomes : @xmath295 thus this operator also creates the soliton and anti - soliton . hence this theory predicts a single particle excitation observable in raman scattering at the same field - dependent wave - vector and frequency as the incommensurate mode observed in neutron scattering . upon allowing for su(2 ) symmetry breaking the radius changes . after making the gauge transformation , a u(1 ) symmetry still survives , corresponding to shifting @xmath117 by a constant . the triplet is now split , with the lowest breather having a different mass than the degenerate soliton antisoliton pair . since the operators @xmath296 have charge @xmath297 with respect to this @xmath298 symmetry , we see that the soliton and anti - soliton are created by the @xmath299 fourier modes of @xmath300 respectively . the breathers can be classified as even or odd with respect to the discrete symmetry @xmath301 . the odd breathers are created by the @xmath302 component of @xmath303 and the even breathers by the @xmath302 component of @xmath304 . it can be shown that even and odd breather alternate in the spectrum of the sine - gordon model . furthermore , the number of breathers increases with decreasing @xmath16 . a third breather drops below the soliton antisoliton ( @xmath305 ) continuum immediately as soon as @xmath16 decreases below the isotropic value , @xmath126 with another one dropping below the @xmath305 continuum each time @xmath306 passes through an integer . the mass of the @xmath307 breather , expressed in terms of the soliton mass , @xmath308 , is : @xmath309 where : @xmath310 thus the odd - numbered breathers contribute single - particle poles to @xmath283 and the even - numbered ones to @xmath311 while the soliton and antisoliton contribute single - particle poles to @xmath284 . in addition various multi - particle continua contribute to the three spectral functions . for a field of 7 t. we estimate @xmath312 . there are 3 breathers at this point , with masses .79 m , 1.45 m and 1.87 m . a resolution limited peak was observed at @xmath313 , at energy .22 mev . we identify this with the soliton ( or anti - soliton ) contribution to @xmath284 ; hence m=.22mev . a resolution limited peak is clearly observable in the neutron scattering data at @xmath302 and an energy of @xmath314mev=@xmath315 . this agrees very well with the prediction for the first breather mass . we may also test the su(2 ) prediction of eq . ( [ intrat ] ) . the su(2 ) symmetry is broken by various small effects as exemplified by the fact that @xmath316 . in particular , this implies that @xmath317 has a second peak , of very low intensity , corresponding to the third breather . ignoring these effects we expect the intensity of the lowest breather peak in @xmath283 to be approximately twice the intensity of the soliton peak in @xmath284 . however , before a comparison can be made with experiment it must be taken into account that the unpolarized neutron scattering cross - section contains an important direction dependence arising from the fourier transform of the dipole - dipole interaction between the neutron and the spins . the cross - section can be written : @xmath318 where @xmath319 is a unit vector in the direction of @xmath320 and the function @xmath321 is slowly varying . thus the soliton , even breathers and odd breathers are weighted by different factors @xmath322 , @xmath323 and @xmath324 , respectively . we also have to consider the variation of @xmath321 in examining the relative intensity of solitons to breathers since they occur at different values of @xmath325 . it must be recalled that @xmath326 here refers to the direction of the effective staggered magnetic field ( and @xmath327 the direction of the uniform field ) . in the neutron scattering experiments the field was along the b - axis . we note that the @xmath27 axis is rotated by about @xmath328 from the crystal axis , @xmath20 . [ we define the rotation angle in the @xmath329- ( @xmath35- ) plane so that @xmath25 axis is @xmath330 rotated from @xmath20-axis . ] strictly speaking , we must take into account the effect of the redefinition of the spin operators , discussed in section ii , that was used to eliminate the dm interaction . letting @xmath331 to be the rotated spin operators , defined in eq . ( [ redef ] ) , and inverting the transformation , we may write the structure function for the original spin operators in terms of the structure function for the rotated operators , which we write as @xmath332 . in this way we obtain : @xmath333 @xmath334 remains 0 due to translation invariance and the @xmath335 are unaffected by the transformation . here @xmath336 and @xmath337 refer to two axes orthogonal to @xmath22 ( _ not _ orthogonal to @xmath338 as in most of this paper . ) we expect the second terms in eqs . ( [ gredef ] ) to be negligible since @xmath339 is small . furthermore , the @xmath340 are small for @xmath341 also making the second terms in eq . ( [ gredef ] ) smaller than the first for @xmath342 near @xmath1 . henceforth we ignore this small correction and simply use @xmath343 . the direction ( and magnitude ) of the effective staggered field depends on both the staggered part of the @xmath10-tensor and dm interaction . since the dm interaction in cu benzoate is unknown , the direction @xmath344 is not known at present . ( the dm interaction in cu benzoate may be estimated from various experimental results based on the present theory . we will discuss this issue in the next section . ) however , the direction @xmath344 can be deduced from the polarization analysis@xcite of the neutron scattering experiment . dender et al . analysed the polarization of the neutron scattering at constant energy @xmath345 and various momentum transfers . for magnetic field @xmath346 b and momentum transfer along the chain @xmath1 , this should probe the lowest ( @xmath347 ) breather . as discussed above , this odd breather is polarized orthogonal to the total effective staggered field . in ref . [ ] it was claimed that the observed scattering is polarized in the @xmath27 direction . also note that a misstatement of the crystal orientation occurred in ref . [ ] so that the wrong sign appeared there for @xmath348.@xcite correcting this error , the polarization is @xmath349 from a - axis , or equivalently @xmath350 from @xmath27-axis . this implies that , for an applied field in b - direction , the total effective staggered field @xmath351 ( @xmath352 ) from a - axis , or equivalently @xmath353 ( @xmath354 ) from @xmath27-axis . [ there is actually another feature of the polarization analysis in ref . [ ] which appears inconsistent with the theory presented here , namely the polarization analysis in zero field . the strong dependence of the intensity on the a - component of the momentum is taken to indicate that @xmath355 . our work ignores any anisotropy in the zero field limit and therefore predicts @xmath356 in that limit . we do not understand the source of this discrepancy at present . ] the constant-@xmath357 ( momentum transfer ) scan experiments , sensitive to the breather modes , were carried out at the fixed momentum transfer : @xmath358 where the lattice constants are @xmath359 , @xmath360 and @xmath361 . here we have again corrected the misstatement@xcite of the crystal orientation in ref . [ ] mentioned above . note that the antiferromagnetic wave - vector , in the chain direction , is actually @xmath362 rather than the normal @xmath363 , because there are 2 cu atoms per unit cell along the c - axis . the crystal axes @xmath20 , @xmath28 and @xmath25 are essentially orthogonal . thus , in the a - b - c system : @xmath364 @xmath319 is rotated @xmath365 from the @xmath20 axis , or @xmath366 from the @xmath27-axis . we note that , this @xmath367 is almost parallel to the direction of the total effective staggered field estimated from the polarization analysis above . the constant-@xmath357 scan for the soliton modes was done for a slightly different momentum transfer @xmath368 however , the direction of this momentum transfer @xmath367 is almost the same as the above and we will ignore the difference . the fact that , an intense first breather peak is observed experimentally , supports the deduction that @xmath319 is nearly parallel to the total effective staggered field . in fact , assuming that @xmath319 is completely parallel to the staggered field , the approximate su(2 ) prediction is that the lowest peak at @xmath104 ( from the first breather in @xmath283 ) should have twice the intensity of the lowest peak at @xmath369 ( from the soliton / antisoliton in @xmath284 ) . this prediction is only a very approximate one due to the breaking of su(2 ) . there could also be corrections from the function @xmath321 in eq . ( [ neut ] ) . experimentally this ratio appears to be about @xmath370 . this is perhaps satisfactory agreement . this agreement is only worsened if @xmath367 deviates from the direction of the staggered field . thus , analyses of polarization and of the scattering intensity of the lowest breather mode are consistent , and apparently lead to the conclusion that the total staggered field for h@xmath371b is almost parallel to the @xmath367 direction of eq . ( [ kdir ] ) used in the constant-@xmath357 scan . on the other hand , in ref . [ ] we discussed a feature in the experimental data at @xmath104 , and @xmath372 . this is very close to the predicted mass of the second breather , which contributes to @xmath311 . a recent calculation , based on integrability of the sine - gordon model , indicates that the relative intensity of the second breather in peak in @xmath311 should be roughly 1/2 of the intensity of the first breather in @xmath283 . however , we expect this to have essentially zero intensity in the neutron scattering cross - section due to the factor of @xmath323 in eq . ( [ neut ] ) . [ in ref . [ ] an apparently good agreement between theory and experiment was obtained because the factors of @xmath373 were not included . in ref . [ ] intensities were not considered . ] a possible resolution of this disagreement is that this `` feature '' at @xmath374 , discussed by two groups of theorists , is just noise . as stated in ref . [ ] `` given the quality of the data , this double - gap conjecture is highly speculative '' . clearly more data is needed to determine whether or not there is really another sharp peak at this frequency . an even more statistically insignificant feature in the data , at @xmath104 and @xmath375 , was discussed in ref . [ ] where it was interpreted as the third breather peak in @xmath283 . both the energy ( @xmath376 ) and the intensity ( very approximately 1/6 the intensity of the first breather ) agree with the predictions of the sine - gordon model . note that in this case the factor of @xmath324 is common to first and third breathers so it does nt affect the intensity ratio . however , once again , considerably more data is needed to determine if there is really a peak at this frequency . another striking feature of the neutron scattering data is a second resolution limited peak at @xmath104 and @xmath377 . this also has a possible interpretation in our field theory approach . it is natural to assume that this peak actually comes from @xmath284 . at this wave - vector , from eq . ( [ strans ] ) , this is proportional to @xmath378 at wave - vector @xmath379 . we expect the continuum limit to hold for some range of wave - vectors close to the gap minimum at wave - vector @xmath380 . thus , at least for weak enough fields , we would expect the soliton to persist as a single - particle excitation in @xmath284 up to wave - vector @xmath1 . its energy should obey the lorentz invariant formula : @xmath381 ( here @xmath342 is measured from the incommensurate wave - vector @xmath382 and @xmath172 is set equal to 1 . ) since @xmath383 this gives approximately @xmath384 , as seen in the experiment . the intensity of this feature in @xmath284 can be easily calculated . the result follows from the fact that the @xmath342 near @xmath1 parts of the spin operators are all lorentz scalars . the matrix element between groundstate and a single particle excited state of a lorentz scalar operator is independent of the momentum of the particle , by lorentz invariance ( assuming a lorentz - invariant normalization of the state ) . it then follows that the soliton and anti - soliton peaks in @xmath284 have an intensity that is proportional to @xmath385 . the energy is approximately four times higher at @xmath104 . we must also take into account that both soliton and anti - soliton are contributing at @xmath104 which increases the intensity by a factor of 2 . thus , we expect the single - particle peak at @xmath386 , @xmath104 to have an intensity approximately 1/2 that of the peak at @xmath387 . experimentally this ratio looks somewhat larger than 1/2 but it must be remembered that the peak at @xmath386 , @xmath104 is sitting on top of a background from @xmath283 . taking this into account , the agreement looks fair . there is actually a possible objection to this argument . the same reasoning would seem to imply sharp peaks near @xmath384 at @xmath387 coming from @xmath283 and @xmath311 . these were not observed experimentally ; at most a small shoulder , was observed beginning at @xmath384 for @xmath387 . this may simply mean that the breathers have merged into the continuum by this wave - vector , ( due to non - relativistic effects not contained in the continuum limit field theory ) whereas the soliton has not . we note that spin - wave theory fails to capture the one - dimensional quantum fluctuation dominated physics in various ways . it predicts a single low energy mode with @xmath388 , instead of @xmath389 with soft wave - vectors @xmath1 and @xmath177 , missing the incommensurate shift . it also predicts another single particle mode at energy approximately @xmath17 , at the same wave - vectors . in the present framework , the only unknown parameters of cu benzoate are the dm vector , which has not been determined directly in previous studies . based on the present theory , we can in principle determine the dm vector from several experimental results . actually , there seem to be no solution that can perfectly fit all the available experimental data , as explained below . presumably , a precise error estimate on an experiment gives a permissible region for the dm vector , and such constraints from several experiments would give a region of possible values of the true dm vector . however , it should be noted that we have been ignoring the interchain effects , irrelevant operators etc . which might be necessary in such a precise discussion . firstly , as argued in sec . ii @xmath22 must lie in the @xmath390 plane , leaving two free parameters . the total staggered field is determined by eq . ( [ huaeff ] ) . here we list the constraints on the dm vector from various experiments . as observed in ref . [ ] , the induced gap is strongly dependent on the direction of the applied uniform field . in the present theory , this is accounted by the field direction dependent constants of proportionality between the uniform and staggered magnetic fields . as discussed in sec . ii , the proportionality constant should be given by the @xmath10-tensors and the dm vector @xmath43 . since the @xmath10-tensors were obtained previously@xcite , the measured gap can be used to determine @xmath43 . the gap is proportional to essentially the 2/3 power of this effective staggered field . the specific heat measurements of ref . [ ] were fit by the authors to the specific heat of a collection of free massive relativistic bosons . the masses were found to scale approximately as @xmath9 with a direction dependent amplitude in the @xmath391 . in ref . [ ] , this ratio was used to estimate the dm vector . very recently@xcite the specific heat of the sine - gordon model was calculated from the thermal bethe ansatz and fit to the cu benzoate data . again a good fit of the gap to @xmath9 was obtained for fields in the @xmath28 or @xmath26 directions with a somewhat different amplitude ratio @xmath392 . ( the velocity , @xmath14 is another parameter in the fit . this may also be determined from bethe ansatz for the s=1/2 chain in a uniform field . a slightly better fit to the specific heat data was obtained in [ ] by letting @xmath14 be a free parameter . ) a reasonable fit was not obtained for the field in the @xmath27 direction where the specific heat data is nearly linear . this suggests that , for h@xmath393 the apparent gap structure was either due to some sort of experimental error or due to other mechanisms@xcite than the effective staggered field . in any case , it seems that the effective staggered field for h@xmath393 is rather close to zero . this implies the cancellation of the effective staggered field coming from the staggered @xmath10-tensor and the dm interaction . it is not quite unnatural , because for the applied field in ac - plane , both the staggered field generated by the staggered @xmath10-tensor and the dm interaction point to b - direction . thus there is a direction in the ac - plane where the cancellation occurs , for a wide range of parameters . actually , the cancellation of the staggered field is also consistent with the electron spin resonance ( esr ) result , which will be explained later . assuming the cancellation of the staggered field for h@xmath393 , and that observed gap for h@xmath371b and h@xmath394 are entirely due to the staggered field , the ratio of the proportionality constants between staggered and uniform fields for h@xmath393 , h@xmath371b and h@xmath394 are @xmath395 . the ratio gives a constraint on the dm vector through ( [ huaeff ] ) . based on several exact results on the sine - gordon field theory and on the @xmath396 heisenberg antiferromagnetic chain , we have determined the magnitude of the gap for a given staggered field @xmath68 . thus , comparing this with the gap estimated from the specific heat measurement , we can fix the proportionality constant @xmath25 between the staggered field @xmath68 and the applied field @xmath17 ( @xmath397 ) . here we use only the result for h@xmath394 , which is presumably most reliable . for h@xmath394 , the gap is very well fit by the power law @xmath398 , without introducing the logarithmic correction . the proportionality constant is given by @xmath399 if @xmath111 and @xmath17 are measured in units of kelvin and tesla , respectively . by comparing this with eqs . ( [ deltah02 ] ) and ( [ a0ex ] ) , where we assume the logarithmic factor to be close to unity for the present case , we obtain @xmath400 . this also gives a constraint on the dm vector through eq . ( [ huaeff ] ) . an anomalous broadening of esr , which is strongly dependent on the field direction , at low temperatures was observed@xcite in cu benzoate . the mechanism of this broadening was left unexplained . however , we have recently developed a field - theory approach to esr on quantum spin chains at low temperature.@xcite according to the theory , the contribution of the staggered field to the esr linewidth is given by @xmath401 this diverges at lower temperature , in agreement with the experiment . the direction dependence can also be explained by the direction - dependent proportionality constant between the effective staggered field and applied uniform field . actually , this is consistent with the previous discussion on the field - induced gap , at least qualitatively . in particular , for h@xmath393 , the low - temperature anomalous part of the esr linewidth vanishes . this implies the cancellation of the staggered field for h@xmath393 . while this appears to contradict the apparent gap found in ref . [ ] , it is rather consistent with more refined analysis discussed in the last subsection . in addition , h@xmath394 gives the largest linewidth , which is consistent with the larger field - induced gap for h@xmath394 . on the other hand , the ratio of the staggered field is not quantitatively consistent with the specific heat measurement . the estimate of the staggered field is somewhat subtle because there are also contributions to the esr linewidth from other sources ( most importantly exchange anisotropy / dipolar interaction ) . the low - temperature anomalous part , which is related to staggered field , appears to be approximately @xmath402 for h@xmath371b and h@xmath394 . this gives the ratio of the staggered field @xmath403 for h@xmath371b and h@xmath394 . this is smaller than expected from the specific heat analysis . as we have discussed in section vi , the analyses on the polarization and intensity of the first breather suggests that the total effective staggered points to @xmath351 ( @xmath352 ) from a - axis , or equivalently @xmath353 ( @xmath354 ) from @xmath27-axis , if the external field is applied in b - direction . this gives another constraint on the dm vector . there are several experimental data which give some constraints on the dm vector , and they are not perfectly consistent . the estimate of dm vector is also sensitive to the assumed form of the g - tensor , extracted from esr measurements.@xcite the experimental data are presumably subject to several errors which has not been identified precisely . we hope that more experimental data will be available in the future to make more precise comparison with the theory . the esr linewidth was measured for various directions of the applied field . thus it is perhaps quite reliable that the staggered field precisely cancels at h@xmath393 . as we have discussed , this is rather consistent with the refined specific heat result . this gives a single constraint on the dm vector . in the linearized approximation , it reads @xmath404 the estimated ratio of the staggered field for h@xmath405b and h@xmath394 was inconsistent between the specific heat and esr . however , it should be recalled that each analysis has its own problem . in the specific heat analysis , an apparent gap structure , which is unrelated to the staggered field , was observed for h@xmath393 . whatever the origin of this gap structure , it is natural to expect similar contributions also for other field directions . unfortunately , we do not know how to estimate these effects at present . on the other hand , there are also contribution to the esr linewidth from other sources than the staggered field , and the subtraction causes some uncertainty , in addition to the estimate of the linewidth itself . the constraint from various experimental result on the dm vector is summarized in fig . ( [ fig : dmest ] ) . for the case of neutron scattering polarization we have included an estimated error bar from the polarization analysis of ref . we have not attempted to estimate error bars in the other cases . we see that a candidate dm vector @xmath406 , which satisfies the most reliable requirement of the cancellation for h@xmath393 , is roughly consistent with all the constraints except for the ratio of the gap between h@xmath371b and h@xmath394 . this suggests that the observed ratio of the gap was wrong , or was affected by other factors than the staggered field . more experimental data are needed to draw a reliable conclusion . in the first subsection we consider the staggered susceptibility , resulting from the application of a staggered magnetic field , using the mapping onto the sine - gordon model . in the next subsection we combine this with the standard uniform magnetization of the s=1/2 heisenberg model to obtain the total physical susceptibility . we first discuss the susceptibility of the sine - gordon model . writing the sine - gordon lagrangian in the form : @xmath407 and adopting units where the velocity is set to 1 , we define the sine - gordon susceptibility as : @xmath408 where @xmath409 is the free energy . the groundstate energy ( i.e. the @xmath410 free energy ) is expressed in terms of the gap as:@xcite @xmath411 this determines the @xmath410 susceptibility using the exact relationship between the coupling constant , @xmath412 and the mass @xmath111 : @xmath413 where @xcite @xmath414^{2/3}\left[{\gamma ( 1/6)\over \gamma ( 2/3)}\right ] \approx 4.82764 . \label{atildedef}\ ] ] this gives the @xmath410 susceptibility : @xmath415 the high - temperature susceptibility is given by : @xmath416 this green s function is normalized to @xmath417 ( at @xmath418 ) . at finite @xmath419 and @xmath420 we have : @xmath421^{1/2}}.\label{gft}\ ] ] here @xmath422 . thus , the susceptibility becomes : @xmath423^{1/2}}={1\over t}\left[{\gamma ( 1/4)\over \gamma ( 3/4)}\right]^2 \approx { 8.75376 \over t}.\label{chisght}\end{aligned}\ ] ] ( this result for the integral can be obtained , by analytic continuation , from the general results of schulz.@xcite ) an integral equation determining the sine - gordon free energy at finite @xmath420 was given in ref . [ ] . we may determine the susceptibility by differentiating twice with respect to @xmath412 . note however that this integral equation actually determines @xmath424 so we must add the zero - temperature part of the susceptibility , given above . the resulting susceptibility , @xmath425 is plotted versus @xmath426 in fig . ( [ fig : susc ] ) . as expected , it agrees quite well with high temperature result @xmath427 down to @xmath428 . it has a maximum at about @xmath429 . the @xmath410 value is given by eq . ( [ chi0sg ] ) , ( [ atildedef ] ) . up to a multiplicative factor and logarithmic corrections , the sine - gordon susceptibility essentially gives the staggered susceptibility of the s=1/2 chain , i.e. its response to a staggered field : @xmath430 in order to determine this factor and estimate the logarithmic corrections we first consider the @xmath410 staggered magnetization of the s=1/2 chain . we refer to the staggered magnetization as @xmath431 : @xmath432 in the continuum limit , @xmath433 since this operator has scaling dimension @xmath187 , a standard rg scaling argument gives the scaling of the staggered magnetization with staggered field : @xmath434 for some function @xmath435 . for weak fields the exponent is approximately 1/3 . in a similar way to our analysis of the gap in the previous section , by combining the exact results for the xxz model with an rg analysis of the marginal operator we may determine the scaling of magnetization with staggered field in the case of zero uniform field : @xmath436^{1/3 } \label{magscale2}\ ] ] and determine the behavior of @xmath437 in eq . ( [ magscale1 ] ) for small field : @xmath438^{1/3}.\label{cas}\ ] ] @xmath431 obeys a standard rg equation relating a change in the cut - off energy scale , @xmath153 , to a change in the coupling constant , @xmath160 : @xmath439m=0.\ ] ] working to linear order in the marginal couplings , as before , we set @xmath440 and use eq . ( [ gamma ] ) . using eq . ( [ rge ] ) , and lowering the cut - off scale to the gap , @xmath111 , this gives : @xmath441^{-1/4 } , \label{mlambda}\ ] ] for some constant , @xmath409 . ( [ hlambda ] ) , and ( [ a0ex ] ) ) this can be written : @xmath442^{-1/3},\ ] ] where @xmath193 is defined in eq . ( [ gaph0 ] ) and determined in eq . ( [ a0ex ] ) . as in the previous sections we obtain the various formulae from the different asymptotic scaling of @xmath443 in the 3 cases : @xmath120 , @xmath225 and @xmath120 with exchange anisotropy @xmath237 , defined in eq . ( [ epsilon ] ) . ( [ lambdaiso ] ) we obtain eq . ( [ magscale2 ] ) with : @xmath444 using eq . ( [ lambdar ] ) we obtain eq . ( [ magscale1 ] ) with : @xmath445^{-1/3 } \left ( \frac{h}{j}\right)^{-2(1 - 2\pi r^2)/3}.\ ] ] using eq . ( [ lambdaz0 ] ) we obtain eq . ( [ cas ] ) with : @xmath446 using eq . ( [ lambdaep ] ) , in the limit @xmath236 , in eq . ( [ mlambda ] ) we obtain an expression for @xmath431 in terms of @xmath111 with exchange anisotropy : @xmath447 we may determine the constant @xmath409 , by comparing to the exact result of ref . [ ) ] . from eq . ( [ prop ] ) : @xmath448>,\ ] ] with the exact formula for @xmath449 give in eq . ( [ a(r ) ] ) . this expectation value is given in terms of the soliton mass , @xmath111 in [ ] , eq . ( 15 ) , with : @xmath450 inserting a power of the spin - wave velocity , @xmath172 by dimensional analysis , and taking the limit of small @xmath237 , we obtain : @xmath451 this gives a result consistent with eq . ( [ manis ] ) and determines the constant , @xmath409 , to be : @xmath452^{-3/2}=2a_0^{3/2}/(3\sqrt{3}\pi ) , \ ] ] where the constant @xmath193 is defined in eq . ( [ gaph0 ] ) and its value is given in eq . ( [ a0ex ] ) . here we have used the exact identity @xmath453 . hence the amplitude of eq . ( [ magscale2 ] ) is given by : @xmath454 we thus obtain the @xmath410 staggered susceptibility of the s=1/2 chain by differentiating @xmath431 with respect to @xmath68 : @xmath455 for @xmath120 . comparing to eqs . ( [ atildedef ] ) and ( [ chi0sg ] ) , we see that : @xmath456 the susceptibility for @xmath457 ( and @xmath120 ) follows from eqs . ( [ gft])-([chisght ] ) . here we use the exact result for the @xmath410 correlation function of the s=1/2 chain:@xcite @xmath458 this differs from the correlation function of the sine - gordon model by a factor of @xmath459 . [ note the factor of 4 difference in the susceptibilities due to the factor of 2 in the interaction term of the sine - gordon lagrangian of eq . ( [ sgnorm ] ) . ] upon going to finite @xmath420 and fourier transforming at zero frequency and wave - vector , we expect the logarithmic factor to become : @xmath460 . thus : @xmath461 comparing to eq . ( [ chia0 ] ) suggests the heuristic formula : @xmath462\over 2(2\pi ) ^{3/2}}\chi_{sg}(t,\delta ) .\ ] ] @xmath463\approx \chi_s$ ] is plotted in fig . ( [ fig : susc ] ) . including a small uniform field , @xmath17 , only makes unimportant changes in these formulas . at @xmath410 , the power of @xmath111 in eq . ( [ chia01 ] ) changes by a small amount ; the argument of the logarithm changes to @xmath464 and the amplitude by a factor of @xmath465 . for @xmath457 we must distinguish two regimes , depending on the relative magnitude of @xmath17 and @xmath420 . for @xmath466 , the power of @xmath420 changes . on the other hand , for @xmath467 we expect to obtain eq . ( [ chialt ] ) . above we considered the staggered susceptibility resulting from ( independent ) staggered and uniform fields . to make any comparison with experiments we must take into account that the effective staggered field is proportional to the uniform field , @xmath468 where the constant of proportionality , @xmath25 is strongly dependent on field direction . the physical susceptibility is conveniently obtained from its thermodynamic definition : @xmath469 where @xmath409 is the free energy . writing @xmath409 as a function of uniform and staggered fields , we must set @xmath260 before taking the @xmath17-derivative . we may calculate the free energy in the rotated spin basis of eq . ( [ redef ] ) , used throughout this paper . noting that the first derivative of @xmath409 with respect to @xmath17 or @xmath68 gives the uniform magnetization , @xmath470 and staggered magnetization @xmath431 respectively ( in the rotated basis ) , we obtain : @xmath471 where we have used : @xmath472 we may ignore the dependence on @xmath68 of the first term and use the standard result for the uniform susceptibility of the @xmath396 chain , @xmath473 . for low fields and temperatures this gives : @xmath474 independent of field and temperature . the second term in eq . ( [ chiphys ] ) is larger than the third so we approximate : @xmath475 where @xmath476 is the staggered susceptibility discussed in the previous subsection . thus , when measuring the physical susceptibility of the present system , one actually probes also the staggered susceptibility @xcite . while the first term , the standard result for the susceptibility of the s=1/2 chain , goes to a finite constant , at @xmath420 an @xmath18 , the second term , resulting from the effective staggered field , is highly singular . at zero field it blows up at @xmath477 as @xmath478 . thus , although it is multiplied by the small constant , @xmath479 , it eventually dominates for low enough @xmath420 . for any finite field , the divergence of the second term is cut off , at essentially the gap energy , @xmath480 , at a value of approximately @xmath481 , as shown in fig . ( [ fig : susc ] ) . at low fields the behavior looks quite similar to a paramagnetic impurity contribution . it can be distinguished from that , however , by its very strong field - direction dependence . the effect is largest for the field in the c - direction when the parameter @xmath482 . the experimental susceptibility@xcite of cu benzoate shows very peculiar behavior at low fields and temperatures . as the temperature is lowered , in low fields , the susceptibility grows . this effect is highly direction dependent with the biggest effect occurring for fields in the c - direction . this effect is cut off by the application of a field . qualitatively , the experimental results look similar to our predictions . however , at a more quantitative level there are many differences . the anomalous low field low t part of @xmath483 is much larger in the experiments than in the theory . for instance , at zero field , a temperature of @xmath484 , and @xmath485 , the value that we find for h@xmath371c , the second term in eq . ( [ chiphys ] ) is still smaller than the first by a factor of about 1/2 . the anomalous contribution could be somewhat larger due to the third term in ( [ chiphys ] ) and the logarithmic correction . however , the anomalous term in @xmath486 observed in cu benzoate at this zero field and low t is about six times _ larger _ than the @xmath473 . this is perhaps too large to be explained by our theory which is purely one - dimensional . furthermore , the detailed experimental dependence on t and h is quite complex . the maximum susceptibility occurs at a field of about 30 g . and t=1.5k . at lower fields or temperatures the susceptibility decreases somewhat . at low fields , two extra low temperature peaks are observed in the susceptibility as a function of t ( in addition to the normal one for an s=1/2 chain at @xmath487 ) . while it is tempting to try to identify the higher-@xmath420 peak with the peak in @xmath476 shown in fig . ( [ fig : susc ] ) , and the lower-@xmath420 peak with the onset of nel order , both susceptibility peaks are broad , unlike what would be expected from a phase transition . ( indeed , no evidence for magnetic order was found from neutron scattering,@xcite down to t=@xmath488k . ) a strong frequency dependence of the low t and h susceptibility was also observed . we expect that a proper theoretical description of the low field and temperature susceptibility of cu benzoate will require the inclusion of inter - chain coupling effects . the interchain super - exchange paths look complicated and it is not even clear what is the sign of the interchain coupling . we remark that a perfectly one - dimensional s=1/2 antiferromagnet with su(2 ) symmetry broken only by the dm interaction has a disordered groundstate . this follows from our spin redefinition in sec . ii which maps the system into an _ easy plane _ xxz model which is well - known to have a disordered groundstate with power - law correlations . we may think of the spins as fluctuating primarily in the plane perpendicular to @xmath22 with a tendency for the cross - product of neighboring spins to be parallel to @xmath22 . this system has a u(1 ) symmetry . interchain coupling would normally be expected to produce long - range order with both an anti - ferromagnetic component and a possible perpendicular ferromagnetic component . for example , for @xmath489 , @xmath490 see fig . ( [ fig : dmclass ] ) . the standard mean field treatment of the interchain interactions @xcite would suggest a critical temperature of order the interchain coupling ( assuming that it is antiferromagnetic ) . on the other hand , when a magnetic field is applied ( not parallel to @xmath22 ) , the u(1 ) symmetry is broken and the phase transition should disappear . @xmath431 and @xmath470 now becomes non - zero even in the purely one - dimensional system and are no longer order parameters for spontaneous symmetry breaking . thus the application of a magnetic field smoothes out the phase transition in this system . surprisingly , neutron scattering experiments in zero field have failed to detect a nel transition . we would like to thank y. ajiro , c. broholm , w. hardy and d. reich for helpful discussions . m.o . thanks t. kato for his help on drawing crystal figures . this research was supported in part by nserc of canada and by the ubc killam fund . dender , p.r . hammar , d.h . reich , c. broholm and g. aeppli , phys . lett . * 79 * , 1750 ( 1997 ) . dender , ph.d . thesis , johns hopkins university , 1998 . m. oshikawa and i. affleck , phys . lett . * 79 * , 2883 ( 1997 ) . i. dzyaloshinskii , j. phys . solids * 4 * , 241 ( 1958 ) . t. moriya , phys . rev . * 120 * , 91 ( 1960 ) . essler and a.m. tsvelik , phys . rev . * b57 * , 10592 ( 1998 ) . essler , cond - mat/9811309 . k. oshima , k. okuda and m. date , j. phys . . jpn . * 44 * , 757 ( 1978 ) . l. shekhtman , o. entin - wohlman and a. aharony , phys . * 69 * , 836 ( 1992 ) . t. sakai and h. shiba , j. phys . japan * 63 * , 867 ( 1994 ) . cardy , _ scaling and renormalization in statistical physics _ , cambridge university press , 1996 . s. lukyanov and a. zamalodchikov , nucl . phys . * b493 * , 571 ( 1997 ) . i. affleck , j. phys . * a31 * , 4573 ( 1998 ) . s. lukyanov , preprint cond - mat/9712314 . i. affleck , d. gepner , h.j . schulz and t. ziman , j. phys . * a22 * , 511 ( 1989 ) . n. m. bogoliubov , a. g. izergin and v. e. korepin , nucl . * b275 * , 687 ( 1986 ) ; v. e. korepin , n. m. bogoliubov and a. g. izergin , _ quantum inverse scattering method and correlation functions _ , cambridge university press ( 1993 ) . griffiths , phys . rev . * 133 * , a768 ( 1964 ) . a. fledderjohan et al . phys . rev . * b54 * , 7168 ( 1996 ) . cabra et al . phys . rev . * b58 * , 6241 ( 1998 ) . hammar et al . phys . rev . * b59 * , 1008 ( 1999 ) . r. dashen , b. hasslacher and a. neveu , phys . ref . * d11 * , 3424 ( 1975 ) . d. reich , private communication . k. okuda , h. hata and m. date , j. phys . . jpn . * 33 * , 1574 ( 1972 ) m. oshikawa and i. affleck , preprint cond - mat/9904199 . baxter , _ exactly solved models in statistical mechanics _ . london : academic press ( 1982 ) . c. destri and h. de vega , nucl . b * 358 * , 251 ( 1991 ) . schulz , phys . rev . * b34 * , 6372 ( 1986 ) . m. fowler and x. zotos , phys . b * 25 * , 5806 ( 1982 ) . m. matsuura and y. ajiro , j. phys . . jpn . * 41 * , 44 ( 1976 ) . dender , d. davidovi , d.h . reich and c. broholm , phys . rev . * b53 * , 2583 ( 1996 ) . see for example , i. affleck , m. gelfand and r. singh , j. phys . * a27 * , 7313 ( 1994 ) ; i. affleck and b. halperin , j. phys . * a29 * , 2627 ( 1996 ) .

in 1970 , efimov described a phenomenon occurring in a three - particle system of identical bosons , in which there is an accumulation of bound states near the point of zero energy and diverging scattering length of the interparticle interaction @xcite . efimov was inspired by the physics of the nucleus of the tritium atom . however , the experimental signature for the existence of such efimov states was not observed in nuclear physics , but in a gas of ultracold cesium atoms @xcite by studying three - body recombination . the reason that properties predicted for nuclear physics few - body systems can be derived from ultracold atoms lies in the fact that the two systems are connected through universal relations , that are irrespective of the specific nature of the interparticle interactions . since the first observations in cesium , the efimov effect has received a lot of attention @xcite . several groups now have observed efimov physics by studying three - body recombination in a variety of atomic systems @xcite . evidence for the existence of trimers appears as some distinct features ( resonances ) in the recombination rate coefficient . in fig . [ fi : efimovschematic ] the energies of the efimov trimers are shown , as a function of the inverse two - body scattering length @xmath0 . a crucial advantage of ultracold atoms with respect to nuclear systems is that the interparticle interactions can be easily manipulated using feshbach resonances @xcite , by varying the magnetic field . the two - body scattering length , which is obtained from the scattering phase shift @xmath1 as @xmath2 , diverges on resonance to infinity , and is typically described by the dispersive formula @xmath3 with @xmath4 the background value of the scattering length , @xmath5 the magnetic field , @xmath6 the resonance position , and @xmath7 the width of the resonance . this possibility to change the scattering length allowed for the observation of distinct efimov features in the three - body recombination rate , connected to two different efimov states in a row @xcite , confirming the scaling law given by @xmath8 with @xmath9 a constant . moreover , recent developments extended the universal few - body physics to the domain of four - body states which was theoretically predicted @xcite and experimentally verified @xcite . in most of the experiments , efimov physics has been observed only indirectly via three - body recombination . due to the short lifetime of these states , it is difficult to probe the efimov states directly . these weakly - bound trimers are not only unstable in collisions with other particles , but they are also intrinsically unstable by itself , as trimers may decay to a deeper lying dimer state and an unbound atom . however , the group of jochim in heidelberg recently succeeded in associating efimov trimers directly via a new and promising approach using radio - frequency fields @xcite . optical lattices might provide more experimental control , and potentially give rise to an enhanced lifetime of efimov trimers . efimov physics can be studied in this way by loading three atoms in each lattice site . when the lattice is deep enough , single sites act as harmonic oscillator potentials where individual trimers can be trapped . a possible advantage is then that the interparticle separation of the trimers can be increased to the scale of the harmonic oscillator length , while their properties are still determined by efimov physics . in this way the overlap of these states with deeply bound dimers is reduced , thereby suppressing the process of three - body recombination . the influence of other limiting processes , where four or more particles involved , will be reduced as well . , indicating dimer and efimov trimer states as a function of the inverse scattering length @xmath0 . the horizontal line @xmath10 indicates the threshold to the three - body scattering continuum ( gray ) . simultaneously , the blue region marks the atom - dimer continuum for @xmath11 , and the green zone indicates the trimer region . the dimer binding energy scales as @xmath12 . only a few efimov states are shown . the figure is not to scale : for clarity reasons we plotted instead of @xmath13 the quantity @xmath14 , and instead of @xmath0 the quantity @xmath15 ( cf . eq . ( [ eq : polarcord])).,width=302 ] jonsell , heiselberg and pethick were the first to introduce an harmonic oscillator term in a three - body system @xcite , by adapting the approach of ref . @xcite . for diverging scattering length , they found exact expressions for the bound states of three identical bosons , while they obtained numerical results for finite scattering length . later , werner and castin @xcite obtained additional analytical results for diverging scattering length , and complimentary first - order corrections for large scattering length . stoll and khler @xcite performed numerical simulations with a tunable separable potential to find the bound states of three particles in a trap . et al . _ found numerical solutions of three particles in a harmonic trap @xcite , mainly to study the non - universal corrections to the efimov effect that come from the effective range . more recently , liu , hu and drummond derived an analytical description of three attractively interacting fermions in a harmonic trap @xcite . a convenient description of the efimov effect in free space is presented in ref . , in which polar coordinates @xmath16 and @xmath17 are introduced , which are related to the wave number or energy parameter @xmath13 and the scattering length @xmath18 via [ eq : polarcord ] @xmath19 using efimov s radial law , it is derived that @xmath20 for a function @xmath21 \to \mathbb{r}$ ] . here @xmath22 is the three - body parameter that defines the trimer energies exactly on resonance as @xmath23 , with @xmath24 the mass of a single atom , and is a result of the short - range behavior of three particles being close together . the combinations of @xmath13 and @xmath17 that yield bound states describe a family of curves that can be mapped on top of each other by multiplying the distance of the curves to the origin by a factor @xmath25 , see fig . [ fi : efimovschematic ] . in this paper , we extend the description of ref . and show how by introducing the harmonic oscillator length as a new parameter , and by using spherical coordinates in parameter space , the scaling behavior in free space very naturally generalizes to a situation with three particles in a harmonic trap . in particular , the curves that are shown in fig . [ fi : efimovschematic ] will be replaced by surfaces that are mapped on top of each other . in the derivation , we use an approach slightly different from earlier work , that calculates the equivalent of the function @xmath26 by solving an ordinary differential equation . additionally , we also include the coupling terms , we investigate the stability of the trimers in the lattice , and we study the non - universal corrections coming from the effective range parameter . we consider a system of three bosons with pair interactions trapped in a harmonic potential . the corresponding hamiltonian is @xmath27 where @xmath28 is the harmonic oscillator length , @xmath29 the trap frequency , and @xmath30 is the interaction between the particles @xmath31 , with @xmath32 a cyclic permutation of @xmath33 . for convenience , the hamiltonian is scaled by a factor @xmath34 . throughout this paper , we will denote the eigenvalues of @xmath16 by @xmath35 , and the corresponding trimer energies by @xmath36 . to find the bound state energies , we closely follow the method described in ref . @xcite and ref . . we denote the permutation operator corresponding to one cyclic permutation of @xmath33 by @xmath37 and transform to the faddeev operator @xmath38 if @xmath39 is an eigenfunction of @xmath40 , @xmath41 is an eigenfunction of @xmath16 , provided it is nonzero . in case @xmath42 , @xmath39 is called a spurious solution . we separate the center of mass motion , and consider the relative system in hyperspherical coordinates . for that , we introduce jacobi - coordinates according to @xmath43 the hyperradius @xmath44 and the hyperangle @xmath45 are defined by @xmath46 the remaining hyperspherical coordinates are given by the unit vectors @xmath47 and @xmath48 . we restrict ourselves to states with zero total angular momentum . the advantage of considering the faddeev rather than the schrdinger operator is that subsystem angular momentum comes in only at second order in the short - range potential @xcite . this motivates us to neglect the subsystem angular momentum , by averaging over the jacobi coordinates @xmath47 and @xmath48 . in this approximation , the schrdinger wavefunction @xmath49 is given in terms of reduced faddeev wavefunctions @xmath50 : @xmath51 in the differential faddeev equation we then make an adiabatic hyperspherical expansion @xmath52 for every value @xmath44 , which is treated as a parameter , the functions @xmath53 of hyperangle @xmath45 form a complete set , and are solutions to the eigenvalue equation on @xmath54 @xmath55\phi_n(r,\alpha ) = - 2 r^2 v(\sqrt{2 } r \sin \alpha ) \left [ \phi_n(r,\alpha ) + \frac{4}{\sqrt{3 } } \int_{|\frac{1}{3}\pi-\alpha|}^{\frac{1}{2}\pi - |\frac{1}{6}\pi-\alpha| } \phi_n(r,\alpha ' ) d \alpha ' \right],\ ] ] with dirichlet boundary conditions @xmath56 . the eigenvalues @xmath57 depend parametrically on @xmath44 . the set of hyperangular functions @xmath53 are generally not orthogonal functions in @xmath45 . therefore , we define @xmath58 which allows us to derive the following coupled set of differential equations for the functions @xmath59 @xmath60 f_n(r ) + \\+ \sum_m \left[2 p_{nm}(r ) \frac{\partial}{\partial r } + q_{nm}(r)\right]f_m(r ) = e f_n(r).\end{gathered}\ ] ] the coupling terms in these equations are defined by [ eq : couplingterms ] @xmath61 this framework is very similar to the work of refs . , and the harmonic oscillator term is taken into account as in ref . @xcite . for now , we neglect the coupling terms conform the adiabatic hyperspherical approximation . however , later on we wil study the coupling between different channels more carefully , as the harmonic oscillator gives rise to degeneracies between the efimov - like solutions in the zeroth channel and the continuum - like solutions of the first and higher channels . + + the effect of the interaction potential @xmath62 is now encoded in the eigenfunctions @xmath57 . parametric solutions for @xmath63 can be found by fixing the parameter @xmath44 , while solving for the angular problem . the functions @xmath57 will then serve as input for the radial equation ( [ eq : coupledfn ] ) , and when the coupling terms are ignored , solutions are obtained by solving an ordinary differential equation ( ode ) . to find an approximation of @xmath63 , one can study solutions @xmath50 of eq . ( [ eq : eqhypwf ] ) for small values of @xmath45 and values of @xmath45 close enough to @xmath64 . matching of the solutions at intermediate values of @xmath45 then yields an approximation of @xmath63 . this procedure has been described in ref . the equation found in this manner is @xmath65 where @xmath66 . note that @xmath67 always solves this equation , but it gives rise to a spurious solution . the right hand side of equation ( [ eq : approxlambda ] ) can be conveniently expanded using the effective range approximation for the s - wave scattering phase shift of the two - body problem @xmath68 with @xmath69 the effective range parameter . when this parameter is neglected , we find the approximating equation for @xmath70 that is at the basis of the derivation in ref . @xcite : @xmath71 by taking into account the effective range as well , one obtains an approximation that is more accurate with respect to the two - body physics near narrow feshbach resonances . this has been proposed in ref . @xcite as a regularization procedure and has been readdressed in ref . we will come back to this issue in section [ sec : effrange ] . + + with the approximation of the channel eigenvalues at hand , we may turn to the hyperradial problem . in the adiabatic hyperspherical approximation , we ignore the coupling between the different hyperradial channels and only take the scattering length into account in the approximation of @xmath63 . in this setting , the efimov effect was derived in ref . @xcite , and repeating it in this context allows for a comparison in the limit of @xmath72 . the uncoupled system of equations is given by @xmath73f_n(r)= 2 e f_n(r),\ ] ] on @xmath74 , with a boundary condition @xmath75 and an appropriate boundary condition on the edge of the inner region @xmath76 where the short - range potentials become important , which is given in terms of the three - body parameter @xmath22 . the exact boundary condition will be specified later . we first concentrate on the zeroth channel , i.e. @xmath77 , which is the only channel that has an attractive potential , and therefore the only channel in which the efimov effect is found . we introduce spherical coordinates @xmath78 according to [ eq : sphercor ] @xmath79 and write down the differential equation for @xmath80 as [ subeq : problemforf ] @xmath81f(\sigma ) = 2\operatorname{sgn}{(\xi)}\sin^2 \vartheta \sin^2 \xi f(\sigma),\ ] ] with long - range boundary condition @xmath82 for small values of @xmath83 , the model does not provide us with a natural boundary condition . solutions of eq . ( [ eq : diffeqforf ] ) have a log - periodic behavior as @xmath84 , given by @xmath85 for some value of @xmath86 . here @xmath87 , and @xmath88 fortunately , the real short - range potentials @xmath30 provide us with a natural boundary condition at @xmath76 encoded by @xmath89 , where @xmath22 is the three - body parameter that accounts for the short - range physics , which was mentioned already in the introduction . this gives rise to a short - range boundary condition at @xmath76 : @xmath90 , @xmath13 and @xmath91 ( blue ) . only three surfaces are shown , but in fact there is an infinite number of them , which all can be mapped on top of each other by multiplying the distance to the origin by powers of @xmath25 . these surfaces are sliced by a plane ( red transparant ) , for a given @xmath92 that corresponds to the height of the plane . the intersection lines ( red ) represent bound state solutions for combinations of the scattering length @xmath18 and the wave vector @xmath13 , for this value of @xmath92 . the picture is not too scale : we used similar scaling relations as used in fig . [ fi : efimovschematic ] in order to show the surfaces more clearly.,width=302 ] we may determine for which parameter combinations a solution exists by a shooting method from large values of @xmath83 . a particular convenient method makes use of a certain phase variable @xmath93 . one can set up an equation for @xmath93 that is much alike the variable phase equation , but is adapted to fit the boundary condition at short distances very well . henceforth , it will play a role similar to the function @xmath26 as we mentioned in the introduction . we introduce the transformation @xmath94 the functions @xmath95 , @xmath96 and @xmath97 may be chosen to our convenience . continuity properties in the parameters are easier to derive considering @xmath93 than when one considers @xmath98 directly . a particular choice @xmath99 is highly compatible with the boundary condition ( [ eq : bcsmallsigma ] ) . the equation for @xmath93 becomes @xmath100 which is a nonlinear first order differential equation . given values for @xmath17 and @xmath101 , we denote by @xmath102 the solution to ( [ eq : forphief ] ) that satisfies @xmath103 . with this definition , the problem ( [ subeq : problemforf ] ) has a solution if and only if @xmath104 consequently , eigensolutions only exist if there exists an @xmath105 such that @xmath106 we would like to emphasize here the similarity with the description of the efimov effect in free space given by equation ( [ eq : descrefimovfreespace ] ) . using the asymptotic behavior of @xmath107 for small arguments , one can easily show that @xmath108 converges to a finite value for @xmath109 . the function @xmath110 can be shown to be continuous over the region . for @xmath111 and @xmath112 $ ] , the efimov effect in free space is retained . hence , @xmath113 is an approximation of @xmath114 . + + [ se : resultssimplemodel ] s@xmath115 . a deeper bound state , with much lower energy , can not be shown in this graph . also indicated is the dimer binding energy ( solid blue line),width=302 ] the trimer solutions are most easily represented by surfaces in three - dimensional @xmath116 space , as defined in eq . ( [ eq : sphercor ] ) . in fig . [ fi : boundstatesurfaces ] such surfaces are shown for the lowest channel , which can be mapped on top of each other by multiplication of the distance to the origin by powers of @xmath25 . fig . [ fi : boundstatesurfaces ] can be regarded as a generalization of the two - dimensional fig . [ fi : efimovschematic ] . when these surfaces are sliced with a plane corresponding to a fixed harmonic oscillator length @xmath92 , we obtain curves that express wave vectors @xmath13 in terms of the scattering length @xmath18 , and an example can be seen in fig . [ fi : channelscombined ] . when taking the limit of @xmath117 , the efimov effect in free space is obtained as depicted in fig . [ fi : efimovschematic ] . in the region @xmath112 $ ] , the intersections of the surfaces will tend to the curves that describe the efimov states . in the region @xmath118 , the energy levels come closer and closer together , expressing the fact that in the limit , this region is filled with continuum states . the intersection of the surfaces with the @xmath119 plane precisely corresponds to the exact results obtained by jonsell _ et al . _ and werner @xcite . as an example , we show in fig . [ fi : channelscombined ] the trimer binding energies of cesium as a function of scattering length with a fixed oscillator length , for the lowest and next to lowest hyperspherical channels . the parameters are @xmath120 @xcite , with @xmath121 the bohr radius , @xmath122 , and @xmath123 , which corresponds to a trap frequency of approximately @xmath124 s@xmath115 . it is interesting to note the change of energy scales in this figure . below the thresholds of the trimer and atom - dimer continuum ( cf . [ fi : efimovschematic ] ) , which would exist for the free - space efimov effect , the bound states show the characteristic scaling law of eq . [ efimovscaling ] . above these thresholds , the bound states show a spacing typical for the harmonic oscillator potential which is approximately @xmath125 . this change of exponential scaling to linear scaling of the trimer energies makes that the accumulation of bound states in the point of zero energy and infinite scattering length disappears . effectively , the efimov trimers are `` pushed '' out of the trimer region through the thresholds and aquire a strong harmonic oscillator state character . are replaced by avoided crossings when the coupling between the zeroth and first hyperspherical channels is taken into account.,width=302 ] there is also a clear difference between the zeroth and first hyperspherical channels . in this respect , it is interesting to note the different limiting behavior of the solutions for small positive and negative scattering length . for small and negative scattering length , there is no weakly - bound dimer state in the interatomic potential , and the solutions have the character of three weakly - interacting atoms in a harmonic oscillator potential . for small and positive scattering length , there is a weakly - bound dimer state , and for the zeroth hyperspherical channel the solutions have the character of a weakly - interacting dimer and an atom in a harmonic oscillator potential . these regimes are connected through solutions which have a strong efimov character . however , for the first hyperspherical channel the situation is different . here the solutions behave more as those of two particles undergoing a feshbach resonance @xcite , which as a result interpolate between two different harmonic oscillator levels . it should be possible to investigate this change of character experimentally , by using similar techniques as have been done by stferle _ et al . _ @xcite . here the creation of dimers was investigated in an optical lattice while sweeping the magnetic field through a feshbach resonance . for the trimers , such adiabatic sweeps through trimer solutions should allow for a creation of states with a character changing between harmonic oscillator- , efimov- and dimer+atom- type solutions . so far , we have applied the adiabatic hyperspherical approximation , in which we completely ignored the coupling between the various hyperspherical channels . while this is generally a good approximation , we see from fig . [ fi : channelscombined ] that there are degeneracies between the zeroth and first channel solutions , which should be lifted when the coupling is taking into account . in order to demonstrate this , we introduce an approximation of the coupling . we calculate the coupling terms @xmath126 and @xmath127 of eq . ( [ eq : couplingterms ] ) using first order approximations of the functions @xmath128 , and find in this way combined solutions which involve both channels . in free - space , the coupling between these two hyperspherical channels is directly related to the process of three - body recombination , where three colliding particles relax into a weakly - bound dimer and a third particle , with a surplus of kinetic energy . interestingly , for our lattice solutions we observe that this coupling transforms the crossings in fig . [ fi : channelscombined ] into avoided crossings , as shown in fig . [ fi : avoidedcrossings ] . this could be experimentally interesting , since using magnetic field sweeps would allow for adiabatic transitions from one hyperspherical channel to an other . of cesium efimov states in a trap , with @xmath129 s@xmath115 and @xmath130 , for the first ( red ) , second ( black ) , third ( green ) and fourth ( blue ) trimer states depicted in fig . [ fi : channelscombined ] . also , the width of a free efimov trimer corresponding to the first state is shown ( black dashed).,width=377 ] the change of character of the trimer solutions , which was discussed above , will have a profound impact on the stability of these states . so far , we implicitly assumed that the trimers in the harmonic oscillator potential are stable . however , when the three particles are close together the trimer can dissociate into a deeply - bound dimer and a third atom , with large kinetic energy . here , we investigate the stability of the trimers by applying the method of braaten and hammer , that regards the trimers as scattering resonances in a dimer - atom scattering process . in this respect , the trimer energy @xmath131 acquires a width @xmath132 , where the latter can be used to calculate the lifetime @xmath133 of the trimers . in this method , an inelasticity parameter @xmath134 is introduced to account for the decay into various deeply bound states . which can be regarded as the imaginary part of a complex phase factor that describes the reflection of hyperradial waves in the inner region . effectively , this comes down to changing the boundary condition at @xmath76 to @xmath135 for given values for the scattering length and the harmonic oscillator length , we are then able to find the associated complex eigenvalues @xmath136 where the trimer energy @xmath137 and the width @xmath138 . as an example , we calculated the linewidths or cesium , with a trapping frequency @xmath139 s@xmath115 , @xmath140 , and @xmath130 @xcite . the short - range boundary condition was implemented at a value @xmath141 . for some of the efimov trimers shown in fig . [ fi : channelscombined ] , the linewidths are shown in fig . [ fi : stabplot ] . the change of character of the trimers can be clearly seen from this graph : for values @xmath142 the states have a harmonic oscillator character , and the widths are relatively small . however , the efimov effect shows up via a strong increase of the widths towards @xmath143 , which is the parameter region where in free - space a new efimov state would enter the tree - body system . it is also the region where a local peak in the three - body recombination rate would occur ( trimer resonance ) . here , we see that at this point the width of the lowest trimer keeps increasing , while for the higher trimer states the width saturates and displays an oscilating behavior . this is not surprising , since the latter trimers are more similar to three - particle continuum states . for comparison , the linewidth of the free - space efimov trimer compared to the lowest lattice trimer state is shown in the region where it exists . these lines overlap mostly , and can be more than two orders of magnitude above the widths for the higher trimer states . at the point where the trimer energies approach the atom - dimer threshold , around @xmath144 , the widths go down again to the level of the higher trimer states . the width of the lattice trimer is there somewhat larger than the free - space trimer , probably due to the additional confinement of the weakly - bound dimer state caused by the lattice . while the efimov effect shows up in the excited states by a clear deviation from the harmonic oscillator levels , these states benefit from a much smaller linewidth than the deeply bound efimov trimers , and they should therefore be easier to study experimentally . also the lowest trimer energy , which most closely follows a free - space efimov state , could be easier to study in experiment . the trimer could first be prepared in the lowest harmonic oscillator level for @xmath145 , and then the scattering lenght can be changed via a magnetic field sweep to the region @xmath146 , where the decay is still manageable . in order to improve the approximation of the channel eigenvalues @xmath63 arising in ( [ eq : eqhypwf ] ) , one might want to include the influence of the effective range @xmath69 . this can be done by making the so - called effective range expansion , i.e. using the expansion in eq . ( [ eq : effrangeexp ] ) up to second order . in literature there have been various instances in which the effective range is taken into consideration in the determination of the three - particle spectrum @xcite . having an additional interaction parameter at hand , raises the question whether in this regime efimov physics in ultracold gases can still be universal . on the experimental side , there are presently also some conflicting observations . in order to understand these issues , we first need to define what we mean by the universal efimov regime . in the above , we have already seen that besides the scattering length , we need at least one additional parameter that fixes at least one of the efimov trimer energies . the other ones can then be derived from this position . this information is provided via the three - body parameter @xmath22 , which can be regarded as a boundary condition for the interaction potentials in the regime where the three particles are very close together . it is related to the energy of the lowest efimov trimer via @xmath147 in the infinite scattering length limit , and with this only additional parameter the efimov physics is universal . of @xmath148 depending on @xmath149 , as calculated in three different ways . the blue lines represent the approximations of the eigenvalues by a truncated fourier expansion . the black lines and red lines are the approximations obtained by ( [ eq : lambdaonlya ] ) and ( [ eq : lambdawitheffrange ] ) respectively . we take for the square well parameters @xmath150 . it can be clearly seen that the effective range gives a much better asymptotic behavior . , title="fig:",width=264 ] + however , some experiments on three - body recombination do not imply universality . while the experiments at rice university did show the universal connection between efimov states in @xmath151li at one side of the resonance @xcite , the universality did not extend to the other side of the resonance . the rice group was not alone in this observation : also in florence @xcite on @xmath152k , in innsbruck @xcite on @xmath153cs , and in tokyo @xcite on @xmath154li this non - universality was observed . on the other hand , in three other experiments universality across the resonance was observed : in bar - ilan it was found that universality does extend across a feshbach resonance in @xmath151li , and even that the short range physics encapsulated in @xmath22 is independent of the nuclear spin configuration @xcite . furthermore universality was also observed in a three - component fermionic @xmath155li gas in heidelberg @xcite and in state college @xcite . now the question arises whether the effective range paramer @xmath69 can explain , in addition to the three - body parameter @xmath22 , the non - universal behavior of some of the above - mentioned experiments . however , it might also be that in a certain regime , the effective range will take over the role of the three - body parameter due to its regularising effect of the radial wavefunctions @xcite . and recently , it has been shown that small experimental deviations from universality @xcite can be explained by effective range corrections using an effective field theory @xcite . as it has been shown by fedorov and jensen @xcite and more recently by platter _ @xcite , the introduction of the effective range in the approximate equation ( [ eq : approxlambda ] ) for the eigenvalues @xmath63 yields the relation @xmath156 to get an idea on the validity of the approximation ( [ eq : lambdawitheffrange ] ) , we have compared the results with the eigenvalues obtained by a truncated fourier expansion for a square well potential with height @xmath157 and width @xmath76 , such that @xmath158 . this expansion is very convenient when one considers the operator @xmath148 , which depends on the hyperradius @xmath44 and is defined via the eigenvalue equation ( [ eq : eqhypwf ] ) as follows @xmath159.\ ] ] the reason for this convenience is that in the basis of fourier expansion functions @xmath160 both the differential operator and the integral operator are diagonal . the matrix elements are then given by @xmath161 where @xmath162 in fig . [ fi : comparisoneffrange ] , we compare the numerically calculated eigenvalues @xmath57 of @xmath148 with results obtained from eq . ( [ eq : lambdaonlya ] ) , which depends on the scattering length only , and with the results from solving eq . ( [ eq : lambdawitheffrange ] ) , which depends both on scattering length and effective range . here the values for @xmath18 and @xmath69 are taken from the effective range expansion of the square well potential . it can be seen that asymptotically , that is for large hyperradius , the inclusion of the effective range in the model yields results that are considerably better than if one only takes the scattering length into account . for relatively small values of the hyperradius , the predictions obtained by the fourier expansion and this simple model start to deviate . hence , we expect that for short hyperradii , additional details of the interaction potential become important , i.e. higher order expansion coefficients beyond the effective range approximation have to be taken into account . for @xmath163 and a fixed value for @xmath18 , the lowest solution @xmath107 to eq . ( [ eq : lambdawitheffrange ] ) can be defined as a smooth function of @xmath44 for @xmath164 . moreover , @xmath165 as @xmath166 . therefore , if we use this approximation in the hyperradial equation , we can find _ regular _ solutions instead of rapidly oscillating ones . this could , at least in principle , remove the need for a renormalization boundary condition in terms of @xmath22 for @xmath166 . in the absence of a harmonic oscillator , and in the limits of small energy and @xmath167 , we will see that the solutions to the hyperradial equation are identical to those when the effective range is not taken into account , except that the value of @xmath22 is determined by @xmath69 by boundary layer behavior for small values of the hyperradius . such a relation between @xmath22 and @xmath69 has been established before @xcite . let us , therefore , study the equation ( [ eq : lambdawitheffrange ] ) more closely . we have plotted the solution @xmath107 for the situation @xmath168 in fig . [ fi : illlambda0 ] , to give a characteristic example . in the region @xmath169 , the value of @xmath107 satisfying ( [ eq : lambdawitheffrange ] ) is close to that obtained by solving ( [ eq : lambdaonlya ] ) . similarly , when @xmath170 , the solution @xmath107 to ( [ eq : lambdawitheffrange ] ) will be close to @xmath171 , where @xmath171 is the solution to @xmath172 around @xmath173 , there is a transition from a boundary layer behavior similar to @xmath174 to behavior obtained by ignoring the effective range ( [ eq : lambdaonlya ] ) . ( purple , solid ) as a function of @xmath175 , for a fixed value of @xmath176 , with @xmath177 . for @xmath178 , the solution is very close to that obtained from ( [ eq : lambdaonlya ] ) ( red , dotted ) . for @xmath179 , there is a boundary layer , in which the solution resembles the solution @xmath174 of ( [ eq : mu ] ) ( blue , dashed).,width=302 ] to find the low - energy behavior of the system , which is needed to find a proper relationship between @xmath69 and @xmath22 @xcite , we have to turn back to the hyperradial equation ( [ eq : hyperradial ] ) . following from the above considerations , the solution @xmath180 that decays exponentially for large hyperradius is to leading order in @xmath181 , and for @xmath182 , equal to @xmath183 , the exponentially decaying solution to the hyperradial equation obtained when the effective range is ignored . this outer solution @xmath183 will have log - periodic behavior as @xmath166 @xmath184 for some @xmath86 . the solution @xmath180 will be similar for @xmath185 , and will therefore for low enough energies also show log - periodic behavior . however , there is a boundary layer for @xmath170 , which makes that we can demand the solution @xmath186 to be regular for @xmath166 , that is , such that @xmath187 for @xmath166 . this effectively determines the value of @xmath86 . to analyze the boundary layer , we introduce the new variable @xmath188 and the function @xmath189 , @xmath190 . the hyperradial equation for @xmath189 reads @xmath191 \tilde{f}_0(\rho ) = 0,\ ] ] and we demand that @xmath192 behaves like @xmath193 for @xmath194 . we consider now the case of a negligible harmonic oscillator potential , i.e. @xmath195 , and take the low - energy limit @xmath196 and a diverging scattering length . the solution will then be equal to @xmath197 , which satisfies @xmath198 f_0^{\text{in}}(\rho ) = 0.\ ] ] for @xmath199 , @xmath200 we determine the phase @xmath201 numerically , by choosing a boundary condition @xmath202 for small enough values of @xmath203 , and by integrating the equation . the boundary condition ensures that we pick up the right solution . the result is @xmath204 . in order to find what value of @xmath22 ( @xmath205 in this limit ) is induced by @xmath69 , we have to match the inner and the outer solutions in the regime where @xmath185 . this yields that @xmath206 . remarkably , this numerical value for @xmath207 differs from the values obtained in ref . @xcite and @xcite , which are @xmath208 and @xmath209 respectively . it is unclear where the difference comes from , however , both papers use models different from the hyperspherical formalism that we used here , where we base our treatment of the effective range parameter on the descriptions in refs . @xcite and @xcite . more research is needed to resolve this discrepancy . now we are in a good position to compare our model with earlier work on the efimov effect in a harmonic potential . in the work of thgersen , fedorov , and jensen , the influence of the effective range is studied on the universal efimov effect @xcite . they considered a system with a gaussian interaction potential , given by @xmath210 where @xmath157 and @xmath76 are constants . the range @xmath76 has been chosen to equal @xmath211 . although the main purpose of ref . @xcite is to show the influence of the effective range , their model contains a harmonic trap as well . in order to make a proper comparison with our model , we take @xmath212 the results are shown in figs . [ fi : topth ] and fig . [ fi : botth ] , and can be directly compared with fig . 1 in ref . @xcite . we observe that the numerical results are reproduced quite closely and we can assign the numerically found bound states to the different hyperspherical channels . we also observed that if we include the above corrections that come from the effective range , the comparison is better than results obtained without these corrections . we studied the efimov effect in a harmonic potential by using the hyperspherical formulation . in the first part of this paper , we studied the universal regime by only taking a large scattering length into account , and by making the adiabatic hyperspherical approximation in that we ignored the coupling between different hyperspherical channels . the model thus obtained is simple in the sense that the binding energies of the trimers result from an ordinary differential equation , with an appropriate boundary condition for the lowest channel encoded by the three - body parameter @xmath22 . considering the lowest channel , we formulate an equation for a phase variable , the exponent of which is the radius in a spherical plot describing a surface of values in @xmath213-space ( inverse scattering length , wavenumber and inverse oscillator length ) for which bound states exists . other such surfaces are obtained by multiplying the distance to the origin by a factor @xmath25 , and there is an accumulation point of surfaces near the origin , analogous to the efimov effect in free space . in the limit @xmath214 , the efimov effect in free space is recovered . interesting differences with the situation of three particles in free space can be observed when we focus at a fixed value of @xmath215 . for parameter combinations where in free - space bound states would exist , the situation is very similar , except that the accumulation of energies only happens up to a typical energy of @xmath216 . the rest of the efimov states are `` pushed out '' into the trimer continuum ( which is of course only a continuum in free space ) . higher hyperspherical channels give rise to additional bound states , and by taking those into account as well we obtain results very close to numerical simulations of three - particle systems @xcite . better qualitative results are obtained by approximating the coupling between the hyperspherical channels , which lifts the degeneracies and creates avoided crossings . the lifetime of the trimers is highest for states with a continuum character , i.e. in the region which we would denote as trimer continuum in free space . the reason for this is the small overlap of these states with deeply bound states . experimental study of efimov - like states , however , could be feasible close to the transition from this continuum region to the trimer region , where the linewidth is still quite small . for a more general efimov model that allows for deviations from the universal regime , a natural step is to improve on the approximation of the hyperangular problem by taking into account the effective range . by doing this , we were able to find a relation between @xmath22 and the effective range @xmath177 if @xmath217 , where @xmath76 is a typical range of the potential . we obtain the same qualitative behavior as ref . @xcite and @xcite where different descriptions were used , however , the actual value of @xmath218 is also very different . it appears that the actual value for this relationship depends strongly on the choice of the model and the corresponding approximations to describe the three - particle physics , and more research is needed to understand this discrepancy . in a different parameter regime for @xmath18 , @xmath69 and @xmath76 , such a relationship between @xmath22 and @xmath69 is not necessary . then we have in addition to @xmath69 , the three - body parameter @xmath22 in the boundary condition for the hyperradial problem to our disposal . by comparing to numerical calculations we find that this description indeed improves the predicted trimer energies . 99 v. efimov , phys . b * 33 * , 563 ( 1970 ) ; v. efimov , sov . . phys . * 12 * , 589 ( 1971 ) ; v. efimov , _ ibid . _ * 29 * , 546 ( 1979 ) . t. kraemer , m. mark , p. waldburger , j.g . danzl , c. chin , b. engeser , a.d . lange , k. pilch , a. jaakkola , h .- c . ngerl , and r. grimm , nature * 440 * , 315 ( 2006 ) . f. ferlaino and r. grimm , physics * 3 * , 9 ( 2010 ) . c.h . greene , physics today * 63 * , 40 ( 2010 ) . s. knoop , f. ferlaino , m. berninger , m. mark , h .- c . ngerl , and r. grimm , nature phys . * 5 * , 227 ( 2009 ) . m. zaccanti , b. deissler , c. derrico , m. fattori , m. jona - lasinio , s. mller , g. roati , m. inguscio , and g. modugno , nature phys . * 5 * , 586 ( 2009 ) . n. gross , z. shotan , s.j.j.m.f . kokkelmans , and l. khaykovich , phys . 103 * , 163202 ( 2009 ) . pollack , d. dries , and r.g . hulet , science * 326 * , 1683 ( 2009 ) . wenz , t. lompe , t.b . ottenstein , f. serwane , g. zrn , and s. jochim , phys . a * 80 * , 040702(r ) ( 2009 ) . t. lompe , t. b. ottenstein , f. serwane , k. viering , a. n. wenz , g. zrn , and s. jochim , phys . lett . * 105 * , 103201 ( 2010 ) . huckans , j.r . williams , e.l . hazlett , r.w . stites , and k.m . ohara , phys . 102 * , 165302 ( 2009 ) . williams , e.l . hazlett , j.h . huckans , r.w . stites , y. zhang , and k.m . ohara , _ ibid . _ * 103 * , 130404 ( 2009 ) . s. nakajima , m. horikoshi , t. mukaiyama , p. naidon , and m. ueda , phys . * 105 * , 023201 ( 2010 ) . g. barontini , c. weber , f. rabatti , j. catani , g. thalhammer , m. inguscio , and f. minardi , phys . lett . * 103 * , 043201 ( 2009 ) . h. feshbach , ann . of phys . ( n.y . ) * 5 * , 357 ( 1958 ) ; * 19 * , 287 ( 1962 ) . e. tiesinga , a. j. moerdijk , b. j. verhaar , and h. t. c. stoof , phys . a * 46 * , r1167 ( 1992 ) ; e. tiesinga , b. j. verhaar , and h. t. c. stoof , phys . rev . a * 47 * , 4114 ( 1993 ) . e. braaten and h .- w . hammer , phys . rep . * 428 * , 259 ( 2006 ) . j. von stecher , j.p . dincao , and c.h . greene , nature phys . * 5 * 417 ( 2009 ) . f. ferlaino , s. knoop , m. berninger , w. harm , j.p . dincao , h .- c . ngerl , and r. grimm , phys . lett . * 102 * 140401 ( 2009 ) . t. lompe , t.b . ottenstein , f. serwane , a.n . wenz , g. zrn , and s. jochim , science * 330 * , 940 ( 2010 ) . s. jonsell , h. heiselberg , and c.j . pethick , phys . lett . * 89 * , 250401 ( 2002 ) . fedorov and a.s . jensen , phys . lett . * 71 * , 4103 ( 1993 ) . f. werner and y. castin , phys . lett . * 97 * , 150401 ( 2006 ) . m. stoll and t. khler , phys . a * 72 * , 022714 ( 2005 ) . m. thgersen , d.v . fedorov , and a.s . jensen , phys . a * 78 * 020501 ( 2008 ) . x. j. liu , h. hu , and p. d. drummond , phys . a * 82 * , 023619 ( 2010 ) . e. nielsen and d.v . fedorov , phys . rep . * 347 * 373 ( 2001 ) . fedorov and a.s . jensen , j. phys . a , * 34 * 6003 ( 2001 ) . l. platter , c. ji , and d.r . phillips , phys . a * 79 * 022702 ( 2009 ) . f. werner , ph.d thesis _ trapped cold atoms with resonant interactions : unitary gas and three - body problem _ , ens , 2008 . t. busch , b .- englert , k. rzazewski , and m. wilkens , found . * 28 * , 549 ( 1998 ) . t. stferle , h. moritz , k. gnter , m. khl , and t. esslinger , phys . lett . * 96 * , 030401 ( 2006 ) . petrov , phys . * 93 * , 143201 ( 2004 ) . gogolin , c. mora , and r. egger , phys . 100 * 140404 ( 2008 ) . n. gross , z. shotan , s. kokkelmans , and l. khaykovich , phys . lett . * 105 * , 103203 ( 2010 ) . n. gross , z. shotan , o. machtey , s. kokkelmans , and l. khaykovich , preprint arxiv:1009.0926 , accepted for comptes rendus physique . c. ji , l. platter , and d. r. phillips , epl * 92 * 13003 ( 2010 ) . b. marcelis , b. verhaar , and s. kokkelmans , phys . lett * 100 * , 153201 ( 2008 ) .

the orion nebula ( m42 ) is the most active site of star formation of the orion molecular cloud and contains many young stars , among them the trapezium cluster . its high mass young stellar members generate an intense ultraviolet radiation field that photodissociates and photoionizes the nearby material . these mechanisms promote the appearance of some complex structures as the so - called proplyds @xcite . proplyds are low mass ysos that are being exposed to an intense ultraviolet radiation field which renders them visible . in orion , there are approximately 160 proplyds @xcite which are being photoionized mainly by @xmath6 ori c , an o6 spectral type star . there are several objects that have been identified as being proplyds not only near the trapezium cluster @xcite but also in other star forming regions ( like ngc 3372 , * ? ? ? most of them share the same features : a bow - shaped head that faces the ionization source , a tail , that is primordially directed away from the source , a young star , that may ( or not ) be visible , and a disk , sometimes seen in silhouette against the hii region ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? early studies of these objects , however , were only able to determine their apparent ubiquity close to the @xmath6 ori c star , their high ionization level and , in later studies , the presence of high velocity structures ( outflows ) associated with these condensations ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the proplyds are presently explained by a set of models which include a photoevaporated wind , an ionization front and a photoionized wind @xcite . @xcite proposed models in which far ultraviolet radiation ( fuv ) and extreme ultraviolet radiation ( euv ) are responsible for the mass loss rates of the proplyds depending on the distance of the proplyd to the ionization source . for the proplyds situated at intermediate distances from the ionizing star , fuv photons penetrate the ionization front and photodissociate and photoevaporate material from the accretion disk surface , generating a supersonic neutral wind that passes through a shock front before reaching the ionization front as a neutral , subsonic wind . at the ionization front , euv photons ionize the wind , and the material is then reacelerated to supersonic velocities . the interaction of this supersonic ionized wind with the star wind generates bow shocks seen in h@xmath2 and in [ o iii]@xmath75007 in some proplyds @xcite . for proplyds closer to the ionizing star , the ionization front reaches the disk surface , and the resulting wind is initially subsonic , becoming supersonic further out . in general , the models cited above are able to explain the main observational features of the orion proplyds . the region close to the ionization front is responsible for most of the emission of these objects . @xcite have shown that heating and dissociation of h@xmath8 can explain the observed [ o i]@xmath76300 emission at the disk surface @xcite . @xcite proposed that neutral gas compressed between the shock front and the ionization front can explain the [ o iii]@xmath75007 emission seen near the ionization front in some objects . @xcite were able to reproduce the observed h@xmath2 intensity profile of the proplyds in orion in terms of accelerating photoevaporated flows . numerical 2d hd simulations with a treatment of the radiative transfer @xcite also reproduce the proplyd morphology and emission , although in this work the treatment of the diffuse radiation is somewhat simplified . there are several open issues about these objects . one of them is related with the calculation of the mass loss rate of the proplyds , that is strongly model dependent and which may pose severe constraints on the age of the orion proplyds ( as well as for @xmath6 ori c ) . as the flux of fuv photons dissociates the molecules in the disk of the proplyds , the sub- or trans - sonic _ wind _ is accelerated at the proplyd s ionization front ( if ) ( see henney et al . 2002 for a discussion ) , leading to derived mass - loss rates of @xmath9 yr@xmath0 which imply a short lifetime for these systems @xcite . it is hard to reconcile such short lifetimes with the fact that the region of the trapezium cluster is populated by several proplyds . improvements in both models and observational techniques have slightly reduced the calculated mass - loss rates for these systems , and @xcite find @xmath10 yr@xmath0 for the 167 - 317 ( lv2 ) proplyd , implying an age of less than @xmath11 yr for @xmath6 ori c , and @xmath12 yr@xmath0 for the 170 - 377 , 177 - 341 , 182 - 413 , 244 - 440 proplyds ( e.g. , * ? ? ? * ) . it is presently known that many proplyds show , besides protostellar features such as accretion disks and a young low mass protostar , the presence of jets . @xcite , in a survey carried out using the wfpc2 camera on the hst , find 23 objects which appear to have collimated outflows seen as one - sided jets or bipolar chains of bow shocks . @xcite also found kinematical traces of jets associated with lv5 ( 158 - 323 ) and gmr 15 ( 161 - 307 ) . the proplyd 167 - 317 shows evidence of the existence of a collimated outflow , detected spectroscopically @xcite . until now , this outflow was detected as a one - sided jet , with a p.a . of @xmath13 and a propagation velocity of about 100 km s@xmath0 . recently , @xcite and @xcite , using the wide field camera of the advanced camera for surveys ( acs / wfc ) on hst , found jets associated with silhouette disks in the outer regions of orion . in this work , we present the first integral field unity ( ifu ) observation of a proplyd the observed object is 167 - 317 , one of the brightest proplyds of the trapezium . we discuss one of the very first results from the gemini multi - object spectrograph , in its ifu mode ( hereafter , gmos - ifu ) , at the gemini south telescope . in 2 , we present the observations and the steps to obtain a calibrated , clean datacube . in 3 we present the spectral analysis for the observed region of the 167 - 317 proplyd , as well as the observed lines and intensity maps . we use the observed line profiles to identify the outflows in the system . in 4 , we present the discussion and the conclusions . the data were taken during the system verification run of the gmos - ifu at the gemini south telescope ( gst ) , under the gs-2003b - sv-212 program , on 2004 february 26@xmath14 and 27@xmath14 . the science field of view ( fov ) is @xmath15 with an array of 1000 lenslets of @xmath16 . the sky is sampled with 500 lenses which are located @xmath17 from the science fov . we used the r831_g5322 grating in single slit mode , giving a sampling of 0.34 per pixel ( @xmath18 km s@xmath0 per pixel at h@xmath2 ) and a spectral coverage between @xmath19 to @xmath20 . the spectral resolution ( i.e. , the instrumental profile ) is 47 km s@xmath0 @xmath21 fwhm @xmath22 km s@xmath0 ( the fwhm decreasing with increasing wavelength ) . a field centered on the 167 - 317 proplyd has been observed with an exposure time of 300s . a 60 s exposure of the standard star the hiltner 600 has also been taken in order to derive the sensitivity function . the data were reduced using the standard gemini iraf v1.6 routines . an average bias image has been prepared using the gbias task . the extraction has been worked out using the gcal flat and the response using the twilight flat . spectra extraction has been performed and wavelength calibration done using an arc taken during the run . the 167 - 317 field and hiltner 600 star have been processed in the same way . the sensitivity function derived from the standard star has been applied to the observed field . finally , data cubes have been built using the gfcube routine . we maintain the ifu original resolution of @xmath23 px@xmath0 although an interpolation was performed in order to turn the ifu hexagonal lens shape into squared pixels . because some of the emission lines are very intense , it was not possible to eliminate cosmic rays using the standard gscrrej routine of the iraf gemini v1.6 package . in order to remove cosmic rays we have then programmed an idl routine . this routine works directly on the data cube by taking out impacts that are above a certain convenient level calculated based on the mean intensity of nearby pixels . the cosmic ray is then eliminated doing a linear interpolation in the wavelength direction . with this procedure , it is possible to keep very intense lines and eliminate lower level cosmic ray impacts that could be misinterpreted as low emission lines . further manipulation of the cube , such as the production of channel maps , emission line maps , dispersion and velocity maps , have been done using starlink , idl routines and fortran programs developed specially for this purpose . the subtraction of the background emission from the proplyd spectrum is a very challenging task @xcite . this problem arises since the background nebular emission is strongly inhomogeneous on an arc - second scale . in this work , we have used standard @xmath24-method in order to fit an intensity versus position plane in a semi - rectangular field located @xmath25 away from the peak proplyd emission in each of the individual velocity channel maps . we assure that we only take samples of the background avoiding the region of the proplyd . these planar fits to the nebular emission are then subtracted from the corresponding velocity channel maps . in figure [ f1 ] , we show the observed field , the object orientation in the plane of the sky , and also the region that defines our background ( region 4 , labeled as r4 in figure [ f1 ] ) . region 4 ( see figure [ f1 ] ) contains 76 spectra which are used to define the background . this figure also shows an internal box ( labeled as r1 , or region 1 ) , for which we have defined an arbitrary @xmath26-coordinate system limited by @xmath27 ( in the east direction ) and @xmath28 ( in the north direction ; see figure [ f1 ] ) . in the following sections , we show maps of spatially distributed physical variables that will be constrained to the limits defined by this box . the spatial positions inside this domain will be defined using the @xmath26-coordinate system described above . regions 2 and 3 ( labelled as r2 and r3 in figure [ f1 ] ) define , respectively , a region near the center of the 167 - 317 proplyd and a region which has a high - velocity , redshifted feature ( see below ) . the lines that we identify in the spectra have already been reported in previous papers of the orion nebula ( e.g , . * ? ? ? * ) . we select 4 different regions to measure the line intensities , namely , regions r1 , r2 , r3 and r4 ( see fig . [ f1 ] and 2 for the coordinate system definition ) . four spectra are then obtained by co - adding the spectra of the pixels included in each region . the spectrum of region 1 is the result of the sum of 187 pixels , for region 2 the sum of 6 pixels , region 3 represents the spectrum of a single pixel and region 4 the sum of 76 pixels . a set of 38 lines has been selected from a manual search for emission features in the spectrum integrated over the selected fields . the intensities of the emission lines are determined by fitting a flat continuum and integrating over the whole emission feature . then , a mean flux was obtained dividing the integrated flux by the total number of pixels of each region . table [ lines ] gives the list of the observed lines together with the mean flux of each region ( after subtraction of the emission from the background , region r4 ) and the ratio to h@xmath2 ( normalized to h@xmath29 ) . the fourth and eigth columns show the background emission and ratios to h@xmath2 , respectively , for each line . the mean fluxes are given in units of @xmath30 erg @xmath31 s@xmath0 px@xmath0 , and are not corrected for reddening . the absolute error of the mean flux is given in parentheses . this error was calculated taking into account the variations of the local continuum . the errors can be very large for the weaker lines . also , the detection of spectral lines in region r3 is affected by the low signal to noise ratio . there are some lines which show stronger mean fluxes in the background , namely , the lines of si iii @xmath32 , si ii @xmath33 , [ ni ii]@xmath34 and [ fe ii]@xmath35 ( see table [ lines ] ) . these lines have low ionization potentials and are expected to appear mainly in regions of lower ionization degree . the lower line intensities within the proplyd indicate that these lines are mainly emitted by the background nebula , and that they are absorbed at least partially by the dust in the proplyd . in figure [ f2 ] , we show the full extracted spectra from the data cube , for regions r1 to r4 as defined in figure [ f1 ] . in this figure , the spectra for regions r1 , r2 , r3 and r4 are depicted from bottom to top in each diagram . the data shown here are not background subtracted . most of the detected 38 lines can be clearly seen in the spectra . it is also evident that the s / n is higher in regions r1 and r4 ( which is due to the increase of the pixel number of these regions ) . here , a third order cubic spline was fitted to the continuum and then subtracted from the data . because of the high order spline polynome subtraction , some minor variations are still present near the h@xmath2 line . figure [ f3 ] depicts ( for region r1 ) , 2d - intensity maps of the lines [ n ii]@xmath75755 , [ n ii]@xmath76548 , h@xmath2 , [ n ii]@xmath76583 , [ he i]@xmath77065 and [ ar iii]@xmath77135 , superimposed by line profiles . the data shown here are background subtracted , as described in 2 . the line profiles can be better seen in figure [ f4 ] , in which we show the mean line profile for each line and for each of the regions r1 , r2 , r3 and r4 , in the same diagram . as could be anticipated , the flux is more intense in region r2 ( dotted line ) , where the proplyd is located , except for the [ n ii]@xmath36 line , which shows a stronger flux in region r3 . for the h@xmath2 and [ n ii]@xmath37 lines the emission of region r3 is more intense than the emission of region r1 . for lines with good s / n , the presence of a redshifted feature starts to become clear . an analisys of the profile components will be given in more detail below . in order to see how the spectra of the different lines change spatially , in figure [ f5 ] we present an intensity _ versus _ position plot for the same lines shown in figures [ f3 ] and [ f4 ] . we show how the intensity of these lines changes along a diagonal line , that crosses region r1 passing through the pixel ( 3,4 ) ( region r3 ) and through the centre of region r2 . we note that there are 2 intensity maxima for the h@xmath2 , [ n ii]@xmath38 and [ n ii]@xmath39 lines , clearly showing in which lines region r3 is clearly visible ( also see figure [ f3 ] ) . we associate region r3 with the redshifted jet of the proplyd , although in figure [ f9 ] we will show that the emission of the jet is not restricted only to this pixel . a redshifted jet associated with the 167 - 317 proplyd was previously detected , with a propagation velocity of @xmath40 100 km s@xmath0 ( e.g. , * ? ? ? * ) . we have also detected this high velocity feature in our data . in order to see this , we examine the behaviour of the moments of the radial velocity distribution . these are the flux - weighted mean radial velocity @xmath41 , the flux - weighted rms width of the line @xmath42 and the skewness @xmath43 , that are given , respectively , by : @xmath44 @xmath45 @xmath46 where @xmath47 the integrated intensity ( see eq . [ eq1 ] and figure [ f3 ] ) shows that if we move away from the proplyd position ( r2 ) , the emission for several lines ( for example , h@xmath2 and [ ar iii ] ) rapidly drops to low values ( close to the background value ) . we also note that the maps for the flux - weighted mean radial velocity km s@xmath0 ( e.g. , * ? ? ? the error in the radial velocity measurements is of the order of 2.5 km s@xmath0 . ] ( not shown here ; see eq . [ eq2 ] ) show that @xmath48 km s@xmath0 . the flux - weighted rms width of the line @xmath49 and the skewness @xmath50 moment of the radial velocity distribution ( see equations [ eq3 ] and [ eq4 ] , respectively ) are depicted in figure [ f6 ] , for the h@xmath2 ( left ) and [ ar iii ] ( right ) lines . both of them were computed after carrying out both background and continuum subtractions . from figure [ f6 ] , we see that there is a clear enhancement of both of these moments from the proplyd position ( r2 region ) towards the r3 region ( i.e. , towards the se direction , at a pa @xmath51 ) . this behaviour of the moments leads us to infer that : 1 ) there is an increase in the line width as we go from the proplyd position to the se direction ; and 2 ) that there are redshifted wings in the line profiles . these results indicate the presence of a redshifted outflow in this region . we note that the same behaviour is seen in the he i @xmath76678.15 , [ n ii]@xmath76583.46 lines ( not shown here ) . in order to confirm the presence of high velocity components in the observed field , as well as to see the behaviour of the photoevaporated proplyd flow and the background emission , we have computed three - component gaussian minimum @xmath24 fits for each position - dependent line profile . the profiles of all emission lines have a major peak at @xmath52 km s@xmath0 , the exact radial velocity of this peak changing with spatial position . in some spectral lines , there is an evident second peak at redshifted velocities ranging from 100 to 150 km s@xmath0 , corresponding to the jet associated with the proplyd . finally , we have been able to detect a blueshifted component , which is fainter than the redshifted components . as an example of our three - component gaussian fit , in figure [ f7 ] we show the data ( full line ) and the fit ( crosses ) , for the h@xmath2 and [ ar iii]@xmath77135 lines in the ( @xmath53,@xmath54)=(6,7 ) position . in this figure , for each emission line , the top - left panel represents the data and the three gaussian fits , the top - right panel shows the main , low velocity component ( and the gaussian fit for this component ) , the bottom - left panel depicts the data minus the main component together with the fits for the blue- and red - shifted components and , finally , the bottom - right panel shows the residual , obtained by subtracting the three - component fit from the observed line profile . the fits depicted in this figure show the presence of redshifted and blueshifted components . the redshifted component is present in several lines in pixels around the se direction , as already mentioned before . on the other hand , blueshifted emission can be found in several pixels to the nw of the proplyd peak emission as can be seen in the left bottom panel of figure [ f7 ] and this could represent the first spectroscopic determination of the presence of a blueshifted counter - jet , with systemic velocities of -60 km s@xmath0 @xmath55 km s@xmath0 . this blueshifted component is also suggested by the fitted profiles of hei @xmath77065 ( not shown here ) . figure [ f8 ] depicts the spatially distributed intensity over the r1 region ( see figure [ f1 ] ) , of the main peak for the [ n ii]@xmath75754 ( top - left ) , [ n ii]@xmath76548 ( top - middle ) , h@xmath2 ( top - right ) , [ n ii]@xmath76583 ( bottom - left ) , hei @xmath77035 ( bottom - middle ) and [ ar iii]@xmath77135 ( bottom - right ) lines . the values shown in this figure were obtained from three - component gaussian fits to the observed line profile . all of the lines peak at the proplyd position , at ( @xmath53,@xmath54 ) = ( 6,7 ) , inside the r2 region , and three of them ( namely [ n ii]@xmath76548 , top - middle ; h@xmath2 , top - right ; and [ n ii]@xmath76583 , bottom - left ) , have a secondary , less intense peak at ( @xmath53,@xmath54)=(3,4 ) ( i.e. , in the r3 region ; see figure [ f1 ] for the definition of the coordinate system ) . this secondary peak is probably related to the jet , since , as we have seen before , the jet propagates from the r2 region towards the se direction . it is also interesting to note that there is a tail of faint emission ( compared with the maximum in a given map ) that extends in the ne direction , and that can be seen in both of the [ n ii ] lines that bracket h@xmath2 ( top - middle and bottom - left maps in figure [ f8 ] ) . in figure [ f9 ] , we show the spatially dependent intensity ( fig . [ f9]a ) and the central velocity ( fig . [ f9]b ) of the redshifted component of the fitted h@xmath2 profile . we limit the maps to the spatial pixels which have a skewness @xmath56 km s@xmath0 . the pixels which satisfy this criterium have a well defined high velocity , redshifted component . these figures show that the high velocity component has an intensity maximum near the center of the proplyd , with an extension towards the se direction ( along the jet axis ) , surrounded by a region of decreasing fluxes . the high intensity spike seen in h@xmath2 , [ n ii]@xmath57 and [ n ii]@xmath58 lines ( figure [ f3 ] ) can be seen here . the jet shows a spread in velocity values , although most of the pixels present velocities around 120 - 140 km s@xmath0 . this figure also shows that the jet velocity decreases with increasing distance from the proplyd . this trend can be seen in the figure [ f9]c , where we show the radial velocity of the redshifted component as a function of distance from the proplyd position , ( @xmath53,@xmath54 ) = ( 6,7 ) in region r1 . to build this figure , we assume that the jet is propagating in a pa @xmath51 position angle . we define a box with one axis aligned with the jet axis , and the second , perpendicular dimension extending two pixels to each side of the jet axis . we then take averages ( of the mean velocity of the redshifted component ) perpendicular to the jet direction and plot the resulting velocity as a function of distance from the proplyd ( see the bottom frame of figure [ f8 ] ) . there is an indication that the jet velocity is slowly diminishing as a function of distance from the source . in figure [ f10]a , we show a map for the r1 region of the line ratio : @xmath59\lambda 6548 + [ { \rm n~ii}]\lambda 6583}{[{\rm n~ii}]\lambda 5754 } = \frac{a_{6548 } \sigma_{6548 } + a_{6583 } \sigma_{6583}}{a_{5754 } \sigma_{5754}},\ ] ] where @xmath60 and @xmath61 are the height and the dispersion of the fitted gaussian profile ( of the main , low velocity component ) . the [ nii](6548 + 6583)/5754 ratio is a classical electron temperature diagnostic of low / medium density nebulae ( @xmath62 @xmath63 ) . at high densities , due to the collisional desexcitations , this ratio can not be used for the determination of the electron temperature , but it turns out to be a quite good electron density diagnostic . this can be seen , for example , from the @xcite equation : @xmath64(6548 + 6583)/5754 = \frac{6.91 \exp ( 2.5 \times 10^{4 } / t_{e})}{1 + 2.5 \times 10^{-3 } ( n_{e } / t_{e}^{1/2})}\ ] ] figure [ f10]b shows the lines corresponding to [ nii](6548 + 6583)/5754 = 9 , 10 , 11 , 80 , 90 , 100 in a @xmath65 plane . we can see that for the values obtained in our case ( line ratios close to 10 ) , the [ nii](6548 + 6583)/5754 ratio is strongly dependent on the electron density , and that the dependence on the electron temperature is small . in figures [ f10]c and [ f10]d , we show the densities computed from the [ nii](6548 + 6583)/5754 ratio for @xmath66 k and @xmath67 k , respectively . although presenting a very complicated pattern , these figures show that the region coinciding with the proplyd center , and extending towards the se direction ( the jet propagation direction ) , shows the lowest line ratio , indicating electron densities ( @xmath68 ) of the order of @xmath69 @xmath63 . this value is at least in qualitatively agreement with the @xmath70 @xmath63 obtained by @xcite considering a @xmath71 k temperature for the ionization front ( that is inside the r2 region ) . we can also note that both figures are similar , presenting few differences in the density values . the extended structure from the r2 region towards the se direction is surrounded by a region in which the nitrogen line ratio reaches values of up to @xmath72 . in this limit , @xmath73 @xmath63 . it is interesting to note that in the clump associated with the redshifted jet , the nitrogen line ratio increases substantially , indicating a decrease in the electron density when compared with the same values near the center of the proplyd . we have also obtained the [ s ii]@xmath74 emission maps . unfortunately , we are not able to use these lines as a diagnostic because neither the ratio of this doublet is a good electron density indicator for this case ( since it is constant for densities higher than 10@xmath75 @xmath63 ) nor do we have a good enough signal to noise ratio for these lines . furthermore , the background subtracted spectra for these two lines give us negative fluxes in the proplyd region ( see table [ lines ] ) . the presence of these negative features in the line profiles is probably related with the presence of dust , as discussed by @xcite . we have presented in this work the first integral field unit ( ifu ) spectroscopic observations of a photoevaporating disk immersed in a hii region . in particular , we have taken advantage of the system verification run of the gemini south telescope multi object spectrograph ( gmos ) to obtain spectra of the 167 - 317 proplyd in the orion nebula . the 167 - 317 proplyd , also known as lv2 ( from the pioneering work from * ? ? ? * ) , is one of the brightest and best studied proplyds . in a single exposure , we took 400 spectra with a spatial resolution of @xmath16 . these spectra have been combined in order to optimize this instrumental feature , and , before discussing of our results , we want to make a few comments concerning the potential use of such ifu data - cubes for these objects . as previously discussed in the literature @xcite , the subtraction of the background emission is a very challenging task . this problem arises since the background near these proplyds ( which are immersed in a highly non - homogeneous , photoionized ambient medium ) is very complex and changes in an arc - second spatial scale . a careful computation of the background contribution to the spectra of these objects is one of the most important tasks in order to have reliable radial velocity measurements for both the proplyd photoevaporated flow and the high velocity features ( jets ) present in these systems . here , we have used the 76 spectra , defined in figure [ f1 ] as region r4 , in order to obtain a planar fit for the background emission . we think that such a definition of the background emission is more precise , since we use several regions to calculate a better function to describe the background . one more thing to point out is that , with spatially distributed spectra , we are able to confidently separate the contribution of the background from the contribution of the object itself . the data from the ifu observation of the 167 - 317 proplyd also allow us to investigate the spatially dependent properties of the outflows associated with this object . in particular , the 167 - 317 proplyd is known to have a redshifted , collimated jet that propagates with heliocentric velocities of @xmath76 km s@xmath0 towards the se direction , and with a spatial extension of @xmath77 ( e.g , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . we find that a prominent , high velocity redshifted component can be detected in some emission lines , particularly in h@xmath2 , in this se direction . the redshifted jet has a trend in its radial velocity ( as a function of distance from the proplyd ) with higher velocities close to the proplyd and lower velocities at increasing distances . we note that @xcite had previously suggested a variation in the jet velocity as a function of distance from the source . there is a subtle peak ( in several emission lines , particularly in [ n ii]@xmath75754 and h@xmath2 ) that is associated with the se jet emission , and located at @xmath78 se of the center of the proplyd [ at ( x , y)=(3,4 ) ; the r3 region in figure [ f1 ] ] . for this intensity peak , and also using the h@xmath2 profiles of the neighbouring pixels , we find a mean heliocentric velocity @xmath79 km s@xmath0 . it is interesting to note that previous hst images and spectroscopic analyses @xcite also reveal a spike in the jet emission at @xmath80 from the proplyd cusp . if we associate these two emission regions as belonging to the same condensation , we can estimate a proper motion of @xmath81 km s@xmath0 , which combined with the @xmath82 km s@xmath0 radial velocity give a full jet velocity of @xmath83 km s@xmath0 ( where the error was inferred from the spatial resolution of our data ) . this velocity is of the same order as the values inferred for herbig - haro jets associated with t tauri stars . however , the exact nature of this spike is unknown . we could relate this spike with a bow shock of the jet , since the line profiles for both are similar @xcite , showing a low velocity , more intense peak together with a high velocity , less intense peak ( similar to the line profiles seen in figures [ f3 ] , [ f4 ] and [ f7 ] ) . we have found evidence for a blueshifted component in several emission lines around the northwest tip of the proplyd position . in particular , the h@xmath2 profiles reveal the presence of a blue - shifted component with systemic velocity @xmath84 km s@xmath0 , which indicates that its velocity is similar though smaller than the one of the redshifted jet . however , the intensities of the blue - shifted features are much lower than the intensity of the redshifted components associated with the se jet : the counter - jet is at least four times less intense than the redshifted jet ( in h@xmath2 ) . the presence of such a blue - shifted component very close to the lv2 proplyd , together with previously reported evidence that ( at several arcseconds to the nw of this proplyd ; see massey & meaburn 1995 ) there is a blueshifted component ( in the [ o iii]@xmath75007 line ) strongly suggests the existence of a faint counterjet . it is interesting to estimate the mass loss rate for the proplyd and the associated redshifted jet . in order to do this , we consider a simple model that assumes that the proplyd flow arises from a hemispherical , constant velocity wind that originates at the proplyd ionization front , at a radius @xmath85 . using the h@xmath2 luminosity @xmath86 , we can then obtain the particle density @xmath87 at this point and the mass loss rate , from : where @xmath90 accounts for the extintion correction , @xmath91 @xmath92 is the effective recombination coefficient for h@xmath2 , @xmath93 is the energy of the h@xmath2 transition , @xmath94 is the mean molecular weight and @xmath95 km s@xmath0 is a typical value for the sound velocity in an hii region . the extintion correction is obtained from the base 10 logarithm of the extintion at @xmath96 ( @xmath97 ) which for the proplyd 167 - 317 is equal to 0.83 @xcite through the relation , @xmath98 @xcite , where here we take @xmath99 = 0.56 . assuming for the ionization front radius a value @xmath100 cm @xcite we obtain @xmath101 @xmath63 and @xmath102 m@xmath4 year@xmath0 . @xcite have obtained a @xmath103 m@xmath4 year@xmath0 @xmath104 , and the small differences between both results may arise from the different observational techniques employed in both cases . where @xmath108 cm is the radius of the blob , @xmath109 km s@xmath0 and @xmath110 cm are the propagation velocity and length of the jet , respectively , derived above . with these values , we obtain for the jet a mass loss rate equal to @xmath111 m@xmath4 year@xmath0 , which is similar to the mass loss rate of typical hh jets from t tauri stars . we have been able to construct a ( [ nii]@xmath36 + [ nii]@xmath37)/[nii]@xmath112 line ratio map . this ratio indicates densities higher than @xmath11 @xmath63 for the proplyd emission region , consistent with values previously reported in the literature . we find a subtle enhancement of this ratio in the region of the redshifted jet . further observations of electron temperature and density diagnostic lines would allow a more accurate determination of the spatial dependence of these important parameters . we thank the anonymous referee for his / her comments and suggestions , which have improved significatively the presentation of the paper . we are greatful to w. henney for enlightning discussions . m.j.v . and a.h.c . would like to thank the staff of the instituto de ciencias nucleares , unam , mxico , for their kind hospitality , as well as for partial financial support during our visit . we would like to thank bryan miller , tracy beck and rodrigo carrasco from the gemini observatory , and bruno castilho from lna / cnpq , for help us in the early stages of ifu - data processing . the work of m.j.v . , a.h.c . and h.p . was partially supported by the bahia state funding agency ( fapesb ) , by the universidade estadual de santa cruz ( uesc ) and by the milenium institute ( cnpq project n. 62.0053/01 - 1-padct iii / milnio ) . this work was partially funded by the conacyt grants 41320-f and 43103-f and the dgapa ( unam ) grants in112602 and in111803 . integrated emission map ( shown with a linear grey scale in arbitrary units ) showing the location of the 167 - 317 proplyd with respect to the scientific field of view ( sfov ) , and its orientation : north is up and east is left , as indicated by the arrows on the top - left corner of the figure . the field has 16 pixels in the east direction and 25 in the north direction . the r4 frame located at the left and top parts of the figure represents the region choosen to define the background , which covers two columns of pixels on both the north and east directions . an internal box ( r1 ) , approximately centered on the object emission defines a coordinate system , and the numbers on the bottom axis and on the left axis of this internal box are referred to in the text as the @xmath53- and @xmath54-coordinates , respectively . this spatial sub - sample of spectra will be used throughout the paper . the @xmath6 ori c star is located at @xmath113 from the proplyd , at a p.a . = 236@xmath114 . four regions are defined in this figure , namely r1 ( the internal box ) , r2 ( the proplyd center ) , r3 ( a clump associated with a redshifted jet ) and r4 ( the background ) . we will refer to these regions throughout the paper . , width=340 ] has been added to the spectra of regions r2 ( @xmath115 ) , r3 ( @xmath116 ) and r4 ( @xmath117 ) in order to produce an intensity offset between the successive spectra . wavelengths ( @xmath7 ) are in , and the fluxes are in units of @xmath30 erg @xmath31 s@xmath0 @xmath0 . , width=566 ] -3truecm ) , showing the integrated flux ( the grey - scale map ) for the following lines : [ n ii]@xmath75755 ( top - left ) , [ n ii]@xmath76548 ( top - center ) , h@xmath2 ( top - right ) , [ n ii]@xmath76583 ( bottom - left ) , [ he i]@xmath77065 ( bottom - center ) and [ ar iii]@xmath77135 ( bottom - right ) . the linear bars give the grey - scale in unit of @xmath118 erg @xmath31 s@xmath0 px@xmath0 . the spectra in each position was background subtracted . superimposed on the intensity maps are the line profiles for each box . see the text for a discussion . the orientation is the same as in figure [ f1 ] ( i.e. , north is up and east is left ) . the original pixels ( see figure [ f1 ] ) have been resampled in order to smooth the emission features , and each box in these panels are slightly larger than the original pixels in fig . title="fig:",width=566 ] -3truecm . each diagram shows the line profile for regions r1 ( continuous line ) , r2 ( dotted line ) , r3 ( dashed line ) and r4 ( dash - dotted line ) . the emission line is indicated on the top - left corner of each plot . the fluxes are in units of @xmath118 erg @xmath31 s@xmath0 per and tey are plotted against the radial velocity ( in units of km s@xmath0 ; not corrected for the heliocentric velocity ) . , title="fig:",width=566 ] -5truecm as a function of the pixel position along a diagonal line that crosses region r1 , beginning at pixel ( 1,2 ) , passing through region r3 ( pixel 3,4 ) and finishing at pixel ( 11,12 ) . each type of line shows the intensity of a different spectral line : the solid line shows [ n ii ] @xmath119 ; the short dashed line , the intensity of [ n ii ] @xmath39 ; the dot - dashed line , h@xmath2 ; the triple dotted - dashed line , [ n ii ] @xmath38 ; the long dashed line , he i @xmath120 and , finally , the dotted line shows [ ar iii ] @xmath121 . in order to show all of the intensities in the same plot , we have divided the h@xmath2 intensity by a factor of 10 . , width=566 ] -2truecm ) and ( [ eq4 ] ) , respectively , for the h@xmath2 ( left ) and [ ar iii]@xmath77135.8 ( right ) lines . the grey scale on the right side of each map is in unit of km s@xmath0 . we note that both integrals show positive values and peak in a se region , indicating the presence of a redshifted component towards the se direction . the velocities are not corrected by the systemic radial velocity , and the integration limits in all maps are from @xmath122 km s@xmath0 . the white pixels have values close to zero . , title="fig:",width=377 ] -4truecm line at position ( @xmath53,@xmath54)=(6,7 ) ( see figure [ f1 ] ) . @xmath123 line profile and the three gaussian fit . @xmath124 main , low velocity line profile and its gaussian fit . @xmath125 the data profile subtracted from the main fitted profile , showing evidence for a blue- and a red - shifted component . @xmath126 the data profile subtracted from the three - component gaussian fit . the residual profile is a random fluctuation around a flux of 0.5@xmath127 erg @xmath31 s@xmath0 . these fluctuations corresponds to 10% of the flux for the blueshifted component , 0.5% of the flux for the main component and 10% of the flux for the redshifted component . b ) the same but for the [ ar iii]@xmath77135 line . , title="fig:",width=453 ] -5truecm erg @xmath31 s@xmath0 px@xmath0 ) as a function of position for the [ n ii]@xmath1285754 ( top - left ) , [ n ii ] @xmath76548 ( top - center ) , h@xmath2 ( top - right ) , [ n ii]@xmath76583 ( bottom - left ) , hei @xmath76678 ( bottom - center ) and [ ar iii]@xmath77135 ( bottom - right ) lines . almost all of the lines peak at the proplyd position , and three of them , namely the [ n ii]@xmath76548 ( top - centre ) , h@xmath2 ( top - right ) and [ n ii]@xmath76583 ( bottom - left lines ) , have a secondary , less intense peak in region r3 ( see figure [ f1 ] ) . , title="fig:",width=566 ] -9truecm -2truecm erg @xmath31 s@xmath0 px@xmath0 , as a function of position in region r1 . b ) the velocity , in km s@xmath0 for the redshifted gaussian fit to the h@xmath2 profiles . c ) the jet velocity variation as a function of distance from the proplyd ( see the text for a discussion ) . the velocities take into account the systemic radial velocity of @xmath129 km s@xmath0 . the error associated with the radial velocity measurement is of the order of 2.5 km s@xmath0 ( see the text for a discussion ) . , title="fig:",width=566 ] -6truecm @xmath130([nii]@xmath36 + [ nii]@xmath37)/[nii]@xmath112 ratio map of region r1 ( see figure [ f1 ] ) of the observed field . @xmath131 a [ @xmath132 diagnostic diagram showing the locci corresponding to different values of the nitrogen line ratio . @xmath133 the density for an assumed temperature of @xmath66 k. @xmath134 the density for an assumed temperature of @xmath135 k. , title="fig:",width=604 ] -8truecm

hadronic processes with a hard scale involved constitute an indispensable tool for probing the qcd dynamics of quarks and gluons , and through the qcd factorization theorems @xcite that separate physics at small and large distances , one may also study the dynamics of soft processes with small momentum transfers . hard quark and gluon interactions at small distances are thus not affected by soft interactions and are described in perturbative qcd . the most problematic part of the process is soft interactions at large distances , where nonperturbative qcd comes into the game and manifests itself as the confinement of quarks and gluons in hadrons and the related hadronization process giving the observable hadronic final states in high energy collisions . diffractive processes are sensitive to the details of nonperturbative qcd dynamics and provide a way to probe the soft and semihard regimes directly . diffractive events are characterized by a leading `` target '' particle , carrying most of the beam momentum , and a well separated produced hadronic system . the `` gap '' in between is connected to the soft part of the event and therefore to nonperturbative effects at a long space - time scale . diffractive deep inelastic scattering ( ddis ) offers a particularly good opportunity to explore the interplay between hard and soft physics due to the precise data from the electron proton collider hera @xcite . ddis in lepton proton collisions involves hard scattering events where , in spite of the large momentum transfer @xmath1 from the electron , the proton emerges essentially unscathed with small transverse momentum , keeping almost all of its original longitudinal beam momentum ( for reviews on ddis , see e.g. refs . the leading proton is well separated in momentum space , or rapidity @xmath2 , from the central hadronic system produced from the exchanged virtual photon s interaction with the proton . thus , this new class of events is characterized by a _ large rapidity gap _ ( lrg ) void of final state particles . rapidity gaps in deep inelastic scattering ( dis ) were discovered by the zeus and h1 experiments at hera @xcite , but the first discovery of hard diffraction was in @xmath3 collisions by the ua8 experiment @xcite . these processes had actually been predicted @xcite by combining regge phenomenology for soft processes in strong interactions via pomeron exchange , with hard processes based on perturbative qcd . by parametrizing the parton content of an exchanged pomeron ( or alternatively diffractive parton density functions ) it is possible to describe the hera data . however , the extracted parton densities are not universal , since when used to calculate diffractive hard scattering processes in @xmath4 collisions at the tevatron one obtains cross sections an order of magnitude larger than observed . schematic illustration of the mechanism for diffractive deep inelastic scattering considered in this paper , with soft gluon exchanges in the final state.,scaledwidth=45.0% ] an alternative dynamical interpretation of hard diffraction was proposed in refs . this soft color interaction ( sci ) model is based on the simple assumption of soft gluon exchanges leading to color rearrangements between the final state partons . variations in the topology of the confining color fields lead to different hadronic final states . the sci model is implemented in monte carlo event generators , e.g. lepto for dis @xcite . the hard part of the process shown in fig . [ fig : ddis - cartoon ] is then calculated in the framework of perturbative qcd with dglap evolution of the parton showers in the same way as in inclusive dis . the large momentum transfer means that the hard subprocess occurs on a spacetime scale much smaller than the bound state proton and is thus `` embedded '' in the proton . the emerging hard - scattered partons propagate through the proton s color field and may interact with it . soft exchanges will dominate due to the large coupling and the lack of suppression from hard gluon propagators . therefore , the momenta of the hard partons are essentially undisturbed the soft , long distance interactions do not affect the hard , short distance process , and the momentum transfer of the soft exchanges can be neglected . however , the exchange of color changes the color charges of the emerging partons such that the confining string - like field between them will have a different topology , resulting in a different distribution of final state hadrons produced from the string hadronization @xcite . in particular , a region in rapidity without a string will result in an absence of hadrons there , i.e. a rapidity gap . the only parameter of this model is the probability for a soft exchange , accounting for the unknown nonperturbative dynamics . remarkably , the sci model is phenomenologically very successful in describing many different processes , both diffractive and nondiffractive @xcite , with only a single parameter @xmath5 for this probability . thus , the sci model captures the essential dynamics of diffraction , but lacks a solid theoretical basis . to understand better what we can learn from the phenomenology of the sci model , we present in this paper a detailed qcd - based mechanism for soft gluon rescattering of final state partons , as illustrated in fig . [ fig : ddis - cartoon ] . this mechanism leads to effective color singlet exchange and thereby to diffractive scattering . inspired by the sci model , the model presented here may be seen as an explicit realization of the earlier attempt @xcite to understand soft gluon exchange in terms of qcd rescattering . our model was initially introduced in a recent letter @xcite , and is here presented in detail . the paper is organized as follows . in section ii we briefly discuss the framework of the dipole approach and motivate our study . in section iii we consider the kinematics of diffractive dis . section iv treats the formalism for generalized unintegrated gluon distribution functions in the diffractive limit . the explicit calculation of the @xmath6 dipole contribution to the diffractive cross section and analytic approximations used are presented in section v. in section vi we study the contribution of the @xmath7 final state . numerical results and comparisons with hera data on the diffractive cross section are given in section vii . finally , in section viii we present some concluding remarks and an outlook . typical contributions to the diffractive dis process are represented by the diagrams in fig . [ fig : ddisa ] . in terms of the four - momenta @xmath8 of the photon , and @xmath9 and @xmath10 of the initial and final proton , the kinematics of the @xmath11 process is defined by the variables @xmath12 where @xmath13 . the invariant mass of the produced system @xmath0 , and the total energy in the @xmath14 center - of - mass system @xmath15 are given by @xmath16 the ddis cross section in general is represented as a function of @xmath17 and the momentum transfer along the proton line @xmath18 . note , that we are working in the forward limit of small @xmath19 . typical diagrams contributing to diffractive dis with the leading order @xmath6 dipole scattering ( left ) and the gluonic @xmath7 contribution ( right ) contributing significantly at @xmath20 ( large @xmath0).,scaledwidth=45.0% ] let us consider first the simplest case of the @xmath6 contribution , which is the leading one for small @xmath0 ( or , equivalently , @xmath21 ) . to compute the diffractive dis amplitude , it is convenient to consider the process in the dipole frame @xcite , where the deeply virtual photon with large virtuality @xmath1 and polarization @xmath22 first splits into a quark @xmath8 and an antiquark @xmath23 with mass @xmath24 , spins @xmath25 and @xmath26 , and flavor @xmath27 , and then the @xmath6 dipole with transverse size @xmath28 interacts with the target proton at impact parameter @xmath29 and dissociates into a final state @xmath30 of invariant mass @xmath0 as shown in fig . [ fig : amplo ] . the photon splitting into the dipole is a qed process and is described by the wavefunctions @xmath31 in the impact parameter space @xcite , where @xmath32 is the fraction of the longitudinal momentum carried by the quark . the amplitude of the dipole nucleon scattering ( denoted as @xmath33 in fig . [ fig : ddisa ] ) is the only unknown non - perturbative object ( for a review , see ref . the dipole picture naturally incorporates the description of both inclusive and diffractive events into a common theoretical framework @xcite , as the same dipole scattering amplitudes enter into inclusive and diffractive cross sections . final states with gluons are suppressed by powers of @xmath34 . however , if @xmath26 becomes small or @xmath1 large , the @xmath6 dipole emits soft or collinear gluons accompanied by large logarithms @xmath35 or @xmath36 which compensate the suppression in @xmath34 @xcite . the @xmath37 pair can emit soft gluons , leading to the dressing up of the quarks , which is parametrized by a scale - dependent constituent quark mass @xmath38 . in general , the more gluons in the final state , the larger the invariant mass produced . the dominant gluon emission from quarks is described by dglap evolution @xcite and is mostly collinear to the radiating quark , so it can not build up a large @xmath0 . the small @xmath20 and large @xmath39 limits can be driven , therefore , only by a semihard gluon radiation from the active gluon ( carrying @xmath40 ) giving rise to a gluonic dipole @xmath41 contribution . these aspects will be discussed in detail below . amplitude of the process @xmath42 at the leading order of perturbation theory.,scaledwidth=40.0% ] diffractive dis at the leading order in @xmath34 is described by the two gluon exchange diagram shown in fig . [ fig : amplo ] . let us first discuss how the longitudinal momentum transfer between the @xmath37 dipole and the proton can be shared between the gluons . the gluon momenta can be sudakov decomposed as @xmath43 cutting the diagram after the first gluon exchange ( picking up the leading poles only ) , in the high - energy limit @xmath44 , we have @xmath45 in the deep inelastic limit @xmath46 , for fixed invariant mass of the intermediate system @xmath47 and bjorken variable @xmath48 , we see that @xmath49 and @xmath50 , thus @xmath51 . on the other hand , when @xmath46 and @xmath52 fixed , we see that @xmath53 . so @xmath54 and @xmath55 , and the first gluon takes all the longitudinal momentum exchange neutralizing the virtuality of the @xmath37 system . the latter kinematical configuration gives the leading contribution to the cross section , whereas the other configurations with equal momentum sharing between the gluons @xmath56 are suppressed by extra propagators . amplitude of the process @xmath42 with all - order resummed soft gluon exchange.,scaledwidth=40.0% ] thus , we consider the asymmetric case with one hard ( perturbative ) gluon carrying most of the longitudinal momentum transfer @xmath40 , and a number of multiple soft screening interactions with total @xmath57 in a color octet state ( in the large @xmath58 limit ) effectively described by the resummed multigluon exchange amplitude as schematically sketched in fig . [ fig : amp ] . the first `` hard '' gluon turns the proton into a color octet proton remnant , which then interacts with a system of soft screening gluons in an octet state and finally recombines into a color singlet corresponding to the leading proton ( or system with invariant mass close to the proton mass ) . these soft gluons can not dynamically resolve a @xmath59 dipole of small size @xmath60 , however , we assume that the gluons interact with the quark current and not with the dipole as a whole . this will be further discussed below in sec . [ interpretation ] . let us first define the kinematics of the process @xmath61 . our primary interest is processes with small momentum transfer @xmath62 . it is convenient to fix the frame of reference in the center - of - mass system of the final states , i.e. the outgoing proton with momentum @xmath10 , and the diffractive system @xmath30 with momentum @xmath63 , @xmath64 . the total @xmath65 c.m.s . energy squared is @xmath66 ( the proton mass @xmath67 is neglected ) . in terms of the cross section variables defined in eqs . ( [ vars ] ) and ( [ vars1 ] ) we write @xmath68 since @xmath69 , we have @xmath70 for any @xmath26 and @xmath1 , so we will first keep @xmath71 and then drop it in comparison with @xmath72 whenever appropriate . the general sudakov decompositions of the final quark / antiquark momenta @xmath73 are @xmath74 where @xmath75 . the on - shell conditions for the quark and antiquark in the final state @xmath76 give @xmath77 finally , @xmath78 leads to @xmath79 where @xmath80 is the transverse quark mass squared . applying eq . ( [ vars1 ] ) , we get the standard relation for the quark transverse momentum @xmath81 and we consider the light quark mass limit @xmath82 . the leading contribution to diffractive dis at hera comes from light quarks , and from now on we do not distinguish between their masses and account for them by one single mass parameter @xmath83 . in the diffractive limit @xmath84 , when @xmath32 and @xmath85 are not very asymmetric , one has with good accuracy : @xmath86 let us now define the quark propagators in the photon splitting wave function . due to the condition @xmath87 the soft screening gluons can not change the longitudinal momenta significantly , but only the transverse momenta . thus , to calculate the hard part of the amplitude let us first neglect these extra screening gluons . we will show below that adding the extra soft gluons leads only to phase shifts ( and their derivatives ) in the transverse coordinates , which are going be resummed to all orders in @xmath34 . in the chosen frame , the momenta of the exchanged hard and the sum of the soft gluons are @xmath88 let us first attach the hard gluon to the lower quark line @xmath89 . then the denominator of the quark propagator between the photon and gluon vertices is @xmath90 where @xmath91 . the first approximation is obtained by substituting @xmath71 from eq . ( [ mn ] ) in the limit @xmath92 and the second by using the limit @xmath93 , realized when @xmath94 , giving a result valid at @xmath95 . when the gluon is attached to the upper gluon line , using momentum conservation @xmath96 , we get analogously @xmath97 which is strictly valid at @xmath98 . it is equal to eq . ( [ pr1fin ] ) with the exchanges @xmath99 and @xmath100 . the expressions ( [ pr1fin ] ) and ( [ pr2fin ] ) will be used for all values of @xmath32 , as is common practice @xcite . this is justified in our asymmetric case @xmath101 because the dominating contribution to the amplitude comes from the configuration that either the quark or the antiquark from the photon is essentially on - shell , and the other carries the negative virtuality of the photon and then absorbs the hard gluon with momentum fraction @xmath40 to become essentially on - shell . in the limit considered , @xmath93 , the quark virtuality @xmath102 is conventionally utilized as the factorization scale @xmath103 of the process , and is expressed in terms of the energy @xmath104 and the transverse momentum @xmath105 as @xmath106 thus , the hard scale depends on both @xmath1 from the space - like photon and @xmath71 from the time - like final state @xmath30 . since these can have any values , the qcd factorization is complicated and the physics may be different in the three cases @xmath107 , @xmath108 and @xmath109 . the quark propagator ( with the hard scale @xmath103 in eq . ( [ muf ] ) ) is antisymmetric , @xmath110 , with respect to reflection between the space - like and time - like regimes , i.e. , with respect to the interchange @xmath111 ( or @xmath112 ) as it should be . the next step is to compute the bilinear spinor combinations @xmath113 in the photon splitting @xmath114 . the photon polarization vectors in the @xmath115 c.m.s . frame have the following general form @xmath116 in particular , by straightforward calculation for the longitudinally ( l ) polarized photon we simply get the following expression @xmath117 which is not dependent on the transverse momenta of the initial quark and antiquark . the transversely polarized case requires a separate discussion . within the dipole picture the diffractive dis process can be basically decomposed into time - ordered stages . first , the space - like photon with @xmath118 fluctuates into a @xmath6 pair , which is then scattered off the target through hard gluon exchange , making the @xmath6-system time - like , and finally the on - shell quark and antiquark scatter softly off the color background field in the proton resulting in a color singlet @xmath30-system with invariant mass @xmath0 . initially , at the moment of the photon fluctuation , the only hard scale is @xmath1 , and the transverse momentum of a produced quark is expressed through this scale as @xmath119 , which is different from the transverse momentum @xmath105 of a quark in the final state defined in eq . ( [ kt ] ) . in particular , the difference between @xmath105 and @xmath120 can depend on the actual momentum transfer @xmath121 for the hard gluon and on the sum of the screening gluons , because according to eq . ( [ kt ] ) the small variation in @xmath105 due to the attached @xmath121 brings a significant change in @xmath0 for small @xmath32 . the relative coefficient between @xmath105 and @xmath120 is @xmath122 , and has to be taken into account when expressing the transversely polarized photon splitting wave function through the final state transverse momenta . this physical argument agrees well with the kinematics corresponding to the diagram shown in fig . [ fig : amp ] for @xmath123 . in the opposite limits @xmath124 or @xmath125 the emission of extra gluons significantly complicates the kinematics , and this will be considered in detail below . as a final result , for the transversely ( t ) polarized photon , we get in the chiral limit @xmath126 the spinor signs @xmath127 here stand for quark / antiquark chiralities , which coincide with the helicity for a quark and have the opposite sign of the helicity for an antiquark . we do not take into account the `` @xmath128 '' and `` @xmath129 '' components since they are small in the bjorken limit and for relatively light quark and antiquark . before calculating the amplitude for the hard and soft gluon exchanges , we note that the exchanged gluons all originate from the proton color field and should therefore be treated through a common description of a general gluon density . the first , hard gluon carries the dynamics through the longitudinal momentum @xmath130 , whereas the soft rescattering gluons carry small momenta @xmath131 but may transfer a color octet charge that screens the color of the first gluon resulting in an overall color singlet exchange . in this sense , the sum of all soft exchanges acts as a single effective gluon exchange between the @xmath6 dipole and the proton remnant ( see fig . [ fig : smallxpdf ] ) . the sum of the multiple soft gluon exchanges can in an effective description be seen as a single gluon exchange between the @xmath6 dipole and the color octet proton remnant , resulting in the color singlet @xmath30-system and a proton in the final state.,scaledwidth=45.0% ] as an appropriate description of the density of hard and soft effective gluons , we use the framework of generalized off - diagonal unintegrated gluon distribution functions ( ugdf ) , which naturally appear in the @xmath105-factorization approach @xcite . within this framework the coupling to a quark is replaced by an off - diagonal ugdf @xmath132 , absorbing a factor @xmath133 and by convention also a gluon propagator @xmath134 in order to keep the ugdf regular as @xmath135 . the absorbed coupling @xmath136 corresponds to the coupling of a screening gluon with virtuality @xmath137 to a quark in the proton , whereas the coupling of the hard gluon to the @xmath6 dipole is purely perturbative and occurs at the hard scale @xmath138 . generalized parton distributions ( gpd ) are not very constrained by data . we use a prescription for the generalized ugdf @xcite , which works well in the description of cdf data on central exclusive charmonium production @xcite . this prescription is motivated by positivity constraints for the collinear gpds @xcite and can be considered as a saturation of the cauchy schwarz inequality for the density matrix @xcite . it incorporates the dependence on the longitudinal momentum fraction @xmath139 and the transverse momenta of the soft gluons in an explicitly symmetric way , @xmath140 with @xmath141 . here @xmath142 is the normal diagonal ugdf , which depends on the gluon virtuality @xmath143 and reduces to the well - known collinear gluon pdf @xmath144 when integrated over this virtuality . this model ( [ sqrt ] ) is a factorization of the generalized ugdf into a hard part depending on a hard scale @xmath138 and on @xmath40 , thus describing the hard gluon coupling to the proton , and a soft part defined at some soft scale @xmath145 and small @xmath57 corresponding to a number of soft gluon couplings . as we will see below , together with the factorization in transverse momentum space , the model ( [ sqrt ] ) provides a qcd factorization of the diffractive amplitude in momentum space . the ugdf @xmath146 introduced above depends on the gluon virtuality , and this dependence is not theoretically well - known for small virtualities . the ugdf is here modeled using the collinear gluon density @xmath147 , fixed at the qcd factorization scale @xmath138 , together with a simple gaussian ansatz for the dependence on the gluon virtuality @xmath143 as @xmath148 where the gaussian width @xmath149 is the soft hadronic scale . as demonstrated below , this scale corresponds to the transverse proton size @xmath150 . the gaussian smearing is then interpreted as the result of many soft interactions in the bound state proton . the factor @xmath151 in eq . ( [ ugdf ] ) is absorbed from the hard subprocess part describing the coupling of a hard gluon to the @xmath6 dipole , and gives us a hint that the off - diagonal ugdf should be proportional to @xmath152 . it is known that at some soft scale @xmath153 collinear pdfs like grv @xcite saturate at small @xmath57 , so one can introduce a function @xmath154 , which is assumed to be slowly dependent on @xmath139 in the case @xmath155 : @xmath156 this factor @xmath157 contains all the soft physics related with soft gluon couplings to the proton . it is interpreted as the square root of the gluon density at very small @xmath57 and soft scale @xmath158 . this is a non - perturbative object , which contributes to the overall normalization and can be determined from data . as will be seen below , the prescription ( [ ugdfsat ] ) is consistent with the hera data for all available @xmath71 and @xmath1 . there is a debate about what the power of the gluon density in the cross section should be ( see @xcite and references therein ) . when squaring the amplitude containing ( [ ugdfsat ] ) , this model leads to a _ linear _ dependence of the diffractive cross section on the gluon pdf . this linear dependence is the same as in the sci model , where it describes both diffractive and non - diffractive events , and provides a continuous transition between the two types of events . this is in contrast to the _ quadratic _ dependence on the gluon density often encountered in two - gluon exchange calculations of ddis @xcite . this arises from another prescription for the off - diagonal ugdf in the asymmetric limit @xmath57 and @xmath159 , which in terms of the diagonal ugdf reads @xcite @xmath160 where the skewedness parameter is an analog of this skewedness factor . ] @xmath161 is roughly constant at hera energies , and gives only a small contribution to an overall normalization uncertainty . the factor @xmath162 can be approximately taken into account in this case by rescaling the @xmath40 argument in the diagonal ugdf as @xcite @xmath163 using the same gaussian ansatz for the intrinsic transverse momentum dependence as in eq . ( [ ugdf ] ) , we see that prescription ( [ rg ] ) leads to a quadratic dependence of the diffractive structure function on the gluon pdf . the unintegrated gluon density in the form ( [ sqrt ] ) reduces to the diagonal form ( [ rg ] ) in the kinematical domain where @xmath164 and the hard gluon is soft , @xmath165 . in this limit there is no qcd factorization , so the hard and soft gluons must be taken into account together on equal footing . this may be the case at low @xmath1 and @xmath71 , when a larger contribution to the cross section comes from relatively soft scales @xmath166 gev , and eq . ( [ ugdf ] ) reduces to @xmath167 where the factor 0.5 appears from the equal momentum sharing between active and screening gluons , since @xmath40 is the sum of all gluon momentum fractions . in this sense , the prescription ( [ sqrt ] ) is more general since it describes both @xmath57 and @xmath164 regimes , and contains prescription ( [ rg ] ) as a limiting case . ( [ ugdfrg ] ) also leads to a cross section with the gluon density in the second power . it is similar to the `` @xmath162-prescription '' ( [ rg ] ) ( if the gaussian smearing like in eq . ( [ ugdf ] ) is adopted ) , and close to its phenomenological form with rescaled argument ( [ rg1 ] ) . this is more conventional in the description of the exclusive processes @xcite , but is valid only for the symmetric case where the two gluon exchanges carry longitudinal momentum fractions close to each other , @xmath164 , and are connected to the same factorization scale @xmath138 . this case can also correspond to the `` no - soft - exchange '' approximation , when the soft rescattering of the on - shell partons in the final state is not taken into account ( then @xmath168 ) . the above discussion leads to the conclusion that whether it is appropriate to use prescription ( [ ugdfsat ] ) or ( [ ugdfrg ] ) depends on what kinematical regime one considers . the problem is that @xmath139 is not strictly constrained by kinematics . its order of magnitude should be collisions , where the screening gluons couple to the triplet / antitriplet charges of the proton remnants , which have predominantly equal momentum sharing @xmath169 , and thus we have @xmath170 . ] @xmath171 thus the @xmath57 regime is realized in the perturbative limit of large factorization scale @xmath138 . in the limit @xmath172 , one instead has @xmath173 , as naturally required by matching the prescriptions of eq . ( [ ugdfsat ] ) and ( [ ugdfrg ] ) . to summarize this section , we have formulated a framework for the gluon density needed as input to the calculation of the diffractive cross section . since this describes soft qcd dynamics in the proton , there are necessarily some uncertainties . the precise hera data on the diffractive cross section are , however , directly sensitive to this gluon density and may , therefore , be used to obtain new information about the gluon pdf at extremely small @xmath139 values and at different scales . we are now fully equipped to derive the amplitude for the dipole proton interaction . the total amplitude for @xmath42 is decomposed into longitudinal ( l ) and transverse ( t ) parts depending on the photon polarization @xmath174 , and each part can be written as a convolution of the hard and soft subprocess amplitudes based on loop integration and cutting rules . starting from the general sudakov decomposition of the total screening gluon momentum @xmath175 with @xmath176 , we can write down the amplitude , for example , for transversely polarized photon as @xmath177 the @xmath178-function product can be rewritten as @xmath179 which takes care of the integrals over @xmath180 and @xmath29 , leading to @xmath181 where , in the frame with @xmath182 , @xmath183 analogously , the longitudinal contribution is @xmath184 to calculate the fourier transform of the total amplitude we use the convolution theorem @xmath185 where @xmath186 denotes the fourier transform of @xmath27 . in our amplitude this convolution is represented by the integral over @xmath121 , while @xmath187 plays the role of @xmath8 . thus , the inverse transformation over the impact parameter @xmath29 is @xmath188 leading to factorization of the amplitude in @xmath29-space as a direct consequence of @xmath105-factorization in impact parameter space . consider first the hard gluon coupling to @xmath8 or @xmath189 shown in fig . [ fig : amp ] in the rotated frame of reference with @xmath190 and @xmath191 , where @xmath192 . in what follows , we use the relation for quark - gluon vertices in the eikonal approximation @xmath193 and for the product of the @xmath194-matrices in the large @xmath58 limit we have @xmath195 the hard part , describing the two possible couplings of the hard gluon to the @xmath6 pair , can then be written as @xmath196 where @xmath120 is the transverse momentum of a quark in the intermediate state ( see fig . [ fig : amp ] ) , and the fourier - transformed hard amplitudes are given by @xmath197 where @xmath198 , and @xmath199 are bessel functions . the function @xmath200 is the gluon density in impact parameter space defined as @xmath201 here , @xmath202 is the generalized ugdf , and @xmath158 the typical soft scale of the process given by the gluon virtuality @xmath203 . the factor containing @xmath204 is introduced in the normalization of the soft part in order to compensate its absorption into the ugdf ( see above ) . inserting @xmath205 from eq . ( [ ugdfsat ] ) , we finally get @xmath206 . \end{aligned}\ ] ] in the fourier transformation we assumed slow evolution of the qcd coupling @xmath204 , as is the case in the analytic perturbation theory discussed in next section or often assumed by freezing the coupling at very small @xmath158 . thus , in the gaussian model ( [ ugdf ] ) , the unintegrated gluon density in the impact parameter space @xmath200 is factorized into a collinear gluon density multiplied by an ( @xmath207)-dependent normal distribution . we now turn to the soft subprocess amplitude , which can be calculated order - by - order as follows . the softness of the color - screening gluons with @xmath131 implies that all intermediate particles are on - shell , and that the dipole size @xmath28 is not changed during the soft interactions . cutting the intermediate propagators we have only the phase shifts with the same origin as in eq . ( [ fourier - hard ] ) , and a dependence on the soft momentum exchanges @xmath208 . in particular , for one and two soft gluon exchanges ( in the large @xmath58 limit ) we obtain @xmath209\,,\\ e^{-i{\bf r}{\bf k}_{\perp}'}m^\text{soft}_{2}=\frac{{\cal a}^2}{2!}\ , e^{-i{\bf r}{\bf k}_{\perp}}\int\frac{d^2\delta'_{2\perp}}{(2\pi)^2 } \frac{1}{{\delta'}^{2}_{1\perp}{\delta'}^{2}_{2\perp}}\\ \times \left[e^{-i{\bf r}{\bf \delta}'_{\perp}}-e^{-i{\bf r}{\bf \delta}'_{2\perp}}- e^{-i{\bf r}{\bf \delta}'_{1\perp}}+1\right]\ , , \end{aligned}\ ] ] where @xmath210 and @xmath211 with @xmath212 in the large @xmath58 limit . for example , the nlo gluonic contribution to the soft part @xmath213 is represented by the four diagrams shown in fig . [ fig:2gresum ] . illustration of resummation of the gluonic contributions to the ( nlo ) soft part @xmath213 ( two gluon exchanges ) in the large @xmath58 limit.,scaledwidth=50.0% ] fourier transformation with respect to @xmath214 leads to @xmath215 where @xmath216 continuing this procedure we see that summing over the number of soft gluons in the final state leads to exponentiation in impact parameter space , so that for the total soft subprocess amplitude we finally get @xmath217 a similar expression was previously derived in the case of scalar abelian gauge theory in ref . note that @xmath218 is independent of the photon polarization in the soft limit of small @xmath208 . as mentioned before , the soft gluon exchanges between the final state partons occur at non - perturbatively small longitudinal @xmath139 and transverse momentum transfer @xmath219 at some soft scale @xmath145 . the strong coupling @xmath204 is not small in this case . there are several approaches for dealing with the landau pole at low momentum transfer ( see e.g. ref . @xcite and references therein ) . we use the infrared - finite analytic perturbation theory ( apt ) @xcite approach to parametrize @xmath204 at @xmath220 . the analytic strong coupling @xmath221 is stable with respect to the choice of the qcd renormalization scheme , higher - order radiative corrections and variations in @xmath222 . apt has also proved to give a quantitative description of light quarkonium spectra within the bethe salpeter approach @xcite and dis spin sum rules at low @xmath1 @xcite . in the one - loop case , the apt euclidean function @xmath223 , i.e. the analyticized first power of the coupling @xmath34 in the euclidean domain , is @xcite @xmath224 , \ ; l=\ln\left(\frac{\delta_{\perp}^2}{\lambda_\text{qcd}^2}\right)\ ] ] where @xmath225 is the first coefficient of the qcd @xmath26-function . since a significant contribution comes from the phase space region with strongly uneven longitudinal momentum distribution between the quark and the antiquark , and where @xmath105 is not very large , the diffractive structure function becomes sensitive to the model of the strong coupling used to calculate @xmath226 , and hence to the typical soft scale @xmath145 of the process . to avoid this problem in practice , in the soft regime we are considering , where @xmath227 , we do not extract the value of @xmath145 but rather fix the coupling at @xmath228 . having all the l and t amplitudes , we can now finally rotate our frame of reference in the transverse plane to one with @xmath182 and @xmath229 , where @xmath105 is defined as in eq . ( [ kt ] ) . in this frame , the variables @xmath29 and @xmath28 are the impact parameter and the @xmath6 dipole size in the final state , respectively . let us turn to the cross section . we are interested in the case when the proton remnant forms a particle in the final state with invariant mass close to the proton mass , so we have a three particle phase space . we may write in general @xmath230 in the large @xmath15 limit the flux factor is @xmath231 . the left hand side can be transformed to @xmath232 the diffractive structure functions have simple relations to the corresponding differential cross sections @xmath233 assuming an exponential @xmath194-dependence @xmath234 of the cross section on the diffractive slope @xmath235 . the @xmath178-functions remove the integrals over @xmath236 and one of the quark momenta , say , @xmath237 , and we get @xmath238 the remaining @xmath178-function removes the integral over @xmath239 . the last phase space integration is rather trivial , @xmath240 finally , in the full phase space we have to take into account an extra factor of two due to the symmetry with respect to the interchange @xmath241 . straightforward calculation leads to the following expressions for the longitudinal and transverse fully - unintegrated diffractive structure functions @xmath242 : @xmath243 where @xmath244 sums over light quark charges @xmath245 , and @xmath246 , \label{jl } \\ j_t = i\alpha_s(\mu_f^2 ) \int d^2{\bf r}d^2{\bf b}\ , e^{-i{\bm \delta } { \bf b}}e^{-i{\bf r}{\bf k}_{\perp}}\,\varepsilon k_1(\varepsilon r)\nonumber \\ \times\,\frac{r_x\pm ir_y}{r } \ , { \cal v}({\bf b},{\bf r } ) \big[1-e^{{\cal a}{\cal w}}\big]\ , . \label{jt } \end{aligned}\ ] ] these are the general expressions of the qcd - based soft multiple gluon rescattering model . we have now derived eqs . ( [ fl][jt ] ) , which describe the diffractive structure function . these have non - perturbative soft gluon exchanges as important ingredients , and to calculate these exchanges we have had to make some model assumptions . some of these assumptions have already been discussed above : we treat the coupling to the quarks using the strong coupling obtained in apt and the coupling to the proton remnant using the function @xmath247 . moreover , we extrapolate perturbation theory and assume a perturbative propagator for the gluons . the infrared logarithmic divergences in these gluon propagators , which appear at each order in the resummation ( see eqs . ( [ resum],[wq ] ) ) , disappear when the gluon exchanges are resummed to all orders ( eq . ( [ m - soft ] ) ) . there is one additional model assumption , as we will explain shortly , but let us first discuss a physical argument based on effective field theory principles , or equivalently , on the uncertainty principle : a gluon with momentum @xmath248 has a `` resolution power , '' or minimal scale of an object it can resolve , of order @xmath249 . put another way , physics should not depend on scales much smaller than the resolution scale . the hard gluon in our calculation can resolve the @xmath6 dipole with transverse size @xmath28 , allowing us to apply perturbation theory to the hard part . on the other hand , the soft gluons do not carry significant longitudinal momentum fractions , and only small transverse momenta @xmath250 , so their `` maximal resolution '' scale is @xmath251 . this means that the screening gluons can not dynamically resolve the internal structure of a small @xmath6 dipole with size @xmath252 . however , in constructing our model , we extrapolate perturbative qcd to the non - perturbative regime and assume that the soft gluons couple individually to the quark and antiquark , since the essential point of the dynamics here is the color exchange and not the momentum transfer . this is in a similar vein to using quark currents in hadronic matrix elements , such as form factors . the underlying quark and gluon dynamics is still important even at very low scales ( see e.g. ref . @xcite for a discussion of this ) . in this way there is a continuous transition between soft perturbative and soft non - perturbative gluons . as they stand , the integrals in eqs . ( [ jl],[jt ] ) exhibit unphysical singularities in the angular integrations . this , however , is because of our model assumption , which so far does not fully take into account the resolution power argument . since the gluons are soft , physics should not depend on the orientation of the @xmath6 dipole with respect to the impact parameter . this will regulate the unphysical singularities in the angular integration in eqs . ( [ jl],[jt ] ) . this will also allow us to evaluate the integrals analytically , and we will use this in our calculations below . we argue that the resulting expressions , eqs . ( [ jl2],[jt2 ] ) below , are the physically correct expressions for @xmath253 and @xmath254 to use in eqs . ( [ fl],[ft ] ) . the expression ( [ vdens ] ) can be considered as a model for the unintegrated gluon density in impact parameter space . in particular , it defines the probability to probe a gluon at impact distance @xmath29 from the proton center with momentum @xmath255 by a hard @xmath37 dipole with small size @xmath252 , where the quarks carry the hard momentum @xmath256 . the process is considered at a factorization scale equal to the quark virtuality @xmath103 . the gluons can not resolve scales below the dipole size @xmath28 . therefore , the gluon density can not depend on the orientation of the @xmath37 dipole with respect to @xmath257 , i.e. , on the angle between @xmath258 and @xmath257 . also , in this gaussian model there is no physical reason for an asymmetry of the ugdf with respect to the direction of the vector @xmath257 . thus we rewrite our expression ( [ vdens ] ) in the following way @xmath259 in the small dipole limit @xmath252 this becomes @xmath260 which will be used below to obtain the formula for the ddis amplitudes . we can check our formalism by taking the small coupling limit @xmath261 , where we can approximate @xmath262 in the longitudinally polarized case , the fourier integrals are then reduced to hankel transforms , leading to @xmath263 thus , in the limit @xmath264 our model successfully reproduces the standard leading - order two - gluon amplitude @xcite and leads to the correct exponential @xmath194-dependence of the cross - section @xmath265 with diffractive slope @xmath266 gev@xmath267 known from hera data @xcite . this gives @xmath268 mev , close to the value of @xmath222 . thus , the gaussian width @xmath149 physically corresponds to the effective transverse size of the proton . however , the strong coupling @xmath204 is not small in the case of small momentum transfers @xmath269 , and we can not calculate the integral in @xmath253 in general form analytically . the soft phase @xmath270 is not in general small in the fourier transformation , and in evaluating the fourier integrals in eqs . ( [ jl],[jt ] ) we should not impose the @xmath271 condition , but rather keep the exponent @xmath272 with imaginary @xmath273 . this produces an extra phase shift in the fourier transform over @xmath258 , coming from the soft gluon exponentiation in the large @xmath58 limit . employing the `` maximum resolution '' argument introduced above , we can write @xmath274 in the longitudinally polarized case the result of the fourier integration over @xmath258 is the hankel transformation of @xmath275 with respect to the momenta @xmath276 and @xmath277 . we obtain @xmath278 \nonumber\end{aligned}\ ] ] in terms of the complete elliptic integrals of the first and second kind , @xmath279 and @xmath280 , respectively . in the forward limit of small @xmath281 we expect @xmath282 , @xmath283 . there is a similar simplification in momentum space for the hard momentum @xmath284 , i.e. , @xmath285 , neglecting the dependence on the direction of the @xmath59 transverse momentum in the isotropic color field of the proton remnant . further , we expand the integrand in eq . ( [ j_l ] ) in @xmath286 , and keep only the leading term in @xmath287 . taking the last fourier integral gives @xmath288 where @xmath289 @xmath290 is a modified bessel function , and @xmath291 . the second term is an nlo contribution since it is proportional to the @xmath34 in @xmath292 , and typically in the forward limit @xmath293 and in the hard momentum transfer limit it is much smaller than the leading term ( we do not consider large @xmath194 , where the whole formalism here does not apply ) . however , this term is the only leading term which survives in the limit when both @xmath294 and @xmath295 , so we have to take it into account . as regards the @xmath194-dependence , the second term in eq . ( [ bint ] ) decreases at large @xmath194 , but not as rapidly as the first term . to good approximation , the integral over @xmath194 in the cross section can be written as @xmath296 the second term must be taken into account when @xmath105 is @xmath297 gev . straightforward calculation leads to the following result for the longitudinal contribution , @xmath298 , \nonumber\end{aligned}\ ] ] and for the transverse contribution , @xmath299,\end{aligned}\ ] ] where @xmath300 is the hypergeometric function and @xmath301 and @xmath302 are the complete elliptic integrals as above . in the large-@xmath0 limit , gluon emission may be important . in principle , gluons may be radiated from both the @xmath6 dipole and the hard gluon . the gluons emitted from the quarks are dominantly soft and move collinearly with the quarks , and do not significantly change the invariant mass of the final system @xmath30 . rather , they dress the quarks to build up their effective mass @xmath83 , which is , in general , a function of the two hard scales @xmath1 and @xmath71 . this mass parameter may be treated as a constituent quark mass . in the current work we do not make predictions for @xmath83 , but instead extract it from data . the scale dependence of the effective quark mass in processes with two hard scales like the one under consideration may be complicated . this will be discussed in connection with the numerical results in sec . [ numresults ] . typical diagram for gluon emission in the ddis final state ( a ) , and its asymptotic limit reducing to the `` gluonic dipole '' contribution for @xmath20 and @xmath303 ( b).,title="fig:",scaledwidth=25.0% ] typical diagram for gluon emission in the ddis final state ( a ) , and its asymptotic limit reducing to the `` gluonic dipole '' contribution for @xmath20 and @xmath303 ( b).,title="fig:",scaledwidth=25.0% ] the small @xmath20 limit is , therefore , driven by gluon emission from the hard gluon , as illustrated in fig . [ fig : qqg ] . the kinematics of the process in the @xmath115 c.m.s . frame , where @xmath304 , is given by the sudakov decompositions @xmath305 with @xmath306 . analogously to the @xmath307 case , we obtain the following expression for the invariant mass @xmath0 in terms of momentum fractions @xmath308 : @xmath309 @xmath0 can also be represented in terms of the invariant mass of the @xmath6 system as @xmath310 the @xmath7-system scatters off the proton by exchanging soft gluons , in the same way as the @xmath6-system above , and also here the gluon exchanges can be resummed . we have two independent transverse momenta , @xmath311 and @xmath312 , outgoing from the hard subprocess , corresponding to impact parameters @xmath313 and @xmath314 . proceeding as for the @xmath315 case , we obtain for the @xmath7 case @xmath316\ , , \end{aligned}\ ] ] where the prefactor @xmath273 was introduced above for the gluon coupling to a @xmath59 dipole , and @xmath317 corresponds to the case with a gluon coupling to a gluon in the @xmath7-system . by explicit calculation of the color factors it can be shown that in the large-@xmath58 limit @xmath318 . this allows us to perform the fourier transformation of the soft part over @xmath319 for any number of exchanged gluons and to resum them in the same way , as for @xmath6 dipole rescattering . for one and two gluon exchanges we have @xmath320\,,\\ e^{-i{\bf r}_1{\bf k}_{1,\perp}'}e^{-i{\bf r}_2{\bf k}_{2,\perp } ' } & \hat{m}^\text{soft}_{2,q{\bar q}g}= e^{-i{\bf r}_1{\bf k}_{1,\perp}}e^{-i{\bf r}_2{\bf k}_{2,\perp}}\\ & \times\frac{{\cal a}^2}{2!}\ ; \big[{\cal w}({\bf b},{\bf r}_1)+{\cal w}({\bf b},{\bf r}_2)\big]^2\,,\ ; \nonumber \end{aligned}\ ] ] where @xmath321 is defined above in eq . ( [ wq ] ) . summing over the number of soft gluons in the final state leads to exponentiation in impact parameter space , i.e. , @xmath322}\big]\ , . \nonumber \end{aligned}\ ] ] let us now focus on the leading asymptotic behavior of the diagram in fig . [ fig : qqg](a ) in the limit @xmath20 . in this limit the hard scale of the process @xmath323 becomes very large . from eqs . ( [ mxg1 ] ) or ( [ mxg2 ] ) we see that the @xmath324 limit is realized when @xmath325 ( more precisely @xmath326 ) , so the invariant mass of the @xmath327 system is @xmath328 where @xmath329 consider first the limit where the gluon transverse momentum @xmath330 is small , such that @xmath331(b ) . in this limit the @xmath59 pair is very small , i.e. , we have strong ordering in impact parameter space , which can be written as @xmath332 . in color space the @xmath59 pair can be considered as a single gluon , and we consider `` gluonic dipole '' scattering off the target . this is consistent with our expression for the corresponding soft part ( [ m - soft - qqg ] ) , which in the limit @xmath333 reduces to @xmath334 , corresponding to the amplitude for soft gluon gluon scattering . this reproduces the conventional @xmath335 dipole result @xcite in the small @xmath336 limit , in which the amplitude of the gluonic dipole scattering differs by a factor of @xmath337 from the amplitude of the @xmath6 scattering . indeed , from our model it follows that in this limit @xmath338 as compared to eq . ( [ altwogl ] ) . however , this limiting case can not give a leading contribution to the diffractive structure function at large @xmath0 because of the smallness of the transverse momentum of the final state gluon @xmath339 . due to eq . ( [ mxg2 ] ) the larger gluon @xmath330 , the larger invariant mass @xmath0 is produced . at the same time , @xmath330 can not be significantly larger than the quark and antiquark transverse momenta @xmath340 . due to momentum conservation , the maximal @xmath0 at fixed @xmath341 occurs in the limit @xmath342 , which corresponds to @xmath343 in impact parameter space , leading to @xmath344 . from eq . ( [ m - soft - qqg ] ) , this corresponds to the situation when only the @xmath59 component of the @xmath7 system scatters off the target with soft part @xmath345 . this purely kinematical argument is compatible with an observation @xcite with respect to models for parton saturation @xcite , that the @xmath7 and @xmath6 dipole contributions should saturate to the same value , i.e. @xmath346 at large invariant masses @xmath0 . in particular , this means that the scattering of the @xmath7 system off the proton can not be reduced to the scattering of the @xmath335 dipole . we argued above that the leading @xmath7 contribution to the diffractive structure function in the large @xmath0 limit comes from on - shell gluon emission from the hard gluon as in fig . [ fig : qqg](a ) . it is clear from eq . ( [ mxg3 ] ) that the relevant limit @xmath347 corresponds to essentially on - shell gluon emission with @xmath325 . the corresponding gluon propagator can be only slightly off - shell to give a leading contribution to the cross section . in this case the @xmath6 pair takes most of the longitudinal momentum of the @xmath30 system , and kinematically there is no symmetry with respect to interchange @xmath348 in such a @xmath349 system , whereas for a @xmath6 dipole this symmetry @xmath241 holds explicitly . if one allows the active gluon to couple to the @xmath6 pair directly , the final state gluon connected to the hard quark propagator can not be on - shell , and we get an extra suppression of the cross section . such a `` symmetry breaking '' in the @xmath7 system does not allow us to reduce it to a symmetric gluonic @xmath335 dipole and consider its soft scattering in the same way as @xmath6 scattering . illustration of the @xmath7 contribution to diffractive dis.,scaledwidth=35.0% ] the corresponding physical situation is illustrated in fig . [ fig : qqgsat ] . the hard virtual photon first fluctuates into a virtual @xmath6 pair , and the leading configuration is when one quark ( antiquark ) takes most of the photon virtuality whereas the other one is almost on - shell . then the most virtual quark ( antiquark ) emits a ( less virtual ) gluon , which interacts with a slightly virtual sea gluon from the proton background field . this last interaction produces an essentially on - shell final state gluon , which contributes to the final @xmath30 system . after the first hard gluon exchange both quarks have similar and small virtualities and scatter off the proton background field . in order to calculate the @xmath7 contribution to the diffractive structure function we include a dglap splitting of the hard gluon ( with longitudinal momentum fraction @xmath40 ) into two gluons one carries momentum fraction @xmath350 and couples to the hard part , and one is on - shell and contributes to the final state in @xmath65 c.m.s . frame as shown in fig . [ fig : qqgsat ] . the diffractive structure function corresponding to the @xmath7 contribution can be then written as ( see e.g. @xcite ) @xmath351 where the integral is regulated in the infrared by the effective gluon mass @xmath352 in the gluon propagator . the factor @xmath353 is due to averaging over the color indices ( in the large @xmath58 limit ) of the extra gluon contributing to the color singlet @xmath30 , and @xmath354 is the gluon gluon splitting function @xmath355\end{aligned}\ ] ] since the @xmath6 contribution is dominated by transverse photon polarization , in our formulation the same is true of the @xmath41 contribution . one could also include more gluons in the final state by applying dglap evolution of the gluon density , and partially populate the rapidity gap by extra hadronic activity from the hadronization of gluons emitted from the hard gluon in the same way as in monte carlo simulations . this would lead to a model describing a smooth transition between diffractive and non - diffractive final states . let us finally comment on another approach to resumming multi - gluon exchange , which results in similar eikonal factors @xmath356 $ ] in the amplitudes . this approach was developed by hautmann , kunszt , and soper @xcite ( hks ) and by hautmann and soper @xcite ( hs ) , and is applicable in both inclusive and diffractive dis . this approach is similar to ours , employing factorization and resummation of soft @xmath194-channel gluons . in diffractive dis , the incoming partonic dipole is assumed to move closely together in the transverse plane before interacting with the color field of the proton . in the hks / hs approach all exchanged gluons are treated on the same footing . these gluons collectively carry color singlet charge and are resummed using a wilson line . in our approach , we use conventional @xmath357-factorization in terms of the unintegrated gluon distribution , and we additionally factorize the `` hard '' gluon , which carries most of the momentum , from the rest of the exchanged gluons , which are much softer ( @xmath57 ) and are resummed to all orders . the resummed gluons collectively carry color octet charge , which combined with the first gluon is required to form an overall color singlet exchange . it would be interesting to examine the connections between the two approaches further . the hera data @xcite on ddis are given in the form of the reduced cross section @xmath358 expressed in terms of the diffractive structure functions @xmath359 . the momentum transfer @xmath194 is integrated over since in most of the data the leading proton is not observed , and diffraction is equivalently defined through a large rapidity gap . the kinematical variable @xmath360 , where @xmath361 gev is the center - of - mass energy of @xmath362-collisions in hera . in fig . [ fig : data ] we compare the latest zeus data @xcite with the numerical evaluation of our model . a generally very good agreement is found , but this needs to be discussed in detail in order to gain understanding of the dynamics involved . as discussed in sec . [ sec - gpd ] , we need the generalized gluon distribution function in the proton , and use the prescription in eq . ( [ ugdfsat ] ) for the ugdf . this reduces the problem to an input of a standard parametrization of the gluon density in the proton , i.e. @xmath363 . here we mainly use the recent cteq6l1 parametrization @xcite , which is in leading order and thereby consistent with our treatment . below we also consider other parametrizations to illustrate the uncertainty at very small @xmath364 and factorization scales @xmath138 . the minimum factorization scale @xmath138 is fixed to be @xmath365 giving rise to a minimum possible fraction of the quark longitudinal momentum @xmath366 in the phase space integral . the physical parameters that are fixed , are the `` soft '' coupling @xmath367 , obtained from infrared - finite analytic perturbation theory ( see sec . [ softpart ] ) and used for the coupling of the soft screening gluons , and the gluon mass @xmath368 adopted as the infrared regulator in the gluon propagator in eq . ( [ fdqqg ] ) . fixing @xmath226 and @xmath369 , the only free parameters in our model are @xmath83 and @xmath370 , representing different soft effects that can not be calculated or safely estimated . the constituent quark mass @xmath83 , which enters the kinematics in sec . iii , accounts for the soft gluon radiation from the @xmath6 dipole and corresponds to forming dressed quarks before hadronization . the soft part @xmath370 of the off - diagonal ugdf in eq . ( [ ugdfsat ] ) can be identified with the square root of the `` soft '' collinear pdf defined at some @xmath57 and @xmath145 which represents the soft scale of the color screening gluons . the sensitivity to these parameters is discussed in the following . the shapes of the model curves are in quite good agreement with the data , except for a few points at extremely small @xmath371 , and small scales @xmath372 and @xmath0 ( see the upper right corner of fig . [ fig : data ] ) . here , we are in the kinematical domain where the uncertainties in the parametrizations of the gluon density of the proton become extremely large , as illustrated in fig . [ fig : pdfs ] . as can be seen , for @xmath373 there are substantial differences between the different gluon parametrizations and the differences become huge for @xmath374 and @xmath375 gev@xmath376 . the reason is that there is no data from inclusive dis or other processes that can measure the gluon density directly in this domain . this gluon pdf uncertainty strongly affects the calculated diffractive structure function at small quark fractions @xmath32 and/or small @xmath1 and @xmath0 , where @xmath138 may drop below @xmath377 gev . in this case @xmath164 , due to eq . ( [ xpr ] ) , so our basic assumptions and qcd factorization itself become less reliable . in principle , the diffractive dis data can be utilized for selecting the best gluon parametrization among those available in the literature or , even better , for making new gluon parametrizations including the data on ddis , which depend directly on the gluon density . @xmath1-dependence of the normalization parameter @xmath378 , see eq . ( [ ugdfsat ] ) , extracted from the hera data , with solid curve obtained using cteq6l1 pdf @xcite and dashed curve using grv94ho pdf @xcite for the gluon density in the proton.,scaledwidth=40.0% ] another signature of such uncertainties is the @xmath1 behavior of the soft part of the ugdf , i.e. @xmath379 , which is shown in fig . [ fig : rg ] for different pdfs . as can be seen , using either the typical leading order cteq6l1 pdf @xcite , which decreases at small @xmath40 and @xmath138 , or the more regular but older grv94ho pdf @xcite ( see comparison in fig . [ fig : pdfs ] ) to perform the fit of our model to zeus data results in quite different fitted @xmath379 . at higher scales , @xmath380 , the soft factor is quite stable at @xmath381 . however , at @xmath372 , the diffractive cross section calculated with cteq6l1 is underestimated by almost an order of magnitude , and in order to get the correct normalization the fitted @xmath370 value grows significantly . this is mostly because of the strong suppression in cteq6l1 at small @xmath382 at scales @xmath383 . in contrast , the fit with grv94ho , which does not decrease at small @xmath40 , leads to a more stable behavior at low @xmath1 , such that @xmath370 essentially becomes an overall normalization constant close to unity . in order to illustrate how uncertainties in the pdfs and in the ugdf prescriptions affect the @xmath40-dependence in comparison with data , we compare our model to the data using both cteq6l1 and grv94ho in the ugdf prescription of eq . ( [ ugdfsat ] ) . this is our `` normal '' prescription that gives a linear dependence of the cross section on the gluon density . for comparison we also use cteq6l1 with the `` old @xmath162''-prescription defined in eq . ( [ ugdfrg ] ) , which makes the cross section depend on the square of the gluon density . [ fig : pdfvsdata ] shows the results in the bins of interest with small scales @xmath0 and @xmath1 . the `` old @xmath162''-prescription leads to an order of magnitude too small diffractive structure function at all @xmath1 , which can not be explained by the expected normalization factor @xmath162 of order unity in eq . ( [ rg ] ) . the corresponding curves in fig . [ fig : pdfvsdata ] have therefore been normalized in order to compare the @xmath40-slopes . these slopes are in reasonable agreement with the data , but there is a tendency for a too large curvature generated by the squared gluon density , in particular at large @xmath1 . for the curves with linear gluon density , the curves with grv94ho lead to better slopes at the smallest @xmath1 and @xmath0 than those with cteq6l1 , but at higher scales they become too steep and can not describe data . this is not surprising since the old grv parametrization from 1994 does not take into account later data from hera and elsewhere , but it provides an interesting alternative due to its more regular behavior at very small @xmath364 at low scales . the curves fitted with the recent cteq6l1 parametrization have better @xmath40-slopes at higher scales and this is therefore the main alternative in fig . [ fig : data ] , in spite of its shortcoming at the very lowest @xmath40 points . the remaining free parameter to discuss is @xmath83 , the effective mass of the quark and antiquark in the @xmath30-system which is used in kinematical relations . in fig . [ fig : meff ] we show fitted values of @xmath83 at different scales @xmath0 and @xmath1 . the diffractive cross section itself is not very sensitive to @xmath83 , which therefore only varies within the physically reasonable interval @xmath384 gev . thus , it is mostly of nonperturbative nature and can be interpreted as a constituent quark mass . nevertheless , @xmath83 depends on both @xmath1 and @xmath0 , indicating that both scales contribute to generating softer gluon radiation . this dependence is , however , non - trivial . indeed , a larger invariant mass @xmath0 provides a larger phase space , which may accumulate more soft collinear gluons , leading to a larger effective quark mass . on the other hand , a harder photon ( large @xmath1 ) can probe a quark at smaller distances , so the @xmath1-dependence of the quark mass obeys renormalization group evolution , i.e.@xmath83 should decrease at larger @xmath1 . these two effects are indeed observed in the description of data ( see fig . ( [ fig : meff ] ) for @xmath385 gev . at larger @xmath0 the situation changes somewhat due to more hard gluon radiation contributing to @xmath0 ( in particular , the @xmath7 contribution becomes important ) . we now investigate the role of the different contributions to @xmath386 in eq . ( [ eq - sigma ] ) , i.e. @xmath6 from longitudinally and transversely polarized photons and the @xmath7 contribution . in our results shown in figs . [ fig : data ] and [ fig : pdfvsdata ] above , they are all included . we find , however , that the leading order @xmath6-dipole contribution dominates in all bins of @xmath0 and @xmath1 and is enough to describe all data for @xmath387 , below which the @xmath7 contribution becomes significant and can be approximated with its leading part calculated via dglap splitting of the first , hard gluon . [ fig : bspecr ] shows the @xmath26-spectra for the different contributions . they all vanish in the limits of @xmath388 , corresponding to large @xmath0 , and @xmath21 , corresponding to small or vanishing @xmath0 with production of resonances , which is not taken into account here , or no available phase space . the transverse @xmath6 contribution dominates over the other contributions . the longitudinal @xmath6 contribution is always small , although it becomes slightly larger at smaller @xmath1 scales . the gluonic @xmath7 contribution becomes relatively larger both at high @xmath1 and small @xmath26 , where it gives an important contribution that must be taken into account . we have in this paper developed a proper qcd framework for diffractive hard scattering , which contains both hard and soft dynamics . the hard part produces a well - defined state of emerging partons , and the soft part is the rescattering of these partons with the color field of the proton remnant . we have demonstrated that , by taking the fourier transform from momentum space to impact parameter space , the overall amplitude can be factorized into separate amplitudes for these hard and soft parts . this provides a substantial simplification for the calculation and is consistent with the physical insight that soft , long distance processes can not affect the hard process occurring on a short distance scale . the hard part is calculated using perturbative qcd , in the same way as for inclusive dis . a perturbative hard scale is provided by the photon virtuality @xmath1 and invariant mass @xmath0 of the diffractive system , and the process thus occurs at a space - time scale much smaller than the proton size . for small @xmath364 , the hard subprocess @xmath389 dominates . this process is mediated by a single gluon exchange taking most of the longitudinal momentum transfer , and leaves a proton remnant consisting of the three valence quarks in a color octet state . the proton remnant carries most of the beam momentum , and is therefore well separated in rapidity from the @xmath37 system . the soft part of the amplitude accounts for the rescattering of the @xmath37 pair ( in a color octet state ) with this remnant . this rescattering is dominated by multiple exchanges of soft gluons , which have larger couplings and less propagator suppression . the result is a negligible change of the momenta of the emerging partons , but an important change of phase is picked up this is the essence of the eikonal approximation . we find that summing over an arbitrary number of exchanged gluons leads to exponentiation of the soft amplitude , which can be written in a closed analytic form free of infrared divergences . the color exchange , treated in the large-@xmath58 approximation , leads to an overall color singlet exchange between the @xmath6 dipole and the proton . these two color singlet systems then hadronize independently separated by a gap in rapidity as characteristic signature of diffractive scattering . by invoking physical considerations based on the uncertainty principle , which limits the possible resolution of small momentum transfers , we obtain simplifications of the otherwise complicated angular relations in the impact parameter space . in essence , the orientation of the @xmath37-dipole relative to the proton color field is physically irrelevant and can be averaged out . in addition to the leading order contribution from the @xmath37-dipole , we have also included the next - to - leading contribution @xmath390 with an extra gluon in the final state . here , we find that the most important contribution is emission of this gluon from the exchanged hard gluon ( in @xmath389 ) , which can be well approximated by leading logarithmic dglap emission . numerical evaluation of the analytical results gives good agreement with the precise hera data on the diffractive deep inelastic cross section . the @xmath37 contribution is indeed dominant , but at @xmath391 , the @xmath390 contribution is important . at very small @xmath382 and scales @xmath392 gev@xmath376 the gluon density in the proton , which is used as input in our calculation , is very poorly known and gives a complication in the comparison with the few hera data points in this extreme region . standard up - to - date parametrizations have a too low gluon density in this @xmath393 region , whereas , e.g. , the old grv94 gluon density does better . since the diffractive cross section depends directly on the gluon density , and not only indirectly via dglap evolution as for inclusive dis , one here obtains an interesting possibility to constrain the gluon density at very small @xmath364 . having demonstrated that our theoretical formalism for ddis does describe hera data , one may then extract the part describing the multigluon exchange process and apply it to other hard scattering processes . this soft rescattering description ought to be universal , due to the factorization of the hard and soft amplitudes . thus , one may apply it together with hard processes in @xmath3 collisions at the tevatron to describe the different hard diffractive processed observed there , and then go to the higher energies at the lhc . one may also apply it to more detailed observables in diffractive dis , such as diffractive dijets or diffractive vector meson production . however , not only diffractive processes are of interest . the soft color interaction ( sci ) model discussed above has previously been successfully applied to both charmonium production and @xmath394-meson decays @xcite , and one may expect the model presented in this paper to have interesting applications in such processes too . moreover , the multi - gluon exchange mechanism will also affect the underlying event , since it effectuates color exchanges that modify the color - string topology and thereby the hadronic final state after hadronization . the underlying event is important in its own right to understand non - perturbative qcd dynamics , and also for understanding of inclusive events when subtracting the standard model background in searches for new phenomena at lhc . finally we note that in deriving the theoretical formalism presented here , we have _ not _ used any assumptions or results from the previous soft color interaction ( sci ) model . our new formalism stands on its own , based on qcd theory and basic physical arguments . the formalism can , however , explain why the simple sci model has been so successful in describing data on diffractive hard scattering and other phenomena . the assumptions of the sci model as well as its major features are essentially what comes out as results of the present paper . of course , our new formalism has a richer dynamical structure and we will therefore attempt to improve the monte carlo implementation of the sci model by replacing its fixed probability for soft gluon exchanges with a mechanism based on the above amplitude for the multiple soft gluon exchanges . this will introduce a non - trivial dependence on the kinematical variables , giving a new level of event - to - event variations . as usual with full event simulation using monte carlo , this will give access to more detailed studies of both the employed theoretical model and its comparison to data in terms of the indicated more elaborate observables . a. v. efremov and a. v. radyushkin , phys . b * 94 * , 245 ( 1980 ) ; riv . nuovo cim . * 3n2 * , 1 ( 1980 ) ; j. c. collins and d. e. soper , nucl . b * 194 * , 445 ( 1982 ) ; j. c. collins , d. e. soper and g. sterman , nucl . phys . b * 261 * , 104 ( 1985 ) ; nucl . b * 308 * , 833 ( 1988 ) ; and phys . b * 438 * , 184 ( 1998 ) ; g. t. bodwin , phys . d * 31 * , 2616 ( 1985 ) [ erratum - ibid . d * 34 * , 3932 ( 1986 ) ] ; j. c. collins , d. e. soper and g. sterman , adv . high energy phys . * 5 * , 1 ( 1988 ) . s. chekanov _ et al . _ [ zeus collaboration ] , nucl . b * 816 * , 1 ( 2009 ) . m. derrick _ et al . _ , phys . b * 315 * , 481 ( 1993 ) ; phys . b * 346 * , 399 ( 1995 ) ; t. ahmed _ et al . _ , nucl . b * 429 * , 477 ( 1994 ) ; nucl . b * 435 * , 3 ( 1995 ) . r. bonino _ et al . _ , phys . b * 211 * , 239 ( 1988 ) . g. ingelman and p. e. schlein , phys . b * 152 * , 256 ( 1985 ) . a. edin , g. ingelman and j. rathsman , phys . b * 366 * , 371 ( 1996 ) ; z. phys . c * 75 * , 57 ( 1997 ) . g. ingelman , a. edin and j. rathsman , comput . commun . * 101 * , 108 ( 1997 ) . a. edin , g. ingelman and j. rathsman , arxiv : hep - ph/9912539 , in proc . ` monte carlo generators for hera physics ' , desy - proc-1999 - 02 p. 280 ; r. enberg , g. ingelman and n. tmneanu , phys . d * 64 * , 114015 ( 2001 ) ; a. edin , g. ingelman and j. rathsman , phys . rev . d * 56 * , 7317 ( 1997 ) ; d. eriksson , g. ingelman and j. rathsman , phys . d * 79 * , 014011 ( 2009 ) . s. j. brodsky , r. enberg , p. hoyer and g. ingelman , phys . d * 71 * , 074020 ( 2005 ) . r. pasechnik , r. enberg and g. ingelman , arxiv:1004.2912 [ hep - ph ] . a. h. mueller , nucl . b * 335 * , 115 ( 1990 ) ; + n. n. nikolaev and b. g. zakharov , z. phys . c * 49 * , 607 ( 1991 ) . a. bialas and r. b. peschanski , phys . b * 378 * , 302 ( 1996 ) ; a. bialas and r. b. peschanski , phys . b * 387 * , 405 ( 1996 ) ; a. bialas , r. b. peschanski and c. royon , phys . d * 57 * , 6899 ( 1998 ) ; s. munier , r. b. peschanski and c. royon , nucl . b * 534 * , 297 ( 1998 ) . m. wusthoff , phys . d * 56 * , 4311 ( 1997 ) . v. n. gribov and l. n. lipatov , sov . j. nucl . * 15 * , 438 ( 1972 ) ; g. altarelli and g. parisi , nucl . b * 126 * , 298 ( 1977 ) ; yu . l. dokshitzer , sov . jetp * 46 * , 641 ( 1977 ) . s. catani , m. ciafaloni and f. hautmann , phys . b * 242 * , 97 ( 1990 ) ; + s. catani , m. ciafaloni and f. hautmann , nucl . b * 366 * , 135 ( 1991 ) ; + j. c. collins and r. k. ellis , nucl . b * 360 * , 3 ( 1991 ) ; + l. v. gribov , e. m. levin and m. g. ryskin , phys . rept . * 100 * , 1 ( 1983 ) . r. s. pasechnik , a. szczurek and o. v. teryaev , phys . d * 81 * , 034024 ( 2010 ) . b. pire , j. soffer and o. teryaev , eur . j. c * 8 * , 103 ( 1999 ) . x. artru , m. elchikh , j. m. richard , j. soffer and o. v. teryaev , phys . * 470 * , 1 ( 2009 ) . m. gluck , e. reya and a. vogt , z. phys . c * 67 * , 433 ( 1995 ) . k. j. golec - biernat and m. wsthoff , phys . d * 59 * , 014017 ( 1999 ) ; k. j. golec - biernat and m. wsthoff , phys . rev . d * 60 * , 114023 ( 1999 ) ; j. bartels , k. j. golec - biernat and h. kowalski , phys . rev . d * 66 * , 014001 ( 2002 ) ; e. iancu , k. itakura and s. munier , phys . b * 590 * , 199 ( 2004 ) . f. hautmann , z. kunszt and d. e. soper , phys . lett . * 81 * , 3333 ( 1998 ) ; nucl . b * 563 * , 153 ( 1999 ) . f. hautmann and d. e. soper , phys . d * 63 * , 011501 ( 2001 ) ; phys . d * 75 * , 074020 ( 2007 ) . a. edin , g. ingelman and j. rathsman , phys . d * 56 * , 7317 ( 1997 ) ; c. brenner mariotto , m. b. gay ducati and g. ingelman , eur . phys . j. c * 23 * , 527 ( 2002 ) ; d. eriksson , g. ingelman and j. rathsman , phys . d * 79 * , 014011 ( 2009 ) .

_ monadic second - order logic _ ( @xmath6 ) is one of the most expressive logics for which the theories of many interesting classes of structures are still decidable . in particular , the infinite binary tree and many linear orders have a decidable @xmath6-theory @xcite and the same holds for many classes of ( finite or infinite ) structures with bounded _ tree - width _ @xcite . furthermore , for every fixed @xmath6-sentence @xmath7 and every class @xmath8 of finite structures with bounded tree - width , there is a linear - time algorithm that checks whether a given structure from @xmath8 satisfies @xmath7 @xcite . examples of monadic second - order expressible graph properties are @xmath9-colourability , various types of connectivity , and planarity ( via kuratowski s well - known characterisation by forbidden configurations ) . a variant of monadic second - order logic called _ guarded second - order logic _ ( @xmath10 ) allows quantification not only over sets of elements but also over sets of edges ( i.e. , tuples from the relations ) @xcite . the above mentioned linear - time algorithms can be adapted to this logic . there are tight links between guarded second - order logic and tree - width.08em : every class of ( finite or infinite ) relational structures with a decidable @xmath10-theory has bounded tree - width . this gives a sort of converse to the above mentioned decidability results @xcite . the proof of this result uses a deep theorem of graph minor theory by robertson and seymour.08em : a set of graphs has bounded tree - width if and only if it excludes some planar graph as a minor @xcite . to compare the @xmath6-theories or @xmath10-theories of two classes of structures we can use _ monadic second - order transductions , _ a certain kind of interpretations suitable both , for monadic second - order logic and , using a detour via incidence structures , also for guarded second - order logic @xcite . in the present article we classify classes of finite structures according to their ` combinatorial complexity ' . ( note that we do not consider decidability issues . ) we will consider two ways to measure the complexity of such classes . on the one hand , we can use their tree - width and its variants . on the other hand , we can compare them via transductions . as it turns out , these two approaches are equivalent and they give rise to the same hierarchy . this indicates the robustness of our definitions and their intrinsic interest . other possible hierarchies , based on different logics , will be briefly mentioned in section [ sect : conclusion ] . let us give more details . an @xmath6-transduction is a transformation of relational structures specified by monadic second - order formulae . as graphs can be represented by relational structures , we can use @xmath6-transductions as transformations between graphs . an @xmath6-transduction is a generalisation of the following kind of operations ( see definition [ def : transduction]).08em : \(i ) the definition of a relational structure `` inside '' another one ( in model theory this is called an _ interpretation_).08em ; \(ii ) the replacement of a structure @xmath11 by the union of a fixed number of disjoint copies of @xmath11 , augmented with appropriate relations between the copies.08em ; \(iii ) the expansion of a given structure @xmath11 by a fixed number of unary predicates , called _ parameters . _ usually , these predicates are arbitrary subsets of the domain , but we also may have a formula imposing restrictions on them . because of the possibility to use parameters , a transduction @xmath1 is a many - valued map in general . ( we may also think of it as non - deterministic . ) each relational structure @xmath11 has several images @xmath12 depending on the choice of the parameters @xmath13 . if @xmath14 we can consider the tuple @xmath15 as an _ encoding _ of @xmath16 in @xmath11 . the transduction @xmath1 is the corresponding _ decoding function . _ each transduction @xmath1 extends in a canonical way to a transformation between classes of structures . if @xmath8 and @xmath17 are classes of relational structures with @xmath18 , we can think of @xmath1 as a way of encoding the structures in @xmath8 by elements of @xmath17 . for instance , every finite graph can be encoded in a sufficiently large finite square grid ( by a fixed transduction @xmath1 ) . every finite tree of height at most @xmath4 ( for fixed @xmath4 ) can be encoded in a sufficiently long finite path . but it is not the case that all finite trees can be encoded by paths ( by a single transduction ) . the purpose of this article is to classify classes of finite relational structures according to their _ encoding powers . _ we will compare classes @xmath8 and @xmath17 of structures by the following preorder.08em : @xmath19 we attack the problem of determining the structure of this preorder . since , at the moment , a complete description of this hierarchy seems to be out of reach , we concentrate on a variant where we replace monadic second - order logic by _ guarded second - order logic . _ in this case , the corresponding hierarchy can be described completely . to obtain a corresponding notion of transduction we can not simply change the definition of an @xmath6-transduction to use @xmath10-formulae since the resulting notion of transduction would not yield a reduction between @xmath10-theories , and it even would not be closed under composition . instead , we will take a detour by combining ordinary @xmath6-transductions with a well - known translation between @xmath10 and @xmath6 . this translation is based on _ incidence structures . _ let us first describe this notion for undirected graphs where it is very natural . there are two canonical ways to encode a graph @xmath20 by a relational structure . we can use its _ adjacency representation _ which is a structure @xmath21 where the domain @xmath22 consists of all vertices of @xmath20 and @xmath23 is a binary relation containing all pairs of adjacent vertices . but we also can use the _ incidence representation _ of @xmath20 . this is the structure @xmath24 where the domain @xmath25 contains both , the vertices and the edges of @xmath20 , and @xmath26 is the incidence relation between vertices and edges . in a similar way , we can associate with every relational structure @xmath11 its incidence structure @xmath27 ( see definition [ def : incidence structure ] ) where the domain also contains elements for all tuples in some relation of @xmath11 . it is shown in @xcite that every @xmath10-formula @xmath7 talking about some structure @xmath11 can be translated into an @xmath6-formula talking about the incidence structure @xmath27 , and vice versa . hence , we can use incidence structures to obtain an analogue @xmath28 of the preorder @xmath29 suitable for guarded second - order logic . we set @xmath30 where @xmath31 . the main result of the present article is a complete characterisation of the resulting hierarchy for classes of finite structures . we show that the preorder @xmath28 is linear of order type @xmath32 . it turns out that every class of finite structures is equivalent to one of the following classes , listed in increasing order of generality.08em : * trees of height at most @xmath4 , for each @xmath33.08em ; * paths.08em ; * arbitrary trees ( equivalently , binary trees).08em ; * ( square ) grids . each of these levels can be characterised in terms of tree decompositions . hence , we also obtain a corresponding hierarchy of complexity measures on structures that are compatible with @xmath6-transductions transforming incidence structures . the upper levels of the hierarchy can be determined easily using techniques from graph minor theory developed by robertson and seymour , such as the notions of a minor and a tree decomposition . in particular , we employ two results characterising bounded tree - width and bounded path - width in terms of excluded minors @xcite . for the lower levels , which consist of classes of bounded path - width , the characterisation is more complicated and requires new results relating tree decompositions and monadic second - order logic . in sections [ sect : preliminaries ] and [ sect : mso ] we give basic definitions . section [ sect : tree decompositions ] collects some known results from graph minor theory . we also introduce a new variant of tree - width and prove some of its basic properties . in section [ sect : tree decompositions and transductions ] we expound the connections between tree - width and monadic second - order transductions . in section [ sect : hierarchy ] we introduce the transduction hierarchy and we state our main theorem . its proof is contained in sections [ sect : hierarchy2 ] and [ sect : hierarchy3 ] . in the first one , we prove that the hierarchy is strict while , in the second one , we show that it covers every class . the final section [ sect : conclusion ] contains some extension of our results to other logics and some open problems in this direction . let us fix our notation . we set @xmath34 : = \{0,\dots , n-1\}$ ] and we write @xmath35 for the power set of a set @xmath36 . we denote tuples @xmath37 with a bar . the components of @xmath37 will be @xmath38 where the length @xmath4 will usually be implicit . we sometimes identify a tuple @xmath37 with the set of its components . for instance , we write @xmath39 to express that @xmath40 , for some @xmath41 . in this article all graphs , trees , and relational structures are finite . we will not repeat this finiteness assumption . a relational structure @xmath11 is of the form @xmath42 with domain @xmath43 and relations @xmath44 . the _ signature _ of such a structure is the set @xmath45 of relation symbols . in some proofs we will also use signatures with constant symbols denoting elements of the domain . we write @xmath46 for the arity of a relation @xmath47 . for a signature @xmath48 , we denote by @xmath49 $ ] the class of all @xmath48-structures . we write @xmath50 for the _ disjoint union _ of the structures @xmath11 and @xmath16 . we mainly consider _ incidence structures . _ these are representations of structures @xmath11 where we have added new elements to the domain , one for each tuple in the relations of @xmath11 . [ def : incidence structure ] let @xmath51 be a structure and let @xmath52 be the maximal arity of a relation @xmath53 . the _ incidence structure _ of @xmath11 is the structure @xmath54 where we extend the domain @xmath43 by @xmath55 and the relations are @xmath56 the class of all incidence structures is @xmath57 : = \set { \fraka_{\mathrm{in } } } { \fraka \in { \mathbb{str}}[\sigma ] } $ ] . note that incidence structures are binary ( i.e. , their relations have arity at most @xmath58 ) . hence , they can be regarded as bipartite labelled directed graphs . one important property of incidence structures is the fact that they are _ sparse , _ i.e. , their relations contain few tuples . let @xmath59 . a structure @xmath60 is _ _ @xmath9-sparse__-sparse hypergraphs _ and _ uniformly @xmath9-sparse hypergraphs . _ what we call @xmath9-sparse is a slight modification of the uniform version of @xcite . ] if , for every subset @xmath61 and all relations @xmath53 , we have @xmath62 every incidence structure is @xmath63-sparse . let us fix our notation regarding trees and graphs . a _ directed graph _ is a pair @xmath64 where @xmath22 is the set of _ vertices _ and @xmath65 is the _ edge relation . thus , graphs are by definition simple ( without parallel edges ) . an _ undirected graph _ is a graph where the edge relation @xmath66 is symmetric . when speaking of a _ graph _ we will always mean an undirected one . we regard a _ coloured graph _ as a relational structure @xmath67 with binary relations @xmath68 and unary relations @xmath69 that encode the colours of , respectively , the edges and the vertices . we allow graphs whose edges and vertices have several colours . trees play a major role in this article . intuitively , a tree is a directed graph @xmath70 with a unique vertex @xmath52 of indegree @xmath71 , called the _ root _ of @xmath70 , such that every vertex is reachable from @xmath52 by a unique directed path . the actual definition we will use is slightly more concrete . it is based on the usual encoding of the vertices of a tree by finite sequences describing the path from the root to the given vertex . in fact , we introduce two notions of a tree.08em : _ order - trees _ and _ successor - trees . _ the latter use the usual edge relation , while the former are equipped with the tree - order instead . let @xmath72 be a set . \(a ) for an ordinal @xmath73 , we denote by @xmath74 the set of all sequences of elements of @xmath72 of length less than @xmath73 . the _ prefix relation _ on @xmath75 is defined by @xmath76 the infimum of @xmath77 and @xmath78 with respect to @xmath79 , i.e. , their longest common prefix , is denoted by @xmath80 . \(b ) a finite prefix closed subset @xmath81 is called a _ tree domain . _ following our intuition that a vertex is represented by the path leading to it , we call the empty sequence @xmath82 the _ root _ of @xmath83 and the maximal elements of @xmath83 its _ leaves . _ the domain of the _ complete @xmath84-ary tree of height @xmath4 _ is @xmath85 . hence , @xmath86 is the empty tree , @xmath87 the one consisting only of the root , and @xmath88 consists of a root and @xmath84 leaves . given a tree domain @xmath83 we can define the _ successor relation _ @xmath66 on @xmath83 by setting @xmath89 in this case we call @xmath78 a _ successor _ of @xmath77 and @xmath77 the _ predecessor _ of @xmath78 . a structure of the form @xmath90 ( and every structure isomorphic to it ) is called a _ successor - tree . _ sometimes it is convenient to replace the successor relation @xmath66 by the tree order @xmath79 . structures of the form @xmath91 are called _ order - trees . _ a _ coloured tree _ is the expansion of a ( order- or successor- ) tree by unary predicates @xmath92 . ( we do not require these predicates to be pairwise disjoint . hence , every vertex may have none , one , or several colours . ) we write @xmath93 for the class of all order - trees @xmath94 with @xmath84 colours . the set of _ leaves _ of a tree @xmath70 is denoted by @xmath95 . \(c ) let @xmath96 be an order - tree . the _ level _ of an element @xmath97 is the number of vertices @xmath98 with @xmath99 . we denote it by @xmath100 . the _ height _ of @xmath70 is the least ordinal @xmath73 greater than the level of every element of @xmath83 . hence , the empty tree has height @xmath71 and the tree with a single vertex has height @xmath63 . the _ out - degree _ of @xmath70 is the maximal number of successors of a vertex of @xmath70 . for successor - trees we define these notions analogously . \(d ) let @xmath70 be a tree and @xmath101 a vertex of @xmath70 . the _ subtree _ of @xmath70 _ rooted _ at @xmath101 is the subtree @xmath102 consisting of all vertices @xmath103 with @xmath104 , i.e. , all vertices below @xmath101 . sometimes it is possible to reduce statements about relational structures to statements about graphs . one way to do so consists in replacing a structure by its _ gaifman graph . _ the _ gaifman graph _ of a structure @xmath60 is the undirected graph @xmath105 with the same domain @xmath43 and with the edge relation @xmath106 monadic second - order logic ( @xmath6 ) is the extension of first - order logic by set variables and quantifiers over such variables . an important variant of @xmath6 is guarded second - order logic ( @xmath10 ) where one can quantify not only over sets of elements but also over sets of tuples from the relations ( see @xcite for details ) . hence , guarded second - order logic over a given structure @xmath11 is equivalent to monadic second - order logic over its incidence structure @xmath27 . + for every @xmath10-sentence @xmath7 , we can effectively construct an @xmath6-sentence @xmath107 such that @xmath108 for every @xmath6-sentence @xmath7 , we can effectively construct a @xmath10-sentence @xmath107 such that @xmath109 throughout the article we will consistently work with incidence structures , thereby avoiding the treatment of guarded second - order logic . in particular , all formulae are tacitly assumed to be @xmath6-formulae . besides @xmath6 and @xmath10 we also consider their _ counting extensions _ @xmath110 and @xmath111 . these add predicates of the form @xmath112 to , respectively , @xmath6 and @xmath10 , where @xmath36 is a set variable and @xmath113 are numbers . all of our results for @xmath10 go through also for @xmath111 , i.e. , for @xmath110-transductions between incidence structures . in section [ sect : conclusion ] we will give a partial characterisation of the hierarchy for @xmath110-transductions of graphs ( not of incidence graphs ) . in this case the availability of counting predicates does make a difference . to state the composition theorem below it is of advantage to work with a variant of @xmath6 without first - order variables . this variant has atomic formulae of the form @xmath114 and @xmath115 , for set variables @xmath116 , where a formula of the form @xmath115 states that there are elements @xmath117 such that the tuple @xmath37 is in @xmath47 . note that every general monadic second - order formula with first - order variables can be brought into this restricted form by replacing all first - order variables by set variables and adding the condition that these sets are singletons . whenever we speak of @xmath6 we will have this version in mind . in particular , the following definition of the rank of a formula is based on this variant . when writing down concrete formulae , on the other hand , we will allow the use of first - order variables to improve readability . we regard every such formula as an abbreviation of a formula of the restricted form . similarly , when we use structures with constants we actually regard each constant as a singleton set . \(a ) the _ rank _ @xmath118 of a formula @xmath7 is the nesting depth of quantifiers in @xmath7 . formulae of rank @xmath71 are called _ quantifier - free . _ \(b ) the _ monadic theory of rank @xmath84 _ of a structure @xmath11 is @xmath119 for a tuple @xmath37 of elements of @xmath11 , we also consider the monadic theory @xmath120 of the expansion @xmath121 . we use the term ` rank ' instead of the more natural ` quantifier rank ' since in section [ sect : conclusion ] below we will consider @xmath110 where the notion of rank has to be adapted for our results to go through . in order to compare the monadic theories of two classes of structures we employ @xmath6-transductions . to simplify the definition we introduce three simple operations and we obtain @xmath6-transductions as compositions of these . [ def : transduction ] ( a ) let @xmath122 be a natural number . the operation @xmath123 maps a structure @xmath11 to the expansion @xmath124 of the disjoint union of @xmath9 copies of @xmath11 by the following relations . denoting the copy of an element @xmath125 in the @xmath41-th component of @xmath126 by the pair @xmath127 , we define @xmath128 for @xmath129 , we set @xmath130 . \(b ) for @xmath131 , we define the operation @xmath132 that maps a structure @xmath11 to all possible expansions by @xmath84 unary predicates @xmath133 . note that this operation is many - valued and that @xmath134 is just the identity . basic @xmath6-transduction _ is a partial operation @xmath1 on relational structures described by a list @xmath135 of @xmath6-formulae called the _ definition scheme _ of @xmath1 . given a structure @xmath11 that satisfies the formula @xmath136 the operation @xmath1 produces the structure @xmath137 where @xmath138 if @xmath139 then @xmath140 remains undefined . \(d ) a _ @xmath9-copying @xmath6-transduction _ @xmath1 is a ( many - valued ) operation on relational structures of the form @xmath141 where @xmath142 is a basic @xmath6-transduction . when the value of @xmath9 does not matter , we will simply speak of a _ transduction . _ note that , due to @xmath132 , a structure can be mapped to several structures by a transduction . consequently , we consider @xmath140 as the _ set _ of possible values @xmath143 where @xmath92 ranges over all @xmath84-tuples of subsets of @xmath43 . for a class @xmath8 , we set @xmath144 \(a ) the expansion by @xmath84 unary predicates corresponds , in the terminology of @xcite , to using @xmath84 _ parameters . _ we will use this terminology , for instance , in the proof of theorem [ thm : interpretable implies tree decomposable ] . \(b ) note that every basic @xmath6-transduction is a @xmath63-copying @xmath6-transduction without parameters . [ exam : gf interpretable ] ( a ) let @xmath48 be a signature and let @xmath52 be the maximal arity of a relation in @xmath48 . the operation mapping an incidence structure @xmath145 $ ] to the structure @xmath146 is a @xmath9-copying @xmath6-transduction where @xmath147 , for @xmath148 , and @xmath129 , for @xmath149 . \(b ) for every fixed number @xmath33 , we describe a transduction @xmath1 transforming a path @xmath150 of length @xmath151 into the class of all trees of height @xmath4 with @xmath152 vertices . we can encode a tree @xmath83 of height @xmath4 with @xmath84 vertices as a finite word @xmath153 of length @xmath84 over the alphabet @xmath34 $ ] as follows . let @xmath154 be the enumeration of the vertices of @xmath83 in lexicographic order , and let @xmath155 be the level of @xmath156 . we encode @xmath83 by the word @xmath157 . a transduction can recover @xmath83 from @xmath153 as follows . each position in @xmath153 corresponds to a vertex . the predecessor of the @xmath41-th vertex @xmath101 is the maximal vertex to the left of @xmath101 whose label is less than @xmath155 . clearly , this predecessor relation is definable in monadic second - order logic . the two most important properties of @xmath6-transductions are summarised in the following lemmas . [ lem : transductions are comorphisms ] let @xmath1 be a transduction . for every @xmath6-sentence @xmath7 , there exists an @xmath6-sentence @xmath158 such that , for all structures @xmath11 , @xmath159 furthermore , if @xmath1 is quantifier - free , then the rank of @xmath158 is no larger than that of @xmath7 . [ cor : transductions operate on theories ] for every quantifier - free transduction @xmath1 and every @xmath131 , there exists a function @xmath160 on monadic theories of rank @xmath84 such that @xmath161 for all transductions @xmath162 there exists a transduction @xmath163 such that @xmath164 . as a further example note that we can use transductions to translate between order - trees and successor - trees . there exists a transduction @xmath1 mapping an order - tree to the corresponding successor - tree . there exists a transduction @xmath165 mapping a successor - tree to the corresponding order - tree . similarly there are transductions translating between a structure and its incidence structure . for every signature @xmath48 , there exists a transduction @xmath1 such that @xmath166 , for all @xmath167 $ ] . the converse statement is a much deeper result and requires the structure in question to be @xmath9-sparse for some fixed @xmath9 . for every signature @xmath48 and all numbers @xmath59 , there exists an @xmath6-transduction @xmath1 such that @xmath168 , for all @xmath9-sparse structures @xmath167 $ ] . we have seen in lemma [ lem : transductions are comorphisms ] that transductions relate the monadic theories of two structures . we also need techniques to relate the monadic theory of a structure to those of its substructures . the disjoint union operation can frequently be used for this purpose ( for a proof of the following theorem see , e.g. , theorem 7.11 of @xcite , or @xcite ) . [ thm : composition for unions ] let @xmath48 and @xmath169 be relational signatures with constants . for every @xmath131 , there exists a ( computable ) binary operation @xmath170 on monadic theories of rank @xmath84 such that @xmath171 for all @xmath48-structures @xmath11 and @xmath169-structures @xmath16 . below we will mainly make use of the following corollary . [ lem : composition for trees ] let @xmath70 be an order - tree and @xmath97 a vertex . suppose that @xmath172 is the order - tree obtained from @xmath70 by replacing the subtree @xmath102 by some tree @xmath173 . let @xmath174 be a tuple of vertices of @xmath70 with @xmath175 , for all @xmath41 . if @xmath37 are vertices of @xmath102 and @xmath176 are vertices of @xmath173 such that @xmath177 then it follows that @xmath178 let @xmath179 be the tree obtained from @xmath70 by replacing the subtree @xmath102 by a single vertex @xmath153 . we define the following auxiliary predicates.08em : @xmath180 we construct a quantifier - free transduction @xmath1 such that @xmath181 if @xmath160 is the function from corollary [ cor : transductions operate on theories ] and @xmath170 the operation from theorem [ thm : composition for unions ] , it follows that @xmath182 as desired . hence , it remains to define @xmath1 . let @xmath183 be the signature of @xmath184 , and @xmath185 the signature of @xmath186 and @xmath187 . for @xmath1 we can use the basic @xmath6-transduction consisting of the following formulae.08em : @xmath188 -- some properties of the transduction hierarchy , which we will introduce in section [ sect : hierarchy ] below , can be deduced from results about graph minors . \(a ) let @xmath189 be an undirected graph and @xmath190 a set of edges . we denote by @xmath191 the reflexive and transitive closure of @xmath192 . note that @xmath191 is an equivalence relation . the graph @xmath193 is obtained by contracting all edges in @xmath192 . formally , we have @xmath194 where @xmath195 is the set of equivalence classes and @xmath196 contains an edge between classes @xmath197 $ ] and @xmath198 $ ] if and only if @xmath197 \neq [ y]$ ] and there are representatives @xmath199 $ ] and @xmath200 $ ] with @xmath201 . \(b ) a _ minor _ of a graph @xmath20 is a graph that can be obtained from @xmath20 by first deleting some vertices and edges and then contracting some of the remaining edges . for a class @xmath8 of graphs , we denote by @xmath202 the class of all minors of graphs in @xmath8 . one central tool in graph minor theory is the notion of a tree decomposition and the related complexity measures called tree - width and path - width . these notions extend in a natural way to relational structures . let @xmath60 be a structure . \(a ) a _ tree decomposition _ of @xmath11 is a family @xmath203 of ( possibly empty ) subsets @xmath204 indexed by a rooted tree @xmath83 such that * for every element @xmath125 , the set @xmath205 is nonempty and connected in @xmath83.08em ; * for every tuple @xmath206 , there is some index @xmath97 with @xmath207 . we call the sets @xmath208 the _ components _ of the decomposition , and @xmath83 is its _ underlying tree . _ the _ height _ of a tree decomposition @xmath203 is the height of @xmath83 , while its _ width _ is the number @xmath209 \(b ) the _ tree - width _ @xmath210 of @xmath11 is the minimal width of a tree decomposition of @xmath11 . \(c ) the _ path - width _ @xmath211 of @xmath11 is the minimal width of a tree decomposition of @xmath11 where the underlying tree is a path . \(d ) the _ @xmath4-depth tree - width _ @xmath212 of @xmath11 is the minimal width of a tree decomposition of @xmath11 whose underlying tree has height at most @xmath4 . \(e ) for a class @xmath8 of structures , we define @xmath213 as the supremum of @xmath210 , for @xmath214 , and similarly for @xmath215 and @xmath216 . \(a ) the @xmath4-depth tree - width of a graph @xmath20 is related to its _ tree - depth _ @xmath217 as introduced by neetil and ossona de mendez @xcite . the tree - depth of a graph @xmath20 is the least number @xmath4 such that some orientation of @xmath20 is a subgraph of some order - tree of height @xmath4 . for a graph @xmath20 , it follows that * @xmath218 implies @xmath219.08em ; * @xmath220 implies @xmath221 . these facts are easy to establish . we will not need them in the following . \(b ) there are some simple relations between @xmath4-depth tree - width , path - width , and tree - width . for every graph @xmath20 , we have @xmath222 ( let us sketch the proof of the last inequality.08em : let @xmath223 be a tree decomposition of @xmath20 of height @xmath4 and width @xmath224 . as the components of a path decomposition of @xmath20 we take all sets of the form @xmath225 , where @xmath226 is a path from the root @xmath227 to some leaf @xmath228 of @xmath83 . ) the next lemma shows that most questions regarding tree decompositions of a structure can be reduced to the corresponding questions about its gaifman graph . for many of the following results it is therefore sufficient to consider graphs . let @xmath11 be a structure . a family @xmath223 is a tree decomposition of @xmath11 if and only if it is a tree decomposition of @xmath229 . @xmath230 is immediate . @xmath231 follows from the fact that every tree decomposition of a clique has one component @xmath208 containing the whole clique . this implies that , for every clique @xmath232 in @xmath229 induced by some tuple @xmath206 , there is some vertex @xmath97 with @xmath233 . hence , every tuple @xmath206 is contained in some component @xmath208 . in order to separate the higher classes of the hierarchy , we shall employ two deep results of robertson and seymour about excluded minors . [ thm : excluded tree theorem ] for each tree @xmath70 , there exists a number @xmath59 such that @xmath234 [ thm : excluded grid theorem ] for each planar graph @xmath235 , there exists a number @xmath59 such that @xmath236 a class of graphs has bounded path - width if and only if it excludes some tree as a minor . a class of graphs has bounded tree - width if and only if it excludes some planar graph as a minor . we also need a variant of these theorems for @xmath4-depth tree - width . the next lemma contains the main technical argument . suppose that @xmath20 is a graph that does not contain a path of length @xmath151 . then @xmath20 has a tree decomposition of height at most @xmath151 and width at most @xmath237 . let @xmath238 be a depth - first spanning order - tree of @xmath20 , i.e. , a spanning tree such that , for every edge @xmath239 of @xmath20 we have @xmath240 or @xmath104 ( for details see , e.g. , @xcite where such spanning trees are called _ normal _ ) . we define a tree decomposition @xmath223 of @xmath20 by setting @xmath241 since @xmath83 is depth - first , it follows that every edge @xmath239 of @xmath20 is contained in some component @xmath242 where @xmath153 is the maximum of @xmath103 and @xmath101 . the height of the tree @xmath83 can be at most @xmath151 since @xmath20 contains no path of length @xmath151 . furthermore , we have @xmath243 . hence , the width of the tree decomposition is at most @xmath237 . [ thm : excluded path theorem ] for each path @xmath150 , there exist numbers @xmath244 such that @xmath245 suppose that @xmath246 and let @xmath151 be the length of @xmath150 . then the preceding lemma implies that @xmath247 . a class of graphs has bounded @xmath4-depth tree - width , for some @xmath4 , if and only if it excludes some path as a minor ( equivalently , as a subgraph ) . a class of graphs has bounded tree - depth if and only if it excludes some path as a minor ( equivalently , as a subgraph ) . we can also compute a bound on the @xmath4-depth tree - width in terms of the @xmath248-depth tree - width . it will be needed in the proof of theorem [ thm : covering for lower part ] below . we say that the tree @xmath249 can be _ embedded _ into a tree @xmath91 if there exists an order - preserving injective mapping @xmath250 , i.e. , if @xmath251 , regarded as relational structure , is isomorphic to an induced substructure of @xmath238 . for instance , we have an embedding as indicated by the labels . if @xmath252 can be embedded in @xmath83 then @xmath252 is isomorphic to a minor of @xmath83 , when we consider @xmath252 and @xmath83 as graphs . let @xmath203 be a tree decomposition and let @xmath253 be a set of edges of ( the successor - tree corresponding to ) @xmath83 . the tree decomposition @xmath254 obtained by _ contracting _ the edges in @xmath253 is @xmath255})_{[v ] \in t / f}\,,\end{aligned}\ ] ] where @xmath256 } : = \bigcup_{u \in [ v ] } u_u$ ] . [ lem : bound on twd - n ] let @xmath20 be a graph and let @xmath257 be a tree decomposition of @xmath20 of width @xmath9 and height at most @xmath258 . if @xmath131 is some number such that the tree @xmath259 can not be embedded into @xmath83 , then @xmath260 . we construct a tree decomposition @xmath261 of height at most @xmath4 and width at most @xmath262 as follows . let @xmath263 be the minimal ( w.r.t . @xmath264 ) set of vertices that contains * every leaf of @xmath83 at level @xmath4 and * every vertex that has at least @xmath84 successors in @xmath265 . since @xmath259 can not be embedded into @xmath83 it follows that @xmath265 does not contain the root of @xmath83 . let @xmath253 be the set of all edges of @xmath83 linking a vertex in @xmath266 to a vertex in @xmath265 . by definition of @xmath265 it follows that ( i ) every vertex of @xmath83 has less than @xmath84 @xmath253-successors.08em ; ( ii ) every path of @xmath83 from the root to some leaf on level @xmath4 contains at least one edge from @xmath253.08em ; and ( iii ) no such path contains two consecutive edges from @xmath253 . the decomposition @xmath267 obtained by contracting all edges in @xmath253 has width at most @xmath268 furthermore , the height of the underlying tree is at most @xmath4 . [ cor : bound on twd - n ] let @xmath8 be a class such that @xmath269 and let @xmath131 . if every structure @xmath214 has a tree decomposition @xmath223 of width @xmath9 and height at most @xmath258 such that the tree @xmath259 can not be embedded into @xmath83 , then @xmath270 . we conclude this section with a lemma that will be useful when constructing transductions @xmath271 that transform a structure into their tree decompositions of height @xmath4 . our construction works for all tree decompositions that are _ strict _ in the following sense . [ def : strict tree decomposition ] let @xmath223 be a tree decomposition of a structure @xmath11 . we define a function @xmath272 by @xmath273 note that @xmath274 is well - defined since , by the definition of a tree decomposition , there is at least one @xmath97 such that @xmath275 . for @xmath97 , we set @xmath276 the tree decomposition @xmath277 is _ strict _ if , for every @xmath97 , * @xmath278 ( equivalently , @xmath279 , where @xmath103 is the predecessor of @xmath101 ) and * if @xmath101 is not the root of @xmath83 , then the subgraph of @xmath229 induced by the set @xmath280 is connected . we conclude this section by a result implying that , for our purposes , it will be sufficient to consider only strict tree decompositions . [ lem : existence of strict tree decompositions ] let @xmath20 be a graph . for every tree decomposition @xmath223 of @xmath20 , there exists a strict tree decomposition @xmath281 of @xmath20 whose width and height are at most those of @xmath223 . by induction on @xmath33 , we will construct a sequence @xmath282 of tree decompositions such that @xmath283 is connected , for every @xmath284 with level @xmath285 . ( recall that @xmath100 denotes the level of @xmath101 , and the root is the only vertex of level @xmath71 . ) furthermore , the restriction of @xmath286 to the set of vertices of level at most @xmath4 will coincide with the corresponding restriction of @xmath287 , and we have @xmath288 , for all @xmath289 with level @xmath290 . it will follow that the sequence has a limit @xmath291 where @xmath292 we start the construction with @xmath293 and @xmath294 . suppose that we have already defined @xmath282 . for every vertex @xmath284 of level @xmath295 we modify the tree decomposition as follows . let @xmath296 be an enumeration of the connected components of @xmath283 . we replace in @xmath286 the subtree rooted at @xmath101 by @xmath84 copies @xmath297 of the subtree , all attached to the predecessor of @xmath101 . for @xmath298 we define @xmath299 . we can do these modifications for all vertices of level @xmath258 simultaneously . let @xmath300 be the resulting tree decomposition . the limit @xmath291 of this sequence satisfies the connectedness requirement of a strict tree decomposition . to also satisfy the other condition we proceed as follows let @xmath253 be the set of all edges @xmath239 of @xmath301 such that @xmath302 . ( note that this implies @xmath303 . ) we construct the tree decomposition @xmath281 by contracting all edges in @xmath253 . the details and the remaining verifications are left to the reader . in this section we relate the material presented in the preceding one to monadic second - order transductions . let us start by showing that there is a transduction computing the minors of a graph . [ lem : minors interpretable ] there exists a transduction @xmath1 such that @xmath304 , for every graph @xmath20 . a minor @xmath305 of @xmath20 is obtained by deleting vertices , deleting edges , and contracting edges . hence , we can encode @xmath305 by four sets.08em : the set of vertices we delete , the set of edges we delete , the set of edges we contract , and a set of vertices containing one representative of each contracted subgraph of @xmath20 ( these vertices serve as vertices of the resulting graph @xmath305 ) . with the help of these parameters we can define @xmath305 inside of @xmath306 by @xmath6-formulae . there is a close relationship between tree decompositions and transductions . [ lem : tree decomposable implies interpretable ] for every signature @xmath48 and every number @xmath59 , there exists a transduction @xmath307 $ ] that maps an order - tree @xmath83 to the class of all incidence structures @xmath27 such that the corresponding @xmath48-structure @xmath11 has a tree decomposition of width at most @xmath9 with underlying tree @xmath83 . suppose that @xmath11 is a structure which has a tree decomposition @xmath223 of width @xmath9 . we prove that @xmath11 can be defined from a colouring of @xmath83 where the number of colours depends only on @xmath48 and @xmath9 . let @xmath308 be an enumeration of all @xmath48-structures whose domain is a subset of @xmath309 $ ] . for each @xmath97 , let @xmath310 be the substructure of @xmath11 induced by @xmath208 . it follows that , for every @xmath97 , we can find some index @xmath311 such that @xmath312 . let @xmath313 be the corresponding isomorphism . furthermore , we associate with each edge @xmath239 of @xmath83 the binary relation @xmath314 \times [ k+1]\,.\end{aligned}\ ] ] we can recover @xmath11 from @xmath83 with the help of the vertex colouring @xmath315 and the edge colouring @xmath47 . we form the disjoint union of all structures @xmath316 , for @xmath97 , and we identify two elements @xmath317 and @xmath318 if @xmath239 is an edge of @xmath83 such that @xmath319 . this can be performed by an @xmath4-copying @xmath6-transduction where @xmath4 is the maximal size of the structures @xmath320 , @xmath321 . we have just seen that we can map a class of trees to a class of structures with these trees as tree decompositions . conversely , if we only consider strict tree decompositions , we can define a transduction mapping a class of structures to the corresponding class of trees . recall the function @xmath272 from definition [ def : strict tree decomposition ] . that assigns to an element @xmath125 the minimal index @xmath97 such that @xmath275 . [ prop : strict tree decompositions definable ] for each number @xmath33 , there exists an @xmath6-formula @xmath322 such that , for every strict tree decomposition @xmath203 of a graph @xmath20 of height at most @xmath4 , there are sets @xmath323 such that @xmath324 given @xmath72 we use the sets @xmath325 of all elements that first appear at level @xmath41 of the tree . in particular , @xmath326 is the root component of the tree decomposition . for @xmath327 , let @xmath328 be the subgraph of @xmath20 induced by @xmath329 . for @xmath330 and @xmath331 we define @xmath332 clearly , the relation @xmath79 is @xmath6-definable with the help of the parameters @xmath333 . we claim that , for @xmath334 , we have @xmath335 @xmath231 suppose that @xmath336 . then @xmath337 . furthermore , @xmath338 is connected since @xmath72 is strict . hence , @xmath338 is a connected component of @xmath339 containing both @xmath340 and @xmath341 . since @xmath342 it follows that @xmath343 . @xmath230 suppose that @xmath343 . then there exists an undirected path @xmath344 in @xmath345 connecting @xmath340 and @xmath341 . since @xmath346 , for all @xmath347 such that @xmath348 , it follows that @xmath344 is contained in some @xmath280 such that @xmath349 . since @xmath340 is a vertex of @xmath344 we must have @xmath350 . furthermore , @xmath337 since @xmath341 is also a vertex of @xmath344 . this implies that @xmath336 . [ thm : strict tree decompositions can be computed by a transduction ] for each constant @xmath33 , there exists a transduction @xmath271 mapping a graph @xmath20 to the class of all ( underlying trees of ) strict tree decompositions of @xmath20 of height at most @xmath4 . let @xmath203 be a strict tree decomposition of @xmath20 , and let @xmath351 be the formula from proposition [ prop : strict tree decompositions definable ] with parameters @xmath323 . we can define the tree @xmath83 underlying @xmath72 as follows.08em : * its root is any element of @xmath326 . * for the other vertices of @xmath83 , we choose one vertex in each @xmath352-class different from @xmath353 . note that @xmath352 is definable with the help of @xmath354 . * the ordering of @xmath83 is defined by @xmath354 . hence , we obtain a transduction with parameters @xmath333 that transforms a graph into a ` candidate ' tree decomposition . via a backwards translation we can write down a formula stating that the candidate given by the parameters @xmath333 corresponds to an actual strict tree decomposition . we omit the details which are standard for this type of construction . in lemma [ lem : tree decomposable implies interpretable ] we have seen how to obtain classes of bounded tree - width from classes of trees . conversely , it is the case that every class obtained from a class of trees via a transduction has a bounded tree - width . [ thm : interpretable implies tree decomposable ] for every transduction @xmath355 $ ] , there exists a number @xmath59 such that , for each @xmath84-coloured tree @xmath70 with image @xmath356 , the structure @xmath11 has a tree decomposition of width at most @xmath9 where the underlying tree is @xmath70 . \(a ) courcelle and engelfriet @xcite have shown that an incidence structure @xmath27 obtained via a transduction @xmath1 from an @xmath84-coloured tree @xmath70 has bounded tree - width . theorem [ thm : interpretable implies tree decomposable ] strengthens this result by proving that , if @xmath27 is the image of a tree @xmath70 , then we can use the same tree @xmath70 as the tree underlying a tree decomposition of the given width . \(b ) lapoire has announced in @xcite a result somewhat related to theorem [ thm : strict tree decompositions can be computed by a transduction ] . he claims that , for every @xmath59 , there exists a transduction that transforms a given graph @xmath20 of tree - width at most @xmath9 to a coloured tree ( like in the proof of lemma [ lem : tree decomposable implies interpretable ] ) that encodes some tree decomposition of @xmath20 of width at most @xmath9 . our result is less ambitious in the sense that we only consider tree decompositions of a fixed height . this enables us to give a precise description of which tree decompositions ( the strict ones ) our transduction returns . note that one can show that , for @xmath122 , there is no such transduction that would return _ all _ tree decompositions of @xmath20 of width at most @xmath9 . we split the proof into several lemmas . as a technical tool we introduce a second kind of hierarchical decompositions of structures and a corresponding notion of width . to simplify the definition we will only consider incidence structures . let @xmath357 be an incidence structure . \(a ) a _ partition refinement _ of @xmath27 is a family @xmath358 of pairs consisting of a subset @xmath359 and an equivalence relation @xmath360 on @xmath361 with the following properties.08em : * the index set @xmath83 is a tree . * for the root @xmath82 , we have @xmath362 * for every internal vertex ( i.e. , non - leaf ) @xmath98 with successors @xmath363 , the sets @xmath364 form a partition of @xmath365 . * @xmath366 , for every leaf @xmath98 . * @xmath367 and @xmath240 implies @xmath368 . * if @xmath103 is an internal vertex of @xmath83 , @xmath369 successors of @xmath103 , not necessarily distinct , and @xmath370 , @xmath371 elements , then @xmath368 implies either @xmath372 note that it follows that , for every element @xmath373 , there is some leaf @xmath98 such that @xmath374 . \(b ) the _ width _ of a partition refinement @xmath358 is the maximum number of equivalence classes realised in some component @xmath361.08em : @xmath375 the _ partition - width _ of the structure @xmath27 is the minimal width of a partition refinement of @xmath27 . the notion of a partition refinement and of partition - width are adaptations of definitions from @xcite . up to a factor of @xmath58 , the partition - width of an incidence structure and its clique - width coincide . let @xmath376 be a structure with domain @xmath377 and a ternary relation @xmath378 its incidence structure is @xmath379 with @xmath380 . we obtain a partition refinement where we have indicated the partition into @xmath360-classes by vertical bars . this partition refinement has width @xmath381 . [ lem : tree decomposition bounded by partition width ] for every partition refinement @xmath358 of an incidence structure @xmath382 , there exists a tree decomposition @xmath203 of @xmath11 with the same underlying tree @xmath83 such that @xmath383 let @xmath384 be the function assigning to every @xmath373 the unique leaf @xmath385 of @xmath83 such that @xmath386 . we claim that the desired tree decomposition @xmath387 of @xmath11 is given by @xmath388 note that the connectedness condition holds since @xmath389 implies that @xmath101 belongs to precisely those components @xmath390 such that @xmath103 lies on the path from @xmath391 to @xmath392 . it remains to prove that @xmath393 . if @xmath394 , for some @xmath395 , then @xmath396 consists of the components of @xmath174 . hence , @xmath397 therefore , we may assume that @xmath398 $ ] . let @xmath399_u : = \set { y \in w_u } { y \approx_u x } \end{aligned}\ ] ] denote the @xmath400-class of @xmath77 . we prove the following bounds . @xmath401_u } = 1 $ ] , for all @xmath402 . @xmath401_u \cap u_u } \leq 2 $ ] , for all @xmath403 . @xmath404 . since @xmath405 it then follows that @xmath406 . \(1 ) let @xmath402 . there is some tuple @xmath407 and some index @xmath41 such that @xmath408 and @xmath409 . we have @xmath410 , for every @xmath411 such that @xmath412 . since @xmath77 is the only such element it follows that @xmath197_u = \{x\}$ ] . \(2 ) let @xmath403 . there is some tuple @xmath407 and some @xmath41 such that @xmath408 and @xmath413 . let @xmath101 be the successor of @xmath103 such that @xmath414 . we have @xmath410 , for all @xmath415 such that @xmath412 . hence , @xmath197_u \smallsetminus w_v = \{x\}$ ] . suppose that there is some element @xmath416_u \cap w_v \cap u_u$ ] . by definition of @xmath390 there is some tuple @xmath417 and some @xmath418 such that @xmath419 and @xmath420 . as above it follows that @xmath197_u \cap w_v = \{y\}$ ] . consequently , we have @xmath401_u \cap u_u } \leq 2 $ ] . \(3 ) let @xmath421 and consider some tuple @xmath407 such that @xmath408 and @xmath422 . set @xmath423 for @xmath424 , it follows that @xmath425 furthermore , we obviously have @xmath426 . it follows that @xmath427 contains at most @xmath428 vertices . [ lem : partition width of simple tree interpretations ] let @xmath355 $ ] be a basic @xmath6-transduction such that , for every @xmath84-coloured order - tree @xmath70 with image @xmath356 , we have @xmath429 then there exists a number @xmath33 such that , for every order - tree @xmath70 , we can find a partition refinement @xmath430 of @xmath431 of width at most @xmath4 . let @xmath432 be the definition scheme of @xmath1 , and let @xmath433 be the maximal rank of these formulae . given @xmath70 we define the desired partition refinement @xmath434 by setting @xmath435 \prefixtext{and } & x \approx_u y \quad\defiff\quad \begin{aligned}[t ] & { \mathrm{mth}}_h(\frakt_v , x ) = { \mathrm{mth}}_h(\frakt_w , y)\ , , \\ & \text{where $ v , w$ are the successors of $ u$ with } x \in w_v \text { and } y \in w_w\ , . \end{aligned}\end{aligned}\ ] ] ( if @xmath103 is a leaf of @xmath83 then @xmath374 and we take the equality relation for @xmath400 . ) note that the index of @xmath360 is finite and that it only depends on @xmath433 and not on the input tree @xmath70 . it remains to show that @xmath436 is actually a partition refinement . first , let us prove that @xmath367 and @xmath240 implies @xmath368 . it is sufficient to consider the case that @xmath103 is the predecessor of @xmath101 . then the general case follows by induction . hence , suppose that @xmath101 is a successor of @xmath103 , that @xmath437 are successors of @xmath101 , and that @xmath438 are leaves with @xmath439 and @xmath440 such that @xmath367 . then we have @xmath441 which , by lemma [ lem : composition for trees ] , implies that @xmath442 consequently , we have @xmath368 . we also have to show that the incidence relation is invariant under @xmath400 . let @xmath369 be successors of @xmath103 and suppose that @xmath438 are leaves with @xmath443 and @xmath444 such that @xmath368 . we distinguish two cases . suppose that @xmath445 and let @xmath446 be an edge . since @xmath447 it follows that @xmath448 hence , @xmath408 iff @xmath410 . now , suppose that @xmath449 and let @xmath450 be an element . since @xmath447 it follows that @xmath451 hence , @xmath452 iff @xmath453 . \(1 ) first , suppose that @xmath355 $ ] is a basic @xmath6-transduction such that , for every @xmath84-coloured order - tree @xmath454 with image @xmath356 , we have @xmath429 it follows by lemma [ lem : partition width of simple tree interpretations ] that there is a number @xmath455 such that , for every tree @xmath70 , we can find a partition refinement of @xmath356 with underlying tree @xmath70 whose width is at most @xmath153 . by lemma [ lem : tree decomposition bounded by partition width ] it follows that @xmath11 has a tree decomposition @xmath223 with underlying tree @xmath70 and whose width is less than @xmath456 . \(2 ) if @xmath1 is a basic @xmath6-transduction such that @xmath457 then we can argue similarly . let @xmath458 be the @xmath6-transduction mapping @xmath70 to the structure obtained from @xmath431 by adding one isolated element for every leaf of @xmath70 that does not correspond to an element of @xmath431 . then @xmath458 is of the form considered in ( 1 ) and we obtain a tree decomposition @xmath223 of @xmath459 . deleting from every component @xmath208 all elements not in @xmath431 we obtain the desired tree decomposition of @xmath431 . \(3 ) suppose that @xmath1 is a non - copying @xmath6-transduction as in ( 2 ) but with @xmath460 parameters . we can regard @xmath1 as a basic @xmath6-transduction @xmath461 $ ] . by ( 2 ) it follows that , for every value of the parameters @xmath92 , the structure @xmath462 has a tree decomposition of the required form . \(4 ) finally , consider the general case . suppose that @xmath1 is @xmath151-copying . given @xmath70 let @xmath463 be the tree obtained from @xmath70 by adding @xmath151 new successors to every vertex of @xmath70 . formally , suppose that @xmath81 , for some finite set @xmath72 . w.l.o.g . we may assume that @xmath464 = \emptyset$ ] . we define the domain @xmath465)^{<\omega}$ ] of @xmath463 by @xmath466)\,.\end{aligned}\ ] ] furthermore , we add new colour predicates @xmath467\,.\end{aligned}\ ] ] note that every element of @xmath431 is of the form @xmath468 where @xmath469 $ ] and @xmath97 . hence , each such element corresponds to a leaf @xmath470 \subseteq t^+$ ] . using the parameters @xmath471 we can construct a basic @xmath6-transduction @xmath472 $ ] satisfying the conditions of ( 3 ) such that @xmath473 . by ( 3 ) , we obtain a tree decomposition @xmath474 of @xmath473 . let @xmath475 be the tree decomposition obtained from @xmath476 by contracting every edge leading to a leaf in @xmath477 . then we have @xmath478 -- the focus of our investigation lies on the following preorder on classes of structures which compares their ` encoding powers ' with respect to @xmath6-transductions . our main result is a complete description of the hierarchy induced by this preorder . it will be given in theorem [ thm : hierarchy ] . let @xmath479 . we define the following relations . @xmath480 if there exists a transduction @xmath1 such that @xmath18 . @xmath481 if @xmath480 and @xmath482 . @xmath483 if @xmath480 and @xmath484 . @xmath485 if @xmath481 and there is no class @xmath486 with @xmath487 . @xmath488 if @xmath489 . the relations @xmath490 , @xmath491 , and @xmath492 are defined analogously to @xmath493 , @xmath494 , @xmath495 by replacing @xmath29 everywhere by @xmath28 . the _ transduction hierarchy _ is the hierarchy of classes @xmath496 induced by the relation @xmath28 . as the class of transductions is closed under composition , it follows that the relation @xmath28 is a preorder , i.e. , it is reflexive and transitive . @xmath28 is a preorder on @xmath497 . we consider the following subclasses of @xmath498 $ ] . ( all trees below are considered to be successor - trees . ) @xmath499 is the set of all complete @xmath84-ary trees of height @xmath4 . @xmath500 is the class of all binary trees . @xmath501 is the class of all trees . @xmath502 is the class of all paths . @xmath503 is the class of all rectangular grids . the following description of the transduction hierarchy is the main result of the present paper . [ thm : hierarchy ] we have the following hierarchy.08em : @xmath504 for every signature @xmath48 , every class @xmath505 $ ] is @xmath491-equivalent to some class in this hierarchy . there is a lot of flexibility in the choice of representatives for the various levels . for instance , we could replace @xmath506 by the class of _ all _ trees of height at most @xmath4 , @xmath501 by @xmath500 or @xmath507 , and @xmath503 by the class of square grids . it is straightforward to show that the above classes form an increasing chain . the hard part is to prove that the chain is strictly increasing and that there are no further classes . we have @xmath508 in the example before lemma [ lem : transductions are comorphisms ] , we have constructed transductions @xmath271 such that @xmath509 . hence , @xmath510 . the remaining assertions follow from the observation that , by lemma [ lem : minors interpretable ] , @xmath511 implies @xmath488 . let us collect some easy properties of the hierarchy . our first result states that @xmath503 is a representative of the top level of the transduction hierarchy . [ lem : g is top ] @xmath49 \sqsubseteq_{\mathrm{in}}\calg$ ] recall that the @xmath512 grid is the undirected graph @xmath189 with vertices @xmath513 \times [ n]$ ] and edge relation @xmath514 before encoding arbitrary structures in such grids we describe a transduction mapping @xmath20 to its directed variant @xmath515 where @xmath516 this can be done with the help of the parameters @xmath517 where @xmath518 then @xmath519 it is easy to write down a formula checking that the parameters @xmath520 and @xmath521 are correctly chosen ( see , e.g. , @xcite ) . to show that @xmath49 \sqsubseteq_{\mathrm{in}}\calg$ ] , suppose that @xmath167 $ ] is a structure with @xmath522 . fix enumerations @xmath523 of @xmath43 and @xmath524 of @xmath192 . by the above remarks , it is sufficient to encode @xmath27 in the directed @xmath512 grid . consider the following subsets of @xmath525 \times [ n]$].08em : @xmath526 \times \{0\}\,,\qquad & p'_r & : = \set { \langle 0,k\rangle } { e_k \in p_r } \ , , \\ e ' & : = \{0\ } \times [ n]\,,\qquad & i'_l & : = \set { \langle i , k\rangle } { ( a_i , e_k ) \in { \mathrm{in}}_l } \,.\end{aligned}\ ] ] then @xmath27 can be recovered from @xmath20 by an @xmath6-transduction using these sets as parameters . @xmath527 . for one direction , note that @xmath528 implies @xmath529 . conversely , each finite tree can be obtained as minor of a binary tree . therefore , we have @xmath530 . we have @xmath531 if and only if @xmath8 is finite and contains at least one nonempty structure . as indicated in the example before lemma [ lem : transductions are comorphisms ] , there exists a transduction mapping an incidence structure @xmath27 to the incidence structure @xmath146 of the gaifman graph of @xmath11 . [ lem : gaifman graph interpretable ] for every class @xmath8 of structures , we have @xmath532 . the next result is just a restatement of lemma [ lem : minors interpretable ] in our current terminology . [ lem : minor equivalent ] for every class @xmath8 of graphs we have @xmath533 . in this section we prove that the hierarchy is strict . using the results of section [ sect : tree decompositions and transductions ] we can characterise each level of the hierarchy in terms of tree - width and its variants . [ thm : characterisation of pwd , twd ] let @xmath505 $ ] . @xmath534 @xmath535 . @xmath536 @xmath537 . @xmath538 @xmath539 . in each case @xmath231 follows from lemma [ lem : tree decomposable implies interpretable ] and @xmath230 follows from theorem [ thm : interpretable implies tree decomposable ] . let @xmath8 be a class of @xmath48-structures . @xmath540 implies @xmath541 . @xmath542 implies @xmath543 . \(a ) suppose that @xmath540 . then theorem [ thm : excluded tree theorem ] implies that @xmath544 . hence , the claim follows from lemmas [ lem : gaifman graph interpretable ] and [ lem : minor equivalent ] . \(b ) suppose that @xmath542 . then theorem [ thm : excluded grid theorem ] implies that @xmath545 . as in ( a ) , the claim follows from lemmas [ lem : gaifman graph interpretable ] and [ lem : minor equivalent ] . [ cor : nothing between p , t , g ] let @xmath505 $ ] . @xmath546 implies @xmath541 . @xmath547 implies @xmath548 . in particular , it follows that the upper part of the hierarchy is strict.08em : [ cor : upper part ] @xmath549 both assertions follow from theorem [ thm : characterisation of pwd , twd ] and corollary [ cor : nothing between p , t , g ] . for the first one , note that we have @xmath550 since @xmath551 . conversely , @xmath552 implies , by theorem [ thm : characterisation of pwd , twd](a ) , that @xmath553 . hence , @xmath554 . finally , if @xmath555 then @xmath534 , by corollary [ cor : nothing between p , t , g](a ) . consequently , we have @xmath556 . similarly , the fact that @xmath557 follows from lemma [ lem : g is top ] . since @xmath558 , theorem [ thm : characterisation of pwd , twd](b ) implies that @xmath559 . finally , we obtain @xmath560 by corollary [ cor : nothing between p , t , g](b ) . let us turn to the lower part . we start with two technical lemmas . let @xmath561 be an order - tree . vertices @xmath562 of @xmath83 are _ horizontally related via _ a vertex @xmath153 if all @xmath156 are at the same level of the tree and @xmath563 , for all @xmath564 . [ lem : large degree is preserved ] let @xmath70 be a coloured order - tree of height @xmath4 , and suppose that @xmath1 is a parameterless @xmath9-copying @xmath6-transduction of rank @xmath52 such that @xmath431 is a successor - tree of height at most @xmath258 . consider vertices @xmath562 of @xmath70 that are horizontally related via @xmath153 and fix some number @xmath565 . let @xmath566 be the successor of @xmath153 with @xmath567 . if , for all @xmath568 , we have @xmath569 then at least @xmath570 elements of the set @xmath571 ( these are elements of the domain of @xmath431 ) are horizontally related in @xmath431 . let @xmath572 be the formula defining the successor relation in @xmath431 between vertices of the form @xmath573 and @xmath574 . by assumption the rank of @xmath572 is at most @xmath52 . first , note that a vertex @xmath575 is on level @xmath433 in @xmath431 if and only if there are indices @xmath576 such that @xmath577 \exists y_0\cdots\exists y_{h-1}\bigl[\land_{i < h-1 } \varphi_{s_is_{i+1}}(y_i , y_{i+1 } ) & \land \varphi_{s_{h-1}l}(y_{h-1},v ) \\ & \land \neg \exists z\lor_{s < k } \varphi_{ss_0}(z , y_0)\bigr ] \end{aligned}\end{aligned}\ ] ] expresses that there exists a path of the form @xmath578 from the root @xmath579 of @xmath431 to the vertex @xmath575 . by assumption on @xmath156 and lemma [ lem : composition for trees ] , we have @xmath580 since the rank of @xmath581 is @xmath582 it follows that @xmath583 hence , all vertices @xmath584 are on the same level @xmath433 in @xmath431 . we prove by induction on @xmath433 that @xmath585 implies that all but at most one of @xmath584 are horizontally related . let @xmath586 be the predecessor of @xmath587 in @xmath431 , that is , @xmath588 we distinguish two cases . first suppose that @xmath589 in @xmath70 . by @xmath590 and lemma [ lem : composition for trees ] , we have @xmath591 for all @xmath41 such that @xmath592 . note that there can be at most one index @xmath41 that does not satisfy this condition since , if we had @xmath593 and @xmath594 , for @xmath595 , then we would have @xmath596 and @xmath562 would not be horizontally related via @xmath153 . it follows that @xmath597 hence , @xmath598 is the common predecessor of all the @xmath587 , except for possibly one of them . ( for an index @xmath41 with @xmath593 our composition argument does not work since in that case @xmath590 does not imply that the theories of @xmath599 and @xmath600 coincide . ) it remains to consider the case that @xmath601 . setting @xmath602 we have @xmath603\,.\end{aligned}\ ] ] since the rank of this formula is @xmath604 it follows that @xmath605\ , , \quad\text{for all } i < m\,.\end{aligned}\ ] ] consequently , we have @xmath606 , for all @xmath41 , and @xmath607 are horizontally related via @xmath153 . since the vertices @xmath608 are on level @xmath609 in @xmath431 , we can apply the induction hypothesis and it follows that all but at most one of then are horizontally related via some vertex @xmath610 . therefore , the same holds for their successors @xmath584 . we denote by @xmath611 the number of functions @xmath612)$ ] with @xmath613 . intuitively , each such function corresponds to a vertex - colouring of the tree @xmath85 with @xmath614 colours . for @xmath615 and @xmath122 , we have @xmath616 for @xmath617 , we have @xmath618 since @xmath619^{<n } } = \sum_{i < n } m^i = m^{n-1 } + \sum_{i < n-1 } m^i$ ] it follows that @xmath620^{<n } } \leq 2m^{n-1}.\end{aligned}\ ] ] therefore , we can bound @xmath621^{<n}}}\end{aligned}\ ] ] from above by @xmath622 and from below by @xmath623 -- [ lem : lower part strict ] @xmath624 for a contradiction , suppose that there exists a transduction @xmath1 such that @xmath625 . let @xmath626 be the class of all order - trees corresponding to successor - trees in @xmath506 , and let @xmath627 be the class of all coloured successor - trees with one colour whose underlying tree is in @xmath628 . since the successor - trees in @xmath506 are @xmath63-sparse we can construct an @xmath6-transduction @xmath629 such that @xmath630 . since @xmath631 we can combine @xmath1 and @xmath629 to a transduction @xmath165 such that @xmath632 . by lemma [ lem : large degree is preserved ] , it follows that there is some constant @xmath633 such that every tree @xmath634 with out - degree at most @xmath9 is of the form @xmath635 where @xmath636 has out - degree at most @xmath637 . ( the out - degree of an order - tree is the out - degree of the corresponding successor - tree . ) suppose that @xmath165 uses @xmath614 parameters . there are @xmath638 colourings of trees in @xmath626 with out - degree at most @xmath637 . on the other hand , there are @xmath639 trees in @xmath640 with out - degree at most @xmath9 . for large @xmath9 it follows that @xmath641 consequently , there is some tree in @xmath640 that is not the image of a tree in @xmath626 . a contradiction . we have shown that the hierarchy presented in theorem [ thm : hierarchy ] is strict . to conclude the proof of the theorem it therefore remains to show that there are no additional classes . we have already seen in corollary [ cor : upper part ] that @xmath501 and @xmath503 are the only classes above @xmath502 . next we shall prove that @xmath642 . [ lem : embeddable trees means interpretable ] let @xmath8 be a class of structures . if , for every number @xmath131 , there exists a structure @xmath214 such that we can embed @xmath259 into every tree underlying a tree decomposition @xmath223 of @xmath11 of width @xmath9 , then @xmath643 . by lemma [ lem : existence of strict tree decompositions ] , it follows that , for every @xmath131 , there is a structure in @xmath8 with a strict tree decomposition of width at most @xmath9 and with an underlying tree @xmath83 into which we can embed the tree @xmath259 . according to proposition [ prop : strict tree decompositions definable ] there is a transduction mapping @xmath8 to the class of trees underlying these strict tree decompositions . hence , there exists a class @xmath644 of successor - trees containing , for every @xmath131 , some tree into which we can embed @xmath259 . hence , @xmath645 . [ thm : covering for lower part ] let @xmath8 be a class of structures . then @xmath646 implies @xmath539 . suppose that @xmath646 . then @xmath647 since @xmath648 . by lemmas [ lem : gaifman graph interpretable ] and [ lem : minor equivalent ] , this implies that @xmath649 . therefore , we can find a path that is not in @xmath650 . by theorem [ thm : excluded path theorem ] , it follows that there are numbers @xmath651 such that @xmath652 . by induction on @xmath151 , we prove that @xmath653 implies @xmath539 . for @xmath654 , there is nothing to do . for @xmath655 , we have @xmath656 , which implies that @xmath657 . consequently , it follows by lemma [ lem : embeddable trees means interpretable ] and corollary [ cor : bound on twd - n ] that @xmath658 . by induction hypothesis , the result follows . by lemma [ lem : tree decomposable implies interpretable ] we obtain the following results . [ cor : nothing between tn and tn+1 ] if @xmath646 then @xmath538 . [ cor : lower part covers ] @xmath659 . to conclude the proof of theorem [ thm : hierarchy ] it remains to show that there are no classes between the lower part of the hierarchy and its upper part . [ lem : calt_n dense below calp ] let @xmath8 be a class of structures . if @xmath660 , for all @xmath33 , then @xmath661 . we show the contraposition . suppose that @xmath647 . we have to show that there is some @xmath4 such that @xmath662 . as in the proof of theorem [ thm : covering for lower part ] it follows that there are numbers @xmath651 such that @xmath652 . hence , we can use lemma [ lem : tree decomposable implies interpretable ] to obtain a transduction @xmath1 witnessing that @xmath663 . by theorem [ lem : lower part strict ] we have @xmath664 . it follows that @xmath665 , as desired . [ cor : noting between lower and upper part ] if @xmath666 then there is some @xmath33 such that @xmath538 . by lemma [ lem : calt_n dense below calp ] , there is some @xmath33 such that @xmath646 . hence , corollary [ cor : nothing between tn and tn+1 ] implies that @xmath538 . together , corollaries [ cor : upper part ] , [ cor : lower part covers ] , and [ cor : noting between lower and upper part ] ( and the fact that every class @xmath8 satisfies @xmath667 ) show that every class of @xmath48-structures is @xmath491-equivalent to some of the classes in theorem [ thm : hierarchy ] . this completes the proof of this theorem . above we have obtained a complete description of the transduction hierarchy for classes of finite incidence structures . the most surprising result is that the hierarchy is linear . at this point there are at least three natural directions in which to proceed . an answer to ( ii ) seems within reach , at least if we restrict our attention to countable structures . although the resulting hierarchy is no longer linear we can adapt most of our techniques to this setting . ( for an example of nonlinearity , note that the class of all countable trees and the class of all finite grids are incomparable . ) concerning question ( iii ) , let us remark that all results above go through if we use @xmath110 instead of @xmath6 . we only need the right definition of rank for @xmath110 . in the proof of theorem [ lem : lower part strict ] we needed the fact that there are only finitely many theories of bounded rank . we can ensure this for @xmath110 by defining the rank as the least number @xmath4 such that for logics much weaker than @xmath6 , on the other hand , it seems unrealistic to hope for a complete description of the corresponding transduction hierarchy . for instance , a related hierarchy for first - order logic was investigated by mycielski , pudlk , and stern in @xcite . the results they obtain indicate that the structure of the resulting hierarchy is very complicated . finally , let us address question ( i ) . when using transductions between structures instead of their incidence structures , we can transfer some of the above results to the corresponding hierarchy . but we presently have no complete description since we miss some of the corresponding excluded minor results . note that all classes in theorem [ thm : hierarchy for mso ] are @xmath58-sparse . for @xmath58-sparse classes @xmath8 and @xmath17 , lemma [ lem : relation between ad and in ] implies that @xmath679 consequently , the result follows from theorem [ thm : hierarchy ] . if we only consider classes of graphs and if we use @xmath110-transductions instead of @xmath6-transductions , then the following result can be used as replacement of theorem [ thm : excluded grid theorem].08em : this eliminates some possibilities for intermediate classes of graphs in the hierarchy of theorem [ thm : hierarchy for mso ] , but to complete the picture we still need analogues of proposition [ prop : strict tree decompositions definable ] and of theorems [ thm : excluded tree theorem ] and [ thm : excluded path theorem ] . furthermore , the techniques of @xcite are specific to graphs ( or , more generally , to relational structures where all relations are binary ) . even with the results of @xcite one can not exclude the existence of a class @xmath8 of arbitrary relational structures strictly between @xmath501 and @xmath503 in the @xmath110-transduction hierarchy . let us make a final comment about relational structures . an incidence structure @xmath27 can be seen as a bipartite labelled directed graph ( see the remark after definition [ def : incidence structure ] ) . furthermore , it is @xmath63-sparse . hence , our results use tools from graph theory , in particular those of @xcite . however , there is currently no encoding of relational structures as labelled graphs that could help to solve question ( i ) above .

parameter estimation is a central issue in system modeling : a typical problem is to start from a certain amount of information on a given system ( e.g. its correlation functions ) and then extract the parameters of a model which is supposed to describe the its properties @xcite . the parameter estimation procedure gives insight on the validity of the model and can suggest the introduction of more appropriate and efficient models . a usual approach is to extract the parameters from an instance of the problem in certain conditions and subsequently testing the model in other instances . from this point of view is useful to deal with systems in conditions where the relation between observables and model parameters is more transparent : e.g. for a statistical mechanics system this corresponds to high / low temperature or field . once the parameters have been estimated , one moves to more interesting parameter regions , where the full complexity of the system shows up . such an approach , when translated into the wide arena of complex systems , generally can not be carried out since no knob such as temperature or field is available , so that we may be faced with the inverse problem in the hardest region . a huge interest in obtaining accurate parameter estimation stems from the current availability of large datasets in several areas of biology , economy and social sciences , to name a few examples dna sequences , stock market time series and internet traffic data ( see more references in @xcite ) . this great amount of data has made even more pressing the quest for efficient models , allowing us to extract and encode the relevant information . various techniques have been developed in order to solve this problem : two general approaches which can be flexibly adapted to the specific problems are bayesian model comparison @xcite and boltzmann - machine learning @xcite . in the past decade a significant contribution to topic of parameter estimation came from the application of typical statistical mechanics techniques which turned out to be very useful in the modeling and study of different fields ranging from neurobiology @xcite to the economy @xcite . the description of a system using statistical models ( and in particular ising - like models ) appears natural in many contexts : e.g. , effective ising models generally arise when the space of states is intrinsically discrete ( e.g. , for dna and proteins ) and , even when this is not the case , some ising variables may be lurking behind the continuous ones . in the statistical physics realm such emergence of effective ising models could occur near a critical point when the microscopic model is in the ising universality class @xcite - but one can also find more subtle examples where discrete ising - like spin degrees of freedom describe some hidden order , e.g. the chiral ordering in frustrated continuous spin models @xcite . a paradigmatic example considered by the statistical physics community in the context of parameter estimation is , of course , the inverse ising problem , i.e. the problem of inferring the values of the coupling constants of a general ising model from the correlation functions . the inverse ising problem has been tackled by numerical and analytical methods , often adapting old techniques to the problem at hand . among these attempts we mention monte carlo optimization @xcite , message passing based algorithms @xcite and thouless - anderson - palmer equations approaches @xcite ( see @xcite for a review ) . field theoretical techniques has been used by sessak and monasson @xcite who perturbatively calculated , in terms of the correlations , expressions for the interaction parameters of a general ( heterogeneous ) ising model with two - body interaction and external field . most of the available results on the inverse ising problem concerns ising models having two - spin interactions : in this context exact methods , solving the inverse ising problem with general multispin interactions , are welcome . we decided to concentrate in this paper on the inverse ising problem in one dimension . the motivation is three - fold : firstly one - dimensionality allows for exact solutions . in this manuscript we indeed present explicit analytical formulas to exactly perform the inversion for one - dimensional ising systems having general multispin interactions . our results therefore provide a theoretical laboratory where different approximate inverse ising techniques @xcite can be benchmarked against the exact results obtained using our method : in the following we compare some other approximate methods with exact results . the possibility of testing approximate methods against exact results in one - dimensional systems is not our only motivation : indeed one - dimensional classical models are often employed to describe the conformational transition of systems , as proteins or dna , naturally possessing an underlying one - dimensional structure . such simple models are found to capture some of the global properties of these complex systems as long as conformational properties are concerned . the existence of exact methods would then help to determine the parameters and the important interactions of effective models describing the properties of such systems . more in detail , the use of one - dimensional statistical mechanics models applied to systems like proteins or dna is usually based on the individuation of a reduced set of states representing the conformational state of a given elementary unit : e.g. in protein systems the states could be chosen as helix , coil and sheet ( for aminoacids belonging to an @xmath0-helix , to a coil and to a @xmath1-sheet , respectively ) . the task is then , given this reduced set of states , to estimate the probabilities to have the consecutive elements in different states @xcite and then our method ( working for ising and potts models ) would then allow for the determination of the parameters of effective discrete models . we notice that in our method we can consider also longer ranged couplings ( _ i_.e . , longer than nearest - neighbour ) emulating interactions among aminoacids distant along the chain , but near in physical space @xcite . another motivation for our work is based on the fact that one - dimensional ising - like models can also be used to deal with stationary time - series of correlated data : as we will later discuss in the conclusions , it is possible to connect stationary time - series of data by using a mapping onto an equilibrium discrete markov chain having finite memory . for this application , the inversion task ( to which we can refer as an inverse markov problem ) consists in extracting from the data the transition probabilities of the associated guessed markov chain : therefore , given the similarity of the two inversion ( ising and markov ) problems , the existence of exact techniques can provide in perspective a way to effectively attack the inverse markov problem . we observe that the method of using markov chains to describe sequences of data may prove useful even in biological realms when statistical properties of e.g. dna sequences are concerned @xcite . in the following we study the one - dimensional inverse ising problem with general finite - range multispin interactions : by finite range @xmath2 we mean that that two spins exceeding the distance @xmath2 do not interact ( this implies that at maximum @xmath2-spin couplings can be present ) . we will then consider the reconstruction of ising models having exponential or power - law two - spin couplings ( and three- or four - spin interactions ) , approximating them with a finite range @xmath2 and checking the validity of the reconstructed couplings . a remark about dimensionality is due : being the dimension set to one , the system can not order at finite temperature . however we show that mean - field like interactions can be included in our formalism , so that one can treat systems having finite - range multispin couplings and long - range mean - field interactions giving rise to finite - temperature transitions . another possibility would be to extend the range of the interaction and perform a so - called finite - range scaling . such a technique has been employed @xcite to the ising model with power - law @xmath3 decaying interactions , a model exhibiting a rich behaviour including a berezinskii - kosterlitz - thouless transition ( for @xmath4 ) @xcite and gaussian and non - gaussian rg fixed points ( in the range of @xmath0 between @xmath5 and @xmath6 ) as the decay exponent @xmath0 is varied @xcite . in this paper we present the solution of the inverse problem for a one - dimensional ising model with finite - range arbitrary interactions , i.e. not restricted to the one- and two - body type . the main result of our paper is formula which expresses the entropy of a one - dimensional translational invariant system ( in equilibrium ) in terms of a sufficiently large number of correlation functions , from which the inversion formula immediately follow . we observe that ising chains are usually treated via transfer matrix method , but when longer range or multispin types of interaction are included the search for the parameters reproducing the observables might become very onerous . our method provides a direct method of estimating the parameters when a sufficiently large number of correlation function is known . the inclusion of many - body interactions may prove useful for the description of complex systems where the two - body assumption is not justified or in more traditional many - body systems with long - range interaction , where the construction of low - energy effective theories quite naturally leads to the appearance of multispin interactions @xcite . the paper is structured as follows : in section [ notation_and_statement ] we introduce our notations and we state the mathematical problem . section [ inversion_formula ] contains our main result on the entropy in terms of the correlation functions and the resulting inversion formula . the obtained result is illustrated on simple problems in section [ simple_examples ] . in section [ realistic_examples ] we examine more complicated examples where the usefulness of our result is shown . we analyze models formally not have finite - range interactions and having exponential or power - law two - spin interactions plus multispin interactions . the data generated by monte carlo simulations are analyzed with our technique which correctly detects the structure of interactions . in section [ mf_lr ] we briefly discuss how the developed formalism may be modified to allow mean - field interactions . finally we draw our conclusions in [ conclusions_and_perspectives ] . the appendix present checks of the obtained findings for small values of the range @xmath2 using the transfer matrix method , and as well as supplementary material on the @xmath7 ising model . we consider a general one - dimensional ising model with multispin interactions defined by the hamiltonian @xmath8 @xmath9 where @xmath10 is the configuration of the @xmath11 ising spins ( @xmath12 ) ; periodic boundary conditions will be assumed so that @xmath13 for @xmath14 . the sums runs over distinct couples , triples and so on ; the temperature dependence is absorbed in the coupling constants : explicitly , @xmath15 , @xmath16 , and so on ( where _ e_.g . @xmath17 is the two - body coupling among a spin in @xmath18 and a spin in @xmath19 - as usual @xmath20 ) . the couplings @xmath21 are assumed to be invariant under translation by @xmath22 spins ( for simplicity we will assume @xmath23 is an integer , but since we are interested in the @xmath24 limit this is not strictly necessary ) : this condition reads @xmath25 ( if the indices on the right hand side exceed @xmath11 , they have to be replaced by the indices equivalent modulo @xmath11 contained in the set @xmath26 ) . finally we assume that the couplings are zero if their indices can not be brought by a translation of a multiple of @xmath22 to a subset of @xmath27 . since the use of the form of the hamiltonian may be cumbersome , it is convenient introduce a more compact notation , rewriting hamiltonian as @xmath28 where @xmath29 is a subset of @xmath30 ( this is encoded in the writing @xmath31 , which stands for `` the range of the interaction is less or equal than @xmath2 '' ) . @xmath22 is the periodicity of the interaction and @xmath32 is an operator associated to the subset @xmath33 ( @xmath34 is the number of elements of @xmath29 ) translated by @xmath35 which acts on the spins as @xmath36 for the null subset @xmath37 we define @xmath38 . the prime in the sum over @xmath29 in indicates that the null subset ( which would contribute just to a constant in the hamiltonian ) is not included and that the terms related by a translation of a multiple of @xmath22 are counted only once , in order to avoid the presence of equivalent operators in the hamiltonian . once the hamiltonian is specified we proceed in the usual calculation of the thermodynamic quantities , defining the partition function @xmath39 the free energy per elementary unit cell in the infinite volume limit ( i.e. @xmath22 spins ) @xmath40 and the correlation functions associated to the operator @xmath29 @xmath41 ( by definition , @xmath42 ) . the inverse problem for the system introduced in the previous section is stated as follows : given the set of correlations @xmath43 determine the couplings @xmath44 . the hamiltonian is the one specified in equation , i.e. the most general finite - range multispin hamiltonian ; in section [ mf_lr ] we will extend this treatment to include long - range mean - field interactions . the calculation is based on the evaluation of the entropy per unit cell @xmath45 characterized by the set of correlation functions @xmath43 . once @xmath45 is known we may compute the free energy @xmath46 where @xmath47 is the energy of a unit cell which is readily evaluated using directly and on a state specified by the set of correlations @xmath43 . the minimization of the above expression yields the inversion formulas : @xmath48 we may state now our main result for the entropy @xmath45 , which is given by @xmath49 the entropy is written in terms of the functions @xmath50 ( to which we may refer as `` entropy at range @xmath51 '' ) , given by @xmath52 where @xmath53 is the configuration of @xmath51 auxiliary ising spins . notice that the sum over @xmath29 now includes every subset , including the null one . the entropy can be shown to be convex in the variables @xmath43 , thus the equation admits a solution , unless some of the @xmath54 s used in the calculation become negative , signaling a set of `` nonphysical '' correlations . we now discuss the derivation of the formula . let us think how the measurement of a correlation @xmath55 is operatively defined : we look at @xmath2 consecutive spins and we perform the measurement . each of the microscopic configuration @xmath56 will occur with a given probability @xmath57 which would give rise to a mean value of @xmath55 given by @xmath58 since we know all of the correlations within the subsets of the @xmath2 spins , the system of the equations above may be inverted giving rise to with @xmath59 . then the boltzmann formula @xmath60 is applied to this set of probabilities obtaining the expression @xmath61 ( always for @xmath59 ) . to derive we calculate the entropy of the unit cell of size @xmath22 , regardless of the state of the remaining @xmath62 spins ; in terms of number of states it is @xmath63 where @xmath64 denotes the number of microstates of a set of spins @xmath65 ( subject to the constraints imposed by the correlations ) . it should be noted that the @xmath62 spins to be traced out can not be chosen at will : by inspection it turns out that the first @xmath62 spins is a good choice . thus taking the logarithm of we obtain our expression for the entropy of a state characterized by the set of correlations @xmath66 . the number of correlations required to specify the state can be shown by simple counting to be equal to @xmath67 . the above procedure can be formally applied also to a finite system of size @xmath2 : this is achieved by letting @xmath68 . in this case the system looks like a set of @xmath69 disjoint assemblies of @xmath2 spins for which the entropy is given by @xmath70 being @xmath71 . this result refers to a general finite system of ising spins with arbitrary heterogeneous couplings without translational invariance and it should be used if one wants to treat datasets obtained from finite heterogeneous systems and extract ising couplings @xcite . in general , the number of needed correlations to be known grows exponentially with the system size : in the case of the present paper , exponentially with @xmath2 . therefore , our result might be of practical importance if @xmath2 is small , @xmath11 is very large ( _ i.e. _ , near the thermodynamical limit ) and the underlying ising model is supposed to be one - dimensional . although the number of required correlation grows exponentially with the range @xmath2 , some simplifications may occur . for example if the system is known to be invariant under reflection @xmath72 , then odd couplings vanish and we do not have to measure the corresponding correlation function which are trivially zero . more generally , if we know that a coupling is equal to a given value @xmath73 then one has the additional ( nonlinear ) equation : @xmath74 thus reducing the number of independent correlation functions . the technique described may be easily adapted to other discrete spin systems as potts or blume - emery - griffith models by using two or more ising spins to encode the state of the discrete spin , although such a mapping may obscure the symmetries of the original model . to conclude this section , we finally observe that it is possibler to show in general the equivalence between the problem we have solved with the problem of finding , from a set of known correlations , the transition rates of an equilibrium - i.e. satisfying detailed balance - markov chain with finite memory : we postpone such discussion to the conclusions . legs may be mapped onto a spin chain with @xmath75 and @xmath76 . the filled circles represent spins which are present in the different operators considered ( and associated to the couplings @xmath77 ) . in the middle part of the figure we represent them on the ladder system , while on the bottom part we show how they look like on the chain . explicitly these operators correspond to the following terms in the hamiltonian : @xmath78 , @xmath79 , @xmath80 , @xmath81 , @xmath82 and @xmath83 legs may be mapped onto a spin chain with @xmath75 and @xmath76 . the filled circles represent spins which are present in the different operators considered ( and associated to the couplings @xmath77 ) . in the middle part of the figure we represent them on the ladder system , while on the bottom part we show how they look like on the chain . explicitly these operators correspond to the following terms in the hamiltonian : @xmath78 , @xmath79 , @xmath80 , @xmath81 , @xmath82 and @xmath83 as a first application of the results presented in the previous section , we consider a model with @xmath84 and @xmath85 ( i.e. the hamiltonian is @xmath86 where @xmath87 is the magnetic field and @xmath88 is the coupling ) . the only independent correlations are the one - body correlator , i.e. the magnetization @xmath89 , and the nearest - neighbour correlator @xmath90 . using the entropy is calculated as @xmath91 which agrees with the expression obtained in @xcite by combinatorial means . in figure [ entropy_g_m ] we plot the entropy : the convexity of @xmath92 guarantees to obtain the field and nearest - neighbour interaction in terms of @xmath93 and @xmath94 . the system we have just described presents no phase transitions , apart from the zero temperature ones which occur at the border of the surface depicted in figure [ entropy_g_m ] ; in section [ mf_lr ] we will see how the addition of a mean - field type interaction is easily included , making phase transitions possible . interestingly on the lines @xmath95 the system is frustrated and our approach readily provides an expression for the ground state degeneracy . differentiation of the entropy allows us to obtain the couplings , field @xmath96 and nearest - neighbour interaction @xmath97 conjugated to @xmath93 and @xmath94 respectively : @xmath98 in appendix [ transfer_matrices ] we examine this example ( @xmath84 and @xmath85 ) and higher range ones ( @xmath99 and @xmath85 ) , explicitly checking the validity of the inversion formula using the transfer matrix method . we will consider now a translationally invariant spin ladder with interaction among the nearest - neighbors of the same and other chain . for simplicity we will restrict ourselves to even interactions , i.e. in the hamiltonian only terms containing an even number of spins enter . as shown explicitly in figure [ ladder_to_chain ] this system may be mapped onto a chain system with @xmath75 and @xmath76 , where the original interactions ( allowed by symmetry ) and the new ones are shown . in terms of our subset notation used in the interaction parameters are defined as @xmath100 . this is easily generalized to ladders made up of more than two chains and higher interaction range , and thus our method is suited to treat general finite - range ladder systems . we point out that the inversion formula allows to explicitly write the relation among the @xmath88 s and @xmath94 s while the transfer matrix approach , e.g. in the simple ladder system described above , already entails the solution of a fourth order algebraic equation . in this section we consider another simple example , in which nearest - neighbour and next - to - nearest - neighbour interactions are present together with a four - spin interaction : denoting the coefficients @xmath101 , @xmath102 and @xmath103 by @xmath104 , @xmath105 and @xmath106 respectively , the hamiltonian reads @xmath107 this hamiltonian can be exactly treated in our framework ; the case @xmath108 ( the @xmath7 model ) is discussed in the appendix . here we aim at comparing approximate inverse ising methods against exact results , focusing in particular on the low - correlation expansion ( lce ) @xcite which is in the following compared with exact findings . we will use the lce discussed in @xcite using as input a finite number of correlations , coherently with what is done in this work ( notice that for the present case our method needs just four correlation functions in order to recover exactly the couplings ) ; the maximal range of two - body correlators for the lce is denoted by @xmath109 . the lce is discussed in @xcite to present an approximate technique for inverse ising models having at most two - spin interactions : since hamiltonian has only two - spin interactions for @xmath110 , we present lce results for the case @xmath110 in the appendix where we discuss the @xmath7 model in detail , presenting the explicit solution using the transfer matrix approach . as expected , for low temperatures ( i.e. large couplings ) , the lce breaks down and , as it can be seen in the right panel of figure [ j1j2sm ] , for moderate temperatures the expansion may settle to an incorrect value of the coupling as the range @xmath109 is increased . in order to further test the performance of the lce against exact results we present in figure [ cv_chi ] results for @xmath110 and @xmath111 : although the lce is developed in @xcite for two - spin interactions ( and the extension to treat multispin interactions is expected to be cumbersome ) , the lce reconstructed two - spin couplings with @xmath112 may partially take into account the effect of the four - spin interaction . to test to what extent this occurs , we consider two observables , susceptibility and specific heat , calculated using the reconstructed couplings : the comparison with the exact results is in figure [ cv_chi ] . as we can see in the left panel the specific heat is more sensitive than the susceptibility to reconstruction errors , even without four - spin interaction ( _ i_.e . @xmath110 ) . this may be traced back to this type of lce inversion procedure which aims at reproducing two body correlators which in this model are required to calculate the susceptibility while the specific heat already contains averages of four body operators . obviously the lce , being not designed to infer models with multispin interaction , gives no hint on the value @xmath106 but in the right panel of figure [ cv_chi ] we apply it for @xmath113 in order to test how it can reproduce the considered observables anyway : we can see that the addition of such an operator reduce the temperature range where the observables are correctly reproduced . as noted above the specific heat is more subjected to errors than the susceptibility at contains higher - body operators . from numerical inspection we saw that the lce rather well performs in the high temperature regime even for relatively large values of @xmath106 , but deviates from exact results at lower temperature even for small values of @xmath106 as shown in figure [ cv_chi ] . is varied . the value @xmath114 is held fixed at the value @xmath5 . the figure on the right includes the four - spin interaction whose coupling is fixed at the value @xmath115 . the inset shows the ratio between the predicted values of specific heat and inverse susceptibility and the exact ones . both figures are obtained keeping @xmath116 correlation functions and using third order loop resummed expansion to reconstruct the couplings , as presented in @xcite.,title="fig : " ] is varied . the value @xmath114 is held fixed at the value @xmath5 . the figure on the right includes the four - spin interaction whose coupling is fixed at the value @xmath115 . the inset shows the ratio between the predicted values of specific heat and inverse susceptibility and the exact ones . both figures are obtained keeping @xmath116 correlation functions and using third order loop resummed expansion to reconstruct the couplings , as presented in @xcite.,title="fig : " ] [ reconstruction ] ( empty symbols ) and the values of the couplings really used to generate the correlations ( filled circles ) - the full line is a guide to the eye . the figures refers to hamiltonian with parameters @xmath117 , @xmath118 , @xmath119 ( left ) and @xmath120 , @xmath121 , @xmath122 ( right).,title="fig : " ] ( empty symbols ) and the values of the couplings really used to generate the correlations ( filled circles ) - the full line is a guide to the eye . the figures refers to hamiltonian with parameters @xmath117 , @xmath118 , @xmath119 ( left ) and @xmath120 , @xmath121 , @xmath122 ( right).,title="fig : " ] ( top ) and the inferred couplings @xmath123 ( bottom ) for the @xmath29 s allowed by translational symmetry ( filled blue circles ) and exact values ( empty purple triangles ) . the black diamonds refer to lce results obtained with the perturbative expansion up to third order @xcite . the figure at the bottom is a logarithmic plot of the absolute values of the @xmath123 s , while the inset gives for comparison the linear plot of the same @xmath123 s . the figures refer to hamiltonian with parameters @xmath117 , @xmath118 . the reconstruction range is @xmath124 . in abscissa are reported the @xmath29 s denoting the various coupling and correlators with the subset notation introduced in section [ notation_and_statement ] , e.g. @xmath125 is the correlation of the subset @xmath126 , @xmath127 is the correlation of the subset @xmath128 and so on . ] in this section we consider examples where we can not access the full knowledge of our system : our inversion procedure will therefore yield approximate results . first , we consider an ising model with an exponentially decaying two - body interaction @xmath129 since the interaction now is not formally of `` finite range '' i.e. it does not vanish for distances beyond a given value of @xmath2 , the transfer matrix method is not viable ( although we still may perform a finite - range scaling in the size of the transfer matrix @xcite ) . the set of synthetic correlation functions is generated by a monte carlo method . of course we will not record all of the correlation function , but we will fix a maximal range @xmath130 , thus we will have to measure on the order of @xmath131 correlation functions . the results for such a reconstruction are shown in figure [ reconstruction ] for two values of the parameters . we see that the agreement improves as the value of @xmath130 is increased ( at the expense of calculating a larger number of correlation functions ) . in figure [ gj1 ] a full set of the correlations and inferred couplings is shown ; if we look at the lower panel , the nonzero couplings are clearly singled out ( even for a value of @xmath130 as low as @xmath132 ) , thus our reconstruction procedure gives useful hints to build a faithful model of an unknown system . in figure [ gj1 ] results obtained with the lce are also reported : one sees that there is a good agreement for the considered value of the temperature between the lce results and the findings obtained using our reconstruction procedure . in order to test the procedure on a system with more - body couplings we consider the hamiltonian @xmath133 which includes three- and four - body interactions . as we see in figure [ gj2 ] , even in this case the reconstruction procedure gives useful hints on the couplings present in the system , although some of the inferred couplings , which were zero in the starting model , are predicted to be of comparable size to the nonzero ones ( especially the @xmath134 coupling ) . this is due to the finite reconstruction range and to the , albeit small ( indeed smaller than the symbols in the figures [ gj1 ] , [ gj2 ] , [ gj3 ] ) , errors in the determination of the correlation . this implies that in order to clearly distinguish the contribution of the different couplings , the correlation should be known with high accuracy . as for modeling purposes , this is not a problem since the values of the coupling obtained give rise to a set of correlations not distinguishable from the original one . for reference we also plot in figure [ gj2 ] the results of the tap equation approach developed in @xcite , which of course provides no information on the multispin couplings , but as far as one - body operators and two - body operators are concerned this approach at the temperature considered in figure [ gj2 ] performs very well . finally we examine a model with power - law decay of the interaction @xmath135 as can be seen in figure [ gj3 ] ( with @xmath136 ) , the results are good also in this case : it can be generally observed that the reconstructed interactions are higher than the exact ones , due to the fact that the interaction within the reconstruction range have to account for the interactions lying outside this range . except for the black diamonds which are obtained with the tap approach @xcite . the figures refer to hamiltonian with parameters @xmath137 , @xmath118 , @xmath138 , @xmath139 . the reconstruction range is @xmath124 . the arrows mark the multispin interactions . ] . measured values of the correlations @xmath55 ( top ) and inferred couplings @xmath123 ( bottom ) for the power - law decaying hamiltonian with parameters @xmath140 , @xmath136 . the reconstruction range is @xmath124 . ] in this section we briefly discuss how our previous results can be used in presence of mean - field long - range interactions , showing that our inversion approach may be used on this class of systems . we consider a system with energy @xmath141 of general form , i.e. a non - linear function of the correlators . the number of couplings entering the energy @xmath141 should still equal the number of independent correlation functions in order to perform , at least in principle , the inversion procedure . such an energy will be denoted by @xmath142 where the index @xmath93 runs over the mean - field couplings . by requiring the free energy @xmath143 to have a minimum when the @xmath144 are set to the known values will give the equations implicitly determining all the couplings , including the mean - filed ones @xmath145 . when the energy is differentiable these equations read @xmath146 this set of equations will in general have multiple solutions or possibly no solutions at all . if more solutions are present the one ( or ones ) rendering the free energy minimal should be chosen . the points where the absolute minimum of the free energy branches or it changes discontinuously will signal a phase transition . as described in the section [ inversion_formula ] some values of the couplings may be known in advance thus reducing the number of equations to be solved . another possibility is that a function of the coupling is fixed ; a notable example being the energy itself , corresponding to the microcanonical description of the system . we remark that in the class of models we consider all of these steps can be carried out exactly since the explicit form of entropy is known . as an example we may consider the first model examined in section [ simple_examples ] ( @xmath84 , @xmath85 ) by adding mean - field two - body couplings . instead of the energy @xmath147 ( where @xmath148 and @xmath149 ) we will set @xmath150 . the appearance of the nonlinear term @xmath151 in the energy is due to the presence of non - local mean - field operators in the spin hamiltonian like @xmath152 . it should be noted that such an operator is not uniquely defined , e.g. the operator @xmath153 and other kac - rescaled nonextensive potentials give rise to the term @xmath154 in the energy density at the thermodynamic limit when evaluated on a state with magnetization @xmath93 and nearest - neighbour correlation @xmath94 @xcite . if we set @xmath155 we get the usual mean - field model , otherwise we obtain a model with competing mean - field and short - range coupling introduced by kardar @xcite , exhibiting a complex phase diagram which shows nonequivalence between the canonical and microcanonical description @xcite ( for a review on inequivalence between ensembles and other issues concerning nonextensive systems see @xcite ) . as already observed the entropy of such a model within our approach is easily computed [ see ] and it could be generalized thus allowing one to treat one - dimensional models possessing multiple competing finite - range and mean - field interactions . in this paper we presented the explicit solutions of the inverse ising problem for a one - dimensional translational invariant model with arbitrary finite - range multispin interactions once a number @xmath156 ( where @xmath2 is the range of the interactions ) of independent correlations is known . when applied to unknown systems this method correctly detects arbitrary interactions ; our results are then applied to systems with a range extending beyond the one set by maximum distance of the spins of the recorded correlation functions , giving useful hints on the interactions that should be kept in an effective model . as an application , we reconstructed the couplings of chain ising hamiltonians having exponential or power - law two - spin plus three- or four - spin couplings . we also discussed the generalization of the method to ladders . mean - field interactions can be also included in the framework , allowing us to describe systems exhibiting phase transitions . the presence of both finite - range ( local ) and mean - field ( nonlocal ) interactions can give rise to interesting competition effects greatly enhancing the descriptive power of the models we can exactly solve with our techniques . our results provide then a theoretical laboratory where different approximate inverse ising techniques can be benchmarked against the exact results obtained using our method : in the paper we performed such comparison is some illustrative examples . the one - dimensional inverse ising problem we have solved in the present paper is analogous to what may be called the inverse markov chain problem : given a specific set of correlations at equilibrium , find the corresponding transition rates . the two inversion ( ising and markov ) problems are related since it is possible to associate an ising model to an equilibrium markov ising chain with finite memory in full generality : to show it , for definiteness let us consider ising variables ( although extensions to other discrete state spaces is straightforward ) . the finite - range @xmath2 in our solution of the inverse one - dimensional ising problem is the counterpart of the finite memory in the inverse markov problem : let the state of the next @xmath22 spins be ruled by the state of the preceding @xmath62 spins . we put the new spins on the right side of the old ones : the time of the markov process is increasing from left to right . the correlations impose constraints on the transition rates : it turns out that the number of independent correlations required to solve the inverse markov problem is the same as the number needed to solve the ( related ) inverse ising problem in one dimension of range @xmath2 and period @xmath22 , i.e. the number of independent correlations is @xmath67 . adapting the procedure discussed in section [ inversion_formula ] which lead to it is possible to compute the transition rates from the state @xmath157 of @xmath62 spins to the state @xmath158 of @xmath22 spins . these transitions are given by @xmath159 where the @xmath160 s are calculated according to formula where the input correlations @xmath66 s are plugged in , and @xmath161 is the configuration of @xmath2 spins obtained by juxtaposing the states @xmath157 ( on the left ) and @xmath158 ( on the right ) . the transition matrix obtained this way is a @xmath162 by @xmath163 matrix ; in order to have a square matrix we have to fold more steps of the the transition matrix until we obtain the probability to go from a set of @xmath164 to the next @xmath164 spins @xcite . the transition matrix satisfies detailed balance by construction , therefore this is a reversible markov chain . this mapping has already been worked out in a discrete different form in @xcite where the connection to discrete statistical models is also discussed . according to the previous discussion , we may thus associate an ising model to the an equilibrium markov ising chain with finite memory in full generality , allowing us to treat systems where one direction ( typically time ) is singled out . as a possible example deserving future investigation one could consider time - series of financial data and try to estimate with the procedure discussed before the transition probabilities of the associated guessed markov chain in order to test the validity of such a description . it is intended - in order to apply the previous results - that the analyzed data should be discretized on a timescale such that nontrivial correlations occur , and that the whole time of observation is such that the system can be reliably considered at equilibrium . obviously a way to encode significant information in an ising variable has to be devised , being this in general a nontrivial task ; for example , we may think of `` up '' and `` down '' spin corresponding to a price raise or lowering respectively . the next step would be to analyze the correlations among different time - series of data to determine if and how correlations among different stocks occur . for example the ising ladder system depicted in figure [ ladder_to_chain ] may reproduce the correlations among two stocks whose state depends on the value of the other stock at the same timestep and on the value of the same or other stock at the previous timestep ( for simplicity , in figure [ ladder_to_chain ] odd interactions are not depicted ) . , it should be noted that the inclusion of the interaction dubbed @xmath165 in figure [ ladder_to_chain ] may reproduce some kind of nontrivial many - body interaction among the stocks . extending the number of chains in the ladder system and/or the range allows us to treat larger sets of stocks with longer correlations in time . we think that studying stationary time - series of correlated data using the techniques presented in this paper ( and as well the mapping onto markov chains discussed in this section ) will be an interesting subject of future research . in perspective , one could apply the method here discussed to datasets and/or statistical mechanics models which are supposed to be described by effective one - dimensional ising chains near the thermodynamical limit . to this aim , one should address in the future a treatment of the case where large errors in the measured correlation functions are present and/or some of the correlations are missing ; our exact result could be a good starting point to move in that direction . next , our results could be extended in higher dimensional cases ( where of course one does not expect to find closed formulas ) , hierarchical or tree - like models . some preliminary results in the two dimensional case seem to indicate that this approach leads to equations resembling the dobrushin - lanford - ruelle ones @xcite . another interesting direction could be to use the a renormalization group approach on the correlation functions , in order to study how the couplings determined by correlations at some scale @xmath2 are related to the ones computed at a larger scale @xmath166 . _ acknowlegments : _ we wish to thank i. mastromatteo and m. marsili for many very useful discussions . this work has been supported by the grants instans ( from esf ) and 2007jhlpez ( from miur ) . in this appendix we work out the transfer matrix for the general translational invariant ( @xmath85 ) ising model with range @xmath167 and we check that the results obtained analytically and numerically with the transfer matrix formalism are fully consistent with the predictions of the formula . we start by briefly recalling the method ( see e.g. @xcite ) . in general the transfer matrix @xmath168 is built by identifying the @xmath169 states of a block of spins @xmath170 with independent and orthogonal vector of a space of dimension @xmath169 such that the matrix elements of @xmath168 are @xmath171 where @xmath172 is the interaction energy among two consecutive blocks of spins @xmath173 and @xmath174 ( @xmath173 is placed to the left of @xmath174 ) , @xmath175 is the interaction energy among the spins belonging to the same block . the vector corresponding to the spin state @xmath173 is denoted , using the ket notation , by @xmath176 . the size of the blocks @xmath170 has to be chosen according to the range and size of the unit cell @xmath22 of the system in order to have all of the interaction terms contained in @xmath177 or @xmath178 . the partition function of the system of size @xmath11 is simply given by @xmath179 so that in the infinite size limit the free energy per unit cell may be written in terms of the largest eigenvalue @xmath180 of @xmath168 : @xmath181 the existence and unicity of @xmath180 is guaranteed by perron - frobenius theorem , being all elements of @xmath168 strictly positive ( for nonvanishing temperature ) . the correlation functions may be obtained just by differentiation of the free energy @xmath182 . we start with the simple example of @xmath84 which has been worked out in section [ simple_examples ] . the two independent couplings will be denoted as usual by @xmath183 ( magnetic field ) and @xmath184 ( nearest - neighbour coupling ) . the @xmath6 by @xmath6 transfer matrix @xmath168 , with the states identification @xmath185 and @xmath186 , reads @xmath187 the free energy is then given by @xmath188.\ ] ] differentiating @xmath189 with respect to @xmath87 and @xmath88 gives the magnetization @xmath93 and the nearest - neighbour correlation @xmath94 respectively : @xmath190 the inversion of the above formulas with respect to @xmath87 and @xmath88 yields the expressions . now we consider a model with only even interactions and range @xmath191 , i.e. containing the coupling @xmath192 and @xmath193 which we dub @xmath7 model : the corresponding hamiltonian is given by with @xmath110 . this model may be mapped onto the previous example ( @xmath84 , @xmath85 ) by introducing `` kink '' variables @xmath194 with the identifications @xmath195 , @xmath196 ( we do not report the corresponding results ) . in figure [ j1j2sm ] we plot the couplings reconstructed according to the low - coupling expansion ( lce ) introduced in @xcite against the exact results which can be found by the method discussed in this paper or by the transfer matrix approach . the lce allows to infer the magnetic fields and the two - body couplings from the two - body correlators and magnetizations . in @xcite the lce has been carried out , in the zero field case , up to seventh order in the correlations with loop resummation . we have used the lce as both the order of the expansion and the number of correlation function we assume to know are increased . the maximal range of two - body correlations which are used as input is denoted by @xmath109 . in figure [ j1j2sm ] we depict the reconstructed couplings @xmath197 , @xmath198 and the exact ones for different values of @xmath109 . the ratio @xmath199 is kept fixed , thus @xmath104 serves as inverse temperature . as expected as the temperature is lowered the agreement gets worse , and it may be noticed that the inclusion of higher order terms in this case does not significantly improve the performance of the inversion : as one can see , the lower order results depicted in the left panel of figure [ j1j2sm ] is more reliable at lower temperatures ( this is the reason why in figure [ cv_chi ] we employed the third order lce ) . in the right panel of figure [ j1j2sm ] we can see how the the increase of @xmath109 improves the quality of the inversion but beyond a given range the reconstructed couplings @xmath200 settle to a value which , in the lower temperature case examined , deviates from the exact one . . the various lines denote how many correlation functions are kept in the reconstruction formulas as indicated . the continuous and dotted line refer respectively to the third and seventh order in the lce of @xcite ( including the loop contributions ) . on the right the ratio of reconstructed @xmath201 ( above ) , @xmath202 ( below ) and the exact value of @xmath104 in terms of the reconstruction range @xmath109 are shown . only the result for the seventh order lce is shown . in both figures the green straight lines are the exact values and the nearest - neighbour coupling @xmath105 is set to the half of @xmath104.,title="fig : " ] . the various lines denote how many correlation functions are kept in the reconstruction formulas as indicated . the continuous and dotted line refer respectively to the third and seventh order in the lce of @xcite ( including the loop contributions ) . on the right the ratio of reconstructed @xmath201 ( above ) , @xmath202 ( below ) and the exact value of @xmath104 in terms of the reconstruction range @xmath109 are shown . only the result for the seventh order lce is shown . in both figures the green straight lines are the exact values and the nearest - neighbour coupling @xmath105 is set to the half of @xmath104.,title="fig : " ] we move on to the next example , the @xmath191 case with no restriction on the symmetry of the couplings . identifying the states as @xmath203 , @xmath204 , @xmath205 , @xmath206 and the couplings as @xmath207 , @xmath208 , @xmath209 , @xmath210 , the transfer matrix reads @xmath211 proceeding as before , we can obtain the entropy and correlation functions . the comparison between the resulting obtained findings and our inversion formula are shown in the left panel of figure [ tm4x48x8 ] . the plot shows the numerically calculated entropy and the analytical one where the numerical correlations are plugged in , for some values of the coupling constants . the entropy in this case is given by the expression ( the subscripts in the @xmath94 s just indicate to which coupling they are conjugated ) : @xmath212 as a last case we consider a translation invariant chain with range @xmath75 : identifying the states as @xmath213 , @xmath214 , @xmath215 , @xmath216 , @xmath217 , @xmath218 , @xmath219 , @xmath220 and the couplings as @xmath207 , @xmath208 , @xmath209 , @xmath221 , @xmath222 , @xmath223 , @xmath224 , @xmath225 , the @xmath226 by @xmath226 transfer matrix reads : @xmath227 where the @xmath228 matrices @xmath229 are @xmath230 results are shown in figure [ tm4x48x8 ] ( we do not write down the entropy for this case ) . by inspecting figure [ tm4x48x8 ] we find that the agreement between the two approaches is complete . we do not report here the other checks we performed for higher values of @xmath2 . ( left ) for the transfer matrix and @xmath231 ( right ) for the transfer matrix . the values of the remaining couplings are specified on the @xmath232-axes of the plots . the arrows point to the relevant @xmath232-axis . the continuous curves are obtained by numerical calculation , while the dots are calculated by means of .,title="fig : " ] ( left ) for the transfer matrix and @xmath231 ( right ) for the transfer matrix . the values of the remaining couplings are specified on the @xmath232-axes of the plots . the arrows point to the relevant @xmath232-axis . the continuous curves are obtained by numerical calculation , while the dots are calculated by means of .,title="fig : " ]

pion multiplicity distribution and their associated fluctuation and correlations have been of interest for several reasons . several models predict large fluctuations such as the disoriented chiral condensate @xcite and in density fluctuations from droplets arising in a first order phase transition @xcite . a well known procedure for studying correlations uses the bose - einstein symmetries associated with pions in a hanbury brown - twiss analysis @xcite . such an analysis gives information about the space time history of the collision through measurements of source parameters . if the density of pions becomes large , bose - einstein correlation may also lead to a strongly emitting system which has been called a pion laser @xcite . the pion laser model has been recently solved analytically by t. csrg and j. zimanyi @xcite . the importance of bose - einstein correlations has also been illustrated in the observation of a condensation of atoms in a harmonic oscillator or laser trap @xcite . previous interest in pionic distributions have centered around the possibility of intermittency behavior @xcite and fractal structure based on parallels with turbulent flow in fluids . a distribution widely used to discuss such features has been the negative binomial ( nb ) distribution @xcite with its associated clan structure @xcite and kno scaling feature @xcite . kno scaling properties have been interpreted in terms of a phase transition associated with a feynman - wilson gas @xcite . various other issues associated with pions include evidence for thermalization @xcite , critical point fluctuations @xcite , fluctuations from a first order phase transition @xcite , charge particle ratios and question of chemical equilibrium @xcite , the behavior of fluctuations in net charge in a qg plasma for transition @xcite . for lower energy heavy ion collisions , multifragmentation of nuclei takes place . the fragment distribution can also be described statistically by considering all the possible partition of @xmath4 nucleons into smaller clusters @xcite . this study gives a tool for the description of nuclear multifragmentation distributions @xcite , nuclear liquid - gas phase transition @xcite , critical exponent , intermittency , and chaotic behavior @xcite of nuclear multifragmentation . the same model can describe pionic distribution . this possibility arises in our approach which is based on feynman path integral methods where symmetrization of @xmath4 bosons or anti - symmetrization of @xmath4 fermions leads to a cycle class decomposition of the permutations associated with these symmetries . the correspondance comes from the identification of clusters of size @xmath5 and the cycles of length @xmath5 in a permutation as discussed below . in nuclear multifragmentation the number of nucleons , @xmath4 , is fixed and thus a canonical partition function approach is appropriate . by contrast , the number of pions , @xmath4 , in particle production is not fixed and thus we need to use a grand canonical partition . using this parallel , we also note that some results from cluster yields can be carried over into particle multiplicity distributions and associated properties . namely , we show the importance of the fisher critical exponent @xmath3 and relate it to one of the parameters called @xmath0 in our approach which has three main parameters @xmath0 , @xmath1 , @xmath2 . moreover , taking special values of @xmath0 , or equivalently @xmath3 , reduces our unified model to various specific cases that are frequently used in particle production phenomenology . quantities that appear in this development can also be related combinants as will be discussed . in so doing we can give a physical significance to the combinants and show how the fisher exponent appears in them and how the resulting hierarchical structure and void scaling relations also depend on its value . in next section a summary of a generalized statistical model based on a grand canonical partition function will be given . this generalized model can be used to study multiplicity distributions associated with particle production such as at rhic . various models of particle multiplicity distribution using the grand canonical partition function will then be developed . in section ii - b , we will summarize the multiplicity distribution and its various moments of all order for a general grand canonical ensemble which we will then use for particle production . the physical meaning of the parameters used in this model is given here . the differences and relations between the canonical nuclear multifragmentation and the grand canonical multiparticle production are sumarized in sect . ii - c . in sect . ii - d and e , moments ( especially the mean and the fluctuation ) in the general grand canonical model are related to variables used in other existing standard models in describing multiparticle production . the connection of the general grand canonical model developed here to existing standard models are summarized in sect . moreover , we derive a generalized model ( hga ) of the grand canonical partition which can further be reduced to a geometric , negative binomial , and lorentz / catalan model . various moments and multiplicity distribution for a generalized hga model are summarized in sect . ii - g and another generalized model ( grw1d ) in sect . ii - h . in sect . [ sectcomp ] , we compare various statistical properties between different models within the generalized hga model and with data for charged particle distribution in jets of l3 and h1 . fluctuation and multiplicity distribution in hga are compared with other standard models of multiparticle production . consider a system composed of @xmath6 different types of species or objects which could be the fragments in a fragmentation or in a cycle class description . any event of such a system can be associated with a vector @xmath7 or @xmath8 where the non - negative integer @xmath9 is the number of individuals of species @xmath5 . for example @xmath9 can be the number of clusters of size @xmath5 or the number of cycles of length @xmath5 in a given permutation of @xmath10 particles . the later is important for bose - einstein and fermi - dirac statistics and particle multiplicity distributions . a general block picture of @xmath11 is shown in fig.[fig1]a . fig.[fig1]b shows how the various partition can be developed as an evolution from successively smaller systems . the number of species @xmath6 can be infinity in general . ( 25,8)(7.5,0.0 ) ( 0.5,0.5 ) ( 12.2,6.2)*a ) * ( 8.0,0.25)*@xmath12 * ( 8.01 , 1.51)(0,1)4.5 ( 8.0 , 5.5)(0.5,0)2 ( 8.0 , 5.0)(0.5,0)2 ( 8.0 , 4.5)(0.5,0)4 ( 8.0 , 4.0)(0.5,0)5 ( 8.0 , 3.5)(0.5,0)5 ( 8.0 , 3.0)(0.5,0)7 ( 8.0 , 2.5)(0.5,0)7 ( 8.0 , 2.0)(0.5,0)7 ( 8.0 , 1.5)(0.5,0)11 ( 8.99 , 5.0)(0,1)1.0 ( 9.99 , 4.5)(0,1)0.5 ( 10.49 , 3.5)(0,1)1.0 ( 11.49 , 2.0)(0,1)1.5 ( 13.49 , 1.5)(0,1)0.5 ( 8.01 , 1.51)(1,0)5.5 ( 11.51 , 2.01)(1,0)2.0 ( 10.51 , 3.51)(1,0)1.0 ( 10.01 , 4.51)(1,0)0.5 ( 9.01 , 5.01)(1,0)1.0 ( 8.01 , 6.01)(1,0)1.0 ( 7.5,6.5)(0,0.01)2(1,0)0.5 ( 9.5,6.5)(0,0.01)2(-1,0)0.5 ( 8.0,6.2)(1.0,0.6)*@xmath9 * ( 7.6,3.2)(0,0.01)2(0,-1)1.7 ( 7.6,4.3)(0,0.01)2(0,1)1.7 ( 7.4,3.5)(0.3,0.6)[r]*@xmath5 * ( 15.5,0 ) ( 10,8)(0,0 ) ( 0.5,0.5 ) ( 8,6.2)*b ) * ( 3.5 , 7.0 ) ( 4.2,7.3)@xmath13 ( 3.6,6.9)(-3,-2)1.05 ( 3.9,6.9)(2,-3)0.85 ( 2.0 , 5)(0,0.5)2 ( 2.7,5.6)@xmath14 ( 2.05,4.9)(-2,-1)0.65 ( 2.4,4.9)(3,-2)1.1 ( 4.5 , 5.0)(0.5,0)2 ( 5.7,5.4)@xmath15 ( 4.52,4.9)(-1,-1)0.78 ( 5.45,4.9)(3,-4)0.95 ( 1 . , 3.0)(0,0.5)3 ( 1.7,4.0)@xmath16 ( 1.1,2.9)(-1,-1)0.75 ( 1.4,2.9)(1,-4)0.3 ( 3.5 , 3.0)(0,0.5)2 ( 4.0 , 3.0 ) ( 4.2,3.8)@xmath17 ( 3.6,2.9)(-4,-3)1.6 ( 3.9,2.9)(0,-1)1.75 ( 4.4,2.9)(3,-4)1.3 ( 6.0 , 3.0)(0.5,0)3 ( 7.1,3.7)@xmath18 ( 6.4,2.9)(-1,-3)0.6 ( 7.3,2.9)(1,-2)1.1 ( 0 . , 0.0)(0,0.5)4 ( 0.7,1.6)@xmath19 ( 1.5 , 0.0)(0,0.5)3 ( 2.0 , 0.0 ) ( 2.15,1.1)@xmath20 ( 3.5 , 0.0)(0,0.5)2 ( 4.0 , 0.0)(0,0.5)2 ( 4.7,0.7)@xmath21 ( 5.5 , 0.0)(0.5,0)3 ( 5.5 , 0.5 ) ( 6.2,0.8)@xmath22 ( 8.0 , 0.0)(0.5,0)4 ( 9.6,0.7)@xmath23 various probability distributions related with this system can be developed by assigning an appropriate weight @xmath24 to each type @xmath5 . a weight @xmath25 is then given to each event @xmath11 and the type of weight that will be considered has the structure : @xmath26 \label{weight}\end{aligned}\ ] ] the @xmath27 are gibbs factorials . such a weight structure appears in feynman s path integral approach and bose - einstein problems @xcite . the @xmath24 will be given below and contains various physical quantities . in sect.[sectiib ] we will show that the @xmath24 s are also the combinants which in turn can be related to the factorial cumulants . summing the weight @xmath25 over all the possible events of @xmath11 , the grand canonical partition function @xmath28 of the system is given as @xmath29 = \exp\left[\sum_{k=1}^n x_k\right ] \label{grandptf } \end{aligned}\ ] ] the last equation holds due to the form of eq.([weight ] ) of the weight @xmath25 , i.e. , the factor @xmath30 is the @xmath9-th order expansion term of @xmath31 . introducing other quantities @xmath32 to each individual entity or group of type @xmath5 , @xmath33 , we can define a canonical partition function @xmath34 with a fixed @xmath4 as @xmath35 \label{canoptf } \end{aligned}\ ] ] with @xmath36 . here @xmath37 is the summation over all events with a fixed value of @xmath4 , i.e. , over a canonical ensemble of a fixed @xmath4 , and the @xmath38 is a summation over all the possible values of @xmath4 ; it becomes @xmath39 for the case of @xmath40 with positive integer @xmath5 . the multiplicity of a partitions is @xmath41 the case @xmath40 is encountered in fragmentation problems and permutation problems . various type of @xmath32 can be used depending on the physics related with the quantity @xmath4 as discussed in ref.@xcite . the physics of the canonical ensemble depends on the choice of the quantity @xmath32 @xcite . if we take @xmath42 , then the canonical partition function @xmath34 is the @xmath43 dependent term of the grand canonical partition function @xmath28 . if we take @xmath44 , then @xmath1 counts the number of clusters @xmath45 explicitly and the @xmath46 dependent term of the grand canonical partition function @xmath28 is the partition function for events with multiplicity @xmath45 . there always is at least one event having @xmath47 , i.e. , the event where all @xmath9 s are zero , @xmath48 . thus @xmath49 if all @xmath32 are non zero positive since then there are no other possible events having @xmath47 . due to the form of the weight @xmath25 given by eq.([weight ] ) the canonical partition function @xmath34 satisfies a recurrence relation @xcite @xmath50 this relation is nothing but the constraint eq.([asum ] ) in terms of the mean @xmath51 using eq.([factma ] ) . for non - zero positive @xmath32 , there is no case having @xmath52 , i.e. , @xmath53 for @xmath52 and thus the @xmath54 can be obtained by the recurence relation of eq.([recur ] ) starting from @xmath49 . in a canonical ensemble of fixed @xmath4 , we can define a probability distribution of a specific partition @xmath11 as @xmath55 with this probability , various mean values , fluctuations and correlations of the number of species @xmath9 can be evaluated as a ratio of canonical partition functions for two different values of @xmath4 such as @xcite @xmath56 thus we have @xmath57 . the recurrence relation of eq.([recur ] ) then follows simply using the fact that @xmath58 . this distribution has been used in describing various fragment distributions in nuclear multifragmentation @xcite . now knowing the partition functions @xmath28 and @xmath34 , we can associate a probability @xmath59 of the system to have a fixed value of @xmath4 in a grand canonical ensemble as @xmath60_{z=0 } \label{pax } \\ z(\vec x , z ) & = & \sum_a z_a(\vec x ) z^a = \exp\left[\sum_k x_k z^{\alpha_k}\right ] \label{grandzxz}\end{aligned}\ ] ] the last step follows from the fact that the @xmath43 power term of @xmath61 is @xmath34 if we put @xmath62 . for particle production @xmath59 is the probability of having @xmath4 particles . thus the generating function @xmath61 of @xmath63 can also be looked at as a grand canonical partition function with the weight @xmath24 replaced to be @xmath64 , where the variable @xmath2 counts @xmath4 explicitly and @xmath65 . if we consider @xmath66 then @xmath2 has two roles ; one as a weight which is assigned the same to each constituent and another as a generating parameter of the probability @xmath59 . for the case that @xmath49 , the void probability @xmath67 @xcite is the inverse of the grand canonical partition function , i.e. , @xmath68 inversely the canonical partition function @xmath34 is the probability @xmath63 normalized or rescaled by the void probability @xmath67 , i.e. , @xmath69_{z=0 } = \frac{p_a(\vec x)}{p_0(\vec x)}\end{aligned}\ ] ] another type of generating function of @xmath63 that is frequently used may be defined as @xmath70 \right ] \label{genu } \\ p_a(\vec x ) & = & \frac{1}{\gamma(a+1 ) } \left[\left(-\frac{d}{d u}\right)^a g(\vec x , u)\right]_{u=1}\end{aligned}\ ] ] we see that @xmath71 , @xmath72 and @xmath73 . once the probability @xmath59 is determined , various statistical quantities can be evaluated . also the grand canonical partition function @xmath74 is given once we know the thermodynamic grand potential @xmath75 this result can be used to study the statistical properties of a system . moreover various moments and cumulants , mean values and fluctuations may be obtained using the generating function @xcite . this subsection gives general expression for various quantities that will be used later when we discuss specific models . since the probability of a specific event @xmath11 in a grand canonical ensemble is given by @xmath76 the mean of a quantity @xmath77 in a grand canonical ensemble is related to the mean of @xmath77 in a canonical ensemble as @xmath78 we can easily show that @xmath79 this result shows that the weight factor @xmath24 in this model is the mean number @xmath80 in a grand canonical ensemble and that the mean multiplicity @xmath81 is related to the thermodynamic potential . the @xmath82-th power moment of @xmath4 and its factorial moments are given simply by @xmath83_{z=1 } \\ \left<\frac{\gamma(a+1)}{\gamma(a - m+1)}\right>(\vec x ) & \equiv & \sum_{\vec n } \frac{\gamma(a+1)}{\gamma(a - m+1 ) } p(\vec x , \vec n ) = \frac{1}{z(\vec x ) } \left[\left(\frac{d}{d z}\right)^m z(\vec x , z)\right]_{z=1 } \end{aligned}\ ] ] similarly the @xmath82-th cumulants , which is the power moments of @xmath32 , and the factoral cumulants are @xmath84_{z=1 } = \sum_{k=1}^\infty \alpha_k^m x_k \label{alpham } \\ f_m(\vec x ) & \equiv & \left < \sum_{k=1}^n \frac{\gamma(\alpha_k+1)}{\gamma(\alpha_k - m+1 ) } n_k\right > = \left[\left(\frac{d}{d z}\right)^m \ln z(\vec x , z)\right]_{z=1 } \nonumber \\ & = & \left[\left(- \frac{d}{d u}\right)^m \ln g(\vec x , u)\right]_{u=0 } = \sum_{k=1}^n \frac{\gamma(\alpha_k+1)}{\gamma(\alpha_k - m+1 ) } x_k \label{factfm } \end{aligned}\ ] ] due to eqs.([grandzxz ] ) and ( [ grandpot ] ) we can see easily that , for the power moments of @xmath32 , @xmath85 the power moments of @xmath32 are directly related to the power moments of @xmath4 measured from the mean @xmath86 , i.e. , the cumulants @xmath87 are same with the central moments of @xmath4 . this simple relation does not hold for @xmath88 but we can evaluate them starting from @xmath89 using the recurrence relation @xmath90 similarly the @xmath82-th factorial cumulants @xmath91 , which is the factorial moments of @xmath32 , can be found using recurrence relation @xmath92 starting from @xmath93 the reduced factorial cumulants @xmath94 defined in ref.@xcite corresponds to the factorial cumulants @xmath91 normalized with mean number @xmath95 as @xmath96 thus with @xmath97 and @xmath98 . the cummulants are directly related with the generating functions as @xmath99 = \exp\left[\sum_{m=1}^\infty \frac{(\ln[1-u])^m}{m ! } \bra\alpha^m\ket(\vec x)\right ] \\ z(\vec x , z ) & = & \exp\left[\sum_{m=0}^\infty \frac{(z-1)^m}{m ! } f_m(\vec x)\right ] = \exp\left[\sum_{m=0}^\infty \frac{(\ln z)^m}{m ! } \bra\alpha^m\ket(\vec x)\right]\end{aligned}\ ] ] the generating function @xmath61 differs from the generating function @xmath100 only by an extra term of @xmath101 in their exponent . the relation between the @xmath24 s and the @xmath102 shows that the @xmath24 s are also the combinants of ref.@xcite . in the approach presented here , the combinants are given an underlying significance through the partition weight @xmath25 of eq.([weight ] ) . in turn the combinants @xmath24 can be related to the factorial cumulants @xmath91 defined by @xmath103 the factorial cumulants @xmath91 are the @xmath82-th order factorial moments of @xmath32 of eq.([factfm ] ) . thus @xmath104 for @xmath40 which is the eq.([factfm ] ) with @xmath105 . the normalized factorial cumulant , i.e. , the reduced cumulant , is @xmath106 for a negative binomial ( nb ) distribution , this result of eq.([kappam ] ) shows that @xmath94 for nb has an hierarchical structure of a distribution at the reduced cumulant level which was realized for the nb distribution in ref.@xcite . this result will be generalized later . also using the above power moments and factorial moments we can study voids and void scaling relation , hierarchical structure , combinant and cummulant properties which will be discussed below . since our approach was first used to discuss multifragmentation and then later extended to include multiparticle production , we briefly mention some of difference between multifragmentation and multiparticle production . we also show how the fisher exponent @xmath3 , which initially appeared in cluster yields around a crititical point , may manifest itself in particle production yields . in nuclear multifragmentation , @xmath40 is the number of nucleons in a fragment and @xmath9 is the number of fragments of size @xmath5 . the total number of fragments is @xmath107 and the total number of nucleons is @xmath108 . in nuclear multifragmentation , the total number of nucleons @xmath4 is usually fixed and we study the distribution in size @xmath5 of the mean multiplicity @xmath109 of fragments of size @xmath5 in a canonical ensemble . in multiparticle production the @xmath4 is the total number ( multiplicity ) of produced particles and is not fixed and we study the multiplicity distribution of produced particles @xmath110 as a function of @xmath4 in a grand canonical ensemble . the mean multiplicity @xmath111 and @xmath51 in a canonical ensemble are related with grand canonical ensemble as @xmath112 the weight @xmath24 can be determined experimentally from the mean multiplicity @xmath80 of cluster size @xmath5 in a grand canonical ensemble . in our initial fragmentation studies @xcite we have used @xmath113 . this choice gives @xmath114 and thus @xmath115 as @xmath116 where the @xmath117 with @xmath118 the chemical potential . from eq.([factma ] ) , @xmath119 , thus @xmath120 as @xmath116 in the case of nuclear fragmentation in the canonical ensemble of fixed @xmath4 for @xmath113 and the yields fall as a power law @xmath121 at a critical point ( @xmath122 ) for this choice of @xmath24 . the yield in the grand canonical ensemble is @xmath123 and this becomes the same as @xmath120 in @xmath116 limit for @xmath124 with @xmath125 . in general , cluster yields fall for large @xmath5 as a power law @xmath126 at a critical point where @xmath3 is the fisher critical exponent . the above choice of @xmath24 can be generalized to @xmath127 so that the cluster yields fall as @xmath126 . the @xmath3 determines the grand canonical partition function @xmath128 = \exp[x \sum_k z^k /k^\tau]$ ] . forms for @xmath24 whose asymptotic behavior is @xmath129 can also be used . in section ii.f we will give various choice for @xmath24 for particle production . for example , the above choice of @xmath24 with @xmath130 gives the frequently used negative binomial distribution with @xmath1 the negative binomial parameter which determines the degree of depature from poisson statistics . other distributions can be generated and their associated @xmath3 dependence will also be given . then we will show how the @xmath3 dependence of the combinants shows up in hierarchical structure and void scaling relations in sect ii.g and in other places . the physical meaning of @xmath1 and @xmath2 will also be discussed for these other cases . since an exact description of particle multiplicity yields is not known , we consider @xmath3 as a free exponent to be determined by comparing various resulting distributions with experiment . this procedure is what is done in cluster yields . in multi - fragmentation distribution , the mean number of clusters @xmath81 is determined by @xmath1 and @xmath2 is directly related with the number of constituent particles @xmath4 . the large @xmath5 behavior of the mean number of cluster @xmath80 gives the fisher power @xmath3 . in multiparticle distribution , @xmath1 and @xmath2 determine the mean particle number @xmath86 or the peak position of the particle number distribution @xmath63 and the fluctuation @xmath131 or the width of the particle distribution . the fisher power @xmath3 determines the large @xmath5 behavior of the weight @xmath24 and offers a new parameter in particle production . @xmath132 in grand canonical ensemble . bose - einstein condensation of atoms in a box of sides @xmath133 of dimensions @xmath134 have @xmath135 with @xmath136 and have @xmath137 . feynman used random walk arguments , the closing of a cycle parallels a closed random walk , to discuss his choice of @xmath24 in his discussion of a superfluid phase transition in liquid helium . thermal emission of pions based on statistical mechanics and equilibrium ideas have been popular descriptions of pions coming from relativistic heavy ion collisions . for thermal models @xcite , the @xmath138 for a cycle length @xmath5 or a cluster of size @xmath5 with @xmath139 a mac donald function . for low temperatures , @xmath140 and the boltzmann factor in mass , @xmath141 , suppresses large fluctuations . in the high temperature limit and/or zero pion mass limit @xmath142 . thus the high temperature limit has @xmath143 and the low temperature limit has @xmath144 . the @xmath24 can be used to generate the pion probability distribution @xmath145 . the thermal models can be combined with hydrodynamic descriptions and an application was given @xcite to @xmath146 gev pb+pb data measured by the cern / na44 and cern / na49 collaborations . the results of ref.@xcite showed a gaussian distribution with a width about 20 % larger than the poisson result . van hove and giovannini have introduced clan variables @xmath147 and @xmath148 to describe a general class of probability distributions , with most discussions of these variables centering around the negative binomial distribution @xcite . these variables are defined as @xmath149 where the mean number of clans is @xmath147 and the @xmath148 is the mean number of members per clan . the @xmath102 is the grand canonical generating function and thus @xmath150 where @xmath24 is the cycle class weight distribution @xmath151 . the @xmath152 is the mean number of total members ( particles ) . the @xmath9 here is the number of clans of size @xmath5 having @xmath5 members . the clan variable @xmath147 is also related to the void probability @xmath153 ; thus @xmath154 . an important function in void analysis is @xmath155 . thus the void parameters , void probability @xmath67 and void function @xmath156 @xcite , are equivalent to the generalized clan parameters @xmath147 and @xmath148 with the equivalence given by : @xmath157 the @xmath158 is the normalized grand potential @xmath159 for the mean @xmath86 . void analysis looks for scaling properties associated with @xmath160 ; specifically , @xmath160 is a function of the combination @xmath161 where @xmath162 is the coefficient of @xmath163 in the fluctuation @xmath164 . since @xmath165 , the variance of @xmath4 in a grand canonical ensemble becomes @xmath166\end{aligned}\ ] ] with @xmath167 , i.e. , the normalized factorial cumulant . since the variance for a poissonian distribution is the same as the mean , @xmath168 , the @xmath169 ; thus , the parameter @xmath170 represents a degree of departure from poissonian fluctuation normalized by mean @xmath86 of the distribution . a well known non - poissonian example is a nb distribution which has @xmath171 and this becomes plank distribution with @xmath172 . thus the variable @xmath1 in @xmath24 determines the strength of the non - poissonian fluctuation term for the negative binomial distribution . in general , @xmath1 and @xmath2 determine the mean particle number or the peak position of the particle multiplicity distribution and the fluctuation or the width of the particle distribution . using the recurrence relation eq.([recfm ] ) for @xmath91 , we can show that @xmath173 - 2 \frac{\xi(\vec x , z)}{\bra a\ket } \end{aligned}\ ] ] a nb distribution has @xmath174 while the lorentz / catalan ( lc ) distribution discussed in ref . @xcite and below has @xmath175 . we will study in sect . [ voidscal ] , @xmath160 _ vs _ @xmath176 , i.e. , the void or clan variable _ vs _ the fluctuation for various choices of @xmath24 summarized in table [ tabl1 ] . the lc model , which has @xmath177 , was shown to be a useful model for discussing an underlying splitting dynamics when ancestral or evolutionary variables @xmath178 and @xmath179 were introduced into @xmath1 and @xmath2 as discussed in ref.@xcite . specifically @xmath180 and @xmath181 giving @xmath182 where @xmath183 . percolation or splitting dynamics with a branching probability @xmath178 and survival probability @xmath184 has a hierarchical topology as shown in fig.[lcfig ] . weighting each diagram by @xmath185 , the evolutionary or ancestral variables are related to the clan variables @xmath186 and @xmath187 . @xmath188 is the number of diagrams of size @xmath5 shown in fig.[lcfig ] . for this case the evolutionary dynamics is just that of the lc model . then with @xmath179 set equal to 1 , @xmath189 , @xmath190 , @xmath191 , @xmath192 , etc . the interpretation of this set of @xmath24 s reads as follows : @xmath193 has 1 surviving line without a branch ( @xmath194 ) and one diagram ( @xmath195 ) , @xmath196 has 1 branch ( @xmath197 ) leading to 2 surviving lines ( @xmath198 ) and one diagram ( @xmath199 ) , @xmath200 has 2 branch points ( @xmath201 ) leading 3 surviving lines ( @xmath202 ) and two diagrams ( @xmath203 ) , @xmath204 has 3 branch points(@xmath205 ) , 4 surviving lines ( @xmath206 ) and 5 diagrams ( @xmath207 ) , etc . in these evolutionary / ancestral variables the @xmath208 which determines @xmath102 is @xmath209 for all @xmath210 . for @xmath211 , @xmath212 is no longer a constant and is @xmath213 . to keep @xmath214 a constant without changing @xmath215 , a @xmath216 was introduced in ref.@xcite . the @xmath217 for @xmath210 and @xmath218 for @xmath211 . for @xmath211 there is a finite probability ( @xmath219 ) that the splitting will go on forever ( @xmath220 ) . in percolation above a certain @xmath178 an infinite cluster is formed and @xmath221 is similar to the strength of the infinite cluster . moreover the sudden appearance of @xmath221 is similar to the behavior of an order parameter in a phase transition . the appearance of @xmath221 can also be interpreted as the sudden appearance of a jet without pion ( @xmath222 ) , if we introduce @xmath223 instead of @xmath216 . since the clan variables are @xmath224 and @xmath225 , then , for the lc model with evolutionary or ancestral varables @xmath179 and @xmath178 , @xmath226 \nonumber \\ \beta & = & n_c \frac{n_c - ( 1 \pm 1)/2}{n_c - 1 } \nonumber\end{aligned}\ ] ] these can be reduced to @xmath227 , @xmath228 , and @xmath229 for @xmath210 while @xmath230 , @xmath231 , and @xmath232 for @xmath233 . since the branching probability @xmath178 varies in @xmath234 , the clan variable @xmath148 has @xmath235 with @xmath236 at @xmath237 and 1 , only one member per clan in average , and @xmath238 at @xmath239 , infinitly many members per clan . the lc model thus connects the clan variable @xmath148 to the probability @xmath178 of branching in the evolutionary or ancestral picture of fig.[lcfig ] or in a percolation model . for poisson processes @xmath240 ( no branching ) , @xmath241 ( only unit cycles and no be correlations ) and @xmath236 ( one member in each clan in average ) . ( 35,7)(1.2,4.0 ) ( 3,7)(0,1)4 ( 2,6.0)@xmath242 ( 1,4.5)unit cycles ( 8.5,9)(0,1)2 ( 8.0,9)(1,0)1 ( 8.5,9 ) ( 8.7,9.4)@xmath178 ( 8.0,7)(0,1)2 ( 9.0,7)(0,1)2 ( 7.0,6.0)@xmath243 ( 5.6,4.5)period 2 cycles ( 13.2,10)(0,1)1 ( 12.5,10)(1,0)1.5 ( 13.2,10 ) ( 13.4,10.4)@xmath178 ( 12.5,7)(0,1)3 ( 14.0,8.5)(0,1)1.5 ( 13.5,8.5)(1,0)1 ( 14.0,8.5 ) ( 14.2,8.9)@xmath178 ( 13.5,7)(0,1)1.5 ( 14.5,7.0)(0,1)1.5 ( 16.8,10)(0,1)1 ( 16.0,10)(1,0)1.5 ( 16.8,10 ) ( 17.0,10.4)@xmath178 ( 17.5,7)(0,1)3 ( 16.0,8.5)(0,1)1.5 ( 15.5,8.5)(1,0)1 ( 16.0,8.5 ) ( 16.2,8.9)@xmath178 ( 15.5,7)(0,1)1.5 ( 16.5,7.0)(0,1)1.5 ( 13.0,6.0)@xmath244 ( 12.2,4.5)period 3 cycles ( 20.9,10)(0,1)1 ( 20.2,10)(1,0)1.5 ( 20.9,10 ) ( 21.1,10.4)@xmath178 ( 20.2,7)(0,1)3 ( 21.7,9.0)(0,1)1.0 ( 21.2,9.0)(1,0)1 ( 21.7,9.0 ) ( 21.9,9.4)@xmath178 ( 21.2,7.0)(0,1)2.0 ( 22.2,8.0)(0,1)1.0 ( 21.7,8.0)(1,0)1 ( 22.2,8.0 ) ( 22.4,8.4)@xmath178 ( 21.7,7.0)(0,1)1.0 ( 22.7,7.0)(0,1)1.0 ( 24.1,10)(0,1)1 ( 23.4,10)(1,0)1.5 ( 24.1,10 ) ( 24.3,10.4)@xmath178 ( 23.4,7)(0,1)3 ( 24.9,9.0)(0,1)1.0 ( 24.4,9.0)(1,0)1 ( 24.9,9.0 ) ( 25.1,9.4)@xmath178 ( 24.4,8.0)(0,1)1.0 ( 25.4,7.0)(0,1)2.0 ( 23.9,8.0)(1,0)1 ( 24.4,8.0 ) ( 24.6,8.4)@xmath178 ( 23.9,7.0)(0,1)1.0 ( 24.9,7.0)(0,1)1.0 ( 27.4,10)(0,1)1 ( 26.7,10)(1,0)1.5 ( 27.4,10 ) ( 27.6,10.4)@xmath178 ( 26.7,8.5)(0,1)1.5 ( 28.2,8.5)(0,1)1.5 ( 26.2,8.5)(1,0)1 ( 26.7,8.5 ) ( 26.9,8.9)@xmath178 ( 26.2,7)(0,1)1.5 ( 27.2,7.0)(0,1)1.5 ( 27.7,8.5)(1,0)1 ( 28.2,8.5 ) ( 28.5,8.9)@xmath178 ( 27.7,7)(0,1)1.5 ( 28.7,7.0)(0,1)1.5 ( 30.8,10)(0,1)1 ( 30.0,10)(1,0)1.5 ( 30.8,10 ) ( 31.0,10.4)@xmath178 ( 31.5,7)(0,1)3 ( 30.0,9.0)(0,1)1.0 ( 29.5,9.0)(1,0)1 ( 30.0,9.0 ) ( 30.2,9.4)@xmath178 ( 29.5,7.0)(0,1)2.0 ( 30.5,8.0)(0,1)1.0 ( 30.0,8.0)(1,0)1 ( 30.5,8.0 ) ( 30.7,8.4)@xmath178 ( 30.0,7.0)(0,1)1.0 ( 31.0,7.0)(0,1)1.0 ( 34.1,10)(0,1)1 ( 33.3,10)(1,0)1.5 ( 34.1,10 ) ( 34.3,10.4)@xmath178 ( 34.8,7)(0,1)3 ( 33.3,9.0)(0,1)1.0 ( 32.8,9.0)(1,0)1 ( 33.3,9.0 ) ( 33.5,9.4)@xmath178 ( 32.8,8.0)(0,1)1.0 ( 33.8,7.0)(0,1)2.0 ( 32.3,8.0)(1,0)1 ( 32.8,8.0 ) ( 33.0,8.4)@xmath178 ( 32.3,7.0)(0,1)1.0 ( 33.3,7.0)(0,1)1.0 ( 25.5,6.0)@xmath245 ( 24.7,4.5)period 4 cycles a connection of the lc model can also be made with a ginzburg - landau theory of phase transitions and a feynman - wilson gas . these connections are discussed in ref . @xcite . where @xmath1 and @xmath2 are related to coefficients in the ginzburg - landau approach . once we identify an appropriate @xmath24 for a physical system , then we may use our general model to study the statistical behavior of the system . the various models used in pion distribution can be related to our general model with @xmath40 by choosing @xmath24 as a term in a gauss hypergeometric series @xmath246 ; @xmath247_m [ b]_m}{[c]_m } \frac{z^m}{m ! } \label{hyperf } \\ { [ a]_m } = \frac{\gamma(a+m)}{\gamma(a ) } & = & \frac{\gamma(a+n)}{\gamma(a ) } \frac{\gamma(a+n + m - n)}{\gamma(a+n ) } \label{amfc } = [ a]_n [ a+n]_{m - n } \end{aligned}\ ] ] usefull values of @xmath248_n$ ] are @xmath249_n = ( 2n)!/(n ! 2^{2n})$ ] , @xmath250_n = n!$ ] , @xmath251_n = ( n+1)!$ ] . considering only positive @xmath5 , we choose @xmath252_{k-1 } [ b]_{k-1}}{[c]_{k-1 } } \frac{z^k}{(k-1 ) ! } = x \frac{\gamma(a+k-1)}{\gamma(a ) } \frac{\gamma(c)}{\gamma(c+k-1 ) } \frac{\gamma(b+k-1)}{\gamma(b ) } \frac{z^k}{(k-1)!}\end{aligned}\ ] ] for this case , the thermodynamic grand potential or the generating function is @xmath253 if we allow jets without a pion , then we may allow @xmath222 also . for such a case , @xmath252_{k } [ b]_{k}}{[c]_{k } } \frac{z^k}{k ! } = x \frac{\gamma(a+k)}{\gamma(a ) } \frac{\gamma(c)}{\gamma(c+k ) } \frac{\gamma(b+k)}{\gamma(b ) } \frac{z^k}{k ! } \\ f_0(\vec x ) & = & f_0(x , z ) = \log z(x , z ) = - \omega(x , z ) = \sum_{k=0}^\infty x_k = x f(a , b ; c ; z)\end{aligned}\ ] ] we can see the only difference of the generating functions between the above two cases is the extra factor @xmath2 for the grand potential . we will mostly concentrate on the first case , i.e. , @xmath254 . using eq . ( [ factfm ] ) or the recurrence relation eq.([recfm ] ) and eqs.([hyperf ] ) ( [ f0hg ] ) , @xmath255_{m } [ b]_{m}}{[c]_{m } } z^{m+1 } f(a+m , b+m ; c+m ; z ) \nonumber \\ & & + x m \frac{[a]_{m-1 } [ b]_{m-1}}{[c]_{m-1 } } z^{m } f(a+m-1 , b+m-1 ; c+m-1 ; z)\end{aligned}\ ] ] the second order normalized factorial cumulant @xmath167 and the void variable @xmath160 are then @xmath256_2 [ b]_2}{[c]_2 } z^2 f(a+2 , b+2 ; c+2 ; z ) + 2 \frac{a b}{c } z f(a+1 , b+1 ; c+1 ; z ) } { \frac{a b}{c } z f(a+1 , b+1 ; c+1 ; z ) + f(a , b ; c ; z ) } \\ \chi(x , z ) & = & \frac{f_0(x , z)}{\bra a\ket } = \frac{f_0(x , z)}{f_1(x , z ) } \nonumber \\ & = & \frac{f(a , b ; c ; z ) } { \frac{a b}{c } z f(a+1 , b+1 ; c+1 ; z ) + f(a , b ; c ; z)}\end{aligned}\ ] ] for some values of @xmath0 , @xmath257 , and @xmath258 , the hypergeometric function become a simple function ; @xmath259 various models of pion distributions can be related with these functions as listed in table [ tabl1 ] . .various models with specific choice of @xmath40 and @xmath24 in hypergeometric series @xmath246 of eq.([hyperf ] ) . here @xmath260 is not included , and thus @xmath261 . here @xmath262 with @xmath105 except for poisson which has a finite @xmath263 . fisher exponent @xmath3 for each @xmath24 as discussed in sect.[sectfrag ] are given too . [ cols="^,^,^,^,^,^,^",options="header " , ] because of a unique role played by the poisson distribution and the form @xmath264 $ ] of the generating function , the cluster distribution @xmath265 is usually taken to be a poisson . however , as noted before , other divisions are possible . using the same approach used above for @xmath266 , we can expand any distribution using a nb instead of poisson , i.e. , with @xmath267 using the form of the generating function @xmath268 for the nb . replacing @xmath2 by @xmath269 , the normalization factor of a new distribution , we have @xmath270^{-x } = \left[1 - { \cal n}(z ) { \cal g}(u=0)\right]^{-x}\end{aligned}\ ] ] the @xmath271 is @xmath272z)}{{\cal n}(z ) } = \sum_k ( 1-u)^k { \cal p}_k\end{aligned}\ ] ] while the @xmath273 is @xmath274z)}{z(x , z ) } = \frac{\left[1 - { \cal n } { \cal g}(u)\right]^{-x } } { \left[1 - { \cal n}\right]^{-x}}\end{aligned}\ ] ] thus we have @xmath275^m = \sum_{m=0}^\infty p_m^{\rm nb}(x , { \cal n } ) [ { \cal g}(u)]^m \nonumber \\ & = & \sum_{m=0}^\infty p_m^{\rm nb } \left[\sum_{j=0}^\infty ( 1-u)^j { \cal p}_j\right]^m = \sum_{m=0}^\infty p_m^{\rm nb } \sum_{\{n_k\}_m } \prod_{k } \left[(1-u)^k{\cal p}_k\right]^{n_k } \\ p_n(\vec x ) & = & \sum_m p_m^{\rm nb}(x , { \cal n } ) \sum_{\{n_k\}_m } \prod_k { \cal p}_k^{n_k}(\vec x ) \label{compnb}\end{aligned}\ ] ] the result of eq.([compnb ] ) shows that the distribution @xmath145 can be written as a compound probability distribution of a negative binomial @xmath276 with another probability @xmath277 distribution generated from @xmath271 . for the case of @xmath278 , which may be considered as the fugacity @xmath279 for a particle with chemical potential @xmath280 , @xmath281 can further be decomposed as @xmath282 i.e. , @xmath277 for this case is poisson @xmath283 . if @xmath284 , then @xmath285 without @xmath222 and @xmath286^x = e^{f_0}$ ] . for @xmath214 given in table [ tabl1 ] , the @xmath277 becomes the same probability with @xmath24 replaced by @xmath287 . as an example the hga with @xmath288\right)$ ] can be decomposed as a sequential process consisting of a nb distribution of clusters with @xmath289 which is then followed by a breakup of clusters distributed with a hga distribution with @xmath290 but without voids , i.e. , with @xmath291 . thus this decomposition separates the parameter @xmath1 assigned to cluster from other parameters . event - by - event studies from high energy collisions are being used to study the details of particle multiplicity distributions as , for example , those associated with pions . such studies not only give information about the mean number of particles produced , but also information about fluctuations and higher order moments of the probability distribution which are important tools for studying the underlying processes and mechanisms that operate . they are also useful in distinguishing various phenomenological models . issues associated with fluctuations play an important role in many areas of physics and departures from poisson statistics are of current interest . one purpose of this paper was an investigation of various models of particle multiplicity distributions that can be used in event - by - event analysis . these various phenomenological models are developed using a general form of a unified model which is based on a grand canonical partition function and an underlying weight arising from feynman s path integral approach to statistical processes . a resulting distribution has three control parameters called @xmath0 , @xmath1 , @xmath2 . the relationships of these parameters to various physical quantities are discussed . one important result is the connection of the parameter @xmath0 to the fisher exponent @xmath3 ; namely @xmath292 for hga . this connection arises from a parallel we developed between the model for particle multiplicity distributions considered here and our previous approach to cluster yields . since an exact description of particle multiplicity distributions is not known , we have considered several cases with different @xmath3 s or @xmath0 s which are contained in our unified description . moreover , many of the existing distributions currently used in particle phenomenology are shown to be special choices of @xmath3 or @xmath0 which appear in a generalized hypergeometric model called hga . these include the poisson distribution coming from coherent emission , chaotic emission producing a negative binomial distribution , combinations of coherent and chaotic processes leading to signal / noise distributions and field emission from lorentzian line shapes producing the lorentz / catalan distribution . using the hga model combinants , cumulants and moments are discussed and a physical significance is given to combinants in terms of the underlying partition weights of a feynman path integral approach to statistical processes . the parameter @xmath0 or fisher exponent @xmath3 is shown to play an important role in the behavior of the combinants which manifest itself in various physical relationhips . the hga model and its associated special cases are used to explore a wide variety of phenomena . these include : linked pair approximations leading to hierarchical scaling relations on the reduced cumulant level , generalized void scaling relations , clan variable descriptions and their connections with stochastic variables and branching processes , kno scaling behavior , enhanced non - poissonian fluctuations . models based on an underlying random walk description are also discussed . in this paper we compared various particle multiplicity distributions within the hypergeometric model hga . our results show that even though various distributions have the same mean and fluctuation , the distribution itself or the underlying mechanism could be different . comparisons within the hga model also show that just comparing void variables @xmath160 and @xmath162 or mean @xmath293 and fluctuation @xmath294 or @xmath295 is not enough to distinguish different models that descibe particle multiplicity data . thus , to find the correct distribution and underlying mechanism from various data more information than just the mean and fluctuation are necessary and new variables should be found which are quite different between different models . for example , higher order reduced factorial cumulants need also to be evaluated such as the third order cumulant @xmath296 . applications of our approach to the charged multiplicity data of l3 and h1 are given . the mean @xmath297 , fluctuation variable @xmath162 , void scaling variable @xmath160 , and third order reduced cumulant variable @xmath296 obtained from these experimnets are compared with various models discussed in this paper . the value of the third cumulant @xmath296 shows that the grw1d model can fit the charge multiplicity data of l3 and h1 . the hga model also gives a fit to the l3 data , but is not as successful as grw1d with respect to the third order reduced cumulant of the h1 data . both have very similar multiplicity distributions as shown in fig.[jetdat ] which compares the two models with the charged particle multiplicity distributions of jets for l3 and h1 data . differences appear in the high multiplicity events with reduced multiplicity @xmath298 . the reduced third cumulant is negative for the h1 data of table [ tabldat ] . negative values of a cumulant require negative values of the parameter @xmath0 which are allowed in grw1d if the parameters @xmath1 and @xmath2 are also taken as negative , but a negative value of @xmath0 is not acceptable in hga since this would require a value of @xmath2 greater than 1 and thus a complex @xmath1 . we also discuss the kno scaling behavior for these data and conclude that the kno scaling behavior would not be clearly exhibited in the jets for l3 and h1 . in this paper we have also generalized the compound distribution that arises from sequential process which may reveal the dynamical structure of the distribution . specifically , the underlying sequential picture involves a two step process where the final distribution arises from the production of clusters followed by a subsequent decay of the clusters . for the hga model , the final distribution is obtained from compounding a poisson distribution of clusters with a nb distribution coming from the decay of each of the clusters . the hga may arise through a three step sequential process of poisson - poisson - logarithmic compound distribution . it is also shown that the hga can arise from a two step sequential process of a nb distribution followed by a new hga with a different mean value . k. rajagopal and f. wilczek , nucl . * b404 * , 557 ( 1993 ) . s. gavin and b. mller , phys . lett . * b329 * , 486 ( 1994 ) . g. bayer and h. heiselberg , phys b469 * , 7 ( 1999 ) . v. heinz and b.v . jacak , ann . nucl . part . , * 49 * , ( 1999 ) . s. pratt , phys . lett . * b301 * , 159 ( 1993 ) . csrg and j. zimanyi , phys . lett . * 80 * , 916 ( 1998 ) . andrews , c.g . townsend , h.j . meisner , d.s . durfee , d.m . korn , w. ketterle , science * 275 * , 637 ( 1997 ) . a. bialas and r. peschanski , nucl . phys . * b272 * , 703 ( 1986 ) ; * b306 * , 857 ( 1988 ) . p. carruthers and c.c . shih , phys . * 127b * , 242 ( 1983 ) . l. van hove , phys . lett . * b232 * , 509 ( 1989 ) . l. van hove and a. giovannini , z. phys . * c30 * , 391 ( 1988 ) . z. koba , h.b . nielson and p. olessen , nucl . * b40 * , 317 ( 1972 ) . p. carruthers and i. sarcevic , phys . b189 * , 442 ( 1987 ) . m. gadzicki and s. mrwczynski , z. phys . * c54 * , 127 ( 1992 ) . m. stephanov , k. rajagopal , e. shuryak , phys . lett . * 81 * , 4816 ( 1998 ) . haiasz , a.d . jackson , r.e . shrock , m.a . stepanov , and j.j.m . verbaarschot , phys . rev . * d58 * , 96007 ( 1998 ) . g. baym and h. heiselberg , phys . b469 * , 7 ( 1999 ) . s. jeon and v. koch , phys . rev . lett * 83 * , 5435 ( 1999 ) ; phys . * 85 * , 2076 ( 2000 ) . m. asakawa , u. heinz and b. mller , phys . lett . * 85 * , 2072 ( 2000 ) . h. heiselberg and a.d . jackson , phys . rev . * c63 * , 064904 ( 2001 ) . mekjian and s.j . lee , phys . rev . * a44 * , 6294 ( 1991 ) . s.j . lee and a.z . mekjian , phys . rev . * c45 * , 1284 ( 1992 ) . s.j . lee and a.z . mekjian , phys . rev . * c45 * , 365 ( 1992 ) . s.j . lee and a.z . mekjian , phys . rev . * c56 * , 2621 ( 1997 ) . s.j . lee and a.z . mekjian , phys . rev . * c47 * , 2266 ( 1993 ) . s.j . lee and a.z . mekjian , phys . rev . * c50 * , 3025 ( 1994 ) . k.c . chase and a.z . mekjian , phys . rev . * c 49 * , 2164 ( 1994 ) ; * 50 * , 2078 ( 1994 ) . s. lupia , a. giovannini , and r. ugoccioni , z. phys . * c59 * , 427 ( 1993 ) . m. gyulassy and s. kaufmann , phys . * 40 * , 298 ( 1978 ) ; j. phys . * a11 * , 1715 ( 1978 ) . mekjian , b.r . schlei and d. strottman , phys . rev . * c58 * , 3627 ( 1998 ) . mekjian , phys . lett . * 86 * , 220 ( 2001 ) . a. mekjian , phys . rev . * c65 * , 014907 ( 2002 ) . p. carruthers and c.c . shih , int . j. modern phys . * a2 * , 1447 ( 1987 ) . f. becattini , a. giovannini and s. lupia , z. phys . * c72 * , 43 ( 1996 ) . r. hwa , in _ proc . of the santa fe workshop on inbtermittency in high energy collisions _ ( world scientific , singapore , 1990 , f. cooper , r. hwa and i. sarcevic , edit ) . s. hegyi , phys . lett . * b309 * , 443 ( 1993 ) ; * b318 * , 642 ( 1993 ) . feynman , _ statistical physics _ ( benjamin / cummings , new york , 1972 ) . j. klauder and e.c.g . sudarshan , _ fundamentals of quantum optics _ ( benjamin , n.y . , 1968 ) . h. biyajima , prog . physics * 69 * , 966 ( 1983 ) . l3 collaboration , lanl preprint hep - ex/0110072 ( 2001 ) . s. aid et al . , h1 collaboration , z. phys . * c72 * , 573 ( 1996 ) ; lanl hep - ex/9608011 ( 1996 ) .

with recent advances in the field of ultra - cold atomic gases , trapped bose - einstein condensates ( bec s ) have become an important system to study many - body physics such as quantum phase transitions . in particular , the ability to dynamically control the lattice structure and the strength of interaction as well as the absence of impurities in bec s confined in optical lattices , have led to the recent observation of the superfluid to mott - insulator transition @xcite . at the same time , there has been a tremendous interest in studying rotating bec s in harmonic traps ; at sufficient rotation an abrikosov lattice of quantized vortices has been observed @xcite and the realization of strongly correlated quantum states similar to the fractional quantum hall states has been predicted to occur at higher rotation rates @xcite . in these proposals , the rotation can play the role of an effective magnetic field for the neutral atoms and in analogy with electrons , the atoms may enter into a state described by the laughlin wavefunction , which was introduced to describe the fractional quantum hall effect . while this approach yields a stable ground state separated from all excited states by an energy gap , in practice this gap is rather small because of the weak interactions among the particles in the magnetic traps typically used . in optical lattices , the interaction energies are much larger because the atoms are confined in a much smaller volume , and the realization of the fractional quantum hall effect in optical lattices could therefore lead to a much higher energy gap and be much more robust . in a recent paper @xcite , it was shown that it is indeed possible to realize the fractional quantum hall effect in an optical lattice , and that the energy gap achieved in this situation is a fraction of the tunneling energy , which can be considerably larger than the typical energy scales in a magnetic trap . in addition to being an interesting system in its own right , the fractional quantum hall effect , is also interesting from the point of view of topological quantum computation @xcite . in these schemes quantum states with fractional statistics can potentially perform fault tolerant quantum computation . so far , there has been no direct experimental observation of fractional statistics although some signatures has been observed in electron interferometer experiments @xcite . strongly correlated quantum gases can be a good alternative where the systems are more controllable and impurities are not present . therefore , realization of fractional quantum hall states in atomic gases can be a promising resource for topological quantum computation in the future . as noted above the fqh effect can be realized by simply rotating and cooling atoms confined in a harmonic trap . in this situation , it can be shown that the laughlin wavefunction exactly describes the ground state of the many body system @xcite . in an optical lattices , on the other hand , there are a number of questions which need to be addressed . first of all , it is unclear to which extent the lattices modifies the fractional quantum hall physics . for a single particle , the lattice modifies the energy levels from being simple landau levels into the fractal structure known as the hofstadter butterfly @xcite . in the regime where the magnetic flux going through each lattice @xmath0 is small , one expects that this will be of minor importance and in ref . it was argued that the fractional quantum hall physics persists until @xmath1 . in this paper , we extent and quantify predictions carried out in ref . . whereas ref . only considered the effect of infinite onsite interaction , we extent the analysis to finite interactions . furthermore , where ref . mainly argued that the ground state of the atoms was in a quantum hall state by considering the overlap of the ground state found by numerical diagonalization with the laughlin wavefunction , we provide further evidence for this claim by characterizing the topological order of the system by calculating chern numbers . these calculations thus characterize the order in the system even for parameter regimes where the overlap with the laughlin wavefunction is decreased by the lattice structure . in addition to considering these fundamental features of the fqh states on a lattice , which are applicable regardless of the system being used to realize the effect , we also study a number of questions which are of particular interest to experimental efforts towards realizing the effect with atoms in optical lattices . in particular , we show that adding dipole interactions between the atoms can be used to increase the energy gap and stabilize the ground state . furthermore , we study bragg spectroscopy of atoms in the lattice and show that this is a viable method to identify the quantum hall states created in an experiment , and we discuss a new method to generate an effective magnetic field for the neutral atoms in the lattice . the paper is organized as follows : in sec . [ section - no_hardcore ] , we study the system with finite onsite interaction . in sec . [ section - chern ] we introduce chern numbers to characterize the topological order of the system . the effect of the dipole - dipole interaction has been elaborated in sec . [ section - dipole ] . [ nu1/4 ] studies the case of @xmath2 filling factor . [ bragg ] is dedicated to explore bragg spectroscopy of the system . [ magnetic ] outlines a new approach for generating the type of the hamiltonian studied in this paper . the fractional quantum hall effect occurs for electrons confined in a two dimensional plane under the presences of a perpendicular strong magnetic field . if @xmath3 is the number of electrons in the system and @xmath4 is the number of magnetic fluxes measured in units of the quantum magnetic flux @xmath5 , then depending on the filling factor @xmath6 the ground state of the system can form highly entangled states exhibiting a rich behaviors , such as incompressibility , charge density waves , and anyonic excitations with fractional statistics . in particular , when @xmath7 , where @xmath8 is an integer , the ground state of the system is an incompressible quantum liquid which is protected by an energy gap from all other states , and in the landau gauge is well described by the laughlin wavefunction @xcite : ( z_1,z_2, ... ,z_n)=_j > k^n ( z_j - z_k)^m _ j=1^n e^- y^2_i/2 [ laughlin ] , where the integer @xmath8 should be odd in order to meet the antisymmetrization requirement for fermions . although the fractional quantum hall effect occurs for fermions ( electrons ) , bosonic systems with repulsive interactions can exhibit similar behaviors . in particular , the laughlin states with even @xmath8 correspond to bosons . in this article , we study bosons since the experimental implementation are more advanced for the ultra - cold bosonic systems . we study a system of atoms confined in a 2d lattice which can be described by the bose - hubbard model @xcite with the peierls substitution @xcite , @xmath9 + where @xmath10 is the hopping energy between two neighboring sites , @xmath11 is the onsite interaction energy , and @xmath12 is the phase acquired by a particle going around a plaquette . this hamiltonian is equivalent to the hamiltonian of a u(1 ) gauge field ( transverse magnetic field ) on a square lattice . more precisely , the non - interacting part can be written as -j_<ij > a^_i a_j ( ^j_i ) where @xmath13 is the vector potential for a uniform magnetic field and the path of the integral is chosen to be a straight line between two neighboring sites . in the symmetric gauge , the vector potential is written as @xmath14 . hence , @xmath0 will be the amount of magnetic flux going through one plaquette . while the hamiltonian in eq . ( [ eq : hamiltonian ] ) occurs naturally for charged particles in a magnetic field , the realization of a similar hamiltonian for neutral particles is not straightforward . as we discuss in sec . [ magnetic ] this may be achieved in a rotating harmonic trap , and this have been very successfully used in a number of experiments in magnetic traps @xcite , but the situation is more complicated for an optical lattice . however , there has been a number of proposals for lattice realization of a magnetic field @xcite , and recently it has been realized experimentally @xcite . @xcite have studied the realization of fractional hall states for a few particles in individual lattice sites . a new approach for rotating the entire optical lattice is discussed in sec . [ magnetic ] . the essence of the above hamiltonian is a non - zero phase that a particle should acquire when it goes around a plaquette . this phase can be obtained for example by alternating the tunneling and adding a quadratic superlattice potential @xcite or by simply rotating the lattice ( sec.[magnetic ] ) . the advantage of confining ultra - cold gases in an optical lattice is to enhance the interaction between atoms which consequently result in a higher energy gap comparing to harmonic trap proposals ( e.g. ref . ) . this enhancement in the energy gap of the excitation spectrum can alleviate some of the challenges for experimental realization of the quantum hall state for ultra - cold atoms . in order to approximate a large system , we study the system with periodic boundary condition , i.e. on a torus , where the topological properties of the system is best manifested . there are two energy scales for the system : the first is the magnetic tunneling term , @xmath15 , which is related to the cyclotron energy in the continuum limit @xmath16 and the second is the onsite interaction energy @xmath11 . experimentally , the ratio between these two energy scales can be varied by varying the lattice potential height @xcite or by feshbach resonances @xcite . let us first assume that we are in the continuum limit where @xmath17 , i.e. the flux through each plaquette is a small fraction of a flux quantum . a determining factor for describing the system is the filling factor @xmath18 , and in this study we mainly focus on the case of @xmath19 , since this will be the most experimentally accessible regime . we restrict ourself to the simplest boundary conditions for the single particle states @xmath20 , where @xmath21 is the magnetic translation operator which translates the single particle states @xmath22 around the torus . the definition and detailed discussion of the boundary conditions will be elaborated in section [ section - chern ] . the discussed quantities in this section , such as energy spectrum , gap and overlap , do not depend on the boundary condition angles ( this is also verified by our numerical calculation ) . in the continuum case , for the filling fraction @xmath19 , the laughlin state in landau gauge @xmath23 is given by eq . ( [ laughlin ] ) with @xmath24 . the generalization of the laughlin wavefunction to a torus takes the form @xcite ( z_1,z_2, ... ,z_n)=f_rel(z_1,z_2, ... ,z_n)f_cm(z)e^-_i y^2_i/2 [ laughlin - torus ] , + where @xmath25 is the relative part of the wave function and is invariant under collective shifts of all @xmath26 s by the same amount , and @xmath27 is related to the motion of the center of mass and is only a function of @xmath28 . for a system on a torus of the size @xmath29 , we write the wavefunction with the help of theta functions , which are the proper oscillatory periodic functions and are defined as @xmath30 } ( z|\tau)=\sum_n e^{i \pi \tau ( n+a)^2 + 2\pi i ( n+a)(z+b)}$ ] where the sum is over all integers . for the relative part we have , f_rel=_i < j ( |i ) ^2 . according to a general symmetry argument by haldane @xcite , the center of mass wave function @xmath27 is two - fold degenerate for the case of @xmath19 , and is given by f_cm(z)=(|2i ) where @xmath31 refers to the two degenerate ground states . this degeneracy in the continuum limit is due to the translational symmetry of the ground state on the torus , and the same argument can be applied to a lattice system when the magnetic length is much larger than the lattice spacing @xmath32 . for higher magnetic filed , the lattice structure becomes more pronounced . however , in our numerical calculation for a moderate magnetic field @xmath33 , we observe a two - fold degeneracy ground state well separated from the excited state by an energy gap . we return to the discussion of the ground state degeneracy in sec . [ section - chern ] . in the continuum limit @xmath32 , the laughlin wavefunction is the exact ground state of the many body system with a short range interaction @xcite . the reason is that the ground state is composed entirely of states in the lowest landau level which minimizes the magnetic part of the hamiltonian , the first term in eq . ( [ eq : hamiltonian ] ) . the expectation value of the interacting part of the hamiltonian , i.e. the second term in eq . ( [ eq : hamiltonian ] ) , for the laughlin state is zero regardless of the strength of the interaction , since it vanishes when the particles are at the same position . to study the system with a non - vanishing @xmath0 , we have performed a direct numerical diagonalization of the hamiltonian for a small number of particles . since we are dealing with identical particles , the states in the hilbert space can be labeled by specifying the number of particles at each of the lattice sites . in the hard - core limit , only one particle is permitted on each lattice site , therefore for @xmath3 particles on a lattice with the number of sites equal to @xmath34 , where @xmath35 is the unit lattice side , the hilbert space size is given by the combination @xmath36 . on the other hand , in case of finite onsite interaction , the particles can be on top of each other , so the hilbert space is bigger and is given by the combination @xmath37 . in our simulations the dimension of the hilbert can be raised up to @xmath38 and the hamiltonian is constructed in the configuration space by taking into account the tunneling and interacting terms . the tunneling term is written in the symmetric gauge , and we make sure that the phase acquired around a plaquette is equal to @xmath39 , and that the generalized magnetic boundary condition is satisfied when the particles tunnels over the edge of the lattice [ to be discussed in sec . [ section - chern ] , c.f . ( [ twist_angles ] ) ] . by diagonalizing the hamilton , we find the two - fold degenerate ground state energy which is separated by an energy gap from the excited states and the corresponding wavefunction in the configuration space . the lauhglin wave function ( [ laughlin - torus ] ) can also be written in the configuration space by simply evaluating the laughlin wave function at discrete points , and therefore we can compared the overlap of these two dimensional subspaces . the energy gap above the ground state and the ground state overlap with the laughlin wavefunction for the case of @xmath19 in a dilute lattice @xmath40 , are depicted in figure . [ tuning_u ] . the laughlin wavefunction remains a good description of the ground state even if the strength of the repulsive interaction tends to zero ( fig . [ tuning_u ] a ) . below , we discuss different limits : first , we consider @xmath41 : if the interaction energy scale @xmath11 is much larger than the magnetic one @xmath42 , all low energy states lie in a manifold , where the highest occupation number for each site is one , i.e. this corresponds to the hard - core limit . the ground state is the laughlin state and the excited states are various mixtures of landau states . the ground state is two - fold degenerate @xcite and the gap reaches the value in for the hard - core limit at large @xmath43 , as shown in fig . [ fig : gap ] ( a ) . in this limit the gap only depends on the tunneling @xmath10 and flux @xmath0 , and the gap is a fraction of @xmath10 . these results are consistent with the previous work in ref . . secondly , we consider @xmath44 . in this regime , the magnetic energy scale @xmath45 is much larger than the interaction energy scale @xmath11 . for repulsive regime ( @xmath46 ) , the ground state is the laughlin state and the gap increases linearly with @xmath47 , as shown in fig . [ fig : gap ] b. thirdly , we study @xmath48 where the interaction is absent and the ground state becomes highly degenerate . for a single particle on a lattice , the spectrum is the famous hostadfer s butterfly @xcite , while in the continuum limit @xmath49 , the ground state is the lowest landau level ( lll ) . the single particle degeneracy of the lll is the number of fluxes going through the surface , @xmath50 . so in the case of @xmath3 bosons , the lowest energy is obtained by putting @xmath3 bosons in @xmath50 levels . therefore , the many - body ground state s degeneracy should be : @xmath51 for example , 3 bosons and 6 fluxes gives a 336-fold degeneracy in the non - interacting ground state . if we increase the amount of phase ( flux ) per plaquette ( @xmath52 ) we are no longer in the continuum limit . the landau level degeneracy will be replace by @xmath53 where @xmath54 is the amount of flux per plaquette and @xmath55 and @xmath56 are coprime @xcite . then , the many - body degeneracy will be : @xmath57 . fourthly , we consider @xmath58 : when u is negative ( i.e. attractive interaction ) in the limit strong interaction regime , the ground state of the system will become a _ pillar _ state . in a pillar state , all bosons in the system condensate into a single site . therefore , the degeneracy of the ground state is @xmath59 and the ground state manifold can be spanned as , _ i ( a^)_i^n | vac . these states will very fragile and susceptible to collapse @xcite . in a lattice , it is also possible to realize the fractional quantum hall states for attractive interaction in the limit when @xmath60 . assume that the occupation number of each site is either zero or one . since the attraction energy @xmath11 is very high and there is no channel into which the system can dissipate its energy , the probability for a boson to hop to a site where there is already a boson is infinitesimally small . therefore , the high energy attraction will induce an effective hard - core constraint in the case of ultra - cold system . the energy of these state should be exactly equal to their hard - core ground state counterparts , since the interaction expectation value of the interaction energy is zero for the laughlin state . the numerical simulation shows that these two degenerate states indeed have a good overlap with the laughlin wavefunction similar to their repulsive hard - core counterparts and also their energies are equal to the hard - core ground state . these states are very similar to repulsively bound atom pairs in an optical lattice which have recently been experimentally observed @xcite . is varied by changing the size of the lattice ( the size in the two orthogonal directions differ at most by unity ) . the laughlin state ceases to be a good description of the system as the lattice nature of the system becomes more apparent @xmath61 . the overlap is only calculated at the positions shown with dots and the color coding is an extrapolation between the points . ] is varied by changing the size of the lattice ( the size in the two orthogonal directions differ at most by unity ) . the laughlin state ceases to be a good description of the system as the lattice nature of the system becomes more apparent @xmath61 . the overlap is only calculated at the positions shown with dots and the color coding is an extrapolation between the points . ] so far we have mainly considered a dilute lattice @xmath62 , where the difference between a lattice and the continuum is very limited . we shall now begin to investigate what happens for larger values of @xmath0 , where the effect of the lattice plays a significant role . [ high - alpha - overlap ] , shows the ground state overlap with the laughlin wavefunction as a function of the strength of magnetic flux @xmath0 and the strength of the onsite interaction @xmath11 . as @xmath0 increases the laughlin overlap is no longer an exact description of the system since the lattice behavior of the system is more pronounced comparing to the continuum case . this behavior does nt depend significantly on the number of particles for the limited number of particles that we have investigated @xmath63 . we have , however , not made any modification to the laughlin wave function to take into account the underlying lattice , and from the calculations presented so far , it is unclear whether the decreasing wave function overlap represents a change in the nature of the ground state , or whether it is just caused by a modification to the laughlin wave function due the difference between the continuum and the lattice . to investigate this , in the next section , we provide a better characterization of the ground state in terms of chern numbers , which shows that the same topological order is still present in the system for higher values of @xmath0 . as a summary , we observe that the laughlin wavefunction is a good description for the case of dilute lattice ( @xmath32 ) , regardless of the strength of the onsite interaction . however , the protective gap of the ground state becomes smaller for weaker values of interaction and in the perturbative regime @xmath64 is proportional to @xmath47 for @xmath65 . in the theory of quantum hall effect , it is well understood that the conductance quantization , is due to the existence of certain topological invariants , so called chern numbers . the topological invariance was first introduced by avron _ _ et al.__@xcite in the context of thouless _ et al . _ ( tkndn ) s original theory @xcite about quantization of the conductance . tkndn in their seminal work , showed that the hall conductance calculated from the kubo formula can be expressed into an integral over the magnetic brillouin zone , which shows the quantization explicitly . the original paper of tkndn deals with the single - particle problem and bloch waves which can not be generalized to topological invariance . the generalization to many - body systems has been done by niu _ _ et al.__@xcite and also tao _ _ et al.__@xcite , by manipulating the phases describing the closed boundary conditions on a torus ( i.e. twist angles ) , both for the integer and the fractional hall systems . these twist angles come from natural generalization of the closed boundary condition . to clarify the origin of these phases , we start with a single particle picture . a single particle with charge ( @xmath66 ) on a torus of the size @xmath67 in the presence of a magnetic field @xmath68 perpendicular to the torus surface , is described by the hamiltonian h_s= , where @xmath13 is the corresponding vector potential ( @xmath69 ) . this hamiltonian is invariant under the magnetic translation , t_s(*a*)=e^i * * ak**^s / where @xmath70 is a vector in the plane , and @xmath71 is the pseudomomentum , defined by @xmath72 the generalized boundary condition on a torus is given by the single - particle translation @xmath73 where @xmath74 and @xmath75 are twist angles of the boundary . the origin of these phases can be understood by noting that the periodic boundary conditions corresponds to the torus in fig . [ torus](a ) . the magnetic flux through the surface arises from the field perpendicular to the surface of the torus . however , in addition to this flux , there may also be fluxes due to a magnetic field residing inside the torus or passing through the torus hole , and it is these extra fluxes which give rise to the phases . the extra free angles are all the same for all particles and all states in the hilbert space , and their time derivative can be related to the voltage drops across the hall device in two dimensions . the eigenstates of the hamiltonian , including the ground state will be a function of these boundary angles @xmath76 . by defining some integral form of this eigenstate , one can introduce quantities , that do not depend on the details of the system , but reveal general topological features of the eigenstates . first we discuss the simplest situation , where the ground state is non - degenerate , and later we shall generalize this to our situation with a degenerate ground state . the chern number is in the context of quantum hall systems related to a measurable physical quantity , the hall conductance . the boundary averaged hall conductance for the ( non - degenerate ) @xmath0th many - body eigenstate of the hamiltonian is @xcite : @xmath77 , where the chern number @xmath78 is given by [ twist ] ) ( b ) redefining the vector potential around the singularities : @xmath79 is not well - defined everywhere on the torus of the boundary condition . therefore , another vector field @xmath80 with different definition should be introduced around each singularity @xmath81 of @xmath79 . @xmath79 and @xmath80 are related to each other by a gauge transformation @xmath82 and the chern number depends only on the loop integrals of @xmath82 around those singularities regions , c.f . , eq.([chern_gauge]).,title="fig : " ] [ gauge ] ) ( b ) redefining the vector potential around the singularities : @xmath79 is not well - defined everywhere on the torus of the boundary condition . therefore , another vector field @xmath80 with different definition should be introduced around each singularity @xmath81 of @xmath79 . @xmath79 and @xmath80 are related to each other by a gauge transformation @xmath82 and the chern number depends only on the loop integrals of @xmath82 around those singularities regions , c.f . , eq.([chern_gauge]).,title="fig : " ] c()= _ 0 ^ 2 d_1 _ 0 ^ 2 d_2 ( _ 1 _ 2^()-_2 _ 1^ ( ) ) [ eq : chern ] , where @xmath83 is defined as a vector field based on the eigenstate @xmath84 on the boundary torus @xmath85 by _ j^()(_1,_2 ) i ^()||^ ( ) . it should be noted that the wave function @xmath76 is defined up to a phase factor on the boundary - phase space . therefore , @xmath76 and @xmath86 are physically equivalent for any smooth function @xmath87 . under this phase change , @xmath83 transforms like a gauge : @xmath88 hence , the chern number integral is conserved under this gauge transformation and it encapsulates general properties of the system . chern numbers has been used extensively in the quantum hall literature , for characterization of the localized and extended states ( ref . and refs . therein ) . in this paper , we use the chern number as an indicator of order in the system . moreover , it enables us to characterize the ground state in different regimes , especially where the calculation of the overlap with the laughlin wave function fails to give a conclusive answer . before explaining the method for calculating the chern number , we clarify some issues related to the degeneracy of the ground state . in some systems , the ground state can be degenerate , this can be intrinsic or it can be as a result of the topology of the system . if the ground state is degenerate , we should generalize the simple integral form of eq . ( [ eq : chern ] ) to take into account the redundancy inside the ground state manifold . for example , in the case of a two - fold degenerate ground state , an extra gauge freedom related to the relative phase between two ground states , and this freedom should be fixed . in other words , as we change the twist angles , we can not keep track of the evolution of both states , since one can not distinguish them from each other . therefore , to uniquely determine the chern number of the ground state(s ) , we should resolve this gauge invariance , which is treated in section [ section - impurity ] and [ gauge - fixing ] . it important to note that the degeneracy in the non - interacting regime is fundamentally different from the degeneracy in the interacting case . in the non - interacting limit , the degeneracy can be lifted by a local perturbation e.g. impurity , while in the hardcore case , the degeneracy remains in the thermodynamic limit @xcite . the latter degeneracy in the ground state is a consequence of the global non - trivial properties of the manifold on which the particles move rather than related to a symmetry breaking which happens in conventional models e.g. ising model . the topological degeneracy is not a consequence of breaking of any symmetry only in the ground state , instead it is the entire hilbert space which is split into disconnected pieces not related by any symmetry transformation . with a general argument , wen @xcite showed that if the degeneracy of a chiral spin system moving on a simple torus is @xmath89 , then it should be @xmath90 on a torus with @xmath91 handles ( riemann surface with genus @xmath91 ) , therefore the topological detergency is an intrinsic feature of the system . in particular , in the context of quantum hall effect , this multicomponent feature of the ground state on a torus has a physical significance , while the single component ground state on a sphere boundary condition gives zero conductance , the torus geometry with multicomponent ground state results in a correct conductance measured in the experiment , since the torus boundary condition is more relevant to the experiment . changing twists angles of the boundary will rotate these components into each other and gives an overall non - zero conductance @xcite . as studied in a recent work by m. oshikawa and t. senthil @xcite , as a universal feature , it has been shown that in presence of a gap , there is a direct connection between the fractionalization and the topological order . more precisely , once a system has some quasiparticles with fractional statistics , a topological degeneracy should occur , which indicates the presence of a topological order . therefore , the amount of the degeneracy is related to the statistics of the fractionalized quasiparticles e.g. in the case of @xmath19 , the two - fold degeneracy is related to 1/2 anyonic statistics of the corresponding quasiparticles . chern number has been also studies for spin @xmath92 system on a honeycomb lattice @xcite for identifying abelian and non - abelian anyons . _ @xcite studies chern number for disordered fermi system using a non - commutative geometry . to resolve the extra gauge freedom related to the two degenerate ground states , we consider two possibilities : i. lifting the degeneracy by adding some impurities , ii . fixing the relative phase between the two states in the ground state . below , we explore both cases . in the first case , we introduce some fixed impurities to lift the degeneracy in the ground state for all values of the twist angles . this is an artifact of the finite size of the system which we take advantage of . in the presence of perturbation , we show that the system has a topological order in spite of poor overlap with the laughlin state . in the second approach , we use a scheme recently proposed by hatsugai @xcite which is a generalized form for degenerate manifolds . in this section , we introduce some perturbation in the finite system in form of local potentials ( similar to local impurities in electronic systems ) to split the degeneracy of the ground state and resolve the corresponding gauge invariance , which allows us to compute the chern number . furthermore , the fact that we can still uniquely determine the chern number in the presence of impurities , shows that the system retains its topological order , even when the impurities distort its ground state wavefunction away from the laughlin wavefunction . in the context of the quantum hall effect , the conventional numerical calculation of various physical quantities such as the energy spectrum , screening charge density profile , wave functions overlaps , and the density - density correlation functions , can not be used for understanding the transport properties of electrons in the presence of impurities ( although useful for studying of isolated impurities @xcite ) . recently , d. n. sheng _ et al . _ @xcite calculated the chern number as an unambiguous way to distinguish insulating and current carrying states in the @xmath93 quantum hall regime which correspond to zero and non - zero chern number , respectively . in this work , a weak random disorder was used to lift the degeneracy of the ground state ( three - fold for @xmath93 ) for a finite number of electrons . the energy splitting between the lowest three states then decreased with increasing number of particles , which indicates the recovery of the degeneracy in the thermodynamic limit . moreover , the mobility gap can be determined by looking at the energy at which the chern number drops towards zero . this energy gap is comparable to the energy gap obtained from the experiment and it is not necessarily equal to the spectrum gap which separates the degenerate ground state from the excited states this shows the significance of chern number calculations for understanding these systems . in a finite system , the coupling to a single - body interaction , e.g. impurities , can lift the degeneracy and one can uniquely determine the chern number for the individual states by direct integration of eq . ( [ eq : chern ] ) . on one hand , the impurity should be strong enough to split the energy in the ground state ( in this case @xmath94 , where @xmath95 denotes the energy of the @xmath96th energy level ) for all values of the twist angles . on the other hand , these impurities should be weak enough so that the energy splitting in the ground state remains smaller than the thermodynamic gap ( in this case @xmath97 ) . to calculate the chern number of individual level , as mentioned in the pervious section , we have to fix the phase of the wavefunction . the method that we explore in this section can be considered as a simplified version of the general method developed by hatsugai @xcite which we will explore in the next section . following kohmoto s procedure @xcite , we assume that the ground state @xmath98 may be expanded for all twist angles on a @xmath99-dimensional hilbert discrete space @xmath100 . if @xmath101 in eq . ( [ eq : chern ] ) is a periodic function on the torus of the boundary condition , then by application of stoke s theorem , the chern number will be always zeros . the non - triviality ( non - zero conductance in the case of quantum hall system ) occurs because of the zeros of the wave function , where the phase is not well - defined . therefore , @xmath102 is not well defined everywhere and its flux integral can be non - zero . to uniquely determine the chern number , we assume that the wave function and the vector field are not defined for certain points @xmath81 in @xmath103 regions on the torus of the boundary condition . for simplicity , we first discuss this procedure , in the case of a non - degenerate ground state . for calculating the integral , we should acquire another gauge convention for the wave function inside these @xmath103 regions , e.g. in a discrete system , we may require an arbitrary element of the wave function to be always real , and thereby we can define a new vector field @xmath104 , which is well defined inside these regions . these two vector fields differ from each other by a gauge transformation ( fig . [ torus ] ) : _ j^()(_1,_2)- _ j^()(_1,_2 ) = _ j ( _ 1,_2 ) , [ eq:1d_gauge]and the chern number reduces to the winding number of the gauge transformation @xmath105 over small loops encircling @xmath81 , i.e. @xmath106 , c()= _ n _ s_n d [ chern_gauge ] . the one - dimensional gauge eq . ( [ eq:1d_gauge ] ) should be resolved by making two conventions . for example , in one convention the first element and in the other the second element of the wavefunction in the hilbert space should be real i.e. transforming the ground state @xmath107 into @xmath108 where @xmath109 is a @xmath99-dimensional vector and @xmath110 is a projection into the ground state and similarly with the other reference vector @xmath111 . since the gauge that relates two vector fields is the same as the one that relates the corresponding wavefunctions ( similar to eq.([eq : gauge_wavefunction ] ) ) , we can uniquely determine the gauge transformation function @xmath82 by evaluating @xmath112 . therefore , the chern number will be equal to the number of windings of @xmath82 around regions where @xmath113 is zero . counting the vorticities has a vigorous computational advantage over the conventional method of direct integration of eq . ( [ eq : chern ] ) . in the direct integration , we need to choose a large number of mesh points for the boundary angles , because of the discrete approximation of derivatives in eq . ( [ eq : chern ] ) , and this makes the calculation computationally heavy . we note that for the system on a lattice , we should exactly diagonalize the hamiltonian which is a sparse matrix as opposed to the continuum case where the hamiltonian is a dense matrix residing on a smaller projected hilbert space ( lowest landau level ) . and @xmath114 , for the first and the second energy state , respectively . the results are for 3 atoms on a 6x6 lattice ( @xmath115 ) in the hard - core limit . ( b ) @xmath116 for the first level has no vorticity . however , for the second level , as shown in ( c ) , @xmath116 has vorticity equal to one associated with regions where either @xmath117 or @xmath118 vanishes . , width=566 ] for removing the degeneracy , in our numerical simulations , we add a small amount of impurity which is modeled as delta function potentials located at random sites with a random strength , of the order of the tunneling energy @xmath10 . this is described by a hamiltonian of the form @xmath119 , where @xmath120 numerates the lattice site , @xmath121 is the atom number operator and @xmath122 is the strength of the impurity at site @xmath120 . we choose reference states @xmath123 and @xmath124 to be eigenvectors of the numerically diagonalized hamiltonian at two different twist angles . in fig . [ impurity ] , vorticities of @xmath125 associated with the first and the second energy level is depicted . it is easy to see that the chern number associated to the two ground states is one . number of vortices may vary for the first and second ground states , but their sum is always equal to one . the hard - core limit ( @xmath126 ) is very similar to the case of fractional quantum hall effect , which in the context of hall systems , means a share of @xmath92 ( in @xmath127 unit ) for each ground state @xcite . when the onsite interaction strength is small ( @xmath128 ) , the thermodynamic gap becomes comparable to the ground state energy splitting @xmath129 , the chern number can not be uniquely determined , and the system does nt have topological order . on the other hand , in the limit of strong interaction ( @xmath130 ) , the total chern number associated to the ground states is equal to one , regardless of the impurity configuration . moreover , in the hard - core limit , although the ground state is not described by the laughlin wavefunction , since it is distorted by the impurity ( in our model it can be as low as @xmath131 ) , the chern number is unique and robust . this is an indication that the topological order is not related to the symmetries of the hamiltonian and it is robust against arbitrary local perturbations @xcite . these results indicate the existence of a topological order in the system and robustness of the ground states against local perturbations . the method developed in the previous section has the graphical vortex representation for the chern number which makes it computationally advantageous compared to the direct integration of eq . ( [ eq : chern ] ) . it can not , however , be applied directly to a degenerate ground state , and therefore we had to introduce an impurity potential which lifted the degeneracy . on the other hand , a significant amount of impurity in the system may distort the energy spectrum , so that the underlying physical properties of the lattice and fluxes could be confounded by the artifacts due to the impurities , especially for large @xmath0 . to address this issues , in this section , we explore a generalized method of the previous section based on refs . and , which works for a degenerate ground state . as arrows for fixed @xmath123 and @xmath124 . ( b ) and ( c ) : surface plots of @xmath132 and @xmath133 ( blue is lower than red ) . @xmath74 and @xmath75 changes from zero to @xmath134 . these plots have been produced for 3 atoms with @xmath135 ( @xmath136 ) in the hard - core limit on a 5x5 lattice . the total vorticity corresponding to each of the reference wave functions ( @xmath123 or @xmath124 ) indicates a chern number equal to one . ] as arrows for fixed @xmath123 and @xmath124 . ( b ) and ( c ) : surface plots of @xmath132 and @xmath133 ( blue is lower than red ) . @xmath74 and @xmath75 changes from zero to @xmath134 . these plots have been produced for 3 atoms with @xmath135 ( @xmath136 ) in the hard - core limit on a 5x5 lattice . the total vorticity corresponding to each of the reference wave functions ( @xmath123 or @xmath124 ) indicates a chern number equal to one . ] as arrows for fixed @xmath123 and @xmath124 . ( b ) and ( c ) : surface plots of @xmath132 and @xmath133 ( blue is lower than red ) . @xmath74 and @xmath75 changes from zero to @xmath134 . these plots have been produced for 3 atoms with @xmath135 ( @xmath136 ) in the hard - core limit on a 5x5 lattice . the total vorticity corresponding to each of the reference wave functions ( @xmath123 or @xmath124 ) indicates a chern number equal to one . ] by generalizing the chern number formalism for a degenerate ground state manifold , instead of having a single vector field @xmath137 , a tensor field @xmath138 should be defined , where @xmath139 for a @xmath140-fold degenerate ground state _ j^(,)(_1,_2 ) i ^()||^ ( ) similar to the non - degenerate case , when @xmath141 is not defined , a new gauge convention should be acquired for the regions with singularities . this gives rise to a tensor gauge transformation @xmath142 on the border of these regions _ j^(,)(_1,_2)- _ j^(,)(_1,_2 ) = _ j ^(,)(_1,_2 ) . following hatsugai s proposal@xcite for fixing the ground state manifold gauge , we take two reference multiplets @xmath123 and @xmath124 which are two arbitrary @xmath143 matrices ; @xmath66 is the ground state degeneracy ( equal to 2 in our case ) . in our numerical simulation , we choose the multiplets to be two sets of ground state at two different twist angles far from each other , e.g. @xmath144 and @xmath145 . we define an overlap matrix as @xmath146 where @xmath147 is again the projection into the ground state multiplet , and consider the regions where @xmath148 or @xmath149 vanishes ( similar to zeros of the wave function in the non - degenerate case ) . hence , the chern number for @xmath140 degenerate states , will be equal to the total winding number of @xmath150 for small neighborhoods , @xmath103 , in which @xmath148 vanishes for fixed @xmath123 and @xmath124 . @xmath74 and @xmath75 changes form zero to @xmath134 . this plot has been produced for 4 atoms with @xmath151 in the hard - core limit on a 5x5 lattice ( @xmath152 ) . although , there are more vortices here compared to fig . [ omega ] , the total vorticity corresponding to each of the trial functions ( @xmath123 or @xmath124 ) indicates a chern number equal to one . ] @xmath153 which is the same as the number of vortices of @xmath154 . it should be noted that the zeros of det @xmath155 and det@xmath156 should not coincide in order to uniquely determine the total vorticity . in fig . [ omega ] , we have plotted @xmath157 , @xmath132 , and @xmath158 , found by numerical diagonalization of the hamiltonian for a mesh ( @xmath159 ) of winding angles @xmath74 and @xmath75 . in this figure , the chern number can be determined be counting the number of vortices and it is readily seen that the winding number is equal to one for the corresponding zeros of @xmath148 ( or @xmath160 ) . .chern number for different configurations in the hard - core limit for fixed filling factor @xmath19 . the laughlin state overlap is shown in the last column . although it deviates from the laughlin state . although the ground state deviates from the laughlin state , the chern number remains equal to one half per state before reaching some critical @xmath161 where the energy gap vanishes . [ cols="^,^,^,^,^,^,^,^",options="header " , ] we have calculated the chern number for fixed @xmath19 and different @xmath0 s by the approach described above . the result is shown in table [ chern ] . for low @xmath32 , we know from sec . [ section - no_hardcore ] that the ground state is the laughlin state and we expect to get a chern number equal to one . for higher @xmath0 , the lattice structure becomes more apparent and the overlap with the laughlin state decreases . however , in our calculation , the ground state remains two - fold degenerate and it turns out that the ground state chern number tends to remain equal to one before reaching some critical @xmath161 . hence , also in this regime we expect to have similar topological order and fractional statistics of the excitations above these states on the lattice . the degeneracy of the ground state is a function of twist angles . the shown results are for 5 atoms on a 5x5 lattice i.e. @xmath162 ( a ) shows first three energy manifolds as a function of the toroidal boundary condition angles , ( b ) shows a cross section of ( a ) at @xmath163 for seven lowest energy levels . the first and the second energy levels get close to each other at @xmath164 . ] the degeneracy of the ground state is a function of twist angles . the shown results are for 5 atoms on a 5x5 lattice i.e. @xmath162 ( a ) shows first three energy manifolds as a function of the toroidal boundary condition angles , ( b ) shows a cross section of ( a ) at @xmath163 for seven lowest energy levels . the first and the second energy levels get close to each other at @xmath164 . ] for the arguments above to be applicable , it is essential that we can uniquely identify a two - fold degenerate ground state which is well separated from all higher lying states . for higher flux densities , @xmath165 , the two - fold ground state degeneracy is no longer valid everywhere on the torus of the boundary condition . in this regime , the issue of degeneracy is more subtle , and the finite size effect becomes significant . the translational symmetry argument @xcite , which was used in the section [ section - no_hardcore ] , is not applicable on a lattice and as pointed out by kol _ et al_. @xcite the degeneracy of the ground state may vary periodically with the system size . some of the gaps which appear in the calculation may be due to the finite size and vanish in the thermodynamic limit , whereas others may represent real energy gaps which are still present in the thermodynamic limit . to investigate this , we study the ground state degeneracy as a function of boundary angles ( @xmath166 ) which are not physical observable and therefore the degeneracy in thermodynamic limit should not depend on their value . in particular , fig . [ level - crossing ] shows the energy levels of five particles at @xmath167 for different values of the twist angles . the first and the second level are split at ( @xmath168 ) , while they touch each other at ( @xmath164 ) . we have observed similar behavior for different number of particles and lattice sizes e.g. 3 and 4 atoms at @xmath169 . in this case , the system seems to not have a two - fold degeneracy . therefore , the ground state enters a different regime which is a subject for further investigation . for having the topological order , it is not necessary to be in the hard - core limit . even at finite interaction strength @xmath170 , we have observed the same topological order with the help of the chern number calculation . if @xmath11 gets further smaller , the energy gap above the ground state diminishes ( as seen in sec . [ section - no_hardcore ] ) and the topological order disappears . we conclude that the chern number can be unambiguously calculated for the ground state of the system in a regime where laughlin s description is not appropriate for the lattice . the non - zero chern number of a two - fold degenerate ground state , in this case equal to one half per state , is a direct indication of the topological order of the system . in sections [ section - no_hardcore ] and [ section - chern ] above , we have investigated the conditions under which the fractional quantum hall effect may be realized for particles on a lattice . the motivation for this study is the possibility to generate the quantum hall effect with ultra cold atoms in an optical lattice but the results of these sections are applicable regardless of the method used to attain this situation . in this and the following sections , we investigate some questions which are of particular relevance to ultra cold atoms in an optical lattice . first , we introduce a long range , e.g. , dipole - dipole , interaction which turn out to increase the energy gap and thereby stabilizes the quantum hall states . we then turn to the case of @xmath2 and show that in order to realize this state , it is essential to have some kind of long range interaction . in an experimental realization of the quantum hall effect on a lattice , it is desirable to have as large an energy gap as possible in order to be insensitive to external perturbations . so far , we have studied effect of the short range interaction , and we have shown that the gap increases with increasing interaction strength , but the value of the gap saturates when the short range interaction becomes comparable to the tunneling energy @xmath10 . in this section , we explore the possibility of increasing the gap by adding a long - range repulsive dipole - dipole interaction to the system . previously , such dipole - dipole interaction has also been studied in ref . + onlinecitecooper2005 as a method to achieve read - rezayi states @xcite of rapidly rotating cold trapped atoms for @xmath171 and as a means to realize fractional quantum hall physics with fermi gases @xcite . the dipole - dipole ( magnetic or electric ) interaction is described by the hamiltonian : h_d - d = u_dipole _ 1i < jn where @xmath172 . @xmath173 are unit vectors representing the permanent dipole moments and the position vectors @xmath174 are in units of the lattice spacing @xmath175 . for simplicity , we assume that all dipoles are polarized in the direction perpendicular to the plane . with time independent dipoles , the strength of the interaction is given by @xmath176 ( or @xmath177 ) where @xmath178 s ( @xmath179 s ) are the permanent magnetic ( electric ) dipole moment . static dipoles will thus give repulsive interaction @xmath180 , but experimentally time varying fields may be introduced which effectively change the sign of the interaction@xcite . for completeness , we shall therefore both investigate positive and negative @xmath181 , but the repulsive interaction corresponding to static dipoles will be the most desirable situation since it stabilizes the quantum hall states . experimentally the dipole - dipole interaction will naturally be present in the recently realized bose - einstein condensation of chromium@xcite which has a large magnetic moment . however , for a lattice realization , polar molecules which have strong permanent electric dipole moments is a more promising candidate . for typical polar molecules with the electric moment @xmath182 debye , on a lattice with spacing @xmath183@xmath184 m , @xmath185 can be up to few khz , an order of magnitude greater than the typical tunneling @xmath186 which can be few hundreds of hz @xcite . versus @xmath0 . the results are shown for @xmath187 , @xmath188 , @xmath189 ] versus @xmath0 . the results are shown for @xmath187 , @xmath188 , @xmath189 ] to study the effect of the dipole - dipole interaction , we again numerically diagonalize the hamiltonian for a few hardcore bosons ( @xmath130 ) , in the dilute regime @xmath190 , while varying strength of the dipole - dipole interaction . the results of this simulation is shown in fig . [ dipole ] ( a ) . after adding the dipole interaction , the ground state in a dilute lattice remains two - fold degenerate , since the interaction only depends on the relative distance of the particles and keeps the center of mass motion intact . if the dipole interaction becomes too strong @xmath191 , the ground state wave function deviates from the laughlin wavefunction , but the topological order remains the same as for the system without dipole interaction . we verified this by calculating the chern number as explained in sec . [ section - chern ] for different values of the dipole - dipole interaction strength @xmath192 and observed that the total chern number of the two - fold degenerate ground state is equal to one . moreover , as it is shown in fig . [ dipole ] ( b ) adding such an interaction can increase the gap : the lower curve corresponds to the hard - core limit discussed in the previous work @xcite and the upper curve corresponds to the system including the dipole - dipole interaction . this enhancement varies linearly with the flux density @xmath0 in a dilute lattice and does nt depend on the number of particles and consequently , it is expected to behave similarly in the thermodynamic limit . one of the impediment of the experimental realization of quantum hall state is the smallness of the gap which can be improved by adding dipole - dipole interaction . in this section , we showed that this improvement is possible and moreover , by chern number evaluation , we verified that adding dipole interaction does nt change the topological behavior of the ground state manifold . so far we have concentrated on the case of @xmath19 . in this section , we briefly investigate the case of @xmath2 . it is expected that the laughlin wavefunction remains a good description for the ground state of a bosonic system for any even @xmath66 , where @xmath193 . following haldane s argument@xcite , due to the center of mass motion , the degeneracy of the ground state is expected to be @xmath66-fold on a torus . similar to the case of @xmath19 , the laughlin wavefunction should be a suitable description for any @xmath66 provided that the magnetic field is weak so that we are close to the continuum limit , i.e. @xmath194 . also the chern number is expected to be equal to one for the q - fold degenerate ground state , which in the context of quantum hall effect means a share of @xmath195 of the conduction quantum @xmath127 for each state in the q - fold degenerate ground state manifold . we have done both overlap and the chern number calculation to check those premises . in the case of @xmath2 , significant overlap occurs at low @xmath196 . the average wave function overlap of four lowest energy eigenstates with the laughlin wavefunction is depicted in figure [ overlap_laughlin_1_4 ] , where we have used a generalization of the laughlin wavefunction for periodic boundary conditions similar to eq . ( [ laughlin - torus ] ) @xcite . we observe that the laughlin wavefunction is a reliable description of the system with @xmath2 but only for much more dilute lattices ( @xmath196 ) compared to @xmath19 where significant overlap occurs for @xmath197 . contrary to @xmath19 , where the gap is a fraction of the tunneling energy @xmath10 , the gap for @xmath2 between the 4-fold degenerate ground state and the next energy level is infinitesimal . the reason for the vanishing gap can be understood in the continuum limit from the argument put forward in ref . : as noted previously the laughlin wavefunction is an exact eigenstate of the hamiltonian with an energy per particle equal to the lowest landau level energy . the energy of the @xmath198 state is thus equal to the @xmath24 state . it thus costs a negligible energy to compress the @xmath2 state to the @xmath19 state , and therefore there is no gap . in an external trap the system will always choose the highest density state which is the @xmath19 state . note , however that this argument only applies to short range interactions . for long range interactions , we expect to see a non vanishing gap . even though that with short range interactions , the gap is very small in our numerical calculations , it is still sufficiently large that it allows us to unambiguously determine the chern number for the ground state manifold as described in sec . [ gauge - fixing ] . as expected the calculation shows that the chern number is equal to one corresponding to a four - fold degenerate ground state consistent with the generalization of the fermionic case in the fractional quantum hall theory @xcite . in fig . [ overlap_laughlin_1_4 ] , the overlap of the first four lowest energy state with the laughlin wavefunction is depicted . in the absence of the dipole interaction , the ground state overlap is significant only for @xmath199 , however , by adding a moderate dipole interaction ( @xmath200 ) , the overlap becomes more significant for a larger interval of the flux density , i.e. @xmath201 . this is due to the fact that state with lower density become more favorable in the presence of a long - range repulsive interaction . we observed that adding a dipole interaction would lead to an improvement of the gap for @xmath19 and make the laughlin overlap more significant for larger interval of magnetic field strength @xmath0 in the case of @xmath2 . therefore this long - range interaction can be used as an tool for stabilizing the ground state and make the realization of these quantum states , experimentally more feasible . on a torus . the dashed ( dotted ) line shows the overlap for the system without ( with ) dipole interaction ( @xmath200 ) . the laughlin state is only a good description for a more dilute lattice @xmath202 compared to @xmath19 . the dipole interaction stabilizes the system and the overlap is more significant for higher values of @xmath203 . ] in an experimental realization of the quantum hall states , it is essential to have an experimental probe which can verify that the desired states were produced . in most experiments with cold trapped atoms , the state of the system is probed by releasing the atoms from the trap and imaging the momentum distribution . in ref . , it was shown that this technique provide some indication of the dynamics in the system . this measurement technique , however , only provides limited information , since it only measures the single particle density matrix , and provides no information about the correlations between the particles . in refs . and a more advanced measurement techniques were proposed , where the particle correlation is obtained by looking at the correlations in the expansion images . in this section , we study bragg scattering as an alternative measurement strategy which reveals the excitation spectrum of the quantum system . in ref . bosonic quantum hall system responses to a perturbative potential is studied . we focus on bragg scattering where two momentum states of the same internal ground state are connected by a stimulated two - photon process @xcite . by setting up two laser beams with frequencies @xmath204 and @xmath205 and wave vectors @xmath206 and @xmath207 in the plane of the original lattice , a running optical superlattice will be produced with frequency @xmath208 and wave vector @xmath209 . ( both frequencies @xmath204 and @xmath205 should be close to an internal electronic dipole transition in the atoms ) . the beams should be weak and sufficiently detuned so that direct photon transitions are negligible , i.e. @xmath210 , where @xmath211 is the frequency of the transition , @xmath212 is the corresponding spontaneous decay rate and @xmath213 are the rabi frequencies related to the laser - atom coupling . in this perturbative regime , the inelastic scattering of photons will be suppressed ; the atom starts from the internal ground state , absorbs one photon from e.g. beam 1 by going to a virtual excited state and then emits another photon into beam 2 by returning to its internal ground state . after this process , the system has acquired an energy equal to @xmath214 and a momentum kick equal to @xmath215 . therefore , the overall effect of the recoil process is a moving ac stark shift as a perturbing moving potential , and the effective hamiltonian represents the exchange of the two - photon recoil momentum and the energy difference to the system and is proportional to the local density i.e. @xmath216 . this process can be used to probe density fluctuations of the system and thus to measure directly the dynamic structure factor @xmath217 and static structure factor @xmath218 . this kind of spectroscopy has been studied for a bec in a trap by blakie _ _ @xcite and zambelli _ et al . _ @xcite and has been realized experimentally in refs . . ] . ] also , bragg spectroscopy in an optical lattice is discussed in ref . in a mean - field approach and also in ref . as a probe of the mott - insulator excitation spectrum . on the other hand , in the context of quantum hall effect , the static structure factor has been studied for probing the magnetoroton excitations @xcite and charge density waves @xcite . the dynamic and static structure factors are given respectively as s(,)=_n , 0| n | ^()| 0|^2 ( -e_n+e_0 ) , s()=_n,0 | n | ^()| 0|^2 = _ n,0 | n | _ e^i | 0|^2 , where the density fluctuation operator is defined as @xmath219 and the coefficient are defined as fourier transforms of the wannier functions : @xmath220 , where the wannier function @xmath221 is the wave function of an atom localized on a site centered at @xmath222 . below , we focus on deep optical lattices , where @xmath223 . in the structure factor , there is a sum over the excited states @xmath224 and ground states @xmath225 and the self - term is thus excluded . the ground state on a torus is two - fold degenerate and therefore in our numerics , we add the contribution of both . since we are working on a discrete lattice , there will be a cut - off in the allowed momentum given by the lattice spacing @xmath226 , where @xmath227 is the distance between lattice sites . [ 3atoms - structure](a ) shows the structure factor for the case @xmath19 for a small @xmath0 calculated from our numerical diagonalizations . in the data presented here , we have chosen @xmath228 and but the result should be similar in other directions in the lattice plane . we see that @xmath218 is modulated at a momentum corresponding to the magnetic length . for the parameters that we have investigated , the general features of the structure factor is independent of the size of the system . we obtain the excitation spectrum shown in fig . [ 3atoms - structure ] ( b ) similar to ref . by the feynman - bijl approximation . in the continuum limit ( @xmath229 ) , we assume that the low - lying excitations are long - wavelength density oscillations and their wave functions can be approximated to have the form @xmath230 . therefore , the variational estimate for the excitation energy is @xmath231 . at zero momentum and at the momentum corresponding to the magnetic length , there are gaps , and we also observe a deviation from the free particle spectrum similar to the magneoroton case as a reminiscent of the wigner crystal . it should be noted that the deviation does not depend on the size or the number of particles in the system . as clearly seen in the fig . [ 3atoms - structure ] the energy spectrum and structure factor deviate from those of free particles , therefore , it could be used as an experimental probe of the system . the structure factor and excitation spectrum imply some general features that are very different from that of the mott - insulator and superfluid states , and can be used a powerful experimental indication of the quantum hall states . recently , there has been several proposals for producing artificial magnetic field for neutral atoms in optical lattices @xcite , however , the implementation of each of them is still experimentally demanding . recently , there has been an experimental demonstration of a rotating optical lattice @xcite which is equivalent to an effective magnetic field ( see below ) . the technique used in this experiment , however , generates a lattice with a large lattice spacing , because it uses laser beam which are not exactly counter propagating . this longer spacing reduces the energy scale in the lattice and thereby also reduces quantities such as energy gaps . here , we shall now introduce an alternative method for generating a rotating optical lattice , which does not result in an increased lattice spacing . this method consists of rotating the optical lattice by manipulating laser beams . in a frame of reference rotating with angular velocity @xmath232 around the @xmath233 , the hamiltonian for a particle of mass @xmath8 in an ( planar ) harmonic trap of natural frequency @xmath234 is @xmath235 at resonance @xmath236 the form is equivalent to the hamiltonian of a particle of charge @xmath66 experiencing an effective magnetic field @xmath237 . therefore , by simply rotating the optical lattice , we can mimic the magnetic field for neutral atoms . to rotate the optical lattice , we propose to set up four acousto - optic modulators ( aom ) and four focusing composite lenses as shown in the fig . [ lattice_rotation ] . by sweeping the acoustic wave frequency , the beams can be focused and make a rotating optical lattice . in an aom , for the first order diffracted light we have @xmath238 , where @xmath239 is the wavelength of sound in the medium , @xmath240 is the light wavelength and @xmath241 is half of the angle between a diffracted beam and the non - diffracted beam ( fig . [ lattice_rotation ] ) . by increasing the frequency of the acoustic wave , the diffraction angle increases . however , the beam should be focused by a large aperture lens so that it always passes the region where we want to make the optical lattice . by focusing a similar but counter - propagating beam , we can make a rotating standing wave ( fig . [ lattice_rotation ] ) . by repeating the same configuration in the transverse direction , we can make a rotating optical lattice . in particular , if the aom is far from the composite lenses , @xmath242 , then @xmath243 and @xmath244 where @xmath245 , where the parameters are defined in fig . [ lattice_rotation ] . if we consider a square lattice with dimensions @xmath246 , the number of magnetic flux given by rotation is n_== where @xmath247 is the atomic recoil frequency . on the other hand , the upper limit for the magnetic field comes from the single bloch band approximation which we made in writing the hamiltonian for the optical lattice . in order for the particles to remain in the first band , the traveling lattice beams should move them adiabatically . from ref . , the adiabaticity condition for a moving lattice with an acceleration @xmath248 equal to @xmath249 at the edge , is @xmath250 , where @xmath251 is the frequency difference between the first and the second band in the lattice . this puts a limit on how large the lattice can become @xmath252 . hence , for @xmath19 with a lattice filling fraction @xmath253 and a typical recoil frequency @xmath254 khz , one can enter the regime of fractional quantum hall effect by rotating the lattice at @xmath255 hz . if a deep optical lattice is used e.g. @xmath256 , the adiabaticity condition is easily satisfied for a lattice of size @xmath257 . the experimentally challenging part will , however , likely be to mitigate the instability of the lattice caused by the thickness of the beam and aberration of the lenses at the turning points i.e. near @xmath258 . an extended study of the realization of the fractional quantum hall states in optical lattices has been presented . we showed that a hamiltonian similar to that of a magnetic field for charged particles can be constructed for neutral ultra cold atoms and molecules confined in an optical lattice . by adding an onsite interaction for the case of @xmath19 , an energy gap develops between the ground state and the first excited state , which increases linearly as @xmath47 and saturates to its value in the hardcore limit @xmath126 . we learned that the laughlin wavefunction is a reliable description of the system for low flux densities @xmath201 . for higher @xmath0 s , the lattice structure becomes more pronounced and a better description of the system can be carried out by investigating the chern number associated to the ground state manifold . the chern number indicates that the system has topological order up to some critical flux density @xmath161 , where the properties of the ground state manifold starts to change . we have also studied @xmath2 , where compared to @xmath19 , the laughlin wavefunction only describes the ground state for lower values of the flux @xmath199 . we showed that a dipole - dipole interaction can enhance the gap and stabilize the system , and therefore make the ground state more experimentally realizable . bragg spectroscopy has been studied as a potential experimental tool to diagnose theses incompressible states . characterization of the ground state by evaluating the chern number , developed in sec . [ gauge - fixing ] , can be generalized to other interesting many - body systems where the conventional overlap calculation fails to work . in particular , this method can be applied to ground states with non - abelian statistics which are appealing candidates for fault - tolerant quantum computation . we thank k. yang , m. greiner , s. girvin and j. i. cirac for fruitful discussions . this work was partially supported by the nsf career award , packard foundation , afsor and the danish natural science research council .

the cluster of galaxies is one of the hottest , most x - ray luminous clusters known . since its discovery by @xcite , it has been the subject of intense and ongoing research @xcite . in particular , chandra observations by @xcite revealed the cluster to be a supersonic merger in the plane of the sky with a textbook example of a bow shock , making this cluster a unique case in which to study hydrodynamical properties of interacting systems . the optical images show that the cluster has two distinct components , and the x - ray analysis reveals that the lower mass sub - cluster has recently exited the core of the main cluster with a relative velocity of @xmath1 . although this relative velocity appears to be unusually large , an analysis of cosmological simulations demonstrates that it is well within the predicted range of the currently favored cosmological model @xcite . due to its unique geometry and physical state , this cluster is the best known system in which to test the dark matter hypothesis @xcite . the observed offsets between the weak gravitational lensing mass peaks and the x - ray gas component give the most direct evidence for the presence of dark matter yet available . using the same observations , @xcite placed upper limits on the dark matter self - interaction cross section . the goal of this work is to obtain a high - resolution , absolutely calibrated mass map with no assumptions on the physical properties of the underlying cluster potential . for this purpose we use hst acs data , incorporating the gravitational lensing information from both multiple image systems ( strong lensing ) and from distortions of background sources ( weak lensing ) . the superb spatial resolution delivered by acs both increases the number density of background sources that can be used for weak lensing and reveals new strong lensing candidates , especially around the subcluster . the joint strong and weak lensing analysis thus benefits greatly from the acs data , enabling us to increase the spatial resolution and signal strength of the mass map in the core of the cluster . encouraged by the success of the combined strong and weak lensing reconstruction method developed in @xcite ( hereafter ) and the results from applying the method to cluster rx j1347.5@xmath21145 @xcite , we proceed to apply this method to . because the strong lensing data are richer in this case ( rx j1347.5@xmath21145 has not been observed with either the acs or wfpc2 cameras prior to this study ) , we improve the method as described in the text . we perform the reconstruction in the following sequence . using acs ( multi - color where available ) hst images , we identify the multiply imaged systems . having a spectroscopic redshift for one lensed system ( from @xcite ) , we then use the predictive power of simple strong lens modeling to estimate the redshifts of the other systems . using positions and redshifts of the strongly lensed images and the shape measurements of the `` weakly '' lensing sources , we perform a combined strong and weak lensing mass reconstruction , thereby significantly improving the constraints on the mass and positions of the main cluster and the colliding subcluster of . in terms of the structure of this paper , in section [ sec : data ] we describe the optical images used in this analysis , the basic image processing and the extraction of the strong ( [ sec : datasl ] ) and weak ( [ sec : datawl ] ) gravitational lensing data . we then infer the mass distribution of cluster from these data in section [ sec : swunited_bullet ] , following a demonstration of our methodology on suitable simulated data in section [ sec : swunited ] . we discuss the possible sources of error in section [ sec : systematics ] and summarize our conclusions in section [ sec : conclusions ] . acs / wfc imaging of the cluster was carried out in cycle 13 ( proposal 10200 , pi jones ) on 2004 october 21 in two pointings with one and three different filters respectively . the two pointings are centered on the main cluster and the subcluster with a small overlap between them . the subcluster was observed in three different filters ( three orbits with f814w , one with f606w , and one with f435w ) , while the main cluster was observed only in the f606w filter ( one orbit ) . the demands placed by the lensing analysis require special care when reducing the images . we use the multidrizzle @xcite routine to align the images . to register the images with the astrometric accuracy needed for lensing analysis , we determine the offsets among the images by extracting high @xmath3 objects in the individual , distortion corrected exposures . we use sextractor @xcite and the iraf routine geomap to identify the objects and calculate the residual shifts and rotation of individual exposures , which were then fed back into multidrizzle . we use `` square '' as the final drizzling kernel and an output pixel scale of @xmath4 ; this is smaller than the original pixel scale of the acs ccd to reduce the impact of resampling on the shape measurements . owing to its complex structure , the strong lensing analysis of the cluster is complicated . among the multiply - imaged systems , only the giant arc on the nw side of the main cluster s cd has a measured redshift ( @xmath5 * ? ? ? corresponding images for multiply - imaged sources were identified by matching both morphologies and surface brightnesses in each of the available acs bands ( f435w , f606w , and f814w ) and ground - based filters , bvr data from magellan and i - band from vlt @xcite . we confirm ( based on photometry and morphology ) that the @xmath6 systems ( labeled a - f in table [ tab : arcs ] ) previously identified by @xcite are indeed multiply imaged . we identify four additional systems ( g - j ) in the subcluster region , where none were previously known . all of these are identified in fig . [ fig : arcslabel ] , and image positions ( corresponding to the peak surface brightness ) are presented in table [ tab : arcs ] . unfortunately , for the greater part of the main cluster , only a single band acs image is available , which makes identification of additional candidates significantly more ambiguous . using these identifications , we perform a parametrized strong lensing reconstruction . at this stage we are not generating a detailed strong lensing model , rather we use this parametrized model only to predict the redshifts of the systems where spectroscopic redshifts are not available ( all systems but a ) and as the initial model guess for the subsequent strong and weak lensing reconstruction . for this reconstruction , we use only the image positions as constraints . the parametrized model consists of two non - singular isothermal ellipses ( nie ) @xcite , for which the scaled surface mass density @xmath7 is given by @xmath8 @xmath9 is defined counterclockwise from due west and @xmath10 is related to the line - of - sight velocity dispersion @xmath11 through @xmath12 . the components are centered on the southern cd galaxy of the main cluster and the brightest cluster galaxy ( bcg ) of the sub cluster respectively ( denoted with white crosses in fig . [ fig : arcslabel ] ; it is not straightforward to identify the bcg of the main cluster , therefore we refer to the two main galaxies as the southern and northern cd ) . we allow the scaling @xmath10 , core radii @xmath13 , ellipticities @xmath14 , and position angles @xmath15 of these to vary . throughout this paper as in @xcite with @xmath16 , @xmath17 being the axes ratio . ] following the prescription of @xcite , we also include the 20 brightest cluster members from the f606w - band ( selected using color information from magellan data ) in the mass model . the galaxies are modeled as non - singular isothermal spheres with a line - of - sight velocity dispersion @xmath18 and core radius @xmath19 following @xmath20 the proportionality constants are allowed to vary as well and below we quote the values of @xmath18 and @xmath21 for a fiducial galaxy with a f606w - band magnitude @xmath22 . the best fit model for this system has values of @xmath23 for the main cluster , @xmath24 for the subcluster , and @xmath25 for the cluster members . we also allow the redshifts of systems b - j to vary . the resulting best fit redshifts are given in table [ tab : arcs ] . to evaluate the angular diameter distances throughout the paper we assume the @xmath26cdm cosmology with @xmath27 , @xmath28 , and hubble constant @xmath29 . for objects close in projection to the main cluster , the redshift estimates are fairly reliable because the well - measured arc a is located there ( we estimate an error of @xmath30 ) . however , the redshift estimates for those close to the subcomponent are less well - determined , and are somewhat degenerate with the adopted mass of the subcluster . we postpone an extension of the methodology described in the following sections that incorporates simultaneous solution for the unknown source redshifts to future work , but here note that : 1 ) the parametrized model allows the generation of a set of source redshifts that , while model - dependent , are self - consistent ; 2 ) the predicted redshifts are consistent with the colors of the objects . we adopt the redshifts estimated here for the combined strong and weak lensing reconstruction , and where possible estimate the systematic error introduced as a result of this simplification . rrrr & 104.63316 & -55.941395 & + & 104.62988 & -55.943798 & + & 104.62954 & -55.941844 & + & 104.63042 & -55.941474 & + & 104.63775 & -55.941851 & + & 104.63338 & -55.945324 & + & 104.64709 & -55.943575 & + & 104.63528 & -55.951836 & + & 104.64008 & -55.950620 & + & 104.64232 & -55.948784 & + & 104.65155 & -55.956671 & + & 104.64778 & -55.958284 & + & 104.56568 & -55.939832 & + g & 104.56402 & -55.942113 & 1.3 + & 104.56417 & -55.944131 & + & 104.56293 & -55.939764 & + h & 104.56133 & -55.942430 & 1.9 + & 104.56189 & -55.947724 & + & 104.56186 & -55.946114 & + i & 104.56052 & -55.942930 & 2.1 + & 104.56141 & -55.944264 & + & 104.56909 & -55.946016 & + & 104.57025 & -55.944050 & [ tab : arcs ] in the weak lensing analysis , we use the f435w , f606w , and f814w exposures for the subcluster , and f606w wherever multicolor acs data are not available . we will only outline the main steps used for this analysis , full details on how the weak lensing catalogs were generated will be given in a future paper dealing with the weak lensing mass measurements of at large radius ( * ? ? ? * in prep . ) . we correct galaxy images for the psf anisotropy and psf smearing closely following the technique described in @xcite . the procedure is based on the ksb algorithm @xcite , in particular we use the modified imcat implementation . the ksb method is formally valid only in the weak lensing regime ( i.e. at radii much larger than where multiply imaged systems form and where the distortions are small ) . however recent simulations from the step project @xcite show that this approach is also valid in the non - weak lensing regime ( i.e. with shear values of @xmath31 ) to the accuracy needed for a single cluster reconstruction . lastly , in the vicinity of the highly elongated arcs we are dominated by the strong lensing signal , thereby minimizing the effects of signal dilution due to imperfect psf correction . we select stars from the half - light - radius vs. magnitude diagram and fit a fifth order polynomial to their measured ellipticities as a function of their position on the drizzled image . after the correction the rms of the stellar ellipticity components @xmath32 and @xmath33 ( as defined in @xcite ) changes from @xmath34 to @xmath35 and from @xmath36 to @xmath37 , respectively , and the mean is shifted to @xmath38 in both cases . the shear is then measured independently in all available filters and a weighted average is calculated from those to arrive at our final , psf corrected ellipticity estimates . we use the @xmath39 overlap between the two f606w pointings to test for systematics in the psf correction . the differences in the shears derived from the two exposures is consistent given the noise level . the acs camera has a psf that varies both spatially and temporally . while the approach described above accounts for the spatial dependence , it does not account for time variations . we investigate the psf time dependence by measuring stellar shapes in individual exposures ( four exposures were taken within each orbit ) . we conclude that the psf is _ not _ stable ( i.e. the global pattern of the psf changes ) even within a single orbit . however our detected signal is much stronger than the typical residuals due to imperfect psf removal caused by the temporal variation ; therefore we use the combined images to measure shapes . in addition , the psf shape of an individual star changes as a function of its radius ( see e.g. @xcite ) ; we therefore match the radius used to measure the stellar correction factors with that for the galaxies . using simulated weak lensing data we identified a constant bias between the measured and estimated shear values ; this bias was calculated for each image set separately and was introduced as a multiplicative shear calibration factor when producing final catalogs . the color cuts applied to each catalog to remove cluster and foreground galaxies were color - color cuts estimated using photometric - redshift templates of galaxies at redshifts @xmath40 . we used templates of all but extreme starburst galaxies , under the assumption that there are not likely to be many faint , low - redshift starburst dominated galaxies , particularly in a cluster . as such , we expect that unless there is an unknown population of dwarf galaxies in the cluster with colors much different from known populations , we will have removed all but a handful of outliers ( extreme starbursts or dust obscured ) . we only use objects which were detected in more than a single band ; in particular we reject all the objects that are detected only in the f606w pointing of the main cluster and were too faint for detection in the magellan images . unfortunately the observations at hand do not allow us to determine the photometric redshifts of the sources used for weak and strong lensing . we would need at least nir data to reliably determine the redshifts , as most of the sources are located at @xmath41 . therefore following @xcite , we estimate the redshifts of the background sources by applying the color and magnitude cuts that remove the likely foreground population to the hdf - s photometric redshift catalog of @xcite . for the remaining sources , we average the ratios @xmath42 ( where @xmath43 , @xmath44 , and @xmath45 are the angular diameter distances to the cluster , source and between the cluster and the source respectively ) to generate a very crude estimate of the individual redshifts , but good average redshifts per population ( provided the redshift distribution of the hdf - s is representative of that of our background sources ) . as a result all of the galaxies in a certain magnitude bin will be given the same redshift . the main reason of this approach , instead of calculating a mean for the whole catalog is the fact that the images have different depths in different parts of the reconstructed area , and this method allows us to account for the sudden jumps in the mean background galaxy redshift across the image boundaries , which would otherwise introduce spurious features in the reconstructions . this method is only approximate , therefore we test the reconstruction method using mock catalogs with only limited knowledge of the redshift of the source population in sect . [ sec : swunited ] . the weak and strong lensing reconstruction performs well under these conditions . in addition , any significant dilution of the signal will be compensated by the strong lensing signal ( if more than one strong lensing system is used ) . the cluster is also at a moderate redshift , therefore changing the sources from redshifts of @xmath46 to @xmath47 would change the lensing strength by @xmath48 . in conclusion we expect the contributions due to errors in the assumed distribution of @xmath42 to be smaller than that expected from the galaxy ellipticity shot noise . finally , we fit the non - singular isothermal sphere nis model to the the averaged tangential ellipticities @xmath49 . both components generate large tangential shear signal . we center circular bins on the southern cd of the main cluster and the bcg of the subcluster . the resulting nis fit for the main cluster results in @xmath50 , @xmath51 , and @xmath52 , @xmath53 for the subcluster . this is an oversimplified model because the cluster is morphologically disrupted and we do not correct for the contamination of the main cluster profile in the sub cluster region ( and vice versa ) . our strong and weak lensing reconstruction is performed using the method described in . the main idea behind the method is to describe the cluster mass distribution using a set of generic model parameters , which we choose to be the projected gravitational potential @xmath54 measured on a regular square grid . the total @xmath55 is defined as @xmath56 where @xmath57 is the contribution from weak lensing , @xmath58 from strong lensing and @xmath59 is the regularization function with regularization parameter @xmath60 . we closely follow the methodology presented in , with the few differences described below . in finite differencing formula for @xmath7 all the calculations in all three papers do however use the correct factor of @xmath61 . ] the most important difference is that we extend the formalism to more than a single set of multiple images . this may appear straightforward because we only need to sum up the contributions to @xmath62 from each system . given @xmath63 multiply imaged systems having @xmath64 images each , we define @xmath62 as @xmath65 where @xmath66 is the covariance matrix and @xmath67 . @xmath68 is the average position of the @xmath69-th source ( calculated from the image positions and given the deflection angle @xmath70 ) , and @xmath71 is the so - called `` cosmological weight '' function as defined in @xcite . in we assumed that the covariance matrix is diagonal and independent of the lens model parameters . as discussed in e.g. @xcite , this approach is not optimal and can lead to solutions that are biased toward models that predict high magnification at the image positions . with more sources at differing redshifts , this problem becomes more apparent because the method likely converges toward a solution with steep gradients in the surface mass density close to image positions ( allowing for the `` fine tuning '' of the reconstruction ) . to overcome such unphysical solutions , we use a better approximation for the error containing the magnification at the image position @xmath72 and obtain the covariance matrix @xmath73 . such a procedure is approximately correct , provided we are close enough to the true model ( see @xcite ) . the covariance matrix is still diagonal ; however we argue that assuming it to be such in the source plane is in fact a better approximation than assuming the covariance matrix to be diagonal in the image plane , as sources are on average more circular than their lensed images . in addition , in practice multiple image constraints are satisfied nearly perfectly and exact values of errors on image positions are of lesser importance . we adapt the method to accommodate data regions of arbitrary size and shape , but the potential is reconstructed only in grid cells containing weak and/or strong lensing data . we tile the observed field with a regular grid and determine which cells contain either weak lensing galaxies or multiple images . this can result in some cells being sparsely populated ( holes in the data , edge effects ) , but the regularization ensures that potential is properly reconstructed . once the data - grid is determined , we ensure that each of the data grid points is surrounded by a @xmath74 grid of points that are included in the reconstruction to calculate the scaled surface mass density @xmath7 , shear @xmath75 , and the deflection angle @xmath76 , using finite differencing , as described in . to find the @xmath55-function minimum , we search for a solution of the following system of equations @xmath77 . this is in general a non - linear set of equations , which we solve in an iterative manner . we again linearize this system as described in , calculating all of the non - linear terms ( for the strong lensing term in particular these are now the ones containing @xmath68 and @xmath72 ) using the information from the previous iteration . we regularize the solution using a `` moving prior '' , gradually updating our knowledge of the model by increasing its complexity and computing its likelihood relative to the previous ( simpler ) model . we begin with an initial model ( @xmath78 , @xmath79 , @xmath80 ) computed on a relatively coarse grid , and then gradually increase the number of grid points , comparing the resulting @xmath7 map with that from the previous iteration , linearly interpolated onto the finer grid . in addition to @xmath7 we now also use the components of the shear @xmath81 and @xmath82 in the regularizing scheme , to avoid the numerical effects described in even more efficiently . the regularization constant was set after experimentation with simulated data . when using many multiply imaged systems , the final reconstruction is even less sensitive to these choices than when only a single system is used ( as in ) ; we discuss this finding in more detail in sect . [ sec : systematics ] . with many multiply imaged systems , we must choose a sufficiently fine grid to be able to calculate the difference in the deflection angle between images of different systems ( if images from more than one system lie within the same grid cell , it is difficult to find a good solution ) . ideally one would like to use adaptive grids for this purpose ( having a relatively coarse grid where only weak lensing information is available and increasing the resolution in the vicinity of the multiple images ) . unfortunately it is computationally difficult to track such a problem and is therefore beyond the scope of this paper , but will be a subject of future work . we tested all the improvements described above using a high - resolution n - body simulation of a galaxy cluster by @xcite . the catalogs are generated as in , although adapted to the higher - resolution data that we use here . the weak lensing simulated data are obtained by placing @xmath83 galaxies ( giving a density of @xmath84 ) randomly on a @xmath85 field . the intrinsic ellipticities @xmath86 are drawn from a gaussian distribution , with each component characterized by @xmath87 . we then draw the redshifts of the background sources , following @xcite , from a gamma distribution @xmath88 with @xmath89 , giving a mean redshift of @xmath90 , and a mode of @xmath91 . we reject the foreground objects , and use the background object redshifts when predicting the lensed shapes . however , when using the catalogs for the reconstruction , we assume _ no _ knowledge of the individual redshifts : instead all sources are assumed to have @xmath46 , in agreement with the poor estimates available for the actual data . the multiple - image systems are generated as described in . four quadruply - imaged systems are generated at source redshifts of @xmath92 . the initial model for the regularization is set to @xmath93 . the resulting reconstruction is given in fig . [ fig : sims ] . it is very encouraging to observe how well the overall shape ( the agreement in the ellipticity and position angle of the main component , and the overall profile ) and its mass are reconstructed . these successes would not be possible with weak lensing data alone , owing to the mass - sheet degeneracy ( i.e. the reconstruction would suffer from the degeneracy of the form @xmath94 , where @xmath95 is an arbitrary constant , see e.g. @xcite ) . in addition with the acs - quality data , we are starting to resolve the small - scale substructure of the potential . we now apply the above mentioned reconstruction method to the strong and weak lensing data of . as described in sect . [ sec : datasl ] , we use 10 distinct multiply imaged systems and @xmath96 `` weakly '' lensed galaxies ( @xmath97 ) . we start with a @xmath98 grid for a @xmath99 field oriented with respect to the negative ra coordinate ( not all of the grid cells contain data and those that do not are excluded from the reconstruction ) . we gradually increase the number of grid points ( for reasons described in sect . [ sec : swunited ] ) in steps of 1 , with the final reconstruction performed on a @xmath100 grid . in our initial model , we use a two - component nie model with parameters as given in sect . [ sec : datasl ] to set the regularization . the possible dependence of the reconstruction on initial conditions is discussed in the next section . the resulting reconstruction is shown in fig . [ fig : swunited ] , together with x - ray surface brightness contours from the 500 ks chandra observation . the two cluster components are clearly detected . in addition , with superb strong lensing data ( due to exquisite acs resolution and multi - color information for the subcluster ) we are starting to resolve multiple components within each individual system . from this reconstruction , we measure the enclosed ( projected within a cylinder ) mass for this system . the main cluster has @xmath101 , while the sub cluster has @xmath102 ( @xmath103 at the cluster redshift ) . the mass errors represent combined errors obtained by bootstrap resampling of the weak lensing catalogs , performing a similar experiment for the strong lensing data , and estimating systematic uncertainties from the mass differences obtained using different initial models ( see discussion in the next section ) . we plot the integrated , azimuthally - averaged mass profiles @xmath104 for the two components in fig . [ fig : profile ] . we fit power - law models ( @xmath105 ) to these profiles , and find logarithmic slopes of @xmath106 for the main cluster and @xmath107 for the subcluster . in terms of the surface mass density @xmath108 , this translates to @xmath109 , and @xmath110 . adopting an isothermal profile ( @xmath111 ) , we find the line of sight velocity dispersions for the main and the sub cluster are @xmath112 and @xmath113 respectively ; in good agreement with the analysis in sect . [ sec : datawl ] . however , as already mentioned in that section , given the size of the errors , the asphericity of the system , and the complicated geometry ( the relative contributions of the main and sub cluster are difficult to disentangle ) , the velocity dispersion and slope measurements should only be used as guidelines . because of this cluster s violent history , we can not expect the assumption of isothermality to hold . therefore ( and for the reasons stated above ) it is also non - trivial to compare our velocity dispersion measurements with the ones from @xcite . their estimate for the main cluster agrees with ours , while the velocity dispersion of a subcluster is much lower in their case . an isolated lens with the low velocity dispersion measured by @xcite for the subcluster would have multiply imaged systems only at radii @xmath114 , a factor of @xmath115 below that observed . the influence of the main cluster can increase the radial range over which multiple images exist , but not to the extent observed here . we conclude that the true velocity dispersion of the subcluster is likely to be smaller than that derived from our mass model ( due to the inclusion of mass from the main cluster ) , but larger than that presented by @xcite . one of the important results of the work of @xcite was the measurement of the offset between dark matter and baryons . we confirm these results : our strong and weak lensing offsets for the mass peaks ( by measuring the peaks in @xmath7-distribution ) and the corresponding errors are given in table [ tab : offset ] . to estimate the significance of this offset we also measured the positions of the two peaks ( for the main and the subcluster ) in each of the 1000 reconstructions using the bootstrap resampled weak lensing catalogs ( described below ) . in none of these realizations are the mass peaks coincident with the corresponding x - ray peaks . the 1-@xmath11 error bars are a factor of @xmath116 and @xmath6 smaller than the offsets for the main and subcluster respectively . if these errors are accurate and gaussian distributed , this result translates into 10-@xmath11 and 6-@xmath11 discrepancies between the corresponding gas and the total mass peaks . these offsets directly demonstrate that dark matter is present and is the dominant mass component of this cluster . lastly , we compare the projected gas mass ( the details of modeling will be given in @xcite ) vs. the total mass within elliptical regions centered on the two peaks of the x - ray emission and the southern cd of the main and bcg of the subcluster , respectively ( see table [ tab : massgas ] ) . the ratio of gas mass to total mass is smaller in the regions centered on the massive galaxies rather than the x - ray peaks , indicating that the gas has been ( to some extent ) stripped , and is no longer coincident with the total mass . in addition , we detect an extension in the total mass map to the nw of arc a. this detection can be partly attributed to the hot x - ray gas ( see fig . [ fig : swunited ] ) . rrr main component & @xmath117 & @xmath118 + main component ( gas ) & 1.06 & 0.84 + sub component & @xmath119 & @xmath120 + sub component ( gas ) & -0.73 & 0.10 [ tab : offset ] to assess the reliability of the reconstructed map , we generate 1000 bootstrap resampled weak lensing catalogs and perform reconstructions on each of these . to further test the reliability of the strong lensing data we create 10 different reconstructions each time removing one of the multiply imaged systems we use . the resulting @xmath7-maps do not change substantially , the main features ( i.e. the ellipticity of the main and the subcluster ) seen in fig . [ fig : swunited ] remain in all of the reconstructions . we also perform a reconstruction using the weak lensing data only ( bottom - left panel of fig . [ fig : weakvsstrong ] ) . from only weak lensing , we clearly detect both the main and sub clusters , but are unable to resolve individual substructures within each with high significance . this failure can in part be attributed to the limited field size used here in contrast to @xcite ( our method at present does not allow us to use larger fields for the reconstruction , adaptive grid methods need to be employed then ) . in addition , the mass estimates from weak lensing reconstruction are unreliable ( due to the the mass - sheet degeneracy ) ; the resulting potential does not reproduce any of the multiply imaged systems observed . traditionally , to overcome the problem of the mass - sheet degeneracy and measure the mass from the weak lensing mass reconstruction , one assumes a vanishing surface mass density at the edge of the field or adopts parametrized model fitting . the former is not applicable here because the field used is too small . the latter will yield reliable mass estimates if the assumed profile is correct , but it is difficult to guess the appropriate model for such a morphologically disturbed cluster . combining strong and weak lensing helps us to break this degeneracy . however , the lack of counter arc candidates and redshifts of multiply imaged systems limits our ability to provide even stronger constraints on the mass distribution . the errors on mass estimates quoted throughout the paper are obtained by summing in quadrature the errors obtained from the reconstructions using bootstrapping of the weak lensing catalogs and removing individual strong lens systems . in addition , we estimate and add systematic uncertainties resulting from differences in the reconstruction when using different initial models ( as explained below ) . in principle , we should incorporate all of these effects simultaneously in the reconstructions , however this is too computational intensive and beyond the scope of this analysis . methods involving strong lensing mass reconstruction with many multiple imaged systems are subject to degeneracies in parameter space . it is difficult to search the parameter space because the @xmath55 function has many local extrema . in our case , there are pronounced , narrow secondary minima where individual systems are reconstructed well by the model ( and the others less so ) and high `` walls '' in @xmath55 space where images happen to form close to the corresponding critical curves . because weak lensing data are noisy , additional , though less pronounced , minima in @xmath55 are created . given that our reconstruction method does not guarantee that we find a global minimum , in principle we would need to check the whole parameter space . this search is unfeasible with any available algorithms , given that the number of parameters ( i.e. , pixels in the mass map ) is @xmath121 . instead , we opt to investigate the use of many initial models . we try both extremes ; i.e. models that are clearly in disagreement with the data ( zero initial conditions , single mass peak centered on the position of the gas ) and models that use alternative methods to reconstruct the data ( two - nie model , two - nie plus cluster members model from strong lensing modeling , weak lensing mass model from @xcite ) . although the details of the reconstruction depend on the initial conditions ( as can be seen from comparing all but bottom - right @xmath7-maps in fig . [ fig : weakvsstrong ] ) , the main features ( se - nw elongation of the main cluster , two mass peaks close to the main cluster cds , an elongation in the mass map toward the gas peak , e - w elongation of the subcluster ) are independent of the particular initial model we use . the resulting @xmath55 contribution of the weak lensing data @xmath122 remains the same for all sets of initial conditions ( @xmath123 for @xmath124 data points ) . the strong lensing contribution is lowest when the two component nie model is used ( @xmath125 ) and highest when the zero model initial conditions ( @xmath126 ) are used . although the absolute values of the @xmath55 have little meaning ( i.e. one can change the regularization parameter to obtain a lower value for @xmath122 because the weak lensing data is noisy ) , the relative values tell us which resulting reconstruction is indeed a better fit to the data ( as the same regularization parameters are used in all cases and it is the fit to the high signal to noise strong lensing data that improves ) . we estimate the systematic uncertainty on our mass estimates by calculating the maximum difference in the resulting mass estimates when using different initial models . we exclude the zero model in this calculation as it does not predict the strong lensing data reliably . the resulting @xmath7-map ( see bottom - left panel in fig . [ fig : weakvsstrong ] ) for this model is suppressed where only weak lensing data are available ( due to the mass - sheet degeneracy ) and produce steep gradients close to arc a and a @xmath127 lower overall mass . however , other features in the reconstruction ( positions and shapes of the mass peaks for the main and the sub - cluster ) remain unchanged . in all cases , the resulting mass peaks `` move '' away from the initial guess ; showing that the regularization is sufficiently flexible to reproduce the features in the data . therefore , we are confident that our results are not biased as a result of the particular choices of regularization and are independent of the initial conditions to the extent presented in fig . [ fig : weakvsstrong ] . finally , the mass measurements presented above depend on the reliability of the redshifts and identification used for strong lensing reconstruction . these ( except for system a ) are not secure and depend upon the model used in sect . [ sec : datasl ] . in particular , the redshifts of the systems around the subcomponents ( and consequently the measurements of its mass ) can change because the subcomponent mass reconstruction does not depend strongly on the modeling around the main component ( where the arc with the known redshift is located ) . the parametrized model used in sect . [ sec : datasl ] to estimate the redshifts of the strong lensing systems likely deviates from the truth model . we tested the dependence of the assumed redshifts by performing 100 reconstructions in which we modified all of the strong lensing system redshifts except for system a. for each reconstruction these uncertain redshifts were changed by assuming a gaussian scatter with @xmath128 and requiring that all sources lie behind the cluster . the resulting mass and position errors were smaller ( by a factor of @xmath1292 ) than the ones quoted throughout the paper , and the mass and position measurements were unbiased ( as expected from such an analysis ) . the mass measurements would change , however , if we systematically increase ( decrease ) the redshifts of all the systems . for example , if we decrease the redshifts of the subcluster systems such that the lowest redshift system of the subcluster ( g ) is at @xmath130 ( and scale the others accordingly ) , then the mass increases by a factor of @xmath131 . if we instead increase all the redshifts , pushing the highest redshift system ( i ) to @xmath132 ) , then the mass of the subcluster will decrease by @xmath133 . note however that these are just estimates , since the redshift estimates of the weak lensing sources also plays a role , and the changes will therefore be even smaller . as a further test , we also perform 10 reconstructions where we remove one strong lens system at a time from the analysis . in fig . [ fig : slsyst ] we show the two that differ the most from the reconstruction that uses all the strong lensing systems . when we remove system i from the reconstruction ( right panel in fig . [ fig : slsyst ] ) the double peaked structure of the subcluster disappears , while the east - west elongation remains . for the main cluster we see the largest deviations from the original map when we remove system d from the reconstruction ( left panel in fig . [ fig : slsyst ] ) . however , these results do not mean that these structures are not real , rather they illustrate some possible systematics when dealing with sparse strong lensing data . we are confident that system d is indeed multiply imaged , and images i2-i3 , at least , are also probably multiply imaged . these tests demonstrate that even though the strong lensing system identification and their redshift determinations are important to the details of the reconstruction , the errors arising from misidentifications / redshift misestimation are within those quoted in the previous section . the predicted redshifts and strong lensing system identification can be tested using spectroscopy . we are planning just such a survey and will therefore be able to test the fidelity of our model in the future . in addition , using spitzer irac data we identify a candidate high redshift object . while bright in all irac bands , the faintness of this galaxy in optical and near infrared bands indicates that it could lie at @xmath134 . it has two components and our predicted critical curve for this redshift ( indistinguishable from the @xmath135 critical curve ) is exactly where it should be if indeed these sources are merging images of the same source ; they are denoted with crosses and plotted together with the critical curve in fig . [ fig : ccurve ] . furthermore , we identify a candidate counter image with similar colors in the irac data just outside the critical curve . all three need to be further analyzed and confirmed ( using spectroscopy and/or nir data ) before we include them in the modeling . we also identify possible counter - images for systems b , j , h , and i as predicted from our model , they are indicated in fig . [ fig : predict ] . if these and high - redshift candidate are confirmed , the predictive power of our model would further indicate that the strong lensing data we use is reliable enough for this work . , the results are shown for the main ( solid line ) and for the subcluster ( dashed line).,scaledwidth=50.0% ] rrrrr main ( gas peak ) & & 2.0 @xmath136 0.2 & 10.8@xmath136 0.6 & 0.19@xmath136 0.03 + main ( s cd ) & & 1.4 @xmath136 0.1 & 16.3 @xmath136 0.9 & 0.09 @xmath136 0.01 + sub ( gas peak ) & & 0.42 @xmath136 0.04 & 2.1@xmath136 0.2 & 0.20 @xmath136 0.06 + sub ( bcg ) & & 0.18 @xmath136 0.02 & 4.3 @xmath136 0.2 & 0.04 @xmath136 0.01 + acs field & 1300 ( 4.9 ) @xmath137 830 ( 3.2 ) & 13 @xmath136 1 & 91 @xmath136 10 & 0.14 @xmath136 0.03 [ tab : massgas ] massive and interacting clusters , while quite rare , are remarkably well - suited to addressing outstanding issues in both galaxy evolution and fundamental physics . however , in order to study the mass distribution , methods relying on hydrostatic ( x - rays ) or dynamical equilibrium are ill - suited for such systems . we have applied a mass reconstruction method based on strong and weak gravitational lensing to the cluster . we use deep , high resolution optical data to identify objects belonging to the same multiple - image systems . the same data are used to obtain weak lensing catalogs allowing us to obtain a strong+weak lensing mass map of the cluster core . our main conclusions are the following : 1 . using the combined strong and weak lensing mass reconstruction we derive a high - resolution , absolutely calibrated mass map ; we get projected , enclosed mass @xmath138 around the main and @xmath139 around the subcluster . we detect the main cluster peak and a distinct mass concentration at the subcluster position , both clearly offset from the location of the x - ray gas in the system ( at 10-@xmath11 and 6-@xmath11 significance for the main and the subcluster respectively - see fig . [ fig : swunited ] and tab . [ tab : offset ] ) . the majority of the mass is spatially coincident with the galaxies , which implies that the cluster mass must be dominated by a relatively collisionless form of dark matter . the high resolution data allow us to significantly detect the shapes of both the main mass component and the subcluster with no prior assumptions on their positions or profiles . we show that the cluster is a very efficient strong lens ; the area enclosed within the critical curve for highest redshift sources is comparable to that of a1689 @xcite . consequently , we plan to conduct a pencil beam survey for high - redshift ( @xmath140 ) objects . at present this is not possible , because we lack both reliable redshift information for the multiple image systems , and multi - color high - resolution imaging data in the main cluster region . the latter is needed to identify counter images for the strong lens systems , and so improve the reliability of the mass map to the south - east of the main cluster . once these data are obtained ( hst cycle 15 ) , we will improve our methodology to include an adaptive grid approach , which will greatly improve the constraints on the location of critical curves for these high - redshift sources . the prospects for such a search are however very promising , as we already have two strong candidate high redshift objects . one is the red arc b , which is constrained to lie at @xmath141 based upon the redshift of arc a and detection in f814w . the other ( see fig . [ fig : ccurve ] ) is spitzer irac - selected with a likely redshift of @xmath134 . using this cluster as a gravitational telescope will therefore allow us to study galaxy properties near the epoch of reionization . the high resolution mass map of this cluster , in addition to enabling the high - z galaxy search , will be valuable for several other ongoing science programs . most significantly , we are currently running simulations to reconstruct the dynamical history of the cluster merger , and provide an independent constraint on the dark matter self - interaction cross - section . the relative masses of the main and the subcluster , and the central mass distribution , two critical parameters for the simulations , were the limiting source of uncertainty prior to this work . we would like to thank marco lombardi , roger blandford , peter schneider , and steve allen for many useful discussions that helped to improve the paper , and mischa schirmer for producing the color image used for fig . [ fig : predict ] . we are grateful to the anonymous referee for valuable comments and suggestions that helped to improve this manuscript , and for pointing out a typo in our previous paper of the series . further we would like to thank volker springel for providing us with the simulations used in the paper . support for this work was provided by nasa through grant number hst - go-10200.05-a from the space telescope science institute , which is operated by aura , inc . , under nasa contract nas 5 - 26555 . mb acknowledges support from the nsf grant ast-0206286 . this project was partially supported by the department of energy contract de - ac3 - 76sf00515 to slac . dc and dz acknowledge support from nasa through grant number ltsa04 - 0000 - 0041 . mm acknowledges support from chandra grant go4 - 5152x . , d. , schneider , p. , aragn - salamanca , a. , bremer , m. , de lucia , g. , halliday , c. , jablonka , p. , milvang - jensen , b. , pell , r. , poggianti , b. , rudnick , g. , saglia , r. , simard , l. , white , s. , & zaritsky , d. 2006 , , 451 , 395 , p. , romer , a. k. , peterson , j. b. , chase , w. , runyan , m. , holzapfel , w. , kuo , c. l. , et al . 2004 , in aip conf . proc . 703 : plasmas in the laboratory and in the universe : new insights and new challenges , 361366 , c. , brown , m. l. , barden , m. , caldwell , j. a. r. , jahnke , k. , peng , c. y. , rix , h .- w . , taylor , a. , beckwith , s. v. w. , bell , e. f. , borch , a. , huler , b. , jogee , s. , mcintosh , d. h. , meisenheimer , k. , snchez , s. f. , somerville , r. , wisotzki , l. , & wolf , c. 2005 , , 361 , 160 , c. , van waerbeke , l. , bacon , d. , berge , j. , bernstein , g. , bertin , e. , bridle , s. , brown , m. l. , clowe , d. , dahle , h. , erben , t. , gray , m. , hetterscheidt , m. , hoekstra , h. , hudelot , p. , jarvis , m. , kuijken , k. , margoniner , v. , massey , r. , mellier , y. , nakajima , r. , refregier , a. , rhodes , j. , schrabback , t. , & wittman , d. 2006 , , 368 , 1323

for the past two decades there has been a revival of interest in issues related to charge migration in dna based oligomers . on one side , charge transfer ( ct ) in dna is supposed to play a key role in self - repair processes of dna damage in natural conditions , i.e. via oxidative stress @xcite . on the other side , dna may have a huge potential application field in the development of complex nano - scale electronic devices with self - assembling properties , since self - assembly , recognition and fully - automatic synthesis of oligonucleotides @xcite can allow for building almost any imaginable two and three dimensional dna motifs ( e.g. dna - origami ) @xcite . however , besides considerable progress in elucidating the possible charge transfer mechanisms efficient in dna , there is still disagreement about the conduction properties of dsdna , for several experiments obtained quite controversial results ranging from insulating @xcite , over semi - conducting @xcite to even metallic - like behavior @xcite . this broad variability is related not only to the poor control of the experimental conditions in earlier transport experiments , but is also due to the intrinsic structural and electronic complexity of dna oligomers ( e.g. specific base pair sequence ) , which makes the formulation of a general theoretical framework very challenging . nevertheless , several experiments appeared recently @xcite which have shown electrical currents in the na range despite the differences in the studied base sequences and in the experimental setups . one central issue which has emerged by the theoretical treatment of charge transport is the necessity to consider the conformational fluctuations of the molecular frame not as a weak perturbation but rather as a crucial factor promoting or hindering charge motion @xcite . it has been meanwhile shown that it is possible to synthesize dna derivatives , which do not necessarily have the double - strand structure of natural dna . thus , c and g - rich dna strands are able to form four - stranded quadruplex structures , like the i - motif structure composed of two parallel hemiprotonated duplexes intercalated into each other into a head - to - tail orientation , and the g4-quadruplex ( also known as g4-dna ) which is formed by either one , two or four g - rich dna strands in a parallel or anti - parallel orientation @xcite . the latter is supposed to play an important role in some biological processes such as in telomeric dna regions , where it inhibits telomerase and hiv integrase @xcite . additionally , g4-dna is known to interact with various cell proteins that cause diseases like bloom s and werner s syndromes @xcite . moreover , g4-dna is cytotoxic towards tumor cells and therefore , might be a key for the design of anti - cancer drugs @xcite . eventually , g - quadruplexes are found to be thermally more stable than dsdna @xcite . gilbert and sen resolved the x - ray structure of the first g - quadruplex and proposed the formation of these unique structures in presence of monovalent alkali ions @xcite , see also fig . [ fig : meth - tetrad](a ) . consequently , g4-dna can be regarded as stacks composed of individual planar g - tetrads ( or g - quartets ) as shown in fig . [ fig : meth - tetrad](b ) , each formed by 4 guanines bound together by 8 hydrogen bonds via hoogsteen pairing @xcite . for detailed reviews on the structural diversity and special mechanical properties of g4-dna based molecules see ref . @xcite . recently , porath _ et al . _ @xcite have shown the larger polarizability of g4-dna compared to dsdna , when attached to gold surfaces . this might be an indicator of improved ct properties and the potential advantage of using g4-dna rather than dsdna in molecular scale devices . eventually , the higher conductance in g4-dna could be attributed to the increased structural stability and higher number of overlapping @xmath0-orbitals . furthermore , experimental work by kotlyar _ et al . _ @xcite revealed that up to 300 nm long g4-wires could be synthesized revealing a considerable stability even in the absence of metal ions . also , theoretical studies have been performed on the stability and rigidity of g4-dna . molecular dynamics ( md ) simulations by pakov _ et al . _ @xcite could confirm the experimental results . for one thing , g4-dna is more rigid than dsdna and for another monovalent ions within the quadruplex cavity are necessary for the stability of short g4-dna . on the other hand , md simulations @xcite pointed out the stability of longer g4 molecules ( with 24 g - tetrads ) in absence of metal ions which is consistent with the synthetic procedure of ref @xcite . electronic structure calculations by di felice and co - workers @xcite showed a higher degree of delocalization of the electronic states in g4 than in dsdna . moreover , the presence of metal ions may contribute additional states supporting ct , though this is a still unresolved issue . guo _ et al . _ @xcite have recently shown using the landauer theory ( coherent transport ) that g4-dna exhibits much larger delocalization lengths at the band center compared to double - stranded poly(g ) dna . surprisingly , it was also found out that the delocalization length can be even enhanced via environment - induced disorder through the backbones . though these studies assume a static atomic structure , they nevertheless suggest that disorder plays an important , non - trivial role in mediating ct in dna . similarly , it is known from other biomolecules such as proteins that electron transport is dominated by non - equilibrium fluctuations which has lately been analyzed by balabin _ et al . _ as demonstrated in previous studies , idealized static structures are not representative when considering ct properties , since dynamical as well as environmental effects were shown to be too important @xcite . it was indicated that dynamic disorder has dramatic effects , since it suppresses ct in homogeneous sequences on the one hand , but can enhance ct in heterogeneous sequences , on the other . for instance , in our previous work @xcite , the conductance of the dickerson dodecamer ( sequence : 5-cgcgaattcgcg-3 ) was found to be almost one order of magnitude larger in solution ( qm / mm ) than in vacuo . interestingly , similar findings have recently been shown by scheer and co - workers , for they obtained for a heterogeneous sequence ( i.e. a and g bases are present ) with 31 base pairs @xcite a two order of magnitude larger conductance in solution compared to the in vacuo measurements @xcite . furthermore , it could be shown that only a minor part of the conformations is ct - active @xcite . therefore , neglecting these significant factors or assuming a purely random disorder distribution can lead to a considerable loss of a vital part of ct - relevant structural and electronic information . relying on our previously developed methodologies to deal with ct in different dna oligomers @xcite , we present here a detailed investigation of charge transport in g4-dna . the paper is organized as follows , in sec . [ sec : meth ] details of the md simulations and of the electronic structure approach are described . subsequently , the md and the electronic structure data are analyzed in sec . [ subsec : res - md ] and [ subsec : res - es ] , respectively . finally in sec . [ subsec : res - landauer ] , ct results for g4 molecules are compared with those of dsdna in solution and in vacuo . more importantly , we point out significant factors which are responsible for the enhanced conductance in g4 molecular wires . the molecules used in this work are based on the x - ray crystal structure of a tetrameric parallel - stranded quadruplex ( 244d ) formed by the hexanucleotide sequence d(tg@xmath1 t ) in the presence of sodium ions . the structure has been resolved at 1.2 resolution by laughlan _ et al . _ @xcite . as shown in fig . [ fig : meth - tetrad](a ) , it contains 2 pairs of quadruplexes ( tg@xmath1t)@xmath1 , for which each pair is stacked coaxially with opposite polarity at the 5 ends . moreover , 9 sodium ions are located inside each stack of quadruplexes illustrating well ordered g4-dna constructs . however , the terminal thymine residues , shown in gray , were not completely resolved because of high thermal disorder @xcite . as indicated in fig . [ fig : meth - tetrad ] only one of these 4 quadruplexes is taken as our basis structure , which corresponds to the parallel strands a , b , c and d from the pdb file 244d . this quadruplex , from now on denoted as ( tg@xmath1t)@xmath1 , will be used for simulations and calculations . furthermore , two longer g4 quadruplexes with 12 and 30 tetrads , denoted as ( g@xmath2)@xmath1 and ( g@xmath3)@xmath1 , are generated by omitting the terminal thymine residues and adding subsequently g4 tetrads with a distance of 3.4 and twisted by 30 . for comparison , corresponding double - stranded b - dnas with base sequence 5-tggggt-3 , poly(g ) denoted as g@xmath2 and g@xmath3 , respectively , and a sequence containing 31 base pairs @xcite ( scheer ) are built with the make - na server @xcite . backbones are indicated as light blue ribbons , central coordinated sodium ions as dark blue spheres , terminal thymine and guanine residues are shown in gray and orange , respectively . the highlighted quadruplex is used as starting structure and also to generate longer g4 molecules ( g@xmath2)@xmath1 and ( g@xmath3)@xmath1 . b ) lewis structure of a single g4-tetrad with c@xmath4 symmetry containing a monovalent metal ion in its center . guanines are bound together by 8 hydrogen bonds via hoogsteen pairing @xcite . the metal ion is coordinated either coplanar by 4 or cubic by 8 o6 oxygen atoms of the respective guanines which depends on the ionic radius of m@xmath5 . ] depending on the helix length the g4 and dsdna molecules are solvated in a rectangular box with 4000 to 11000 water molecules using the tip3p model @xcite and periodic boundary conditions ( pbc ) are applied . in order to neutralize the total charge of the system , due to the negatively charged backbones , an appropriate number of sodium counterions ( na+ ) are added . concerning the quadruplexes , we carried out 4 simulations for ( tg@xmath1t)@xmath1 and ( g@xmath2)@xmath6 , respectively , firstly without any central monovalent alkali ions within the quadruplex , and then in presence of either lithium ( li+ ) , sodium ( na+ ) , or potassium ( k+ ) ions within the respective g4 molecules . therefore , either 3 ( ( tg@xmath1t)@xmath1 ) or 11 ( ( g@xmath2)@xmath6 ) of these ions are subsequently placed in the center between two g4 tetrads coordinated by o6 oxygen atoms ( see also fig . [ fig : meth - tetrad](b ) ) . nevertheless , the longer ( g@xmath3)@xmath1 quadruplex is simulated only twice in absence and presence of central sodium ions . all simulations are carried out with the gromacs software package @xcite using the amber parm99 forcefield @xcite including the parmbsc0 extension @xcite . after a standard heating - minimization protocol followed by a 1 ns equilibration phase , which is discarded afterwards , we performed 30 ns ( 50 ns for ( g@xmath3)@xmath1 ) md simulations with a time step of 2 fs . snapshots of the molecular structures were saved every 1 ps , for which the charge transfer parameters were calculated with the scc - dftb - fo approach as described in sec . [ subsec : meth - hop ] . in this section we will shortly describe how the electronic structure of g4-dna molecules is mapped to a coarse - grained transfer hamiltonian using the fragment orbital ( fo ) approach . the method has already successfully been applied to dsdna molecules @xcite . a more detailed description of the methodology can be found in ref . @xcite . to begin with , the electronic structure of the dna system is written in an effective tight - binding basis ( fragment basis ) as : @xmath7 the onsite energies @xmath8 and the nearest - neighbor hopping integrals @xmath9 characterize , respectively , effective ionization energies and electronic couplings of the molecular fragments . the evaluation of these parameters can be done very efficiently using the scc - dftb method @xcite combined with a fragment orbital ( fo ) approach @xcite : @xmath10 and @xmath11 the indexes @xmath12 and @xmath13 correspond to the fragments of the quadruplexes which , in this model , are constituted by single guanine bases . accordingly , the molecular orbitals @xmath14 and @xmath15 are the respective highest - occupied molecular orbitals which are obtained by performing scc - dftb calculations for these isolated fragments , i.e. the individual guanine bases . using a linear combinations of atomic orbitals ansatz @xmath16 the coupling and overlap integrals can be efficiently evaluated as : @xmath17 and @xmath18 @xmath19 and @xmath20 are the hamilton and overlap matrices in the atomic basis set as evaluated with the scc - dftb method . note , since @xmath21 is built from non - orthogonal orbitals @xmath14 and @xmath15 , which is for various problems not suitable , we apply the loewdin transformation @xcite : @xmath22 the effect of the environment , the electrostatic field of the dna backbone , the water molecules and the counter - ions ( including the central alkali ions within the quadruplex ) , is taken into account through the following qm / mm hamiltonian : @xmath23 @xmath24 are the mulliken charges in the qm region and @xmath25 are the charges in the mm region , the dna backbone , counterions and water molecules . the coupling to the environment is therefore explicitly described via the interactions with the @xmath25 charges . in the following , we will denote the calculation set - up based on the complete expression in eq . [ eq : qmmm ] as qm / mm ; neglecting the last term will be denoted as `` vacuo '' . the electronic parameters @xmath8 and @xmath9 are evaluated for every snapshot of the simulations . subsequently , the transfer hamiltonian is used to calculate the ct in the quadruplexes as described in sec . [ subsec : meth - tc ] . once the transfer hamiltonian has been constructed , the transport properties will be calculated using landauer theory for each snapshot along the md trajectories . for this , we consider a two - terminal set up where the g4 oligomer is contacted to left ( l ) and right ( r ) metallic electrodes through the four terminal bases . a central quantity to be computed is the molecular ( g4 ) green function , which can be obtained through a dyson equation : @xmath26 here @xmath27 and @xmath28 are self - energy matrices characterizing the coupling to the electrodes . to simplify the calculations we do not take the full energy dependence of the self - energies into account , but rather use the so called wide band limit , where @xmath27 and @xmath28 are replaced by energy - independent parameters @xcite : @xmath29 we set @xmath30 and @xmath31 to 1 mev . within the landauer approach , the transmission function @xmath32 for a given set of electronic parameters is then obtained as : @xmath33\end{aligned}\ ] ] where @xmath34 and @xmath34 are the broadening matrices calculated as the anti - hermitian part of @xmath35 : @xmath36\end{aligned}\ ] ] using the former expressions , the conformational ( time ) dependent electrical current will be simply calculated by : @xmath37 the i - u characteristics presented in this work should be interpreted only qualitatively , for the fermi energy ( @xmath38 ) is artificially placed as average of the onsite energies for all guanine sites and for each snapshot , respectively . the reader should note that the current could exhibit quite different shapes depending on where @xmath38 is located . however , only the current - voltage gap is affected by @xmath38 , whereas the maximum current obtained at high voltages is not altered . the transmission function and the current are evaluated for every snapshot of the md simulation in order to obtain statistical average quantities . however , we are aware of the limits of the coherent transport model which is valid only in the adiabatic regime . hence , we assume that the timescales of the ct process is shorter than the fastest dominant structural fluctuations . this issue has also been addressed in ref . @xcite . first of all , the ( tg@xmath1t)@xmath1 quadruplex from the crystal structure 244d has been simulated in order to validate further simulations with 12 and 30 tetrads . corresponding rmsd time series in absence and presence of central ions as well as molecular snapshots after 30 ns simulation time can be found in sec . si.a . in ref . based on the crystal structure 244d we built up an artificial all - parallel stranded guanine quadruplex ( g@xmath2)@xmath6 without terminal thymine residues and 12 stacked guanine tetrads ( for details see [ subsec : meth - md ] ) . as for ( tg@xmath1t)@xmath1 , we carried out 4 simulations , one in absence and three in presence of 11 lithium , sodium and potassium ions initially placed in the center of two adjacent g4 tetrads . the rmsd time series and molecular snapshots after 30 ns md simulation for ( g@xmath2)@xmath6 and poly(g ) ( g@xmath2 ) are given in fig . [ fig : rmsd2 ] and fig . [ fig : strucs-12g4 ] , respectively . ) @xmath1 in absence and presence of centrals ions li@xmath5 , na@xmath5 and k@xmath5 , b ) comparison with double - stranded g@xmath2 . ] the rmsd values show that without central ions the quadruplex can not be considered as a g4 structure in equilibrium . in fact , the rmsd increases for the whole simulation time from 1.5 to 3 indicating considerable structural changes . in contrast , the rmsd for structures containing central ions reveal rather consistent values around 1 to 1.4 with only minor fluctuations suggesting rigid g4 structures with only slight structural deviations . compared to the ( tg@xmath1t)@xmath1 simulations , the type of ions within the quadruplex has a lesser influence on the structural deviations , since the rmsd values are rather similar . in order to validate these small structural deviations and therefore the rigidity of g4-dna we carried out a rmsd calculation for double - stranded g@xmath2 ( fig . [ fig : rmsd2](b ) ) . generally , the rmsd values and their fluctuations for the dsdna molecule are substantially larger than those for the quadruplex structures . this clearly supports the notion that guanine quadruplexes are much more rigid than corresponding dsdna and therefore g4-dna reveals less dynamical disorder which might promote ct . ) @xmath1 after 30 ns md simulation in absence and presence of centrals ions li@xmath5 ( green spheres ) , na@xmath5 ( blue spheres ) and k@xmath5 ( purple spheres ) . ] the snapshots in fig . [ fig : strucs-12g4 ] exhibit that ( g@xmath2)@xmath6 in absence of central ions is not completely destabilized which has not been the case for ( tg@xmath1t)@xmath1 ( see sec . si.a . in ref . the inner tetrads roughly maintain a regular g4 structure , whereas the outer ones show a considerable disordered behavior with no planarity at all , even the g4 hydrogen bridge pattern has been entirely destroyed . recently , cavallari _ et al . _ have shown in their md simulations that long g4 quadruplexes can be stable without central ions @xcite . this finding was supported by experimental results in which longer g4 quadruplexes could be synthesized and characterized in absence of monovalent ions @xcite . therefore , an additional 50 ns simulation of a quadruplex with 30 tetrads ( g@xmath3)@xmath1 in absence of central ions is carried out . the simulations clearly exhibit that the major part of the quadruplex forms a quite regular structure with adequate parallel stacking . therefore , it seems to confirm previous results that large g4 wires in absence of central ions become more stable , reasonably due to increased stacking interaction . however , like for the dodecamer , towards the ends the quadruplex is slightly deformed , especially at the 3 terminus . moreover , rmsd values reveal that the quadruplex is not completely structural balanced . the major part of structural fluctuation originates from the 5 and 3 ends which is shown by four groups of large peaks in root mean square fluctuations per atom ( rmsf ) . the corresponding data is shown in fig . s2 in ref . @xcite . the quadruplexes which contain central monovalent ions in fig . [ fig : strucs-12g4 ] exhibit highly regular g4 structures indicating again the strong rigidity and stability of these molecules . generally , the ions appear to prefer different coordinations within the quadruplex cavity , that is lithium favors the planar position within the tetrads while sodium and potassium are most likely to be found in the center of two g4 planes . however , one sodium ion in the middle of the quadruplex is located in - plane and two others are rather closer to one of the corresponding tetrads than sandwiched between them . also , lithium ions are sometimes found slightly apart from the in - plane position . this might be an indicator for the longitudinal mobility of central sodium and lithium ions within guanine quadruplexes . in contrast , potassium ions are rather fixed in their center position between two tetrads , most likely because of their larger ionic radii compared with lithium and sodium . the resulting larger rigidity of g4 compared to dsdna plus the preferred locations of central ions within the quadruplex are in perfect agreement with findings obtained by pakov _ et al . _ as well as previous work by cavallari _ et al . _ @xcite . notwithstanding , the ions are equally capable of moving out of the quadruplex through the minor groove and the 5 and 3 ends . in our simulations 3 sodium , 3 lithium and 2 potassium ions went out of the g4 dodecamer , respectively . sometimes , these ions remain in the minor groove for several ps . likewise , solvent molecules are able to infiltrate the quadruplex via the minor groove and the 5 and 3 ends provided that these locations are not already occupied by initially placed ions . usually , the infiltration of solvent molecules into the helix will not significantly destabilize the structure as long as there is at least a minimum amount of central monovalent cations left in the quadruplex . otherwise , the interior solvation might destroy the regular 4-stranded structure as can be seen for the structures without central ions in fig . [ fig : strucs-12g4 ] . these observations are consistent with results of previous works @xcite . before calculating ct characteristics , we have to focus on the electronic parameters , especially the molecular orbitals involved in the hole transfer process . similarly to our previous work on ct parameters in dsdna @xcite we carried out benchmark calculation of g4 tetrads with dftb and compared to dft ( pbe and b3lyp ) as well as hf calculation . since we intend to deal with hole transfer we are most notably interested in the highest molecular orbitals which ought to have @xmath0-symmetry in order to have sufficient mo overlap along the quadruplex . tab . [ tab : mo - ideal ] shows energies of the 5 highest occupied mos for a idealized g4 structure with c@xmath4-symmetry obtained after geometry optimization with b3lyp/6 - 31g(d , p ) . corresponding visualization of mos can be found in tab . s3 in ref . @xcite . .energies of highest occupied molecular orbitals for a idealized g4 tetrad with c@xmath4-symmetry , comparison between dftb , dft and hf , for the latter the 6 - 31g(d , p ) basis set is used , all values in ev . [ cols="<,^,^,^,^,^",options="header " , ] as has been discussed above , the central ions stabilize the g4 structure significantly , the structural fluctuations of g4 with ions are much lower compared to g4 without ions or dsdna . the effect of ions on the electronic parameters is twofold : first of all , the onsite energies are shifted down by about 0.5 ev but , surprisingly , the enhanced structural stability does not lead to smaller fluctuations of the onsite parameters . they are in the order of 0.4 ev , similar to the situations in dsdna , as discussed recently @xcite . these large parameter variations are introduced by solvent fluctuations ( and not by fluctuations of the dna / g4 structure itself ) , therefore , they are not affected by the higher structural rigidity . the t1 values , on the other hand , are even lower in the g4 structures with ions . this would indicate , that these structures conduct even less , when compared to four strands of dsdna . however , the t2 and t3 values in g4 are still large , indicating that interstrand transfer can occur quite frequently . this opens a multitude of pathways for charge transport in g4 ( compared to dsdna ) , which will be the key to understand g4 conductivity , as discussed in more detail below . as a first result , we do not see any indication that the higher structural stability of g4 with central ions will lead to a higher conductivity due to a reduced dynamical onsite disorder or due to increased electronic couplings , i.e. because of somehow better stacking interactions owing to the more regular structure . therefore , the higher conductivity must have different reasons . the more stable structure leads to smaller couplings , in contrast to prior expectations @xcite . a more detailed analysis of onsite energies and electronic couplings is given in tab . s6-s11 in ref . @xcite . the parameters @xmath39 and t@xmath40 analyzed in the previous section are now used to evaluate the charge transport properties using the approach described in sec . [ subsec : meth - tc ] . first , we focus on static structures to get insight in particular ct features for g4 quadruplexes and also to have a reference for the following calculations which are applied to the whole md trajectory . later on , we are interested in ensemble averages , for single snapshots or static structure can not elucidate the ct process in dna . subsequently , we will compare the ct properties of g4-dna with those of double - stranded dna and distinguish the differences in ct efficiency by means of conformational analysis . additionally , we determine the effects of the mm environment and in particular of the central ions on ct in g4 quadruplexes . primarily , we compare t(e ) and i(u ) for three static structures : i ) idealized g4-dimer as used in tab . [ tab : hop - g4 ] , ii ) the two central tetrads of the crystal structure 244d and iii ) double - stranded idealized b - dna ( g@xmath41 ) , i.e. two stacked g - c base pairs ( 36 twist and 3.4 rise ) . as can be seen from fig . [ fig : static - strucs ] there are only slight structural differences between the idealized g4-dimer and the stacked tetrads of the crystal structure 244d . fig . [ fig : g4-trans - ideal ] exhibits the corresponding transmission functions showing eight resonances due to the eight g sites of the transport model . however , since the idealized g4-dimer has a very regular structure these eight peaks lie in a very narrow energy range of about 0.15 ev and have similar high transmission values of at least 0.9 . on the other hand , the eight peaks of the two central tetrads of 244d lie in a broader energy range of about 0.26 ev and have lower transmission values ranging from 0.4 to 0.8 . this indicates , how small structural deviations lead to a broadening and a decrease of the transmission function . nevertheless , we know from the values in tab . [ tab : hop - g4 ] that the t1 and t2 couplings are much larger for the crystal structure than for the idealized structure , thus the cause for decreased conductivity has to be the onsite energy disorder . ) , i.e. two stacked g - c base pairs ( 36 twist and 3.4 rise ) . ] , comparison of transmission function @xmath32 and current @xmath42 for idealized g4-dimer , the two central tetrads of crystal structure 244d and poly(g ) ( g@xmath41 ) . note , only the stacked g bases are considered , whereas the cytosine bases and the backbone composed of sugar and phosphate groups are neglected . ] [ fig : g4-trans - ideal ] also shows the corresponding current - voltage characteristics . the current for the g4 structures is significantly larger than for g@xmath41 . this is what we would expect , because we compare two idealized molecular units , one with two relevant transmissive states and the other with eight within the relevant energy window . note that in all current calculations shown in this work , the fermi energy is artificially placed as the mean value of onsite energies for each snapshot respectively . generally , the conductance for the idealized double - stranded poly(g ) dna would become twice as large once the cytosine states are included , though it is well established that the c states have only a minor impact in hole transfer due to their higher oxidation potential than g. previous studies revealed that idealized static structures can not exhibit reasonable ct properties in dna , since dynamical as well as environmental effects were shown to be crucial @xcite . dramatic effects are induced by dynamic disorder , for it suppresses ct in homogeneous sequences on the one hand , but can also enhance ct in heterogeneous ( random ) sequences on the other . moreover , only a minority of conformations appears to be ct - active as has been indicated in ref . @xcite . therefore , neglecting these significant factors or assuming purely random distributions for dynamical disorder leads to a considerable loss of a vital part of the ct in dna . on this account the ct properties are evaluated for every snapshot along classical md trajectories which then leads to ensemble averaged quantities , that is the average transmission function @xmath43 and the current @xmath44 . in fig . [ fig : compare - dsdna1 ] @xmath43 and @xmath44 are shown for both quadruplex molecules ( with central sodium ions ) , the two central tetrads of ( tg@xmath1t)@xmath1 ( a ) and ( g@xmath45)@xmath6 ( b ) as well as their corresponding double - stranded poly(g ) analogues g@xmath41 and g@xmath45@xcite . because of the substantial onsite energy fluctuations of about 0.4 ev due to the dna environment and dynamics , the transmission function reveals large broadening for both dna species . the transmission maxima for the quadruplexes are shifted to lower energies by about 0.3 ev due to the presence of the central sodium ions . basically , the average transmission is strongly reduced compared to the idealized static structures in fig . [ fig : g4-trans - ideal ] . nevertheless , the maximum of @xmath43 for the two central tetrads in ( tg@xmath1t)@xmath1 is almost five times larger compared to g@xmath41 . accordingly , the maximum current is about 4.4 times larger in the quadruplex . considering the octamers in fig . [ fig : compare - dsdna1](b ) the conductance difference between g4 and poly(g ) even increases . here , the average transmission in the relevant energy range for the quadruplex is to a great extent two orders of magnitude larger than those for the poly(g ) sequence . the poly(g ) spectra shows much larger spikes at certain energies . these spikes indicate the strong impact of dynamics and may be explained by the existence of few charge transfer active conformations , which dominate the average transmission , i.e. few conformations , which feature a high transmission . this may indicate that even longer sampling ( more than 30 ns ) would be required to converge the spectra . notwithstanding , it also reveals that in dsdna the average current is dominated to a much larger degree on few non - equilibrium structures , as concluded from our earlier work @xcite . as a result , the average current @xmath44 for ( g@xmath45)@xmath6 is almost two orders of magnitude larger than for g@xmath45 suggesting that the enhancement of ct in g4 with respect to poly(g ) might grow with increasing dna length . thus , additional sets of ct calculations for longer dna species with 14 and 20 tetrads ( base pairs ) , respectively , are carried out . the corresponding data was obtained from md simulations of a quadruplex ( g@xmath3)@xmath1 ( including central na@xmath5 ) , and double - stranded dnas g@xmath3 and a heterogeneous sequence containing 31 sites @xcite recently used by scheer and co - workers in a ct - measurement @xcite ( note , for ct calculations only the 14 ( 20 ) central sites are used ) . as expected , the transmission strongly decreases for both dna species . however , in g4 this effect is not as strong as in dsdna . for instance the current for ( g@xmath46)@xmath1 is only about 1 order of magnitude lower than for ( g@xmath47)@xmath1 , although expanding to ( g@xmath48)@xmath1 the current drops significantly additional 3 orders of magnitude . in contrast , @xmath44 for poly(g ) and the `` scheer '' sequence decreases by more than 10 orders of magnitude by increasing the number of base pairs from 14 to 20 . this indicate there is a much stronger distance dependence of ct in dsdna , hence the notion of coherent ct for longer molecular wires might be considerably more likely in g4 than in dsdna . admittedly , the landauer formalism used in this work performs well for short dna species ( less than 10 sites ) , where the transport is assumed to be at least partially coherent . on the other hand , it clearly fails for long dna sequences , hence the ct results for the longer molecules should be interpreted only qualitatively and with caution . for instance , the currents obtained for the 14mer and 20mer of both dsdna molecules poly(g ) and the scheer sequence are orders of magnitude smaller than pa which is far beyond any measurable range . the complete data is given in fig . s3 in ref . @xcite . the significant conductance difference of g4 and dsdna may not be attributed to the fact that g4 is composed of four poly(g ) like wires . to analyze this further the ct in ( g@xmath47)@xmath1 with central sodium ions ( full md ) is compared to various models , in which i ) only intra- and interstrand@xcite couplings are non - zero ( intra+inter ) and ii ) the quadruplex is separated into its 4 single strands , for which the intrastrand transport is calculated independently and added up afterwards ( @xmath494s ) . furthermore the results are analyzed with reference to g@xmath45 ( 4xpg ) . note , for comparison the ct quantities of poly(g ) are multiplied by 4 . as appears from fig . [ fig : transport - fullvsintrainter ] , the average transmissions for the two models : full md and intra+inter in the spectral support region clearly reveal the largest plateaus , also showing similar peak structures . however , both quadruplex models exhibit similar moderate fluctuations . by contrast , the sum of the four single g4 strands ( @xmath494s ) shows much larger fluctuations , comparable with those for g@xmath45 ( 4xpg ) . moreover , the average transmission for @xmath494s is significantly reduced and is even slightly lower than 4xpg . interestingly , in @xmath494s are barely ct - active conformations , i.e. single dominating peaks like in 4xpg which sometimes even outreach the maximum transmission of the quadruplex ( full md ) . this might reflect the structural differences between quadruplexes and dsdna , since the four strands in the quadruplex are not as flexible as those in poly(g ) ( see also rmsd fluctuations in fig . [ fig : rmsd2 ] ) , hence in poly(g ) the structural phase space is much larger and therefore , several high - transmissive structures can arise . on the other hand , the more rigid quadruplex covers only a smaller conformational phase space which indeed ensures a large number of structures showing moderate ct properties for each single strand , respectively . this clearly indicates , that the most important factor for the enhanced conductance in g4 are the interstrand couplings between the 4 strands in the quadruplex . thus , if there are conformations in which the 4 isolated channels are not transmissive , there is a considerable probability that ct might occur via coupling between the individual stands . as can be extracted from tab . [ tab : hop - g4 ] those interstrand couplings are sufficiently large with about 0.01 - 0.02 ev . as a consequence , at each snapshot there is a substantial amount of pathways over which the ct might occur through the quadruplex . these findings are supported by the i - v characteristics in fig . [ fig : transport - fullvsintrainter ] . the current for the intra+inter model almost matches that of the full md with 2.0 and 2.3 na , respectively . if we switch off the interstrand couplings in the quadruplex ( @xmath494s ) the current will drop down to 0.08 na which is more than order of magnitude smaller . a slightly larger current of 0.11 na is obtained for 4xpg . for further insights into the different ct properties between g4 and dsdna we make use of conformation analysis . for that purpose , we investigate the amount of conformations which dominate the average ct . additionally , the distribution of transmissions for the multitude of conformations is explored for both ( g@xmath47)@xmath1 and g@xmath45 . the results are presented in fig . [ fig : transport - conf - g4vspg ] . panel a ) evidently indicates that there are substantially more conformations contributing to the average ct in g4 than in poly(g ) . consequently , virtually every 10th g4 conformation is ct - active ( about 3000 out of 30,000 ) , whereas only 128 ( again out of 30,000 ) single non - equilibrium structures characterize the average ct in poly(g ) . ) @xmath1 ( na@xmath5 ) and g@xmath47 , a ) : number of conformations that make up 90% of the average transmission maximum @xmath50 , b ) : probability distribution functions ( pdfs ) of @xmath51 for ( g@xmath47)@xmath1 ( na@xmath5 ) and the two models 4xpg and @xmath494s as used in fig . [ fig : transport - fullvsintrainter ] . ] second , as demonstrated in the transmission probability distribution functions ( pdfs ) in fig . [ fig : transport - conf - g4vspg]b ) , the distribution width for the g4 quadruplex is considerably narrower compared to 4xpg and @xmath494s . note the x - axis is scaled logarithmic . in addition , the g4 pdf is significantly shifted to higher transmission . this underscores that the majority of poly(g ) structures is not transmissive , yet a few single conformations are responsible for the average ct . on the other hand , for @xmath494s the major part of conformations reveal higher transmission than in 4xpg , but the single dominating conformations are missing . clearly , this explains that these single high - transmissive poly(g ) conformations are the reason for a better conductivity in 4xpg compared to @xmath494s . the higher average t1 values in dsdna are due to few highly conducting conformations . the smaller average of the g4 t1 couplings therefore resembles a more stable structure , however , not leading to a higher conductance , as could be argued beforehand . the advantage of g4 over dsdna is due to the existence of non - negligible interstrand couplings in g4 . the amount of high - transmissive structures is remarkably increased compared to double - stranded poly(g ) dna . [ [ contact - effects ] ] contact effects + + + + + + + + + + + + + + + in most of the conductance experiments for dna the molecules are only connected with one strand to the respective left and right contacts . for instance , in a very recent experiment by scheer _ a single stranded nucleotide with sequence 5-(t@xmath52g@xmath53[ttaggg]@xmath54t@xmath52)-3 ( t@xmath52 denotes modified thymine residues ) , which is known to form stable quadruplexes , has been attached between two contacts @xcite . therefore , the question arises whether the ct in an all - parallel stranded quadruplex differs if only 2 or even 1 strand of the quadruplex are coupled to the left and right contacts , respectively . for that purpose , the ct in ( g@xmath47)@xmath1 ( na+ ) is calculated for various contact models : i ) all 4 strands are connected to the left and right electrodes , respectively ( 1 - 4 ) , ii ) only one strand is connected to both contacts ( 1-intra ) and iii - v ) one strand is attached to the left electrode while one of the remaining is contacted to the right one ( 12-inter , etc . ) . as can be seen in fig . [ fig : transport - contact - ana ] , reducing the number of connected strands from 4 to 1 leads to an decrease in @xmath43 for the relevant energy range of about 1 order of magnitude . however , the transmission for the 1-intra and inter models is very similar indicating a minor significance for which strand or strands are connected to the contacts . this finding is also supported by the i - v characteristics , for the current ranges around 0.2 na for these models which is roughly one order of magnitude smaller as though all 4 strands are attached ( 1 - 4 model ) . ) @xmath1 ( na@xmath5 ) : average transmission and current calculated for various contact models i ) all 4 strands are connected to the left and right electrodes , respectively ( 1 - 4 ) ii ) only one strand is connected to both contacts ( 1-intra ) , iii - v ) one strand is calculated to the left electrode while one of the remaining is contacted to the right one ( 12-inter , etc . ) , and vi ) comparison with poly(g ) . ] interestingly , there seems to be an increase in the transmission fluctuations if the quadruplex is contacted through only one strand at each end . notwithstanding , the transmission plateau for the 1-stranded ( intra and inter ) contacted models is still about 1.5 orders of magnitude larger compared to those for poly(g ) resulting in an average current which is again 1 order of magnitude smaller with 0.025 na . thus our results suggest that independent on the various contact linking schemes of g4 and dsdna one might expect a higher conductivity for the quadruplex . certainly , the optimal conductance for all - parallel stranded g4-dna is ensured if all four strands are coupled to the contacts . [ [ effect - of - dna - environment - on - transport ] ] effect of dna environment on transport + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as discussed above , the major part of the dynamical disorder is induced by the qm / mm environment , i.e. the last term in eq . [ eq : qmmm ] which is built of the mm charges of dna backbone , solvent and counterions . previous results have indicated that the disorder due to the dna environment might not only suppress ct in homogeneous sequences like poly(g ) , rather it is able to enhance ct in random sequences like the heterogeneous dickerson dodecamer @xcite . very recent experiments by scheer _ _ confirmed this notion , since they found the current for a 31 base pair sequence @xcite to be 2 orders of magnitude smaller in vacuo than in aqueous solution @xcite . therefore , it might be interesting to investigate , what happens if we switch of the environment for ct in g4-dna , also with respect to the effect in dsdna . as can be seen in fig . [ fig : transport - envvsvacuo](a ) , the transmission maximum for ( g@xmath47)@xmath1 in vacuo is about 2 orders of magnitude larger than with the qm / mm environment . besides , its broadening and also the fluctuations are significantly reduced and the plateau is shifted to higher energies due to the neglect of the electrostatic interaction with the sodium ions within the quadruplex . the transmission for poly(g ) in vacuo shows basically the same features , for the broadening is likewise strongly reduced and the maximum is located in the same energy range as for g4 in vacuo , although it is 2 order of magnitude smaller . in general , the transmission for g4 including the qm / mm environment clearly reveals the largest broadening which might indicate that the central sodium ions have an additional strong influence on the dynamic disorder due to longitudinal mobility within the quadruplex , which is not the case in dsdna . despite all that , the current at high voltages is larger for g4 with qm / mm than for poly(g ) in vacuo . basically , for heterogeneous sequences ( not shown here ) a reduced transmission in vacuo is found , which is caused by energy gaps between a and g states . notwithstanding , for dna molecules with no static energy gaps like double - stranded poly(g ) and g4-dna ( both with uniform dna bases ) , the qm / mm environment is most likely to increase the dynamical disorder , thus will suppress ct compared to the vacuo model . as a result , there is no significant difference in the effect of the dna environment on ct for g4 and poly(g ) dna . this is also underscored by conformation analysis given in panel b ) and c ) in fig . [ fig : transport - envvsvacuo ] , which indicate that there are considerably more ct - active conformations in vacuo than with qm / mm environment for both dna species g4 and poly(g ) . moreover , for g4 in vacuo nearly every conformation appears to be high - transmissive , since the average maximum transmission @xmath50 increases almost linearly with the number of conformations . thus , the ct in vacuo is only marginally affected by single non - equilibrium conformations , rather the whole ensemble of g4 conformations seems to be ct active which is demonstrated in the pdf of transmission . one last issue remains when considering ct in g4-dna that is the effect of ions within the quadruplex channel . the structural influence of these ions has already been addressed in detail in sec . [ subsec : res - md ] . as could be seen in sec . sii.b in ref . @xcite we did not find the central ions to contribute states in the relevant energy range for hole transfer in g4 , therefore , the effect will only be investigated electrostatically . in fig . [ fig : transport - ion - ch ] the average transmission and current is shown for the quadruplex simulations of ( g@xmath47)@xmath1 in absence and presence of either lithium , sodium and potassium ions . obviously , the transmission maximum is only marginally affected by the presence of different types of central ions . furthermore , as expected the transmission function is shifted to lower energies if central ions are present . however , for potassium the transmission is to be found slightly reduced , also the broadening is not as large as for the other species . this is also reflected in the pdf of transmission maxima which can be found in fig . s4 in ref . @xcite . ) @xmath1 in absence and presence of monovalent central ions li@xmath5 , na@xmath5 and k@xmath5 . ] as a consequence , the average current for the simulation with central potassium ions is half as large as for those with lithium and sodium ions which might be attributed to the different mobilities of li@xmath5 and na@xmath5 compared to k@xmath5 . interestingly , there is no significant difference for ct in absence and presence of ions , although it is known from sec . [ subsec : res - md ] that the g4 molecules without central ions exhibit significant destabilization . once more this supports the notion that the enhanced conductance in g4 may not exclusively be explained in terms of higher structural rigidity , i.e. less dynamical disorder , rather it is the multitude of ct pathways via interstrand couplings that brings on an increased number of high - transmissive conformations . apparently , those interstrand and in - plane couplings t2 and t3 are not altered by the presence of central ions , i.e. by a more rigid quadruplex structure . in this work , we have investigated the conductivity of g4 with respect to dsdna using classical md simulations , combined qm / mm methods to compute ct parameters and landauer theory to compute transmission and i - u characteristics . these properties have been evaluated for phase - space ensembles of g4 and dsdna structures . the approach adopted here to calculate the transport characteristics at a given time ( landauer theory ) has clear limitations related to the underlying assumption of tunneling transport . though this may be an efficient pathway over very short segments of the molecules under study , its validity becomes questionable with increasing length . this becomes quite evident when investigating sequences as used in recent experiments e.g. by scheer and co - workers @xcite , where our methodology would predict currents order of magnitude smaller than those found in the experiments . we are currently developing a methodology in order to describe transport in complex materials in a more general way @xcite . nevertheless , the landauer calculations reported here give interesting insight into particular properties of g4 , especially when compared to dsdna . first of all , g4 with central ions is structurally much more stable than dsdna and has more @xmath0-contacts along the chain @xcite . at first sight , one may argue that this leads to improved electrical conduction , however , it turns out that the large fluctuations found for dsdna lead to highly conducting structures , which dominate the transport . therefore , a dsdna molecule conducts better than one strand of g4 , or equivalently , the higher conduction of g4 is not due an increased structural stability of its single strands . the phase space of the single strands still contains a vast majority of conformations , which are not ct active . it is the ability of g4 to allow for a large number of conformations due to the interstrand couplings t2 and t3 which makes this species better conducting . if the pathway along one strand is blocked , e.g. when one t@xmath40 along the chain vanishes , many other conduction channels may be viable due to interstrand hopping . at the end , g4 has a much larger number of ct active conformations than four dsdna ( poly(g ) ) offers . this is the basis of the higher conductivity of g4 . this advantage is even maintained , when contacted only at one g - site , instead of four sites . as a second point , dsdna or g4 may be exerted to strain due to the contacting procedure . here , clearly the higher stability of g4 may help to maintain a conducting conformation , while the conduction in dsdna may be much more easily disrupted @xcite . in a recent work , we have predicted dsdna with non - uniform sequences ( i.e. not poly(g ) or poly(a ) ) to conduct better in solution than in gas phase . this is due to the larger structural fluctuations in solvent , which introduce conducting structures @xcite . this finding has been confirmed by a recent experiment @xcite . for homogeneous sequences like poly(g ) or g4 however , a higher conductance in gas phase should be found @xcite . finally , our results suggest that the presence of central metal ions within the quadruplex has only a marginal impact on hole transfer in g4 , although they are vital for the stability and rigidity of g4-dna . this work has been supported by the deutsche forschungsgemeinschaft ( dfg ) within the priority program 1243 `` quantum transport at the molecular scale '' under contract el 206/5 - 2 and cu 44/5 - 2 , by the volkswagen foundation grant nr . i/78 - 340 and by the european union under contract ist-029192 - 2 . note , generally , the two last tetrads ( base pairs ) at the 5 and 3 end are not considered for ct calculations to avoid end effects , although the simulations were done with 12 tetrads and base pairs for g4 and dsdna , respectively .

when a quantum field theory is constructed on a curved background spacetime , the gravitationally induced vacuum polarization typically induces a non - zero vacuum expectation value for the stress - energy tensor @xcite . it should be pointed out that fully self consistent calculations are horrendously difficult ; consequently most calculations in the literature are carried out in the test - field limit . ( wherein the vacuum polarization is _ not _ permitted to back - react on the geometry via the einstein field equations . ) even in the test - field limit , most known results are obtained by numerical ( rather than analytic ) computations . _ perturbative _ self - consistent solutions around flat spacetime have recently been investigated by flanagan and wald @xcite , building on earlier work of horowitz @xcite , and horowitz and wald @xcite . another type of energy condition , based on the `` quantum inequalities '' , is investigated for a schwarzschild background in recent papers by ford and roman @xcite . ford and roman have also discussed the anec and awec in ( 1 + 1)dimensional and ( 3 + 1)dimensional evaporating black holes ( unruh vacuum ) @xcite , and for the ( 1 + 1)dimensional boulware vacuum . in contrast , the present paper works strictly in ( 3 + 1 ) dimensions and treats the equilibrium hartle hawking vacuum state . now we know , on rather general grounds ( the existence of the hawking radiation effect and concomitant violation of the _ classical _ area increase theorem of black hole dynamics ) that at least some of the energy conditions must be violated at or near the event horizon of a black hole . in this paper i shall explore these issues in a little more detail . i shall continue to work in the test - field limit : i discuss the effect of the gravitational vacuum polarization on various energy conditions . in particular , i discuss the point - wise null , weak , strong , and dominant energy conditions ( nec , wec , sec , and dec ) , the averaged null energy condition ( anec ) , and shall furthermore introduce and discuss several new , potentially interesting , energy conditions ( pnec , scri nec , and scri anec ) . for the purposes of this paper , i restrict attention to the most well - studied curved - space quantum field theory : the conformally - coupled massless scalar field on schwarzschild spacetime , in the hartle hawking vacuum . for this geometry and vacuum state one has both ( 1 ) a useful analytic approximation to the gravitational polarization page s approximation @xcite , and ( 2 ) numerical estimates of the vacuum polarization see the numerical calculations of howard @xcite , howard and candelas @xcite , and anderson , hiscock , and samuel @xcite . ) having established the basic machinery in this paper , a study of the boulware vacuum will be presented in a companion paper , while further studies of the unruh vacua should be straightforward , although tedious . ( for the current state of affairs in the unruh vacuum , see @xcite . ) similarly , extensions to non - conformal couplings , massive scalars , and reissner nordstrm geometries should be straightforward . to set the stage , recall some basic definitions + * definition [ nec ] * * definition [ anec ] * here @xmath0 is a generalized affine parameter ( see @xcite ) for the null curve , the tangent vector being denoted by @xmath1 . note that in almost all applications it suffices to consider null geodesics , rather than generic null curves . however , isolated discussions of the anec on non geodesic null curves do occur in the literature @xcite the extended definition presented here allows one to compare and contrast the present results with other calculations . as a practical matter i shall often replace the word `` inextendible '' by the phrase `` inextendible past the event horizon '' . this is a purely pragmatic decision based on a lack of numerical data inside the event horizon , coupled with the general feeling in the community that page s analytic approximation will become progressively worse as one approaches the central singularity . ( thus my studies of the anec can more precisely be referred to as studies of a `` truncated anec '' . ) in an earlier publication @xcite , ( see also @xcite ) , i derived a no - go theorem for the anec : i showed that the anec is violated ( in the test - field limit ) whenever the background spacetime has a non - zero _ scale anomaly_. specifically , i proved the + * anec no - go theorem * this no - go theorem , since it depends on a special case of the quantum field theoretic ( conformal ) anomaly , is known to be independent of the choice of vacuum state . ( one can easily generalize this argument to include non - conformally coupled fields , see @xcite . ) unfortunately , a brief calculation suffices to show that the tensor @xmath2 vanishes on schwarzschild spacetime . indeed it vanishes on any ricci - flat spacetime ( or indeed on any spacetime conformal to an einstein spacetime ) . thus schwarzschild spacetime avoids the no go theorem enunciated above , and it becomes a matter of explicit checking to see whether or not the anec is satisfied or violated on this geometry . note that in doing this explicit checking we will have to be cognizant of our particular choice of vacuum state , and indeed our results will also depend on the particular class of null geodesics under consideration . in performing the explicit checks alluded to above , i have found it useful to define several new energy conditions that are in some sense intermediate between the nec and anec : * definition [ pnec ] * now , if one does not place any further constraints on the set @xmath3 the resulting definition , while general , is not particularly useful . ( it is best used as a diagnostic tool to probe interesting regions of the tangent bundle . ) a suitable restriction is : * definition [ scri nec ] * ( this definition basically says that one should not worry too much if violations of the nec occur only when one looks along null geodesics that never make it out to null infinity . ) we shall soon see that in the hartle hawking vacuum the gravitational vacuum polarization of a conformally - coupled massless scalar field residing on the schwarzschild background geometry everywhere satisfies this scri nec energy condition , though it does not everywhere satisfy the nec . we shall also use the pnec to investigate the regions in phase space ( the tangent bundle ) where suitable positivity conditions hold . finally , for completeness , i enunciate + * definition [ scri anec ] * ( this definition basically says that one should not worry too much if violations of the anec occur only along null geodesics that never make it out to null infinity . after all , null infinity is where you want to work to establish results such as the penrose woolgar positive mass theorem @xcite or the friedman schleich witt topological censorship theorem @xcite . ) from the previous comment regarding scri nec it automatically follows that , in the hartle hawking vacuum , the gravitational vacuum polarization of a conformally coupled scalar field residing on the schwarzschild background geometry satisfies this scri anec . we shall also see that not all null geodesics satisfy the anec itself . in the particular case of conformally coupled scalar fields residing on schwarzschild spacetime one has a lot of information about the vacuum polarization . by spherical symmetry one knows that @xmath4.\ ] ] where @xmath5 , @xmath6 and @xmath7 are functions of @xmath8 , @xmath9 and @xmath10 . note that i set @xmath11 , and choose to work in a local - lorentz basis attached to the fiducial static observers ( fidos ) . page s analytic approximation gives @xcite a polynomial approximation to the stress energy tensor : @xmath12 , \\ % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % 2 3 4 5 6 % 6 r 9 r 12 r 15 r 18 r 99 r % 3 + --- + ---- + ----- + ----- + ----- - ----- % r 2 3 4 5 6 % r r r r r % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % \tau(r ) & = & - p_\infty \left [ 1 + 2\left({2m\over r}\right ) + 3\left({2m\over r}\right)^2 + 4\left({2m\over r}\right)^3 + 5\left({2m\over r}\right)^4 + 6\left({2m\over r}\right)^5 + 15\left({2m\over r}\right)^6 \right ] , \\ % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % 2 3 4 5 6 % 2 r 3 r 4 r 5 r 6 r 15 r % 1 + --- + ---- + ---- + ---- + ---- + ----- % r 2 3 4 5 6 % r r r r r % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % p(r ) & = & + p_\infty \left [ 1 + 2\left({2m\over r}\right ) + 3\left({2m\over r}\right)^2 + 4\left({2m\over r}\right)^3 + 5\left({2m\over r}\right)^4 + 6\left({2m\over r}\right)^5 - 9\left({2m\over r}\right)^6 \right ] , \\ % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % 2 3 4 5 6 % 2 r 3 r 4 r 5 r 6 r 9 r % 1 + --- + ---- + ---- + ---- + ---- - ---- % r 2 3 4 5 6 % r r r r r % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % & = & + p_\infty \left [ 1 + 2\left({2m\over r}\right ) + 3\left({2m\over r}\right)^2 \right ] \left [ 1 + 4\left({2m\over r}\right)^3 - 3\left({2m\over r}\right)^4 \right ] . \end{aligned}\ ] ] here i have defined a constant ( to be interpreted as the pressure at spatial infinity ) by @xmath13 note that i have explicitly expanded the functions given in references @xcite as polynomials in @xmath14 to exhibit the fact that all components of the stress energy are well behaved at the horizon . it is perhaps somewhat surprising that the rather messy rational _ quotients _ of polynomials exhibited in @xcite reduce to such relatively nice compact expressions . these formulae have also been checked for consistency with the brown ottewill extensions of the original page approximation ( * ? ? ? * see esp . see also elster @xcite . if one prefers , an alternative form for the page approximation is @xmath15 + q(r ) \left [ \matrix{+3&0&0&0\cr 0&+1&0&0\cr 0&0&0&0\cr 0&0&0&0\cr } \right ] \right\}.\ ] ] the polynomials @xmath16 and @xmath17 being @xmath18 and @xmath19 the trace of the stress energy tensor is given by @xmath20 this result , because it is simply a restatement of the conformal anomaly , is known to be exact . at the event horizon , page s analytic approximation gives @xcite @xmath21.\ ] ] the explicit numerical calculations of howard @xcite , howard and candelas @xcite , and anderson , hiscock , and samuel @xcite show that page s analytic approximation is reasonably good the worst deviations occur in the immediate vicinity of the event horizon where the anderson hiscock samuel numerical analysis gives @xmath22.\ ] ] a convenient table of numerical results , covering the range @xmath23 to @xmath24 may be found on page 2541 of reference @xcite . ( near the event horizon , the howard candelas data is believed to be accurate to two significant digits . the accuracy is expected to improve with increasing @xmath8 , as the numerical results converge upon the page approximation . ) additionally , i have obtained via private communication the more intensive numerical data of anderson hiscock samuel @xcite , which covers the range @xmath23 to @xmath25 at much higher resolution . ( near the event horizon , the anderson hiscock samuel data is believed to be accurate to three significant digits . ) i have collated the data and used it to construct suitable interpolating functions to be used for the numerical aspects of the following discussion . to construct the interpolating functions i use the anderson hiscock samuel data in the range @xmath23 to @xmath25 , and the howard candelas data in the range @xmath25 to @xmath24 , augmented by the fact that we know the exact result at @xmath26 . using mathematica , i have fit the data using third - order interpolating polynomials in the variable @xmath27 . ( note that @xmath28 conveniently runs from @xmath29 at the horizon to @xmath30 at spatial infinity ; @xmath28 is also _ the _ natural variable to use when numerically analyzing the page approximation . ) of course , there are may other papers extant which calculate the stress - energy tensor in somewhat different configurations . see , for example , . more recently , in a very interesting development , matyjasek @xcite has developed a curve - fitting analysis that fits the numerical data to high accuracy . in this paper i prefer to work directly with the numerical data . in this paper i am only presenting the minimum requirements for the particular job at hand and it is clear that with additional effort more information can be extracted by mining this additional vein of data . outside the event horizon , the nec reduces to the pair of constraints @xmath31 page s approximation yields @xmath32.\ ] ] ( note the factorization ! ) consequently @xmath33 is always explicitly positive outside the horizon . furthermore @xmath34.\ ] ] numerically finding the roots of this sixth order polynomial shows that @xmath35 is negative from @xmath23 to @xmath36 . thus the page approximation suggests that the null energy condition is violated in the range @xmath37 $ ] . turning to the numerical data , one easily verifies that that outside the horizon @xmath38 , while @xmath39 at the horizon . this is in complete agreement with page s analytic approximation . on the other hand , inspection of the numerical data indicates that @xmath40 for @xmath41 . ( this number was obtained by fitting third - order interpolating polynomials to the numeric data ; and then numerically finding the root . ) this is qualitatively , though not quantitatively in agreement with page s analytic approximation . thus the numerical data indicate that the null energy condition is violated in the range @xmath42 $ ] . inside the event horizon , the radial coordinate becomes timelike , and the roles played by @xmath43 and @xmath44 are interchanged . the nec reduces to the pair of constraints @xmath45 unfortunately , inside the horizon i am reduced to reliance upon the page approximation by a total lack of numerical data . now there are some subtleties associated with extrapolating the page approximation inside the event horizon . the existence of a killing vector that points in the @xmath46 direction is absolutely critical , ( even if the @xmath46 direction is no longer timelike ) . thus if you wish to extrapolate the page approximation inside the event horizon it is necessary to limit attention to an _ eternal _ black hole ( that is , the maximally extended kruskal szekeres manifold ) . i should remind the reader that in the hartle hawking vacuum the stress - energy is at least known to be regular at the horizon , so the page approximation should not be too far wrong just inside the event horizon of an eternal black hole . on the other hand , one does expect the page approximation to get progressively worse as one moves further in toward the singularity , so one should probably not take these results too seriously once one is far inside the horizon . i should especially warn the reader against extrapolating the page approximation to the interior of an astrophysical black hole : we really do not expect the interior of an astrophysical black hole to have the same killing vectors as the exterior , but instead expect the interior to be fully dynamical . ( in addition we do not expect an astrophysical black hole to even to be in the hartle hawking quantum state . ) the present calculations inside the horizon are strictly limited to eternal black holes and are exhibited merely because they are do - able , and because they may be suggestive of the actual situation . inspection of page s analytic approximation indicates that @xmath47 for all @xmath48 , so this part of the nec is satisfied inside the event horizon . on the other hand @xmath49 which is everywhere negative , ( both inside and outside the horizon ) . thus the page approximation suggests that the null energy condition is violated throughout the interior of the black hole . this is enough to inform us that , ( assuming the reliability of the page approximation in this matter ) , the weak , strong , and dominant energy conditions are also violated throughout the entire interior of the black hole . outside the horizon , the weak energy condition is equivalent to the three constraints @xmath50 the last two constraints have already been discussed they are simply the null energy condition . in the page approximation , finding the roots of the relevant sixth - order polynomial shows that the density constraint is violated for @xmath51 . therefore the weak energy condition is violated in the range @xmath52 $ ] . in contrast , the numerical data indicate that violations of the weak energy condition extend to the region @xmath53 $ ] . inside the horizon , the weak energy condition is equivalent to the three constraints @xmath54 the last two constraints have already been discussed ( when dealing with the nec ) . the violation of the nec is already sufficient to tell us that the wec is violated everywhere inside the horizon , but we can add in passing that the page approximation additionally implies that @xmath44 is everywhere negative inside ( and outside ) the horizon . the page approximation suggests that the weak energy condition is violated throughout the interior of the black hole . outside the horizon , the strong energy condition is equivalent to the three constraints @xmath55 we have already looked at the first two constraints when discussing the null energy condition . the third condition is always satisfied outside the horizon ( since both @xmath56 , and @xmath57 in this region ) . this is true both in the page approximation , and by appeal to the numerical data . thus the strong energy condition is violated in the same region as the null energy condition . inside the horizon , the strong energy condition is equivalent to the three constraints @xmath58 we have already looked at the first two constraints when discussing the null energy condition , and thereby know that ( the page approximation suggests that ) the strong energy condition is violated throughout the entire interior of the black hole . for completeness i point out that the first condition is satisfied throughout the interior , the second of these conditions is violated throughout the interior , while the third condition also fails throughout the interior . outside the horizon , the dominant energy condition is equivalent to the three constraints @xmath59 ? \nonumber\\ & & p(r)\in [ -\rho(r),+\rho(r)]?\end{aligned}\ ] ] we can rephrase this as @xmath60 using the page approximation , and restricting attention to the range @xmath61 $ ] , one has : @xmath62 ; \nonumber\\ & \rho(r)-\tau(r ) < 0 & \qquad r\in\emptyset ; \nonumber\\ & \rho(r)+\tau(r ) < 0 & \qquad r\in[2m,2.77256 m ] ; \nonumber\\ & \rho(r)-p(r ) < 0 & \qquad r\in[2m,2.58512 m ] ; \nonumber\\ & \rho(r)+p(r ) < 0 & \qquad r\in[2m,2.18994m].\end{aligned}\ ] ] pulling this all together , the page approximation suggests that dominant energy condition fails in the region @xmath63 $ ] . the numerical data implies quantitative though not qualitative modifications . in the range @xmath61 $ ] one finds : @xmath64 ; \nonumber\\ & \rho(r)-\tau(r ) < 0 & \qquad r\in\emptyset ; \nonumber\\ & \rho(r)+\tau(r ) < 0 & \qquad r\in[2m,2.992 m ] ; \nonumber\\ & \rho(r)-p(r ) < 0 & \qquad r\in[2m,2.628 m ] ; \nonumber\\ & \rho(r)+p(r ) < 0 & \qquad r\in[2m,2.298m].\end{aligned}\ ] ] pulling this all together , the numerical data indicate that the dominant energy condition fails in the region @xmath65 $ ] . note that this is suspiciously close to @xmath66the unstable circular photon orbit and _ might _ be trying to tell us something . if one actually calculates @xmath67 at @xmath66 one gets @xmath68given the expected three - significant - digit numerical reliability of the data this is ( questionably ) compatible with zero . if one takes this suggestion seriously , it would imply a hidden ( accidental ? ) symmetry in the stress tensor at @xmath66 : @xmath69?\ ] ] it is quite possible however , that this is purely a numerical accident . finally , i point out that this conjectured property certainly does not survive the introduction of non - conformal coupling , nor is there any particular reason to expect it to survive the introduction of non - zero rest mass . this conjecture also most definitely does not hold in the boulware or unruh vacuum states @xcite . inside the horizon , the dominant energy condition is equivalent to the three constraints @xmath70 ? \nonumber\\ & & p(r)\in [ -\tau(r),+\tau(r)]?\end{aligned}\ ] ] we can rephrase this as @xmath71 we already know that @xmath72 inside the horizon ( in fact everywhere ) . so , provided the page approximation is not misleading in this regard , the dominant energy condition is violated throughout the interior of the black hole . no additional information comes from the other conditions . to analyse the partial null energy condition introduced earlier in this paper , consider a generic null vector inclined at an angle @xmath73 away from the radial direction . then without loss of generality , in an orthonormal frame attached to the @xmath74 coordinate system , @xmath75 ignoring the ( presently irrelevant ) overall normalization of the null vector , one has @xmath76 + [ \tau + p ] \sin^2\psi).\end{aligned}\ ] ] we have already seen that @xmath77 is positive outside the event horizon . on the other hand the page approximation gives @xmath78 which is everywhere negative . a glance at howard s numerical data confirms that the numerical data also satisfies @xmath79 . thus the partial null energy condition _ fails _ for those radii @xmath8 , and those angles @xmath73 , such that @xmath80 \equiv \sqrt { \rho(r)-\tau(r ) \over |\tau(r ) + p(r)| } .\ ] ] that is , the partial null energy condition _ fails _ for @xmath81 $ ] . defining @xmath27 this critical angle is ( in the page approximation ) given by @xmath82.\ ] ] for @xmath83 ( @xmath84 ) there are no real solutions to this equation . at @xmath85 one has @xmath86 , while @xmath87 moves monotonically to zero as @xmath88 . ( see figure [ fig1 ] where @xmath89 is plotted as a function of @xmath28 . ) thus , sufficiently near the horizon , almost all directions violate the partial null energy condition . as one moves further away from the horizon the violations of the partial energy condition are confined to null vectors that are progressively more and more transverse , finally at @xmath85 the violations of the partial null energy condition disappear completely . using the numerical data will modify the precise location where this behaviour manifests itself , but will not qualitatively modify this picture . the violations of the partial null energy condition vanish outside @xmath90 ( @xmath91 ) , and the numerically determined values of @xmath89 are superimposed on figure [ fig1 ] . inside the event horizon , one should consider a generic null vector inclined at an angle @xmath92 away from the @xmath46 direction ( which is now spacelike ) . then without loss of generality , in an orthonormal frame attached to the @xmath74 coordinate system , @xmath93 one should now consider the quantity @xmath94 + [ \rho + p ] \sin^2\tilde\psi).\end{aligned}\ ] ] note that inside the horizon @xmath95 is strictly positive , while @xmath96 is strictly negative . in fact , the page approximation yields @xmath97.\ ] ] one deduces that violations of the partial null energy condition occur whenever @xmath98 \equiv \sqrt { \tau(r)-\rho(r ) \over |\rho(r)+p(r)| } \ ] ] consequently , violations of the partial null energy condition are confined to the range @xmath99 $ ] . again defining @xmath27 this critical angle is ( in the page approximation ) given by @xmath100.\ ] ] one has , at @xmath101 , @xmath102 , while @xmath103 falls to zero as @xmath8 approaches @xmath104 . qualitatively : near ( but inside ) the horizon almost all null directions suffer violations of the partial energy condition , while near the central singularity somewhat fewer directions violate the partial null energy condition . again , i remind the reader to not take the page approximation too seriously as one approaches the singularity . i now turn attention to the asymptotic null energy condition ( scri nec ) that i introduced earlier in this paper . discussing this energy condition will require some standard results concerning the null geodesics of the schwarzschild spacetime . ( see , for instance , such standard textbooks as misner thorne wheeler @xcite , wald @xcite , or weinberg @xcite . ) there are two conserved quantities for null geodesic motion in schwarzschild spacetime ; the energy and angular momentum . using these conservation laws the affine parameter can be chosen in such a way that @xmath105 the parameter @xmath106 is the angular momentum per unit energy . if the null geodesic reaches asymptotic spatial infinity then this parameter can also be interpreted as the `` impact parameter '' . however , there is a large class of null geodesics that never reaches spatial infinity , for these null geodesics the notion of `` impact parameter '' is at best an abuse of language . the angle between the null geodesic and the radial direction is given by @xmath107 we are ultimately interested in the quantity @xmath108 + [ \tau + p ] \sin^2\psi ) \nonumber\\ & = & \{[\rho -\tau ] + [ \tau + p ] ( 1 - 2m / r)a^2/r^2\},\end{aligned}\ ] ] but will need to go through a few preliminaries . the radial motion of null geodesics is governed by @xmath109.\ ] ] so the turning points , @xmath110 , are given by the cubic @xmath111 if the impact parameter is small , @xmath112 , then it is a standard result that there are no turning points : the null geodesic either plunges into the future singularity , or emerges from the past singularity of the maximally extended schwarzschild spacetime . ( for @xmath113 the geodesic may make a large number of `` orbits '' before crossing the event horizon . ) if the impact parameter is marginal , @xmath114 , then it is a standard result that @xmath66 , corresponding to the ( unstable ) circular photon orbit . if the impact parameter is large , @xmath115 , then it is a standard result that there are _ two _ turning points at physical values of @xmath8 . one of these turning points lies in the range @xmath116 , while the other lies in the range @xmath117 . note that if @xmath118 then the three mathematical roots of the cubic are approximately @xmath119 , and @xmath120 $ ] . the two physical roots are approximately @xmath121 and @xmath122 $ ] . the first of these turning points , @xmath116 , corresponds to the obvious class of null geodesics with high impact parameter those geodesics that come in from spatial infinity , bounce off the angular momentum barrier , and return to spatial infinity . ( for @xmath123 the geodesic may make a large number of `` orbits '' before returning to spatial infinity . ) the second of these turning points , @xmath117 , corresponds to a completely separate class of null geodesics with high `` impact parameter''these geodesics emerge from the event horizon at @xmath124 , with high angular momentum , make a large number of `` orbits '' before reaching their maximum height above the event horizon , and then make an equally large number of `` orbits '' before returning to re - cross the event horizon at @xmath125 . for these geodesics the use of the phrase `` impact parameter '' to describe the parameter @xmath106 is most definitely an abuse of language . to analyze the asymptotic null energy condition , start by first considering the quantity @xmath126 which i shall refer to as the nec density . ( remember that @xmath127 is an explicitly known function of @xmath8 and @xmath106 . ) those null geodesics that come in from infinity and return to infinity never get closer to the origin than @xmath66 . see , for instance , @xcite . inspection of either the page approximation or of the numerical data indicates that @xmath5 , @xmath128 , and @xmath7 are all positive for @xmath129 . thus the asymptotic null energy condition is satisfied along all null geodesics that come from , and return to , infinity . for other classes of null geodesics it proves convenient to rewrite the quantity @xmath130 as : @xmath131 + [ \tau + p ] ( 1 - 2m / r)a^2/r^2\}.\ ] ] now , for incoming null geodesics with smaller than critical impact parameter , the null geodesic may circle the black hole a large number of times , but is guaranteed to ultimately plunge into the event horizon @xcite . this makes the analysis a little more subtle . inspection of either the page approximation or the numerical data shows that ( outside the horizon ) @xmath77 is always positive , while @xmath132 is always negative . now use the fact that for an infalling null geodesic @xmath112 . since @xmath132 is negative , this implies ( for this class of null geodesics ) a lower bound @xmath133 + [ \tau + p ] ( 1 - 2m / r)27m^2/r^2\}.\ ] ] inspection of page s approximation indicates that this lower bound is strictly positive ( zero at the horizon ) . in fact @xmath134.\ ] ] ( note the factorization ! ) thus the page approximation suggests that the asymptotic null energy condition holds on all infalling null geodesics . ( at least until one crosses the horizon ! ) this lower bound @xmath135 is plotted in figure [ fig3 ] . by time reversal , this suggests that the asymptotic null energy condition also holds on all outgoing null geodesics that reach infinity . with respect to the numerical data , one can easily evaluate @xmath135 numerically . the results are superimposed on figure [ fig3 ] . while there are marked differences between the analytic approximation and the numerical data , both curves are seen to be strictly positive outside the horizon . we conclude that scri nec is satisfied . for _ all _ null geodesics that reach spatial infinity the nec density @xmath130 is strictly positive everywhere outside the event horizon . for completeness , and future reference , i point out that the page approximation yields @xmath136 . \label{eq_pi_r_a}\ ] ] again , note the factorization ! now turn attention to the trapped null geodesics . ( these are null geodesics with high `` impact parameter '' @xmath137 , trapped in the region @xmath138 $ ] . ) while these trapped null geodesics are , by definition , not directly relevant to the scri nec , the tools developed above permit us to gain additional insight into the pnec on trapped null geodesics . pick some value of @xmath8 in the range @xmath139 . then @xmath130 is guaranteed to be negative if one chooses @xmath140 ( setting @xmath27 is again a convenient choice of variables for both analytic and numeric work . in this section we will only be interested in the region @xmath141 $ ] . ) note that what we are doing is guaranteeing that with these definitions the pnec is violated for the region of the @xmath142 plane above the curve @xmath143 . using page s approximation , this critical impact parameter is given by @xmath144 the critical impact parameter implied by the numerical data was also determined and both curves are plotted on figure [ fig4 ] . it should be noted that , for a given value of @xmath117 , one can not choose @xmath106 arbitrarily : in order for a trapped null geodesic with impact parameter @xmath106 to ever reach radius @xmath8 it is necessary that @xmath106 be small enough . indeed one must have @xmath145 this may be thought of as a kinematic bound on trapped null geodesics : the region of the @xmath142 plane above the curve @xmath146 is kinematically inaccessible to trapped null geodesics . ( note that the region below @xmath114 is also kinematically inaccessible . ) the relevant curve is plotted in figures [ fig5 ] and [ fig6 ] , where it is overlain with @xmath147 as obtained from figure [ fig4 ] . by looking at where these curves cross one another we can draw some general conclusions . in the page approximation , now restricting attention to the region outside the event horizon : 1 . some trapped null geodesics ( those with @xmath148 = [ 5.19615 m , 6.1101 m]$ ] ) _ never _ experience nec density violations . that is : @xmath149 is _ satisfied _ along the entire portion of the null geodesic that lies outside the event horizon . ( and therefore the [ truncated ] anec must be satisfied on this entire class of null geodesics . ) 2 . some trapped null geodesics ( those with @xmath150 ) _ always _ experience nec density violations . that is : @xmath151 along the entire [ truncated ] null geodesic . ( and therefore the [ truncated ] anec must be violated on this entire class of null geodesics . ) all other trapped null geodesics ( those with @xmath152 $ ] ) will experience some nec violations when they get sufficiently close to the event horizon . ( and therefore investigating anec violations on this class of null geodesics requires more work . ) 4 . note that @xmath153 corresponds to a null geodesic that reaches a maximum radius @xmath154 , a number that we have seen before ( when discussing the nec and pnec ) . use of the numerical data implies quantitative though not qualitative changes . we observe that 1 . the set of trapped null geodesics which _ never _ experience nec density violations is much smaller those with @xmath155 = [ 5.19615 m , 5.276 m]$ ] . ( the [ truncated ] anec must be satisfied on this entire class of null geodesics . ) 2 . some trapped null geodesics ( those with @xmath156 ) _ always _ experience nec density violations . ( therefore the [ truncated ] anec must be violated on this entire class of null geodesics . ) all other trapped null geodesics ( those with @xmath157 $ ] ) will experience some nec violations when they get sufficiently close to the event horizon . ( and therefore investigating anec violations on this class of null geodesics requires more work . ) 4 . note that @xmath158 corresponds to null geodesic that reaches a maximum radius @xmath159 , a number that we have seen before ( when discussing the pnec ) . the message to be extracted is this : the numeric data and analytic approximations are in good qualitative agreement with each other . null geodesics that reach null infinity are well behaved ( scri nec is satisfied ) , but _ some _ of the trapped null geodesics encounter nec violations ( and anec violations ) . we are finally in a position to pin down precisely violations of the anec itself . although significant information can already be extracted by using the point - wise energy conditions already discussed , beyond a certain stage explicit use of the affine parameterization must be invoked . write the anec integral as @xcite @xmath160 + [ \tau + p ] \sin^2\psi ) \ ; \xi^2 \ ; d\lambda , \\ & = & \int_\gamma \pi(r , a ) \ ; \xi^2 \ ; d\lambda.\end{aligned}\ ] ] note that the _ integrand _ is proportional to the quantity @xmath130 which has already been extensively discussed in the context of the scri nec . the arguments previously adduced for the scri nec can be carried over wholesale to the scri anec . in particular : * for null geodesics that come in from spatial infinity and return to spatial infinity @xmath161 the _ integrand _ is everywhere positive and so the anec holds on this entire class of geodesics . ( note that satisfaction of the anec does not arise from the trivial observation that the heat bath contributes an asymptotically constant and positive energy density far from the black hole rather one has the stronger statement that the integrand itself is positive along the entire geodesic . ) * the fact that the integrand is positive implies that anec will also be satisfied for curves with transverse smearing @xcite , or curves with arbitrary positive weighting functions @xcite . * for infalling or outfalling null geodesics that reach asymptotic spatial infinity @xmath162 the integrand is everywhere positive and anec is satisfied . [ caveat : i stop the anec integral once it reaches the horizon . ] * this may be interpreted as follows : because scri nec is satisfied , it automatically follows that scri anec is satisfied . * for the unstable circular photon orbit @xmath163 the integrand is everywhere positive and anec is satisfied . * however , for trapped null geodesics @xmath164)$ ] the integrand is no longer necessarily positive , and this is the only case for which we will explicitly need to look at the `` weighting function '' @xmath165 appearing in the anec integral . observe that the `` weighting function '' appearing in the anec satisfies ( * ? ? ? * , equations ( 12.60)(12.63 ) ) @xmath166 although this calculation was carried out for the schwarzschild geometry the result remains true for an arbitrary static spacetime the anec integral is simply the time - average along a null geodesic of the local - lorentz nec integrand where the time average is to be taken with respect to the natural static time coordinate . for actual calculations it is much more practical to re - express this as an integral with respect to the radial variable @xmath8 by using @xmath167 so that @xmath168 now this observation appears to weight the region near the event horizon very heavily because of the explicit pole at @xmath23 . however , the integrand @xmath130 has a zero at the event horizon in the page approximation one discovers an explicit factor of @xmath169 [ _ cf . _ equation [ eq_pi_r_a ] ] , while the numerical data also exhibits a first - order zero in @xmath130 at the horizon . thus for trapped null geodesics one may write @xmath170 and have some confidence that the integral actually converges at the lower bound @xmath23 . as a penultimate step , recall that calculating @xmath171 involves solving a cubic . it is more convenient to parameterize the trapped geodesic by calculating the impact parameter in terms of the maximum height attained by the null geodesic : @xmath172 so for these trapped null geodesics one finds the anec integral is @xmath173/[(1 - 2m / r_{\rm max})r^2]}},\end{aligned}\ ] ] to actually evaluate this integral numerically it is useful to change variables to @xmath27 , and @xmath174 , with the result that @xmath175/[z_0 ^ 2(1-z_0)]}},\end{aligned}\ ] ] with a little work , the square root can be seen to factorize explicitly @xmath176 this integral , though singular at the lower limit ( @xmath177 , corresponding to @xmath178 ) , is now certainly finite . while mathematica has resources to deal with @xmath179 singularities at the endpoints of the integration range , it reacts badly to such singularities when the location is chosen dynamically . that is : @xmath180 is handled badly . for this reason the change of variables @xmath181 , while being a formal mathematical identity , leads to much better numerical behaviour for the integral : @xmath182 this integral was determined numerically both for the page approximation and for the anderson hiscock samuel numeric data . the results are graphed in figure [ fig7 ] . remember that it is meaningless to force this particular integral out of the range @xmath138 $ ] , corresponding to @xmath183 $ ] thus we have ( finally ! ) managed to characterize the precise class of null geodesics on which the [ truncated ] anec is violated . the critical values of @xmath184 , @xmath185 , and @xmath106 are , for the page approximation : @xmath186 for the numerical data , one obtains : @xmath187 note that these numbers are qualitatively reasonable and in agreement with the violations of the point - wise energy conditions , and the numerical investigation of the pnec . furthermore the anec violations in the numerical data are seen to extend out to larger distances than the anec violations in the page approximation , in agreement with the general trend . though i have quoted four - significant - digit accuracy for the numerical data one should probably not take anything past the second significant digit too seriously . in fact , given the vagaries of numerical analysis on singular integrals ( and the inherent uncertainties in the anderson hiscock samuel data ) it is conceivable that the exact critical impact parameters might be @xmath188 and @xmath189 respectively . if one considers arbitrary non - geodesic curves the anec loses much ( though not quite all ) of its power : by a judicious choice of non - geodesic null curve one could try to remain in a region where the nec is violated and thereby `` trivially '' violate the anec @xcite . for instance , along any circular null curve at fixed @xmath8 ( _ not _ a geodesic except in the case of @xmath66 ) the anec integral is still proportional to @xmath96 . inspection of the numeric data indicates that @xmath40 for @xmath41 . ( page s analytic approximation giving a slightly different result , @xmath190 for @xmath191 . ) thus the anec is _ not _ satisfied for this particular class of non geodesic null curves . investigation of the properties of the averaged null energy condition is of considerable interest to diverse applications in semiclassical quantum gravity . it is now abundantly clear that , in the test - field limit , semiclassical quantum fields do _ not _ generally satisfy the anec : indeed anec violations are related to the existence of a non - zero scale anomaly @xcite . even if the scale anomaly vanishes , this does not necessarily imply that the anec is satisfied : one has to do a case by case analysis . as an example , this paper investigates the situation in schwarzschild spacetime . the analysis presented herein is somewhat of an attempt to crack a walnut with a sledgehammer in the sense that it is a collection of rather general techniques applied to a rather particular problem it is certainly true that this type of analysis can now be carried forward to other geometries , other quantum fields , and other vacuum states by by straightforward but tedious computation . * point - wise energy conditions * * inside the event horizon , with suitable caveats regarding the applicability of the page approximation , almost any energy condition you can think of will be violated . * * between the event horizon @xmath192 and the unstable photon orbit @xmath193 many of the energy conditions are violated , in a series of onion - like layers . * * outside the unstable photon orbit @xmath193 all energy conditions are satisfied . * * all null geodesics that reach asymptotic infinity are well - behaved : if you look _ along _ the null geodesic you never see nec violations . * averaged null energy condition * * if your null geodesic ever reaches infinity , the [ truncated ] anec is definitely satisfied , and is satisfied for a non - trivial reason : the _ integrand _ is strictly positive all the way from the event horizon to null infinity . * * between the event horizon @xmath192 and the unstable photon orbit @xmath193 some of the trapped null geodesics violate the [ truncated ] anec . * * if you are willing to look at non - geodesic null curves even more violations of the anec can be found . 1 . it would be very nice to have an analytic understanding of the precise role played by the unstable photon orbit the numerical evidence is suggestive but not definitive . 2 . for that matter it would be nice to know if the relevance of the unstable photon orbit generalizes to other geometries . 3 . it would be nice to go beyond the numerics ; to develop some exact analytic arguments that go beyond the page approximation . generalizations to the boulware vacuum will be presented in a companion paper . generalizations to the unruh vacuum , other quantum fields , non - conformal couplings , particle masses , and the reissner nordstrm geometry will be straightforward if tedious . the new energy conditions i introduce , scri nec and scri anec , are interesting in that they focus attention on null infinity . and null infinity is where all the interesting details arise in the friedman witt @xcite topological censorship theorem , and the penrose sorkin woolgar version of the positive mass theorem @xcite . i suspect that with a little more work suitable generalizations of these theorems can be constructed in terms of the scri anec . finally i should point out that even though the various energy conditions are violated in many regions , this does not give one a completely free hand to design spacetime geometries to taste : it seems quite likely that the `` quantum inequality '' approach of ford and roman @xcite will allow us to place constraints on spacetime geometries even if all the usual types of energy condition fail .

one of the most active areas of condensed matter theory is the development of new algorithms to simulate and predict the behavior of materials exhibiting strong correlations . recent developments of the powerful many - body approach , the dynamical mean field theory ( dmft ) @xcite and its cluster extensions @xcite , hold a great promise in being able to accurately predict physical properties of this challenging class of materials . in recent years , the dmft method has substantially advanced our understanding of the physics of the mott transition and demonstrated its power to explain such problems as the structural phase diagrams of actinides @xcite , phonon response @xcite , optical conductivity @xcite , valence and x - ray absorption @xcite and transport @xcite of some of the archetype materials with strong correlations . the success of the dynamical mean field theory in the context of the electronic structure revitalized the search for fast and flexible impurity solvers which could treat hund s coupling and spin - orbit coupling of the parent atomic constituents in the crystal environment of the lattice . further , the newly developed cluster extensions of dmft @xcite require faster impurity solver which could access low temperature strong correlation limit . many impurity solvers were developed over the last few decades , including hirsh - fye quanum monte carlo method @xcite , exact diagonalization @xcite , non - crossing approximation @xcite , and its extensions like one - crossing approximation @xcite or sunca @xcite , iterative perturbation theory @xcite , wilson s numerical renormalization group @xcite expansion around the atomic limit @xcite , and many others . each method has some advantages and disadvantages , but at the present time , there is no method that works efficiently and produces accurate solutions for the green s function in all regimes of parameters . concentrating only on the most often employed method , numerically exact hirsch - fye quanum monte carlo , the following weaknesses limit its usefulness in many realistic materials and clusters of strongly correlated models : i ) it can not treat realistic multiplet structure which is very important in strongly correlated @xmath0 materials . ii ) the discretization of the imaginary time leads to considerable systematic error @xcite and requires extrapolation to infinitesimally small time slices . iii ) the low temperatures regime in the strong correlation limit of large @xmath1 is computationally very expensive since it requires many time slices and infinite @xmath1 models like t - j model are unaccessible . all of the above mentioned shortcomings of the conventional quantum monte carlo algorithm are eliminated by the novel continuous time quantum monte carlo method @xcite . in addition , the new method is even much faster for most of applications we tested , including cluster dmft for the hubbard model and application of lda+dmft to the actinides . the basic idea and its implementation for the hubbard model was recently published in ref . . further extensions to more general impurity model was implemented in ref . . first demonstration of the power of our implementation was presented in ref . by detail low temperature study of sodium doped cobaltates and in ref . to study plutonium valence . here we want to describe the powerful new implementation of the method for applications to realistic materials with complicated multiplet structure , including all interaction terms : hubbard @xmath1 , hund s coupling and spin - orbit coupling . the extension to clusters of few sites and superconducting state within cluster dmft will be addressed and its power will be demonstrated by studying the low temperature coherence of both the hubbard and the t - j model . the kondo volume collapse transition in elemental cerium will be reexamined showing the valence histogram of the alpha and gamma phase of the material . more technical details of the implementation are given in the appendix [ appa ] . in this section , we will explain the main steps of the recently developed quantum monte carlo method with emphasis on the generalization to clusters and multiplets of real materials . the cluster dynamical mean field approach can be conveniently expressed by the functional of the local green s function @xcite @xmath2 = { \mathrm{tr}}\log(g_0^{-1}-\sigma)-{\mathrm{tr}}[\sigma g ] + \phi[g_{loc}].\end{aligned}\ ] ] here @xmath3 stands for the green s function of the cluster or single site under consideration and is in general a matrix of the size equal to the number of sites times the number of correlated orbitals per site . further , @xmath4 is the non - interacting green s function @xmath5 and contains all quadratic terms of the hamiltonian including periodic ionic potential of the crystal ( @xmath6 ) . the interacting part of the functional @xmath7 $ ] contains _ all _ two particle irreducible skeleton diagrams inside the chosen cluster , i.e. , @xmath8/\delta g$ ] . within the dmft method , the summation of all the diagrams is achieved by solving the corresponding quantum impurity problem @xmath9 e^{-s_{c } - \int_0^\beta d\tau\int_0^\beta d\tau ' \sum_{{\alpha}{\alpha } ' } \psi_{\alpha}^\dagger(\tau)\delta_{{\alpha}{\alpha}'}(\tau,\tau')\psi_{{\alpha}'}(\tau ' ) } \label{zimp}\end{aligned}\ ] ] where the anderson impurity model hybridization @xmath10 term plays the role of the generalized weiss field that needs to be added to the cluster effective action @xmath11 to obtain the local green s function . the self - consistency condition which determines this weiss field ( hybridization @xmath10 ) is @xmath12 in the general impurity problem defined in eq . ( [ zimp ] ) the electrons in the cluster @xmath11 hybridize with a matrix of weiss fields @xmath13 . in the case of single - site dmft for an @xmath0 material , this hybridization is a @xmath14 matrix while in cluster dmft for plaquette , it is @xmath15 matrix . in some cases , hybridization can be block diagonalized as we will show in the appendix [ appa ] below on the example of cellular - dmft for the normal state of the hubbard model . however , in the superconducting state of the same model , the hybridization acquires off - diagonal components due to anomalous components of self - energy . the cluster part of the action @xmath11 can be very complicated and the power of this method is that it can treat arbitrary interaction within the cluster . for real materials study , the most important on - site terms are the hund s couplings which are usually expressed by the slater integrals @xcite ( @xmath16 , @xmath17 , @xmath18 in case of @xmath0 electrons ) . in addition , there is spin - orbit coupling @xmath19 and crystal field splittings as well as various hoppings and non - local interactions within the cluster . the continuous time impurity solvers are based on the diagrammatic expansion of the partition function and stochastic sampling of the diagrams . two expansions were recently implemented : the expansion around the band limit @xcite and the expansion in the hybridization strength @xcite . the later seems to be superior in the strongly correlated regime due to substantial reduction of the size of matrices that need to be manipulated @xcite and , most importantly , the empirical finding is that the minus sign problem in this approach is severely reduced or maybe even eliminated @xcite . the expansion in hybridization strength has a long history starting with the famous non - crossing approximation @xcite and various extensions of it such as oca @xcite , ctma @xcite , sunca @xcite . all these approximations can be viewed as partial summation of the same type of diagrams . with stochastic sampling , the summation of essentially all the diagrams is now possible . the only weakness of the approach is that it works exclusively with the imaginary time green s functions and analytic continuation to real axis and access to real axis self - energy , for example , is still hard to achieved . however , we believe that the substantially enhanced precision of the method , as compared to hirsh - fye qmc , will make this step easier . the idea of expanding partition function in terms of the hybridization with the conduction band , dates back to work of g. yuval and p.w . anderson @xcite . in this work , mapping the hybridization expansion to coulomb gas model lead to one of the first breakthroughs in the area of the kondo problem . generalization to asymmetric anderson model was published by h.d.m . haldane in ref . . similar approach was later used in exploring the physics of the generalized hubbard model in the context of dmft @xcite . in this work , the hybridization expansion of the partition function was analyzed by renormalization - group analysis technique . an early implementation of the related idea to sum up the terms of the partition function by monte carlo sampling was implemented in ref . to solve the two two - channel anderson impurity model . _ sampled partition function : _ in the new quantum monte carlo method , the partition function is expanded in terms of hybridization strength @xmath10 and the resulting diagrams are summed up by stochastic metropolis sampling . taylor expansion of the impurity partition function eq . ( [ zimp ] ) gives @xmath20 e^{-s_{c}}\sum_k \frac{1}{k ! } \left [ \sum_{{\alpha}{\alpha } ' } \int_0^\beta d\tau\int_0^\beta d\tau ' \psi_{{\alpha}'}(\tau')\psi^\dagger_{\alpha}(\tau)\delta_{{\alpha}{\alpha}'}(\tau,\tau ' ) \right]^k\end{aligned}\ ] ] by separating the cluster contribution from the bath contribution the partition function can be cast into the form @xmath21 e^{-s_{c}}\sum_k \frac{1}{k ! } \int_0^{\beta}\prod_{i=1}^k d\tau_i \int_0^{\beta}\prod_{i=1}^k d\tau_i ' \sum_{{\alpha}{{\alpha}}'}\prod_{i=1}^k\left[\psi_{{{\alpha}}'_i}(\tau_i')\psi_{{\alpha}_i}^\dagger ( \tau_i)\right ] \times \prod_{i=1}^k \delta_{{\alpha}_i{{\alpha}}'_i}(\tau_i,\tau_i ' ) . \label{z_f}\end{aligned}\ ] ] it becomes clear that partition function is a product of two terms : the average over the cluster states @xmath22 and average over the bath degrees of freedom @xmath10 . it was pointed out in ref . that naive sampling of the above diagrams would run into a very bad minus sign problem . the reason is that crossing diagrams ( vertex corrections to famous non - crossing approximation to anderson impurity model ) can have either sign and thus weights that correspond to crossing diagrams could be negative . the ingenious idea proposed in ref . is to combine all the diagrams of the same order , crossing and non - crossing , into a determinant . mathematically , this can be expressed by @xmath23 where @xmath24 e^{-s_{c}}$ ] and average of the operator is @xmath25 e^{-s_{c } } o$ ] . this is the central equation of the continuous time monte carlo sampling around the atomic limit . to derive eq . ( [ z_sample ] ) from eq . ( [ z_f ] ) one needs to permute time integration variables @xmath26 in all possible ways . permutation of fermions gives minus sign in an odd permutation . this minus sign can be absorbed in the minus sign of the product of hybridizations , resulting in the determinant of hybridizations . _ simulation : _ the set of diagrams , which are associated with the set of imaginary times @xmath27 and corresponding band indexes @xmath28 are visited by monte carlo ( metropolis ) algorithm with the weights given by eq . ( [ z_sample ] ) . the effect of the hybridization @xmath29 is to create a kink in the time evolution of the cluster , i.e. , to destroy one electron at time @xmath30 on the cluster and create another electron at some other time @xmath26 on the cluster . the number of kinks is always even due to particle number conservation . two monte carlo steps which need to be implemented are : i ) insertion of two kinks at random times @xmath31 and @xmath32 ( chosen uniformly @xmath33 ) , corresponding to a random baths @xmath34 , @xmath35 . ii ) removal of a two kinks by removing one creation operator and one annhilation operator . many other steps can greatly reduce the sampling time , for example displacing a randomly chosen operator ( either @xmath22 or @xmath36 ) to a new location chosen uniformly @xmath33 . the double step of inserting or removing two kinks is also possible and is relevant when off - diagonal components of @xmath10 are dominant . the detailed balance condition requires that the probability to insert two kinks at random times @xmath26 , @xmath30 , being chosen uniformly in the interval @xmath33 , is @xmath37 \label{padd}\ ] ] where @xmath38 is the number of baths , @xmath39 is the current perturbation order ( number of kinks/2 ) , @xmath40 is the cluster matrix element @xmath41 and @xmath42 is the ratio between the new and the old determinant of baths @xmath10 and can be evaluated using usual shermann - morrison formulas . the factors of @xmath43 enter because of the increase of the phase space when adding a kinks ( increase of entropy ) while the factor @xmath44 comes from factorials in eq . [ z_sample ] . similarly , the probability to remove two kinks , chosen randomly between @xmath45 $ ] is @xmath46 . \label{premove}\ ] ] an important simplification occurs if the hybridization is block diagonal . since @xmath47 vanishes for some combination of @xmath48 the determinant in eq . ( [ z_sample ] ) can be written as a product of smaller determinants , one for each block of hybridization . a specially simple case occurs when all the blocks are of size one , the partition function becomes a product of @xmath38 terms , where @xmath38 is number of all baths @xmath49 the probability to add two new kinks eq . ( [ padd ] ) now depends on the number of kinks of this particular type @xmath50 ( rather than total perturbation order @xmath39 ) and the dimension of the bath subspace @xmath38 becomes unity . in general the probability to add two kinks is @xmath51 , where @xmath52 is the number of bands which form an off - diagonal sub - block in hybridization @xmath10 and need to be treated simultaneously in one determinant @xmath53 in eq . ( [ z_sample ] ) . the size of hybridization determinants can thus be greatly reduced if hybridization @xmath10 is block diagonal . the cluster term @xmath54 however can not be broken into separate contributions for each bath , rather the full trace needs to be computed numerically . it is therefore essential to find a fast way to compute the cluster average @xmath55 in eq . ( [ z_sample ] ) . _ exact diagonalization of the cluster : _ it is crucial to evaluate the cluster trace in eigenbase of the cluster hamiltonian and take into account the conservation of various quantum numbers , such as the particle number , the total spin , and the total momentum of the cluster states . typical contribution to the cluster part of the trace that needs to be evaluated at each monte carlo step , takes the form @xmath56 where the matrix elements are @xmath57 and @xmath58 are eigenvalues of the cluster . the actual order of operators in eq . ( [ evolution ] ) depends on their time arguments and creation operator is not necessary followed by annhilation operator . the bottleneck of the approach is that the number of cluster states @xmath59 is very large ( for example , single site dmft for the @xmath0 shell requires 2@xmath60 states ) . consequently , the matrix elements @xmath61 are in general very large matrices and the typical diagram order is inversely proportional to temperature ( see fig . [ fig1 ] ) therefore one typically needs to multiply few hundred large matrices at each monte carlo step . it is clear that this is very impractical and the progress can be achieved only by various trick which we have implemented * most of matrix elements vanish . a fast algorithm is needed to determine which matrix elements are nonzero . * symmetries of the problem can be taken into account to reduce the size of the @xmath61 matrix . * the number of trial steps is usually much bigger ( 100 times ) than the number of accepted steps and the insertion or removal of a kink is very local in time operation . it is convenient to store the product eq . [ evolution ] ( from both sides , left and right ) and when trying to insert a new kink , recompute the trace only between the inserted times @xmath31 and @xmath32 . * during simulation , the probability for visiting any cluster state can be recorded and can be used in the next step to eliminate the irrelevant atomic states from the trace in eq . [ evolution ] . the cluster base can hence be adjusted dynamically to describe the particular regime studied by the minimum number of relevant cluster states . to illustrate the method , let us consider a concrete example of the cluster of one band model ( hubbard or t - j ) in the normal state . the bath index @xmath34 runs over cluster momenta @xmath62 and spin orientation @xmath63 . the eigenstates of the cluster can be written in a form @xmath64 , where @xmath65 is total number of electrons in the state , @xmath66 and @xmath67 are the total spin and its @xmath68 components , while @xmath69 is the total momentum of the cluster state and @xmath70 stands for the rest of the quantum numbers . _ concept of superstates : _ in this base , the matrix elements of the creation operator are greatly simplified @xmath71 and @xmath72 . it is therefore convenient to group together states with the same @xmath73 and treat the rest of the quantum numbers as internal degrees of freedom of a cluster superstate @xmath74 . the superstate @xmath75 is multidimensional state with internal quantum numbers @xmath76\rangle\equiv|\{s,\gamma\rangle\}$ ] . it is then clear that creation operator acting on a state @xmath75 gives a unique state @xmath77 and it is enough to store a single index array @xmath78 to figure out how the hilbert subspaces are visited under application of a sequence of creation and annhilation operators such as in eq . ( [ evolution ] ) : @xmath79 . this sequence is very often truncated in few steps only , because most of the sequences contain either multiple application of the same creation or annhilation operator ( pauli principle ) or because they lead to a state outside the base ( for example @xmath80 or @xmath81 ) . once the nonzero matrix elements are found by this simple index lookup , the value of the matrix element needs to be computed by matrix multiplication . by this breakup of the hilbert space and introducing superstates @xmath75 , the largest matrix which needs to be treated for the t - j model on a plaquette is @xmath82 and for the hubbard model it is @xmath83 , thus substantially smaller than the original @xmath84 and @xmath85 dimensional hilbert space . to compute the trace in eq . ( [ evolution ] ) we start with unity matrix in each subspace of a superstate @xmath75 and apply both the time evolution operator @xmath86 ( by multiplying each row of a matrix with its time evolution ) and the kink ( by multiplication with the matrix @xmath87 or @xmath88 ) . the operation of @xmath61 brings us to the next superstate @xmath89 where we repeat both the time evolution and the kink application . after @xmath39 steps , the trace of a matrix gives the desired matrix element . _ storing the time evolution : _ the number of kinks is proportional to inverse temperature @xmath90 and kinetic energy of the system @xmath91 ( see eq . ( [ avek ] ) ) . it thus becomes large at low temperatures . however , an insertion of a kink with large time difference ( @xmath92 with large @xmath93 ) has a very low probability . the reason is that pauli principle forbids insertion of the pair @xmath94 if another kink of the same species @xmath95 is between the two times @xmath96 . at the same time , @xmath97 is like @xmath98 peaked at small times @xmath93 and falls of at large times making the long time intervals rare . the insertion of a kink is thus fairly local in time operation , therefore it is convenient to store a whole chain of products that appear in eq . ( [ evolution ] ) from both sides , left and right , to make trial step very cheap . it takes only few matrix multiplications ( almost independent of temperature ) to compute the trace in eq . ( [ evolution ] ) . when the move is accepted , the trace needs to be updated which takes somewhat more time . however , the acceptance rate is typically small and on average does not require much computational time . _ adjusting the cluster base : _ finally , the ultimate speedup can be achieved by dynamically adjusting `` the best '' cluster base . the probability for a cluster state @xmath59 can be computed during simulation . for a given diagram with particular configuration of kinks , the probability for a cluster state @xmath59 is proportional to its matrix element defined in eq . ( [ evolution ] ) , i.e. , @xmath99 the sampled average of this quantity gives probability for cluster eigenstate @xmath59 . a large number of cluster states have very small probability and can be eliminated in simulation to ultimately speed up the simulation . it is important that the probability for any cluster state depends on the particular problem at hand and the program adjusts the base dynamically after a few million of monte carlo steps . _ green s function evaluation : _ like in other impurity solvers which are based on the expansion of the hybridization ( nca , oca , sunca ) , the green s function is computed from the bath electron t matrix . using equation of motion , it is easy to see that the bath green s function @xmath100 is connected to the local green s function @xmath3 through the following identity @xmath101 or summing over momenta @xmath102 this green s function @xmath103 is equal to the ratio between the determinant of @xmath104 s @xmath105 where one row and one column is added to the bath electron determinant . the reason for this simple form is that the conduction band is non - interacting and thus obeys wicks theorem . here we used the definition @xmath106 using shermann morrison formulas to express enlarged determinant by the original determinant of @xmath104 , @xmath103 becomes @xmath107 finally , comparing eq . ( [ gl1 ] ) and ( [ gl2 ] ) we see that @xmath108 and in imaginary frequency @xmath109 this equation is the central equation of the approach since it relates local green s function with the quantities computed in qmc importance sampling . the equation shows that only matrix @xmath110 needs to be stored and manipulated in simulation . from the above consideration , it is clear that one could also add two rows and two columns to matrix @xmath104 and compute the two particle vertex function in similar way without much overhead . _ fast updates : _ the green s function can be updated in linear time ( rather than quadratic ) . when adding a construction and annhilation operator at @xmath26 and @xmath30 , adding a column at @xmath111 and row at @xmath112 to matrix @xmath113 , leads to the following relation between the matrix @xmath114 and @xmath115 @xmath116 here one row and one column of zeros is added to @xmath117 to match the size of @xmath118 . the arrays @xmath119 and @xmath120 are given by @xmath121 where @xmath122 and @xmath123 is @xmath124 finally , the local green s function becomes @xmath125 it is clear from eq . ( [ dgn ] ) that only linear amount of time is needed to update the local green s function . when removing two kinks of construction and annhilation operators at @xmath111 and @xmath112 , old and new matrix @xmath126 are related by @xmath127 green s function therefore becomes @xmath128 finally , the exponential factors @xmath129 do not need to be recomputed at each monte carlo step since all `` old '' times can be stored and the exponents need to be computed only for the new pair of times and only at each accepted move . in the present implementation , the algorithm to sample directly @xmath130 is sufficiently fast that it does not introduce any performance costs . since it does not introduce systematic error in binning @xmath98 we believe it is superior than the alternative implementations which sample @xmath98 . _ large frequencies and moments : _ similarly to the hirsch - fye qmc , the low frequency points of green s function converge very fast to the exact value while the high - frequency points , when sampled directly , contain a lot of noise . it is therefore not very useful to sample large frequencies in the above described way . usually we sample 200 - 300 matsubara points while the rest are replaced by the high frequency moments of the self - energy computed analytically . the high frequency moments of the self - energy are computed from the green s function moments , which , in general , take the following form @xmath131\cdots ] ] , \psi^\dagger_{\beta}\right\}\right\rangle\ ] ] to compute the moments within the present approach , few operators need to be sampled in simulation . in the one - band model , only density is required , but in more complicated situation , higher order density - density and exchange terms enter . in general , green s function moments can be expressed by the average of a few equal time operator @xmath132 , which take particular simple form in the local eigenbase @xmath133 the lowest order self - energy moments can then be computed in the following way @xmath134 _ sampling of other quantities : _ equal time operator like density or those in eq . ( [ gmoments1 ] ) and ( [ gmoments2 ] ) can be straightforwardly computed in simulation . to improve the precision of simulation , operators are averaged over all times , i.e. , @xmath135\ ] ] the contribution to @xmath136 of each particular diagram is @xmath137 \label{o1}\ ] ] where @xmath138 is the cluster part of the trace eq . ( [ evolution ] ) , @xmath139 are the operators in eq . ( [ evolution ] ) which appear before @xmath140 and @xmath141 contains operator with time arguments greater than @xmath140 . @xmath39 is the number of kinks in the diagram @xmath142 . as we mentioned before , these time evolution operators ( @xmath139 , @xmath143 ) are stored and regularly updated during simulation . further simplification is possible , if operator @xmath136 commutes with @xmath144 @xmath145.\ ] ] if operator @xmath146 is one of the good quantum numbers of a superstate ( one of @xmath73 in our example above ) , matrix @xmath147 is proportional to identity matrix in each superstate @xmath148,m_2[i]}=\delta_{m_1,m_2 } o_i$ ] . starting with a superstate @xmath149 , kinks of a diagram @xmath142 generate a sequence of superstates @xmath150 , as explained above . the equal time average of cluster conserved quantity therefore simplifies to @xmath151 where @xmath152 is probability for a superstate @xmath75 defined in eq . ( [ pm ] ) and @xmath153 is the value of the conserved quantity in each superstate @xmath154 in the interval @xmath155 . this time difference is the time each superstate `` lives '' in the particular diagram . in this way , the electron density @xmath156 and average magnetization @xmath67 is computed to very high precision . similar simplification in evaluating the total spin susceptibility , defined by @xmath157 where @xmath67 is the total spin of the cluster ( or atom ) leads to @xmath158 where @xmath159 is the total spin of the cluster in cluster eigenstate @xmath160 . if @xmath146 does not commute with @xmath161 , the integral ( [ o1 ] ) needs to be evaluated . in cluster eigenbase it takes the form @xmath162 _ sampling the total energy : _ the average of the potential energy @xmath163 can be computed from the average of the local energy since @xmath164\end{aligned}\ ] ] here we concentrate on the case of general single site dmft rather than the cluster extensions . the reason is that kinetic energy in cluster extensions depends on the periodization scheme and we will not go into this details here . kinetic energy of the general single site dmft @xmath165 $ ] can be computed by @xmath166 + { \mathrm{tr}}[(\mu+e_{imp})n]\end{aligned}\ ] ] the total energy is therefore given by @xmath167 + \mu n\end{aligned}\ ] ] the first term @xmath168 can be computed very precisely in simulation . the sampled quantity @xmath136 is just the energy of an atomic state and can be simply obtained from the probabilities for atomic states @xmath169 . computing kinetic energy from the green s function gives in general worse accuracy because the high - frequency behavior of the green s function can not be directly sampled and augmentation with analytically computed tails is necessary . however , it is simple to show that the average value of the perturbation order is related to the average of the kinetic energy @xmath170 \label{avek}\ ] ] where @xmath171 is the average perturbation order and @xmath172 is temperature . the later quantity is directly sampled in the present algorithm and it is just the center of gravity of the histogram ( presented in fig . [ fig1 ] ) . finally , the total energy @xmath173 is given by @xmath174 all quantities in this equation can be computed to very high accuracy and since low temperatures can be reachable in this method , the entropy can be obtained by integrating the specific heat . _ superconductivity : _ the power of the method can be further demonstrated by studying the superconducting state of the strongly correlated systems at low temperature with essentially no performance cost . by employing the nambu formalism , the translationally invariant cluster methods ( for details on cellular - dmft on a plaquette see appendix [ appa ] ) results in @xmath175 two dimensional baths , where @xmath175 is the number of cluster momenta . namely , the baths @xmath176 and @xmath177 are coupled through the anomalous component of hybridization and require simultaneous treatment in @xmath178 in eq . ( [ z1b1 ] ) . the determinants are on average twice as large as in normal state , however , the cluster part of the trace in eq . ( [ z1b1 ] ) remains unchanged . since most of the time is usually spend in evaluating the local part of @xmath179 , the performance is not noticeably degraded in superconducting state . in typical run presented below , the histogram is peaked around @xmath180 which is equal to the order of a typical diagram . in the translational invariant representation employed here , the size of a typical determinant in eq . ( [ z1b1 ] ) is only @xmath181 and using fast - update scheme presented above , the trace over the bath states is not expensive part of the algorithm . _ hund s coupling and spin - orbit coupling : _ in materials with open @xmath0 orbitals , the multiplet effects are very strong and @xmath182 approximation is not adequate . simultaneous inclusion of hund s coupling and spin - orbit coupling in dmft method is crucial for quantitative description of actinides @xcite . minimal local hamiltonian for lanthanide and actinide materials is @xmath183 here @xmath184 is the impurity level without the spin - orbit coupling since later is included explicitely . the hund s coupling and spin - orbit coupling take the following form @xmath185 c^{j m_j}_{lm,\sigma}c^{j m_j}_{l m',\sigma ' } f_{lm\sigma}^\dagger f_{lm'\sigma'}\end{aligned}\ ] ] here @xmath186 are spheric harmonics , @xmath187 are clebsch - gordan coefficients and @xmath188 are slater integrals . the @xmath189 , usual hubbard @xmath1 , is commonly computed by constrained lda , while the rest of the slater integrals ( @xmath190 , @xmath191 and @xmath192 ) are computed using atomic physics program @xcite . finally , the spin - orbit strength @xmath193 is computed within lda program and needs to be updated during self - consistent lda+dmft calculation . exact diagonalization of the atomic hamiltonian leads to eigenstates with conserved number of particles @xmath65 , total angular momentum @xmath194 and its @xmath68 component @xmath195 . for the one electron base , it is more convenient to chose the electron total angular momentum @xmath196 rather than @xmath197 since the construction and annhilation operators conserve @xmath68 component of total spin @xmath198 superstates @xmath75 are chosen in a way to simplify the matrix elements of the creation operator @xmath36 . in this case , a convenient choice is @xmath199 in the absence of crystal field splitting , the hybridization @xmath10 is a diagonal matrix ( all @xmath200 baths are one - dimensional ) . however , strong crystal fields splitting generates off - diagonal components of @xmath201 and in general , the determinant @xmath178 in eq . ( [ z_sample ] ) can not be broken into small determinants as in eq . ( [ z1b1 ] ) . we implemented cellular dmft for both the hubbard and the t - j model on a plaquette . the energy scales are given in units of @xmath202 , hubbard @xmath1 is fixed at @xmath203 while @xmath194 of the t - j model is set to @xmath204 . we simulated @xmath205 monte carlo steps per processor and results were averaged over 64 processors . one dmft step for the hubbard model takes approximately 45 minutes on 1.7@xmath206ghz pc processor therefore each monte carlo step requires on average one million clock - ticks ( 0.9 mflops ) . . ] fig . [ fig1 ] shows histograms ( probability distribution for the perturbation order @xmath39 ) for few dopings of the hubbard model @xmath207 at @xmath208 . the average perturbation order is increasing with doping since the electrons are getting more delocalized ( absolute value of the kinetic energy is increasing ) and the creation of kinks becomes less expensive . it is of the order of 450 in overdoped regime , but the typical size of the determinant is only 112 in superconducting state or 56 in normal state . we recently addressed the problem of coherence scale in the t - j model @xcite and we found , using nca as the impurity solver , that the imaginary part of the cluster self - energy @xmath209 , which plays the crucial role in the optical conductivity and transport , becomes very large at optimal doping and consequently the coherence scale vanishes around optimal doping . here we extend this study to the hubbard model using much lower temperature . we will show that the system become strongly incoherent at optimal doping and the maximum of @xmath210 tracks the maximum scattering rate in both the hubbard and the t - j model . [ fig2 ] shows the imaginary part of the self - energy @xmath211 at few different dopings and temperature @xmath212 which is around the superconducting critical temperature of this approach . the system is still in normal state . it is clear that the self - energy at large frequencies is a monotonic function of doping and is largest at the mott transition @xmath213 . however , the low frequency region is distinctly different and the crossing of self - energies is observed around @xmath214 . low doping as well as large doping self - energies can be extrapolated to zero , while at optimal doping self - energy remains of the order of unity even around the critical temperature . on imaginary axis . temperature @xmath212 is around the critical temperature but still in normal state . at low frequency , the most incoherent self - energy corresponds to optimal doped and not to the underdoped system . ] self - energy on the imaginary axis for both the t - j ( left ) and hubbard ( right ) model showing the large scattering rate at optimal doping . the lower panel shows the anomalous self - energy for the same models and doping levels and can be used to locate the optimal doped regime . ] the upper panels of figure [ fig3 ] show the same self - energy at small frequency for both the t - j ( left ) and the hubbard ( right ) model . in this figure , temperature is @xmath215 and is far below @xmath210 therefore all curves extrapolate to zero and the system becomes coherent in superconducting state . however , the reminiscent of the strong incoherence at optimal doping is the large slope of imaginary part of @xmath216 which induces very small quasiparticle residue in this regime . the precise position of the optimal doping , characterized by the largest low frequency anomalous self - energy , is different in the t - j and in the hubbard model ( see lower panel of fig . [ fig3 ] ) . it is around @xmath217 in the former and around @xmath218 in the later . however , the large scattering rate is ultimately connected with the largest anomalous self - energy and hence largest critical temperature in both the t - j and the hubbard model . in our view , the most important advantage of the new quantum monte carlo method is that it can treat the realistic multiplet structure of the atom . it was shown in ref . that hubbard term only ( @xmath219 ) , leads to sever underestimation of the interaction strength in actinides and for realistic values of @xmath219 , dmft predicts heavy fermion state in curium rather than magnetic state . negligence of the multiplet structure of plutonium misses the fine structure of the quasiparticles ( two peaks around 0.5 and 0.85 ev ) and more importantly , predicts only weakly correlated metallic state in delta phase of plutonium . to demonstrate the advantage of the new method , we recomputed the localization - delocalization transition in the archetype material exhibiting kondo collapse , namely the @xmath220 transition of elemental cerium . since the number of electrons in cerium fluctuates between states with zero , one and two electrons in the @xmath0-shell , the number of atomic states that needs to be kept is relatively small hence solving the impurity problem requires very little computational power in this case . in fig . [ fig4 ] we show the `` valence histogram '' @xcite of the two phases of cerium , i.e. , the projection of the density matrix to the eigenstates of the atom . the plot shows the probability to find an @xmath0 electron of cerium in any of the atomic eigenstates and demonstrates how strongly is the atom fluctuating between atomic states . the typical fluctuating time is inversely proportional to the kondo temperature of the phase , being around 2000@xmath221 in the alpha phase and around 80@xmath206k in the gamma phase . the itinerant alpha phase histogram is peaked for many atomic states , including the spin - orbit split @xmath222 and @xmath223 singly occupied states as well as the empty state . on the other hand , the local - moment gamma phase is peaked only at the ground state of the singly occupied sector with @xmath222 spin showing that the dmft ground state closely resembles the atomic @xmath224 ground state . electron in the solid spends appreciable time in a few atomic configurations . the height of the peak corresponds to the fraction of the time the @xmath0 electron of the solid spends in one of the eigenstates of the atom , denoted by the total spin @xmath194 of the atom . we summed up the probabilities for the atomic states which differ only in the z component of the total spin @xmath225 . the @xmath226 axis indicates the energy of atomic eigenstates in the following way : @xmath227 and @xmath228 , where n is between 0 and 2 . ] we generalized the recently developed continuous time quantum monte carlo expansion around the atomic limit @xcite to clusters treated within cellular dmft or dynamical cluster approximation as well as to materials that require realistic hund s coupling and multiplet splitting of the atomic state . we explained the steps necessary for the efficient implementation of the method . our low temperature data in the strongly correlated regime of the hubbard and t - j model show the efficiency of the method and demonstrate its superiority compared to the conventional hirsh - fye quantum monte carlo method . the long - standing problem of adequate treatment of the multiplet splitting within dmft is resolved . this splitting is crucial in actinides and its omission can lead to wrong prediction of the magnetic nature of the dmft ground state solution . we showed that the optimal doped regime in both the t - j and the hubbard model is characterized by the largest scattering rate and the system becomes more coherent in the underdoped and overdoped regime . the precise position of the optimal doping , as determined from the maximum in anomalous self - energy , is different in the two models . however , the strong incoherence is always found at the doping corresponding to maximum @xmath229 . we computed the valence histogram across the cerium alpha to gamma transition with emphasis on the multiplet splitting of the atomic states . we showed that the atom fluctuates between many atomic states in itinerant alpha phase and both the @xmath222 and @xmath223 spin - orbit split states have large probability in the ground state of the system . the empty state and the doubly occupied states , which are substantially split by hund s coupling , acquire a finite probability . the gamma phase , on the other hand , shows well defined valence @xmath230 and charge fluctuations become rare . the author wish to thank g. kotliar for very stimulating discussion and careful reading of the manuscript . for the present continuous time monte carlo method , it is convenient to pick the base such that the local quantities are diagonal . cellular dmft uses a generalized open boundary conditions and therefore , in general , momentum is not a good quantum number . however , for small clusters , such as plaquette , all sites are equivalent and translational invariance is still obeyed . the local green s function and hybridization in momentum base take the diagonal form @xmath231 in superconducting state @xmath232 is a @xmath233 matrix in nambu notation . although the local quantities are diagonal , non - interacting hamiltonian at general @xmath234-points is not . the self - consistency condition in superconducting state takes the following form where we defined @xmath236 and assumed @xmath237 here @xmath238 is the normal on - site self - energy , @xmath239 is the nearest - neighbor and @xmath240 is the next - nearest neighbor self - energy and @xmath241 s are the anomalous components of the self - energy . for d - wave symmetry , @xmath242 and @xmath243 vanish and @xmath244 . the advantage of this formulation of the cellular dmft is that the hybridization becomes block diagonal and hence the determinants ( @xmath245 ) which enter eq . ( [ z_sample ] ) can be broken up into separate contribution for each momentum @xmath69 point like in eq . ( [ z1b1 ] ) . a. georges , g. kotliar , w. krauth and m.j . rozenberg , rev . phys . * 68 * , 13 ( 1996 ) . g. kotliar , and d. vollhardt , physics today * 57 * , 53 ( 2004 ) . g. kotliar , s.y . savrasov , k. haule , v.s . oudovenko , o. parcollet , c.a . marianetti , rev . mod . phys . * 78 * , 865 ( 2006 ) . savrasov , g. kotliar , e. abrahams , nature * 410 * , 793 ( 2001 ) . savrasov , k. haule , and g. kotliar , phys . lett . * 96 * , 036404 ( 2006 ) . x. dai , s.y . savrasov , g. kotliar , a. migliori , h. ledbetter , e. abrahams , science * 300 * , 5621 , 953 - 955 ( 2003 ) . k. haule , v. oudovenko , s.y . savrasov , and g. kotliar , phys . . lett . * 94 * , 036401 ( 2005 ) . j.h . shim , k. haule , and g. kotliar , cond - mat/0611760 . oudovenko , g. palsson , k. haule , g. kotliar , and s.y . savrasov , phys . b * 73 * , 035120 ( 2006 ) . h. kajueter and g. kotliar , phys . lett . * 77 * , 131 ( 1996 ) . r. bulla , phys . lett . * 83 * , 136 ( 1999 ) ; r. bulla , t. a. costi , and d. vollhardt , phys . b * 64 * , 045103 ( 2001 ) . x. dai , k. haule , and g. kotliar , phys . rev . b * 72 * , 045111 ( 2005 ) .

high precision doppler spectroscopy is an important tool to investigate many astrophysical processes in the universe . measurement of tiny doppler shifts from radial velocity ( rv ) can provide insight about the stellar interior ; reveal the existence of unseen planets around other stars ; help us probe the variability of fundamental constants of nature and provide useful information of binary and pulsating stars @xcite . in particular , the radial velocity technique has been immensely successful in detecting hundreds of planets with different mass range and orbital periods @xcite . the main challenge and focus of contemporary research is to find earth - analogs within the habitable zone of star where life sustaining conditions could prevail . recently , the doppler data from eso s two precision rv instruments , namely harps and uves , was decisive in unveiling the presence of 1.3 earth - mass rocky planet orbiting proxima centauri - the nearest known star to the sun @xcite . typically , doppler precision of @xmath3 cms@xmath2 or better is required to address many compelling science cases including the detection of terrestrial - size planets around sun - like stars @xcite . leading rv observing campaigns are carried out with state - of - the - art echelle spectrograph , carefully designed and built to operate under tightly regulated thermovac conditions @xcite . a high resolution spectrum of star in visible / nir band is recorded on a 2d array of ccd pixels . to extract the rv signal from time series data , spectrograph is wavelength calibrated with a laboratory source of well known emission or absorption lines . in calibration process several lines of known wavelengths are identified and a quadratic or high order polynomial fit is used to invert the pixel coordinates into accurate wavelength units . two commonly used wavelength reference in astronomy are thorium - argon ( th - ar ) emission lamp and iodine absorption cell @xcite . despite averaging the shift over thousands of spectral lines , the best rv precision obtained from these sources is limited to @xmath4 ms@xmath2 . the rv uncertainty of traditional sources arises from wavelength inaccuracies of individual lines , spectral contamination , aging of the lamp , limited spectral coverage , line blending effects , uneven distribution and varying intensity of lines across spectrograph band @xcite . an ideal calibrator choice for enabling sub - ms@xmath2 rv precision in next generation spectrographs is a broadband laser frequency comb ( lfc ) . the lfc consists of thousands of equispaced lines produced from a femtosecond mode - locked laser @xcite . the entire comb spectrum is stabilized by anchoring to atomic clock with well established radio - frequency locking technique . since past decade different versions of lfcs have been tested and validated @xcite . new developments are still awaited to overcome a couple of technical and operational difficulties . it might take a while for lfc to become a simple and affordable turn - key instrument for routine wavelength calibration at most observatories . a simple and cost effective approach to produce reference spectra is to use cavity resonance lines of a fp etalon @xcite . a plane parallel fp cavity when illuminated with a broadband white light source acts as a periodic wavelength filter . the transmitted light consists of evenly distributed lines that result from multiple beam interference inside the cavity @xcite . the cavity lines are separated by free spectra range @xmath5 , where @xmath6 is the light speed and @xmath7 is the cavity length . many variants of fps ( e.g. solid , air spaced and fiber based ) have been successfully built and tested for astronomy calibration @xcite . the positional accuracy of fp lines critically depends on stability of the cavity itself . any noise affecting the cavity length would randomly displace the resonant lines thus directly impairing the achievable rv precision . the impact of temperature and pressure fluctuations on the cavity length is generally eliminated by mounting the fp inside a thermally controlled evacuated chamber which is capable of achieving @xmath8 1 ms @xmath2 short - term stability @xcite . a passively stabilized fp apparatus producing reference spectra has no built in mechanism to track systematic offsets or drifts in cavity length occurring over time scales of weeks , months or even years . regardless of good passive isolation , cavity drift could still occur from temperature residuals , material aging or changes in optical properties of the coatings @xcite . the radial velocity @xmath9 and the fractional length and frequency stability of the cavity can be expressed as @xmath10 . for instance , a 1 nm change in a 1 cm long cavity corresponds to @xmath855 mhz frequency shift of cavity lines in visible ( @xmath11 nm ) region , which amounts to 30 ms@xmath2 rv uncertainty . to reach a rv precision @xmath12 cms@xmath2 , the cavity length should be stable to @xmath13 level or better . for long - term usability ( several months to years ) of astro - fp , research efforts are now being directed at controlling and tracking the time dependent cavity drift with advance techniques @xcite . measurement of tiny dimensional changes in cavity length calls for advanced metrology based on laser frequency standards . a drift tracking or active length - control requires cavity - locking to a frequency stabilized diode laser or vice - versa @xcite . in cavity - lock setup , the frequency stability of the laser is expected to be transferred to the stability of the fp cavity lines . the main task of frequency or phase locking mechanism is to keep cavity transmission point in resonance with frequency of the laser @xcite . a short - length , low - finesse(@xmath14 ) cavity , while useful for generating broadband reference lines for astronomy spectrographs , is not ideal for laser metrology . normally , a robust laser - lock requires a sharp discriminant signal which depends on the laser power and the cavity linewidth @xcite . the linewidth @xmath15 of the cavity should be sufficiently narrow to produce a steeper error signal . the inverse dependence of @xmath7 and @xmath16 on the linewidth @xmath17 implies a clear advantage of longer and high finesse cavities . another practical difficultly of utilizing astro - fp in laser applications is the cavity geometry . a confocal cavity design is more appropriate for locking purpose because it greatly simplifies beam alignment and mode - matching conditions for single frequency laser beam . on the contrary , a collimated white light beam is easily coupled into a short - length planar cavity producing reference spectra @xcite . these considerations show incompatible requirements of a cavity for laser metrology and astronomy calibration . clearly , a single fp cavity is inadequate for serving dual purpose of ` _ _ wavelength calibration _ _ ' on one hand and ` _ _ drift tracking _ _ ' on other . we propose to circumvent this problem by constructing two adjoining cavities of equal length but of different finesse on a single block of ultra low expansion ( ule ) glass . the low finesse _ astro - cavity _ would produce the calibration spectra while the high finesse _ lock - cavity _ would facilitate frequency locking to implement stability tracking mechanism . the length of two cavities is defined by the length of monolithic spacer block . here , the implicit assumption is that external perturbations , if any , act uniformly on the entire fp block , thus causing equal and identical length changes in both cavities . in fact , drift measurements would only be accurate so long the relative length variation between two cavities is well below @xmath18 . under this premise alone the _ lock - cavity _ would function as a reliable proxy for measuring the stability of the _ astro - cavity_. departure from this scenario would result in intractable measurement errors . mm on top are provided for the venting purpose . ( b ) the front view showing the spacer dimensions along with two cavities of diameter @xmath19 mm each and standard hr mirrors with @xmath20 mm each . ( c ) the isometric view . the cavity axis is orientated along the @xmath21-direction and the lab vertical is along @xmath22-direction . ] the main objective of this paper is to numerically examine the validity of our approach . we identify the temperature and vibration induced noise conditions that impart unequal length changes in two adjoining cavities constructed on a monoblock spacer . these studies are important to understand the level of performance one can realistically expect to achieve in the proposed device . we use finite element method to numerically simulate the impact of small thermal gradients and orientation dependent vibration noise that could deforms two cavities in dissimilar manner . from here on , the paper is organized as follows . the dual fp design and laser - lock setup for cavity drift tracking is discussed in section 2 . in section 3 , the mathematical formulism and fea model for cavity deformation is described . the vibration analysis of horizontally mounted cavity is presented in section 4 . a proposed vertical mounting configuration and numerical optimization procedure to reduce the vibration sensitivity of the fp device is given in section 5 . thermo - elastic model is used to calculate the temperature sensitivity of the fp device . effect of temperature gradient and cavity length offset is presented in section 6 . numerical accuracy of fea results is discussed in section 7 and finally , the summary of the work is given in section 8 . fabry - perot cavities are used in diverse range of applications in many areas of fundamental and applied research @xcite . ultrastable cavities are also key to gravitational wave detection , testing of lorentz invariance and experiments in future space missions @xcite . for wavelength calibration , the fp cavity spacing and mirror reflectivity @xmath23 are selected based on spectrograph requirements . the cavity spacing @xmath7 is chosen so that reference lines are easily resolved by spectrograph in the detector plane . generally , a short cavity ( e.g. , @xmath24 cm , @xmath25 ghz ) is required to generate optimally spaced comb lines to calibrate a high resolution ( @xmath26 ) astronomy spectrograph . the reflectivity ( @xmath27 ) of broadband coated mirrors yields a low finesse @xmath28 . the resulting cavity linewidth @xmath29 is such that it is optimally sampled by the detector pixels . to produce reference lines that are 15 ghz apart would require a planar cavity of length 1 cm . the main component of the proposed device is a two - channel fp etalon . it consists of two cylindrical cavities of diameter 12 mm each , bored adjacent to each other on a rectangular block ( @xmath30 mm@xmath31 ) of ultra low expansion ( ule ) glass . the cavity mirrors are made of fused silica ( fs ) and optically contacted on either side of the spacer block . we assume the optical contacts to be sufficiently strong and free from any imperfections . this is a reasonable assumption as the modern techniques of joining optical elements have become more robust @xcite . a cad geometry of the device is shown in fig . 1 . the low - finesse _ astro - cavity _ ( fp1 ) is to provide astronomy calibration lines and high - finesse _ lock - cavity _ ( fp2 ) is to provide a light channel for robust laser - lock . the cavity length @xmath32=1 cm is chosen to produce 15 ghz apart reference lines to calibrate a typical high resolution ( 100,000 ) echelle spectrographs . to maintain sufficient rigidity of the entire structure , a small 5 mm gap is provided on all sides between mirrors and spacer edges . cavity mirrors are assumed to be flat . replacing flat mirrors with curved mirrors will not alter our analysis in any major way , except a very small change in curvature due to differential thermal expansion . generated by electro - optic modulator ( eom ) . the rf electronics following the photodiode ( pd ) provides necessary feedback to keep one of the sideband frequencies , e.g. @xmath33 ( shown in red ) , locked to the peak of cavity transmission line ( shown in blue ) . for clarity , other ancillary components are not shown . ] the conceptual principle of dual fp cavity - lock is illustrated in fig . 2 . the cavity drift can be tracked in several ways , but often requires two lasers @xcite . however , the same goal can be accomplished using a simpler approach of _ sideband locking _ which needs only one laser @xcite . in this method the frequency @xmath34 of the laser is pre - stabilized to rb - cell using standard saturation absorption @xcite . a tunable electro - optic modulator ( eom ) creates radio frequency ( rf ) sidebands @xmath35 around the central frequency @xmath34 of the laser . one of the sidebands is then locked to the nearest resonance peak of fp2 cavity using modified pound - drever - hall ( pdh ) method . the rf frequency @xmath36 of the oscillator driving eom is servo - locked to cavity . this technique preserves the intrinsic stability of main laser beam @xmath34 that is locked to rubidium while exploiting the sideband tunability to track the cavity drift . to estimate the drift , the frequency of the radio oscillator has to be readout independently by a frequency counter or rf spectrum analyzer . a fractional length change ( @xmath37 ) about @xmath18 of the cavity block corresponds to rv error of 10 cms@xmath2 or a rf frequency shift of few hz in an eom operating at 1 ghz . measuring a ghz rf signal with an accuracy of few hz ( amounting to few ppb frequency resolution ) is a challenging task . however , the desired precision can be reached with many cycles of repeated measurements . alternatively , optical - beat measurement with two lasers yields much better sensitivity when the cavity is used in reflection mode , i.e. pdh locking @xcite . for example , a fractional change @xmath18 of the cavity length would induce a readily measurable frequency excursion of @xmath38 khz in the optical - beat signal obtained with pdh setup using 780 nm laser . the exact implementation details will be worked out in a separate study . the resonance stability of fp cavity is determined by the stability of optical length between each mirror pairs . different noise sources e.g. , seismic and acoustic vibrations , material inhomogeneities , fluctuations in temperature , pressure and refractive index of the air etc influence the cavity stability . the impact of pressure on refractive index fluctuations is largely eliminated by placing the cavity inside a vacuum tank . a properly designed vacuum system also attenuates acoustic noise . our focus in this study is on temperature and vibration perturbations which affect the long- and short - term stability of the fabry - perot cavity . to achieve a desired tracking accuracy the length changes @xmath39 in astro - cavity fp1 must strongly correlate with the length changes @xmath40 in lock - cavity fp2 . how well @xmath39 and @xmath40 correlate would largely depends on cavity geometry and the way noise is transmitted into the experiment . normally , the thermal and acoustic noise couples into cavity through mounting and support system . the main purpose of this paper is to estimate noise levels for which the differential length change in two cavities , @xmath41 becomes significant . in other words , when @xmath42 , the change in rf lock - point of fp2 would accurately reflect the length changes in fp1 . uncorrelated cavity deformations would render the laser - lock technique ineffective . the shape of an elastic body can deform under the application of a force that may arise from constant acceleration ( e.g. gravity loading ) , floor - vibrations , shocks or acoustic pickup . in most cases , acceleration experienced by a cavity is in the form of vibration perturbations . these perturbations are time - dependent and contain a large range of frequencies that excite various mechanical eigenmodes of the cavity . the exact nature of interaction depends on wavelength of excitation relative to the physical size of the cavity @xcite . in high - frequency limit ( small wavelength ) , a full dynamic analysis is needed to obtain a complete information of local strain and displacement of the cavity points . since the excitation wavelengths at higher frequencies are significantly smaller than the cavity size , the conventional numerical discretization methods for dynamic analysis are inefficient and require significant computer resources . though from practical viewpoint , the high frequency vibrations are less of a problem . they can be effectively attenuated in lab using vibration isolation tables , suitable damping materials , soft cushioning and multi - stage suspension . in low - frequency limit ( @xmath43 hz ) the dynamic problem of cavity deformation can be reduced to static analysis @xcite . in this regime the excitation wavelength become longer than the cavity size and all points in the cavity respond to the applied force in unison manner . the low - frequency collective motions ( compression or stretching ) of the particles can be approximated by quasi - static equilibrium resulting from application of a static force or a constant acceleration . traditionally , the vibration or acceleration sensitivity of a cavity is expressed in normalized units of acceleration due to gravity @xmath44 [ ms@xmath45 . in this work we also follow the same convention . elastic solids change shape under the influence of external force . the elastic deformations can be quantified by knowing the displacement of material points inside the body . mathematically , elasticity problem can be formulated in terms of field equations which govern the relations among displacements , strains and stresses inside a loaded object . the basic field equations for linear isotropic solids comprise @xcite : * @xmath46__strain - displacement relation ( compatibility equations ) _ _ @xmath47 which completely specifies the strain @xmath48 in terms of displacement @xmath49 and its derivatives . + * @xmath46__equations of equilibrium _ _ @xmath50 which is a statement of newton s 2nd law for static cases where displacement has no time dependence and derivatives of stress tensor @xmath51 and body forces @xmath52 add to zero . + * @xmath46__constitutive equations _ _ ( hooke s law ) @xmath53 which establishes general stress - strain behaviour via material parameters , i.e. , young s modulus @xmath54 and poisson s ratio @xmath55 . . the entire structure is vertically compressed and horizontally expanded from original configuration . ( c ) the z - displacement and tilt of the cavity mirrors ( magnified by @xmath56 ) from their nominal positions . the black outline indicates undeformed shape of the cavity . the color scale represents the cavity displacement in @xmath21-direction . ] but for simple geometries , the solution of general system of eqs.(1)-(3 ) by analytical methods is almost impossible . sometimes further simplifications are made by developing a reduced set of field equations solely in terms of stresses or displacements . one such formulism is to eliminate stresses and strains from eqs.(1)-(3 ) and express the displacement in a simplified form , known as navier s equation . the vector form of navier s equation is @xcite : @xmath57 where @xmath58 is shear modulus and @xmath59 is _ _ lam__s constant . equation ( 4 ) is usually solved with displacement boundary conditions of the form : @xmath60 , where @xmath61 denotes the boundary points on the surface of the body and @xmath62 are prescribed displacement values . as a rough approximation , the cavity deformation problem is similar to stretching or compression of a prismatic bar when acted upon by an external force . for illustration , we consider a simple rectangular block of length @xmath63 resting horizontally as shown in fig . such a block will be compressed by its own weight with body forces @xmath64 and @xmath65 , where @xmath66 is the mass density and @xmath44 is acceleration due to gravity . this problem can be solved by direct integration of field equations . to prevent the translation of the bar we fix the central point @xmath67 at the lower end of the bar , that is , @xmath68 and @xmath69 . further , the rotation of the bar about @xmath67 and about an axis parallel to @xmath67 is prevented by setting @xmath70 at point @xmath67 . final expressions for three displacement components can be written as : @xmath71\\ w & = & \nu \frac{\rho g\,z}{e}(l - y)\end{aligned}\ ] ] where @xmath66 is the material density , @xmath44 is the acceleration , @xmath55 is poisson s ratio , @xmath63 is dimension of the bar along the direction of the force . from eqs.(5)-(7 ) and fig . 3(a ) we note that points on the y - axis are only vertically compressed , i.e. , @xmath72/2e$ ] . for other points on the bar , the vertical contraction @xmath9 is accompanied by a finite stretching @xmath73 and @xmath74 in other two orthogonal directions . the coupling of the displacement field is mediated by non - zero value of poisson s ratio . the example of a prismatic bar illustrated in previous section has an analytical solution due to simple geometry . the actual structure of the fp is quite complex . it has two large cavities and four mirrors attached to the spacer . the dimensions as well as mass of the mirrors are also comparable to the spacer . using analytical means to find a closed - form solution for complicated geometry is extremely difficult task . we , therefore , used finite element analysis ( fea ) to solve our 3d problem numerically . the fea is by far the most widely used and versatile numerical technique for simulating deformable solids . of horizontal ( x - cut ) and vertical ( y - cut ) probe points . the displacement of any other point on the cavity mirrors is confined to the gray area between the curves . @xmath75 denotes the displacement of the mirror centers . ] we chose comsol multiphysics as our fea simulation software . it provides a complete and integrated environment for creating , analyzing and visualizing multiphysics models . built - in cad tools and the _ structural mechanics _ module was used for constructing a 3d model of the fp cavity . additional quantities and model inputs ( e.g. boundary constraints , body load , material type and properties etc . ) for stress and displacement analysis were defined in _ solid mechanics _ interfaces of the _ structural _ module . the relevant material properties used in this study are listed in table 1 . the mesh settings determine the resolution of the mesh element used for discretizing the model . in all simulations we used default tetrahedral mesh shape . extra - fine mesh size was chosen to minimize the discretization error . for our study , we selected the stationary solver since our load , stress and deformation do not vary with time . the _ solid mechanics _ kernel numerically solves the navier s equations ( eqs.(4 ) ) and computes displacements , stresses , and strains over the entire fp domain . the deformation of a material point ( @xmath76 ) in fp geometry is quantified by calculating the displacement ( @xmath77 ) from the undistorted shape . in subsequence analysis , our prime interest will be in displacement component @xmath78 which is along the axis of the cavity , i.e. @xmath21-direction . .mechanical and thermal properties of the materials used in simulations [ cols= " < , < , < , < , < " , ] the fp is mounted horizontally in @xmath79-plane as shown in fig . 1 . the cavity axis is along @xmath21-direction . since the cavity length is short we have not considered any discrete support for horizontal mounting . instead we let the spacer block rest on the bottom plane . first we investigate a simple case where the fp unit is deformed under its own weight , i.e. , force of gravity @xmath44 acting uniformly on entire geometry in @xmath80 direction . fixed boundary constrain @xmath81 is imposed on the bottom @xmath79-plane of the spacer to prevent rigid body motion in fea simulations . cavity deformation in @xmath82-plane due to gravity load is illustrated in fig . in accordance with eqs . ( 3)-(5 ) , the gravity - induced compression of the cavity in vertical direction also causes it to expand in @xmath21-direction , thus increasing the mirror separation . in order to numerically compute the cavity length changes , we extract the displacement component @xmath74 for each mirror surface that is bonded to the spacer block . the numerical probe points are defined along @xmath83 and @xmath22-direction ( see the fig . 4 inset ) on the interior surface of the mirrors forming cavities . the numerical displacement @xmath74 of the probe points , namely the @xmath83-cut and @xmath22-cut , is shown in fig . 4 . in simulations the origin of the coordinate system was chosen to coincides with the geometrical center of the fp spacer block . a near overlap between the fp1 and fp2 probe points indicate that both cavities are deformed identically under the force of gravity . it is also evident from fig . 4 that displacement @xmath74 is not uniform across the cavity cross - section . the horizonal bulge shown by concave shaped red curves , is symmetric about the mirror centers where the displacement is maximum ( @xmath84 m ) and reduces to about @xmath85 m towards to cavity edge . the displacement @xmath74 of the vertical probes , i.e. , the @xmath22-cut shown in blue , is asymmetric . it is minimum ( @xmath86 m ) at the cavity top edge , attains maxima ( @xmath87 m ) at about 3 mm below the mirror centers and reduces slightly towards the edges . this asymmetry in displacement is because of the unequal weight along vertical direction and a small @xmath22-tilt caused by outward buckling of the spacer and the mirrors . the @xmath21-displacement of any other point on the mirror surface is bound by gray area between the blue and red curve . since the displacement @xmath74 is not uniform across the reflecting face of the mirror we express the effective cavity length change @xmath88 for each cavity as the difference in displacement of the mirror centers . finally , the quantity of interest for us is differential length change between two fp cavities , i.e. @xmath89 . for the present case , we find @xmath90pm and @xmath42 . this is expected as the applied gravity load is symmetric and both cavities are deformed equally . with gravity - like acceleration vector @xmath91 oriented in ( a ) @xmath92-plane , ( b ) @xmath82-plane and ( c ) @xmath93-plane of the dual cavity . for each plane the angle @xmath94 and corresponding 2d projection of the fp is indicated in the inset figures . the orientation angle @xmath94 was swept from @xmath95 in step of @xmath96 . ( d ) the differential length change @xmath97 between the two fp cavities as a function of sweeping angle . the y - scale to the right gives corresponding rv tracking error ( cms@xmath2 ) in each case . ] now we examine cavity s response to a gravity - like static force that is orientation dependent . this exercise is aimed at understanding the relative sensitivity of the fp to direction dependent noise perturbations . in simulations , we swept the acceleration vector @xmath91 in different planes and computed differential length change @xmath98 for each orientation . the acceleration @xmath91 oriented in a given @xmath99-plane , can be expressed in its components along @xmath100 and @xmath101 directions , i.e. , @xmath102 , where for consistency we have chosen @xmath103 . using parameter sweep feature of the comsol we simulated the impact of different orientation of @xmath91 in the @xmath99-plane . cavity length change @xmath88 and @xmath98 as a function of orientation in different planes is shown in fig . the impact of orientation dependent force can be understood by analyzing its response on cavity length separately along each direction . a few noteworthy observations are : * when the acceleration is along the @xmath83-direction ( @xmath104[@xmath105 in fig . 5(a ) and @xmath106[@xmath107 in fig . 5(c ) ) the sheared strain in fp is produced in such a way that fp1 cavity mirrors are deflected inward [ outward ] and fp2 cavity mirrors are deflected outward [ inward ] . such response can be attributed to unequal shear strain produced because of the constrained boundary and geometrical asymmetry of the cavity in @xmath83-direction . a significant differential length change ( @xmath108pm ) occurs due to shortening of one cavity and elongation of the other . this amounts to spurious rv tracking errors about 140 cms@xmath2@xmath109 or equivalently a frequency shift of 1.8 mhz@xmath109 at 782 nm laser . * horizontally mounted dual fp is largely immune to relative length change in vertical vibrations . when the acceleration is oriented in @xmath22-direction ( @xmath106[@xmath107 in fig . 5(a ) and @xmath104[@xmath105 in fig . 5(b ) ) it imparts equal length changes ( @xmath110 pm ) to each cavity . since both cavities respond in same way , the relative length change between two cavities is nearly zero . therefore , a tracking mechanism would measure a correct drift caused by vertical acceleration . * when the acceleration is in @xmath21-direction ( @xmath106[@xmath107 in fig . 5(b ) and @xmath104[@xmath105 in fig . 5(c ) ) each mirror is displaced along the cavity axis by same amount without causing any relative drift i.e. , @xmath111 . and the middle mounting hole - pair is denoted by @xmath112 . the concentric through - holes ( dia @xmath113 1 mm ) are shown to provide clearance for the vertical suspension with wires . ] v from the proceeding analysis we conclude that proposed fp device is immune to differential length changes if the acceleration noise is oriented along the cavity length and in the vertical direction . however , the noise coupling in @xmath83-direction can induced non - zero differential length changes ( @xmath114 ) between two cavities . in such case , the instantaneous drift measured by laser - lock setup will not be same as the drift of astro - cavity . based on the type of noise some strategies can be adopted to overcome this limitation . for example , a typical noise spikes ( an impulse excitation ) only lasts for a fraction of second and temporarily deforms the elastic cavity to produce uncorrelated tracking error in drift measurement . after the noise pulse has elapsed the elastic cavity would return to its original shape and laser - lock would resume the correct measurements . in such case , the actual drift has to be estimated from a series of measurements by carefully discarding the erroneous data . if the source of noise is constant then setup has to be properly shielded . for example , the measurement error due to differential length changes in the proposed device can be reduced below cms@xmath2 level by attenuating the noise by @xmath115 . the vibration sensitivity of a cavity can be greatly reduced by mounting it vertically . in this scheme the cavity axis is aligned along gravity and fp is held close to the spacer mid - plane as illustrated in fig . the major advantage of this configuration comes from symmetric distribution of mass above and below the cavity mid - plane . the top and the bottom mirrors are equally displaced by force of gravity in downward direction thus resulting in near - zero change in cavity length . a mounting mechanism can be devised to hold the cavity at suitable supporting positions -also called airy - points . the cavity spacing can be made invariant by optimizing the size , location and depth of the supporting points . mm and the optimized hole depth was found to be hd=4.76 mm . ( d)-(f ) mirror displacement at different diameter @xmath112 of the central hole pair . in these simulations @xmath116 mm and hd=4.76 mm and the optimized diameter @xmath112 of the central hole pair was found to be 7.34 mm . ] three pairs of mounting holes are partially drilled to provide discrete support - points at the under - side of the rectangular spacer . these holes are drilled up to the cavity mid - plane as shown in fig . . the flat bottomed - surface of each hole provides a smooth contact area for cavity to rest on mounting posts from below . alternatively , the fp can be hanged in air by using suspension wires from the above . the retaining disks attached to each end of the wire and smoothly inserted into the mounting holes will wear the cavity weight uniformly . the concentric through - holes ( dia @xmath113 1 mm ) are only shown to provide necessary clearance for wires . the diameter of the mounting holes at the corners ( @xmath117 , see fig . 6(b ) ) is chosen to be 5 mm to maintain a sufficient wall thickness from edges . in vertical mounting airy - points are not located exactly at the geometric mid - plane of the cavity . this is because the compression in upper - half of the cavity is not exactly countered by stretching in lower - half . in addition there is a weight imbalance around the supporting points . the material removed to form mounting holes reduces the weight of the lower - half of the spacer block . in order to correctly locate vertical position of the airy points , we first varied the mounting hole depth ( hd ) in our simulations and determined the gravity induced displacement of top and bottom mirrors . the cavity length change is inferred from the displacement of probe points located on the inner surfaces of mirrors as discussed in previous section . which corresponds to fractional length change @xmath118 below @xmath119 . ] three test cases with different displacement trends are shown in fig . 7(a)-(c ) . when the supporting surface is below the cavity - mid plane , e.g. hd = 4 mm in fig . 7(a ) , the vertical displacement of the upper mirrors is higher as seen from corresponding @xmath83- and @xmath22-cuts . the situation is reversed in fig . 7(c ) where the supporting location hd = 6 mm was moved above the cavity - mid plane . the null point occurred at hd = 4.76 mm in fig . 7(b ) for which the displacement of top and bottom mirrors become nearly equal and the cavity length becomes invariant . figure 8 shows the rms separation between the top and the bottom cavity mirrors at various hole depths . the fractional length change at optimal hole - depth for both cavities is below @xmath120 . if we assume manufacturing tolerance of @xmath121 m , the rms difference still remains below @xmath122 pm , indicating the robustness of the design against machining errors . this tolerance limit is well within the current technology of precision manufacturing . in vertical mounting scheme with acceleration in ( a ) @xmath92-plane , ( b ) @xmath82-plane and ( c ) @xmath93-plane . coordinate axis are visually depicted in the inset images . for each case the orientation angle @xmath94 was swept from @xmath95 in step of @xmath96 . ( d ) the differential length change @xmath97 between two fp cavities as a function of orientation angle @xmath94 . the @xmath22-scale to the right gives corresponding radial velocity error ( cms@xmath2 ) . ] shape of the profile along @xmath83- and @xmath22-directions in fig . 7(a)-(c ) also indicates that displacement is not uniform across the cavity cross - section . a linear @xmath83-tilt is added onto a radially symmetric sag . this tilt is caused by bending of spacer in the middle that bears larger mass concentration . we enlarged the diameter @xmath112 to remove extra material for balancing and equalizing the weight distribution in middle part of the spacer . figure 7(e)-(f ) demonstrates how the tilt component ( x - cut ) changes sign when @xmath112 was varied from 6 mm to 9 mm . at optimum diameter ( @xmath123 mm in fig . 7 ( e ) ) , the tilt component was largely eliminated and the rms difference between @xmath83 and @xmath22-radial profile was reduced below @xmath124 m. having found an equilibrium plane for support , we also estimated the direction dependent noise sensitivity of the vertical mount . for fea simulations , the mechanical displacement of the mounting holes was constrained and the fp geometry was subjected to a sweeping acceleration vector in different planes . the calculated cavity length variations @xmath88 and differential length change @xmath98 is plotted in fig . 9 . for acceleration along @xmath83 , the vibration sensitivity to differential length change has reduced by a factor of seven compared to horizontally mounted case . for acceleration along @xmath22 and @xmath21-direction , the rv error is below 1 cms@xmath2g@xmath2 . since the vibration sensitivity is significantly reduced in vertical mounting scheme , the requirements for external noise isolation and damping are also relaxed . thermally induced material expansion or contraction is also a major source of noise against the dimensional stability of optical materials . a desired length stability of fp cavity can be assured by choosing appropriate spacer / mirror material with low thermal expansion coefficients . to minimize the impact of temperature fluctuations , the reference cavities are also operated near their zero - crossing temperature @xcite . the entire setup is enclosed inside an evacuated and thermally shielded environment . in this section we investigate effect of temperature fluctuations on our fp cavities . to first approximation , the cavity length changes in fp1 and fp2 will strongly correlate if temperature variations within the enclosure are uniform . in that case the tracking mechanism would expectedly register a correct drift . however , small temperature gradients inside the enclosure will expose two cavities to a locally different environment , producing dissimilar length changes in fp1 and fp2 . this would contaminate the tracking signal with intractable radial velocity errors . was varied from 0 mk to 100 mk . ] to analyze the temperature sensitivity of the fp we again used fea to solved the thermal - stress problem with comsol multiphysics . in our thermal model , realistic temperature gradients were simulated by imposing appropriate temperature conditions on selected boundaries of the spacer block . a steady - state numerical solution then establishes a linearly varying temperature and stress fields between the selected sides . temperature sensitivity also depends on the choice of the spacer and mirror material . in these studies , corning ule was used as cavity spacer while three glass - ceramics namely , zerodur , silica and corning ule were tested for mirror substrates . two worse case scenario producing unequal cavity length changes are shown in fig . 10 . the temperature sensitivities ( @xmath125 per mk ) of the device can be determined from the slopes of the plots . among the chosen spacer - mirror pairs , the ule - fs cavity has the largest sensitivity @xmath126 mk @xmath2 ( rv error @xmath113 3.81 cms @xmath2 mk @xmath2 ) while ule - ule has the least sensitivity @xmath127 mk @xmath2 ( rv error @xmath113 0.27 cms@xmath2mk@xmath2 ) for temperature gradient shown along spacer edge in fig . as see in fig . 10(b ) , the temperature sensitivity is almost halved for the gradient considered along diagonal direction . the contribution of cte to cavity length change from mirror substrate is substantially large compared to the contribution from spacer material . the main drawback of fused silica is its large cte which places stringent requirements on the design of thermal control system of the cavity . for example , to keep the rv error of dual cavity fp below 1 cms@xmath2 level , a temperature uniformity at sub - mk level has to be ensured . this requirement is relaxed to few - mk if both spacer - mirror pair are made with extremely low expansion materials such as zerodur and ule . a new method of building temperature insensitive fs cavity mirrors is expected improve their performance in precision metrology @xcite . despite large cte , fused silica is considered as best performing material for cavity mirrors . this is because fs has very small mechanical losses at room temperature which helps to lower cavity s thermal noise floor which is about 2 - 3 times better compared to the ule or zerodur @xcite . in ideal case of equal length cavities i.e @xmath128 , temperature fluctuations will not produce any relative length change between two cavities . in practice , it is not possible to manufacture cavities of equal length with arbitrary precision . a small offset in lengths may arise from machining errors and/or imperfections in the optical bonding between the mirrors and the spacer . to study this effect we created cavities with different length offset and calculated their differential expansion from the thermal model . the length offset was introduced by shortening the spacer thickness @xmath129 of fp1 without altering @xmath130 . in the fea simulations the temperature of the entire fp block was varied uniformly in step of 1 mk . figure 11 shows thermally induced differential length changes between two cavities for different length offsets . clearly , a 0.1 k difference from the temperature set point can produce @xmath131 pm differential expansion ( the green curve in fig . 11 ) between two cavities if the initial length offset was @xmath132 m . this amounts to @xmath133 cms@xmath2 rv tracking error . these calculations reinforces the need of high thermal stability ( few mk ) and cavity length matching better than @xmath121 m for keeping the influence of temperature changes on rv error below 1 cms@xmath2 . k to @xmath134 + 0.1 k in incremental step of 1 mk . the spacer material used in simulation is ule and mirror material is fused silica . various length offsets @xmath135 are given in @xmath58 m . ] in fea , the numerical solution of a problem is approximated by partitioning the cad geometry into smaller parts . the accuracy of the numerical solution is intrinsically linked to the mesh size . by refining the mesh , the solution becomes more accurate but this is achieved at the cost of finite computational resources and time . in a realistic fea model , the difference between exact and approximate solution should be minimized and error should not exceed the tolerance limit defined by some accepted criteria . we used the default tetrahedral mesh settings with adaptive mesh refinement in our fea software to discretize the cavity model . the refinement algorithm globally adjusts the mesh size to provide sufficient resolution around all regions of interest within the geometry . there is no generic rule to verify the numerical accuracy of fea model -especially in complex problems where exact solution is not known a priory . in order to evaluate the discretization error due to finite mesh size , we compare the relative displacement @xmath74 of fp1 and fp2 mirrors for a simple gravity loading case ( see fig . 4 ) already discussed in section 4.1 . from the application of uniform load and symmetry consideration , exact solution requires the cavity displacement in fp1 and fp2 to be equal . any difference in displacement can be attributed to numerical inaccuracies originating from meshing errors . figure 12 shows the displacement data ( @xmath83-cut ) plotted for input cavity mirrors at three different mesh densities . we note that compared to the smaller mesh sizes , the magnitude of the displacement is slightly underestimated at _ ` coarse mesh'_. in addition , a small relative difference can also be seen in fig . this difference reduces progressively with decreasing mesh size ( see fig . 12(b ) ) and becomes insignificant for ` _ _ extra - fine mesh _ _ ' shown in fig.12(c ) where two curves completely overlap . in each case the numerical error is estimated from the rms difference between the two curves . the discretization rms error for extra - fine mesh is @xmath136 m. this amounts to @xmath137 error for cavity length stability targeted at @xmath122 m level . all fea simulations in this paper were carried out with extra - fine mesh . another possible source of error is uncertainties in thermal and mechanical properties of the ceramic material listed in table 1 . typically , the material parameters can have about @xmath138 deviation from their nominal values . we varied the material parameters in our simulation by @xmath139 but did not find any noticeable difference in vibration or temperature sensitivities . the calculated deviation was of the same order as the discretization errors . this is not unexpected since our fp is lightweight and cavity length is short . for the same level of uncertainties in poisson ratio , youngs modulus and material density , a vibration sensitivity of @xmath140 was achieved in longer cavities @xcite . -data probes on the input mirrors of fp1 and fp2 for which the cavity displacement is numerically computed . the default element size ( min , max ) for each case in also indicated . the green curve ( @xmath22-scale to the right ) shows the difference between the two plots . the rms error for coarse , fine and extra - fine meshing is @xmath141 m , @xmath142 m and @xmath143 m , respectively . ] the doppler survey for many astronomy programmes requires high resolution spectrograph with exceptional long term stability in rv precision . in this paper we have proposed a double cavity fp design for aiding high precision doppler measurements in astronomy . the proposed device has two parallel optical cavities embedded inside a 10 mm thick monolithic ule block . a low - finesse _ astro - cavity _ is meant to generate multiline reference spectra for spectrograph calibration while a high finesse _ lock - cavity _ is needed to provide steep discriminant signal for tracking the cavity drift . external noise can have severe detrimental effect on the performance and reliability of the tracking measurements . an accurate measurement of drift is possible only if the noise interventions are common to both cavities . we have used fea to analyze fp s sensitivity to mechanical vibrations and temperature fluctuations . a 3d fea simulations were carried out for the fp structures comprising five domains ( 4 mirrors + 1 spacer ) . an optimized cavity design with proper mounting and support system helps insulate the noise . two mounting geometries were investigated for tracking errors resulting from differential length changes between the cavities . the horizontal geometry showed vibrational sensitivity up to @xmath144@xmath109 ( rv errors @xmath145 cms@xmath2 ) for acceleration directed along the longer edge ( @xmath83-axis ) of the block . the noise along @xmath22- and @xmath21-direction is common and does not contribute to differential length change . for achieving 1 cms@xmath2 rv stability of the device the horizontal mounting configuration may require a well thought strategy to insulate the vibration noise down to @xmath146 . cavity lengths can be made invariant by mounting the fp vertically . we worked out a vibration insensitive design by numerically optimizing the hole - depth and diameter of the mounting holes . in the optimized design the tilt - component was eliminated and a clear null point was achieved for gravity induced displacement between the mirrors . we also studied the impact of thermal instability and choice of mirror / spacer material on the tracking performance . a steady - state thermal analysis was carried out to evaluate the effects of thermal gradient . numerical simulations show that temperature gradient within the fp enclosure has to be accurately controlled and relative length offset between two cavities should be minimized to avoid differential length changes that may cause spurious drift and unintended tracking errors . the proposed dual fp device can be easily fabricated with available machining precision of @xmath147 m and readily achievable thermal stability of few mk . this work is also relevant to cavities used in laboratory experiments designed to test lorentz invariance where a pair of orthogonal fp cavities is employed to detect anisotropy in the speed of light @xcite . the entire optical setup is put on a rotating table . a frequencies beat - note between two lasers , each locked to one of the two orthogonal cavities , is examined while the entire setup is continuously rotated on the turntable . any possible anisotropy is expected to modulate the laser beat signal at twice the rotation rate . however , the discrimination of the anisotropic signal from competing noise artifacts is not trivial . even though the length of two cavities is defined by a single ule block , detection ambiguities may arise , e.g. , from periodic deformation of cavity length caused by gravitational and centrifugal forces . the approach outlined in our studies might be useful to understand and model various systematics producing relative drift in cross - cavities . rkb acknowledges the research support for this project ( reference no . emr/2014/000941 ) from science and engineering research board ( serb ) , department of science and technology ( dst ) , india . a.r . acknowledges the research support from european research council under the fp7 starting grant agreement number 279347 and funding from dfg grant re 1664/9 - 1 . huke , p. , holzhuter , h. , & reiners , a. [ 2015 ] `` transfering rb+ hyperfine - structure stability to a fabry - perot resonator used as a frequency standard for astronomical spectrographs , '' in _ proc . spie _ * 9526*,98260i-7 . probst , r. a. , lo curto , g. , avila , g. , _ et al . _ [ 2014 ] `` a laser frequency comb featuring sub - cm / s precision for routine operation on harps , '' in _ spie astronomical telescopes & instrumentation , _ pp . 91471c-91471c . schfer , s. , & reiners , a. [ 2012 ] `` two fabry - perot interferometers for high precision wavelength calibration in the near - infrared , '' in _ spie astronomical telescopes & instrumentation _ , pp . 844694 - 844694 .

it has been suggested by sen @xcite that the energy gap between unstable d - brane configurations and the stable vacuum should be computable using witten s cubic string field theory ( sft ) @xcite . the level truncation scheme of @xcite has appeared to lead to very good agreement with the expected results in the context of the decay of bosonic d - branes @xcite and unstable d - branes in superstring field theory @xcite . lower dimensional d - branes have been constructed as tachyonic lump configurations on bosonic d25-brane @xcite . support for sen s conjecture that lower dimensional d - branes can be identified with tachyonic lump solution of string field theory on bosonic d25-brane also has come from the noncommutative limit of the effective field theory @xcite . various toy models of tachyon dynamics have been useful tool for understanding the realization of sen s conjectures @xcite . despite the success of cubic sft in leading to very good agreement with the expected results , it is still not clear why the calculations in the level - truncated string field theory converge so rapidly to the correct answer for quantities like vacuum energy . it is also not clear how to study the nonperturbative vacuum using this approach . an alternative method has been suggested in @xcite which says that renormalization group ( rg ) analysis of worldsheet theory in first quantized approach can be used to show that the mass of the tachyon lump on a dp - brane is equal to that of a d(p1)-brane . this leads to the idea that boundary string field theory ( sft ) as was originally proposed by witten and shatashvili @xcite may efficiently describe open string tachyon condensation on d - branes in bosonic string theory @xcite ( also see @xcite . it has been pointed out in @xcite that boundary sft can provide an exact verification of sen s conjectures . based on @xcite , one can compute the action exactly taking a simplest tachyon profile as boundary operator of ghost number zero which is quadratic in the space - time coordinate . one can also describe the lumps corresponding to the lower dimensional d - branes and calculate their tension that agrees with the expected result exactly . one strong point of the results of @xcite using level truncation scheme in cubic sft is that it gives a definite picture of the tachyon profile as superposition of @xmath2 for different @xmath3 with definite coefficients producing soliton of finite width . on the other hand the boundary sft analysis of @xcite does not say how the higher tachyon harmonics get mixed with the cosines of different @xmath3 to produce finite size soliton profile . the reason for this is that the mass parameter flows to infinity in the ir in their particular choice of coordinates starting with the simple choice of the tachyon profile initially . this is because the rg equations in that particular choice of coordinates become linear in coupling constants . since the width of the soliton is given by inverse of the mass parameter which flows to infinity in the exact description , the width is zero . also it is difficult to see in this setup how the theory at ir fixed point can describe configurations of more than one d - brane . we should note that although the theory flows to infinity in the ir , the zamolodchikov metric on the space of worldsheet theories measures the distance between the uv and ir fixed point to be finite . in this paper we will try to address these issues in the worldsheet approach by choosing the initial tachyon profile on a circle on bosonic d25-brane world - volume . the basic setup is similar to @xcite . the choice of boundary perturbation , as we will discuss in the next section , is motivated from the analysis of @xcite . in our choice of coordinates in the space of worldsheet field theories the rg analysis provides the boundary conformal field theory to flow to a nearby ir fixed point . we will consider only the first three tachyon modes . the values they flow to after rg flow appears to be in good agreement with the values computed in @xcite of tachyon modes of the string field at the stationary point of the potential . the plan of the paper is the following . in section 2 we briefly review the worldsheet rg scheme in the context of boundary sft . we comment on the boundary entropy that measures the corresponding d - brane tension describing the boundary cft . we highlight the basic setup of @xcite that leads to a particular choice of the tachyon profile . in section 3 we perform the rg analysis in detail choosing a particular ansatz for the green s function of fast moving modes of the scalar field . the rg equations are obtained up to third order in coupling constants . the easy but tedious parts of the calculations are given in the appendix . in section 4 we plot the lump profile that appears from the rg analysis of the previous section . following the method of affleck and ludwig @xcite we calculate the boundary entropy in the leading order . our result satisfies the @xmath4-theorem of @xcite . the boundary entropy is calculated to an accuracy of about @xmath0 of the exact result . in section 5 we analyse the multicritical behaviour of the potential in the ir and argue that the theory in the ir is that of a configuration of two d24-branes . we make an analogy with the kondo problem and argue that the process of the formation of the lump due to tachyon condensation corresponds to the underscreened kondo effect . exact screening occurs when the theory on the brane rolls down to the nonperturbative closed string vacuum . overscreening , on the other hand leads to a picture similar to dielectric effect . we comment on @xmath1 symmetry restoration . section 6 contains discussions on further related issues that are beyond the scope of the paper . if the spectrum of a point in the moduli space has relevant operators in the ir , the point is said to be unstable . the unstable point then might be discarded from the moduli space . in other words , this results in the appearance of unstable directions in the effective potential . ir instability indicates that we are in a _ false _ vacuum . the obvious question will be which point ( or vacuum ) will replace this unstable point , or in other words , how to resolve the ir instability . often regions in moduli space contain d - brane configurations related by t - duality . the above issue can be addressed in these sectors by adding some ir relevant boundary perturbations . one important feature of this boundary deformation is that the bulk theory always remain conformal . flows caused by a relevant boundary operator appear as open string tachyon condensation . flowing to ir on the worldsheet is equivalent to approaching a classical solution of spacetime theory . as a result of the flow , at some points in moduli space , boundary conditions are changed from neumann to dirichlet . the reverse process indicates nonunitarity . the direction of the flow is determined by _ boundary entropy _ defined by @xcite g_a = 0|a , the disk partition function of the boundary state @xmath5 associated to the perturbed theory . the phases of @xmath6 and @xmath5 can be chosen such that the above quantity is real and positive for any boundary state . it is shown in @xcite in first order in conformal perturbation theory that boundary entropy , @xmath4 decreases along rg flows , suggesting the so called @xmath4-_theorem _ similar to zamolodchikov s @xmath7-theorem @xcite . in this context , the case of sine - gordon boundary perturbation with single frequency is studied in @xcite . in fact @xmath4 measures the tension of the d - brane that describes boundary conformal field theory at the corresponding fixed point @xcite ( see also @xcite ) and the @xmath4-theorem implies the minimization of the action in the space of _ all two - dimensioanal worldsheet field theories_. the worldsheet formulation of string field theory is _ manifestly _ background independent depends on the closed string background . ] . in @xcite a gauge invariant background independent spacetime string field theory action is defined as the solution of the following equation : = _ 0 ^ 2_0 ^ 2 _ i()\{q , o( ) } _ , [ seqn ] where @xmath8 is the brst charge and and @xmath9 is the boundary parameter of the disk . the operator @xmath10 has ghost number one : @xmath11 , where @xmath12 is a general boundary perturbation describing the space of boundary perturbations : @xmath13 . the normalization constant @xmath14 is fixed in @xcite by comparing the on - shell three tachyon amplitude computed from the background independent open string field theory with the same computed in chern - simons string field theory : k = t_p , [ norm ] @xmath15 being the corresponding dp - brane tension . the correlation function in the above expression is defined in terms of the perturbed worldsheet action on the disk . the above action is defined up to an additive constant . although the boundary perturbation @xmath12 does not depend on ghosts , the theory with perturbed action is not renormalizable . but in the case of tachyon condensation , it is renormalizable . so boundary string field theory works well for studying tachyon condensation problem . but it is not clear how to define a general off - shell amplitude due to uv divergences on the worldsheet . an alternative definition of s is given by the metric on the space of worldsheet theories @xcite = -^jg_ij , [ s - g ] where @xmath16 is the corresponding beta function which acts like a vector field on the space of worldsheet theories . in @xcite a very important relation between the action @xmath17 and the partition function on the disk is demonstrated up to second order in coupling constant : _ = -^i+z ( ) . [ s - z ] since on - shell the beta function vanishes , the action is same as the partition function . this fixes the additive ambiguity in ( [ seqn ] ) . the above relation implies that all symmetries of the worldsheet partition partition function are also symmetries of the spacetime action . it is further argued that @xcite the spacetime action @xmath17 is nothing but the boundary entropy @xmath4 . although background independence is manifest in the worldsheet formalism , it is lost once we compute @xmath17 or @xmath4 perturbatively . but if the above relation between the spacetime action and the partition function is true in all orders in coupling constant , it is not lost . on the other hand , the chern - simons string field theory is not manifestly background independent . to achieve this one is forced to work in a truncated version of hilbert space of the first quantized theory restricting the string field to a background independent subspace for studying the classical lump solution . in our renormalization group analysis of the problem of formation of the tachyonic lump , we follow the basic setup of @xcite where the @xmath18 coordinate is taken to be on a circle of radius @xmath19 . we take @xmath20 to be the scalar field on the string worldsheet associated with @xmath18 . on a d25-brane @xmath20 satisfies neumann boundary conditions . the conformal field theory associated with the field @xmath20 has central charge , @xmath21 . before going into the rg analysis let us briefly discuss the setup of @xcite . for details on the background independent subspace of the complete fock space see @xcite . a basis of states of this theory is conveniently formed by grouping the states into verma modules . the states contained in the verma module are obtained out of primary state @xmath22 by acting with the associated virasoro generators @xmath23 . linear independence of the states to form the basis is achieved by removing null states from the spectrum . in fact the spectrum is free of null states for @xmath24 and an irrational @xmath19 value . in order to achieve a successful truncation of the hilbert space one has to restrict the primary states of the boundary conformal field theory along @xmath18 to be even under @xmath25 and to be trivial primaries of the conformal field theory of the fields associated to the rest of the coordinates of bosonic string theory . as a result the appropriate primary states are : ( 1 ) the zero momentum primaries that are even under @xmath25 ( also removing the null states ) , and ( 2 ) the vacuum state @xmath26 with @xmath27 . let us recall that an open bosonic string propagating in the tachyon background is described by the following action s = dsdt ^ab_a__b^+_-^ dk t(k ) e^ik , where @xmath28 is the uv cutoff and @xmath29 is the tachyon field with momentum @xmath30 . in the present case where @xmath18 is compact , the momentum is descrete and we are motivated to take the boundary perturbation to be of the form ( respecting @xmath25 symmetry ) : dt t[(t)]=_n=0^_ndt ( ( 0,t ) ) , where @xmath31s are tachyon modes . the zero mode @xmath32 is the identity operator and is always tachyonic . the higher modes may not be tachyonic depending on the radius @xmath19 . in our analysis we only consider the first three tachyon modes ( _ i.e. _ @xmath33 ) . hence the problem is similar to the boundary sine - gordon model with two frequencies , which typically allows to hit a nearby ir fixed point producing a lump profile of finite size . the boundary conformal field theory can be described by a gaussian model with gapless excitation spectrum and the correlation functions of bosonic exponents follow power laws . this behaviour implies that the correlation length is infinite and the system is in its critical phase . under boundary perturbations , usually the correlation functions are affected differently on different scales and the long distance asymptotics get affected the most . certain perturbations may cause only tiny changes in the uv , but changes the ir behaviour profoundly . in the rg picture this is observed as a growth of the coupling constant associated with the perturbation . a slow decay of correlation functions gives rise to divergences in the perturbation series . if the influence of the perturbing operator grows on large scales , the perturbation is _ relevant_. consider the perturbed action s = s_0 + dt o_(t ) where @xmath34 is the action of the system at criticality and @xmath35 is the perturbing field with scaling dimension @xmath36 , @xmath37 . the perturbation with zero conformal spin and scaling dimension @xmath36 is relevant if @xmath38 and _ irrelevant _ if @xmath39 . the case @xmath40 is the _ marginal _ one and its effect on scaling dimensions of correlation functions depends on the sign of the coupling constant @xmath41 . consider the path integral on the disk with a worldsheet action s = s_0 + dt t[(t ) ] [ totalaction ] where @xmath34 is the free field action on the disk describing an open and closed string conformal background . the perturbing boundary field @xmath42 $ ] is a tachyon profile and @xmath20 is the scalar field along the string worldsheet associated with the compactified coordinate @xmath18 . the above action describes a renormalization group flow from a theory where all 26 bosonic directions are neumann describing a d25-brane to the theory where @xmath18 direction is dirichlet describing a d24-brane . instead of using the callan - symanzik formalism we follow the wilsonian renormalization scheme where we put the theory on a lattice . we assume that the fourier transform of the field , @xmath43 , is defined in the brillouin zone and choose a cut - off , @xmath44 . the aim is to start moving towards larger distances ( or lower energies ) by integrating out the fields with shorter and shorter wavelengths . the procedure is based on the decomposition of the boundary field with a cutoff into a slow moving ( long - wavelength ) and a fast moving ( short - moving ) component , @xmath45 , where we have split the brillouin zone @xmath44 into a wide region @xmath46 and a narrow slice @xmath47 . the original field is given by _ ( t ) & = & e^it _ ( ) + & = & _ ||< d2 e^it _ s ( ) + _ <|| < d2 e^it _ f ( ) . the next step will be to perform a partial path integration in the partition function over the fast moving part and representing the result in terms of an effective model for the slow moving field . if the model is renormalizable , which is known to be the case for sine - gordon model that we will consider below , the effective action will have the same structure as the original one , but with a new set of coupling constants . this procedure is repeated several times , and after each step the form of the original model is reproduced ( upto irrelevant terms ) . the relations between the bare and the renormalized couplings then lead to renormalization group ( rg ) equations . the gaussian part @xmath34 is additive under the above decomposition . hence , given the cutoff , the partition function is given by z_&= & d_sd_f e^-s_0[_s ] e^-s_0[_f]e^-s_i[_s+_f ] + & = & z_fd_s e^-s_0[_s ] e^-s_i[_s+_f]_f where @xmath48 is a nonsingular contribution of fast moving components to the partition function . the effective action involving the slow moving part of the field is given by s_eff[_s]=s_0[_s]+s_i eff[_s ] where s_i eff[_s ] = -e^-s_i[_s+_f]_f it is clear that the effective action preserves all ir singularities . assuming the coupling to be small , we expand the above effective interaction perturbatively upto cubic order s_i eff[_s ] & = & s_i _ f+1 2(s_i_f^2-s_i^2 _ f ) + & & + ( 1 6s_i^3 - 1 2s_i_f s_i^2 _ f + 1 3s_i_f^3 ) + o(^4 ) [ sieff ] now we consider the effect of switching on first three tachyon harmonics : @xmath49 , and @xmath50 . the scalar perturbation is of _ double sine - gordon _ type and reads symmetry of the coupling @xmath31s in writing down the perturbation which might be more appropriate in order to compare the results with that of @xcite . however we chose perturbation of the type ( [ double - sine - gordon ] ) as this allows us to hit the desired multicritical ir fixed point from the stable direction . however we will invoke the @xmath51 symmetry to show multicriticality in section 5 . the opposite sign in front of the identity operator does not modify the analysis for other operators . the coupling in front of the identity operator appears linearly in the beta function and it does not mix with other couplings . also as we will see in the next section , boundary entropy does not get a contribution from it . ] s_i [ _ ] = _ 0dt * 1 * - _ 1dt _ ( t ) - _ 2dt 2_(t ) . [ double - sine - gordon ] where @xmath52 defines the conformal dimensions of our theory and the scaling dimension is @xmath53 for the first harmonic and @xmath54 for the second harmonic . the tachyon zero mode is the identity operator with zero scaling dimension . rescaling @xmath52 : @xmath55 results in the relation @xmath56 . following @xcite , we take the radius @xmath57 . in this radius the first mode is least relevant and the perturbation by the second mode is least irrelevant , which , after being added to the cutoff theory , should improve the rg result . such a procedure should lead to a good result even when only first few irrelevant terms are included . the effects of highly irrelevant operators are highly damped by their rapid decay . using the above definition of the effective action we perturbatively calculate contributions in each order in coupling constant , @xmath41 . then in order to restore the original cut - off to @xmath58 we rescale the energy @xmath59 and the time @xmath60 so that the effective action is of the same form as the bare one but with a renormalized strength of coupling . now we calculate the expression ( [ sieff ] ) term by term . 0.5 cm : in first order in couplings , it is given by s_i_f & = & _ 0dt _ f - _ 1dt _ f + & & - _ 2dt 2_f . [ order1double ] the expression for correlation function of bosonic exponents reads _ n e^i_n(t_n)=e^-_m > n _ m_n g(t_m , t_n)e^-1 2_m _ n^2 g(t_m , t_m ) . [ exp - cor ] the terms containing green s functions of coinciding arguments are singular in the continuous limit . but in our regularized theory they are finite . the green s function has the following well known form g(t_m , t_n ) = 1 4 ln(r^2 ) , [ gf ] where @xmath61 and @xmath28 are ir and uv cut - off respectively . substituting into ( [ exp - cor ] ) we get the result for the correlation function of bosonic exponents : _ n e^i_n ( t_n)= _ m > n ( 1+(t_m - t_n)^2 a^2)^_m_n 4 ( r a)^-(_n_n)^2 4 . [ exp - cor1 ] using the above formulae we obtain e^i_f(t)_f = e^-^2 2_f(0)^2_f = e^-^2 2g_f(0,0 ) , where -^2 2g(0,0 ) = -^2 2_0^df ( ) = -^2 4 ln , [ gengf ] and -^2 2g_f(0,0 ) = -^2 2_-d^df ( ) = di ( ) = -^2 4(d ) = -^2 4dl . [ fastgf ] hence ( [ order1double ] ) turns out to be s_i_f & = & _ 0 * 1 * -_1(1-dl ) _ s(t ) + & & - _ 2(1-dl ) 2_s(t ) + o(dl^2 ) , [ order1doublefinal ] where @xmath62 are small dimensionless couplings . 0.5 cm : 0.5 cm we now turn to the quadratic contribution to ( [ sieff ] ) . imposing translational invariance it can be written as & & 1 2(s_i_f^2-s_i^2_f ) + & = & dt_1 dt_2 \ { e^-^2 g_f(0,0 ) ( 1-e^-^2 g_f(t_2-t_2 ) ) + & & + e^-^2 g_f(0,0 ) ( 1-e^^2 g_f(t_2-t_2 ) ) } + & & + dt_1 dt_2 \ { e^-4 ^ 2 g_f(0,0 ) ( 1-e^-4 ^ 2 g_f(t_2-t_2 ) ) 2 + & & + e^-4 ^ 2 g_f(0,0 ) ( 1-e^4 ^ 2 g_f(t_2-t_2 ) ) 2 } + & & + e^-dl dt_1dt_2 \{(1-e^-2 ^ 2g_f(t_1-t_2 ) ) + & & + \{(1-e^2 ^ 2g_f(t_1-t_2 ) ) } . [ order2double1 ] to evaluate the correlation functions of the bosonic vertices we have used ( [ exp - cor ] ) . to evaluate the green s function for fast moving modes we use the following scheme . instead of considering just the momentum shell for fast moving modes , if we consider all momenta upto cutoff @xmath58 then using ( [ exp - cor1 ] ) we get e^i(t_1 ) e^i(t_2)= ( 1+(t_1-t_2)^2 a^2)^^2 4 ^ -^2 , which , like ( [ gengf ] ) , leads to -^2 g(0,0)-^2 g(t_1-t_2 ) = ^2 4 . hence the analogous treatment for fast moving modes follows by performing the integration in the narrow slice only . using the expression for @xmath63 given by ( [ fastgf ] ) we arrive at the following useful expression for fast moving components -^2 g_f(t_1-t_2 ) -^2 2dl(ln||+1)+o(dl^2 ) , [ fastgf1 ] where we have assumed that @xmath64 . it is clear that g_f(t_1-t_2)=f(r ) dl+o(dl^2 ) , where @xmath65 . if we adopt a sharp momentum cut - off prescription , @xmath66 will be bessel function of order zero , @xmath67 , which has a long oscillating tail and does not fall off rapidly on increasing its argument . however , as was shown in @xcite , in a smooth cut - off procedure @xmath66 is truly short - ranged , essentially nonzero at @xmath68 . this can be seen in ( [ order2double1 ] ) , the functions like @xmath69 are also short - ranged . this allows us to introduce the center - of - mass coordinate @xmath70 and relative coordinate @xmath71 and expand the cosines of ( [ order2double1 ] ) in @xmath72 . the detail calculation is shown in the appendix . here we only mention some of the necessary facts . for example , consider the cosine @xmath73 \approx 1-{\beta^2r^2 \over 2}(\partial_{\tilde r}\phi_{s\lambda'}(\tilde r))^2 $ ] , where the first term contributes to renormalization of the free energy . in our case that generates the rg contribution to identity operator . on the other hand , the second ( gradient)@xmath74 term is an irrelevant term with a factor proportional to the uv cut - off @xmath28 in front of the renormalized coupling . here we see a striking difference between bulk and boundary rg flows , where in the latter case the gradient term is responsible for renormalization of the constant @xmath52 . collecting the @xmath75 , @xmath76 , and @xmath77 contributions from the appendix , we arrive at the following complete second order expression & & ( s_i_f^2-s_i^2_f ) + & = & _ 1 ^ 2dl 2_s ( t ) -_1 ^ 2dl * 1 * + & & + _ 2 ^ 2dl 4_s ( t ) -_2 ^ 2dl * 1 * + & & + 2 ^ 2dl ( 3_s(r ) -_s(r ) ) . [ order2doublefinal ] 0.5 cm : 0.5 cm the calculation of the cubic contribution to ( [ sieff ] ) is similar and the details are given in the appendix . here we give the result only : & & 1 6s_i^3_f-1 2s_i_fs_i^2_f + 1 3s_i_f^3 + & = & _ 1 ^ 3()dl 3_s ( t ) + _ 1 ^ 3 ( ) dl _ s ( t ) + & & + _ 2 ^ 3()dl 6_s ( t ) + _ 2 ^ 3 ( ) dl 2_s ( t ) . [ order3doublefinal ] collecting all the results up to third order from ( [ order1doublefinal ] ) , ( [ order2doublefinal ] ) and ( [ order3doublefinal ] ) we express the effective boundary action as s_i eff[_s ] & = & _ 0 * 1 * -_1(1-dl ) _ s(t ) -_2(1-dl ) 2_s(t ) + & & + _ 1 ^ 2dl 2_s ( t ) -_1 ^ 2dl * 1 * + & & + _ 2 ^ 2dl 4_s ( t ) -_2 ^ 2dl * 1 * + & & + 2 ^ 2dl 3_s(r ) -2 ^ 2dl _ s(r ) + & & _ 1 ^ 3()dl 3_s ( t ) + _ 1 ^ 3 ( ) dl _ s ( t ) + & & + _ 2 ^ 3()dl 6_s ( t ) + _ 2 ^ 3 ( ) dl 2_s ( t ) . [ sieffdouble(semifinal ) ] in order to restore the original cut - off to @xmath58 we rescale the energy @xmath59 . in order to keep the product @xmath78 intact , we have to rescale time in the opposite way , @xmath60 . the effective action is of the same form as the bare one but with a renormalized strength of coupling . hence neglecting the @xmath79 terms , s_i eff [ _ ] & = & _ 0(1+dl ) * 1 * -_1(1+(1-)dl ) _ s(t ) + & & -_2(1+(1-)dl ) 2_s(t ) + & & + _ 1 ^ 2dl 2_s ( t ) -_1 ^ 2dl * 1 * + & & + _ 2 ^ 2dl 4_s ( t ) -_2 ^ 2dl * 1 * + & & + 2 ^ 2dl 3_s(r ) -2 ^ 2dl _ s(r ) + & & + _ 1 ^ 3()dl 3_s ( t ) + _ 1 ^ 3 ( ) dl _ s ( t ) + & & + _ 2 ^ 3()dl 6_s ( t ) + _ 2 ^ 3 ( ) dl 2_s ( t ) . + & & [ sieffdouble(final ) ] note that although initially we considered perturbation with tachyonic zero , first and second modes , the non - linear rg flow equations force the coefficients of various other boundary fields with higher harmonics to evolve . we see from above expression that the third and fourth harmonics evolve at second order and the third and sixth harmonics evolve at third order . also notice that there is no mixing between @xmath80 and @xmath50 at third order in the coupling . the beta functions can be extracted from the above expression and are given by _ 0 & = & = _ 0 -_1 ^ 2-_2 ^ 2 , + _ 1 & = & = -(1-)_1 -_1_2 + _ 1 ^ 3 , + _ 2 & = & = -(1-)_2 + _ 1 ^ 2 + _ 2 ^ 3 . [ betafn2 ] now we take @xmath57 , where the perturbation with @xmath80 and @xmath50 becomes least relevant and irrelevant respectively . perturbation with the identity operator ( associated with @xmath32 ) in the rg picture just adds a constant to the action . typically the coefficient of the least relevant operator grows at the fastest rate at the beginning , and drives the system towards the fixed point . the final shape of the soliton is determined by where this fixed point is . rescaling of @xmath81 in ( [ betafn ] ) results in _ 0 & = & = _ 0 -_1 ^ 2-_2 ^ 2 , + _ 1 & = & = -_1 -_1_2 + _ 1 ^ 3 , + _ 2 & = & = _ 2 + _ 1 ^ 2 + _ 2 ^ 3 . [ betafn1 ] typically single a sine - gordon boundary perturbation does nt lead to a nearby fixed point and the theory typically flows to infinity . typically @xmath41 flows all the way from 0 to @xmath82 under renormalization . the boundary conditions look like neumann at very high energy ( uv ) but like dirichlet at low energy ( ir ) . the field @xmath20 satisfy neumann boundary conditions close to the boundary , but feels dirichlet boundary conditions instead far from it . but as in the next section we will see that sine - gordon model with two frequencies we considered above has a nearby fixed point . the situation is very analogous to string field theory . suppose we are trying to construct the lump solution on a circle of radius @xmath19 . in this case we need to start with an initial tachyon field for which the coefficient of @xmath83 is non - zero , and then use the equations of motion of string field theory iteratively to improve the solution . this generates all higher harmonics as well as the tachyon zero momentum mode as we saw in our rg analysis as well . but if from the beginning we introduce the tachyon zero momentum mode in the initial configuration , then the iterative process drives the solution towards the vacuum solution where the tachyon is constant instead of the one lump solution . in the rg picture perturbations with the identity operator are generally regarded as uninteresting , as they just add a constant to the action . also , in the peturbative definition via correlation functions , the zero mode has no effect . in this section we will calculate the boundary entropy perturbatively in leading order and compare the result with its exact value . we follow the method of @xcite . let us recall the rg equations ( [ betafn2 ] ) obtained in the last section ( up to second order in coupling constants ) _ 0 & = & = _ 0 -2(1-y_1)_1 ^ 2 - 8(1-y_1)_2 ^ 2 , + _ 1 & = & = -y_1_1 -8(1-y_1)_1_2 , + _ 2 & = & = -y_2_2 + 2(1-y_1)_1 ^ 2 , [ betafn ] where @xmath84 is the scaling dimension of the corresponding harmonic . the above rg equations satisfies right @xmath85-expansion behaviour according to @xcite . the first step will be to solve these equations for the bare couplings , @xmath86 , @xmath87 , and @xmath88 , as functions of renormalized couplings at the scale set by @xmath61 , @xmath89 , @xmath90 , and @xmath91 . the pfaffian differential equation obtained from the above set of equations is ( d_0+d_2 ) + [ _ 0 - 8(1-y_1)_2 ^ 2-(4y_1 - 3)_2]d_1 = 0 , [ pfaffian ] which is needed in order to satisfy the integrability condition given by & & [ y_1_1 + 8(1-y_1)_1_2][24(1-y_1 ) _ 2 + 5y_1 - 3]+8(1-y_1)_1 [ _ 0 - 8(1-y_1)_2 ^ 2-(4y_1 - 3 ) _ 2 ] + & & + [ y_1_1 + 8(1-y_1)_1_2][1-y_1 - 8(1-y_1 ) _ 2 ] = 0 . [ integrability ] for @xmath92 , we have the following relation between @xmath86 and @xmath88 64(1-y_1)^2_2 ^ 2 + 8(1-y_1)(1 + 2y_1)_2 + 8(1-y_1 ) _ 0 - 2y_1(1 - 2y_1)=0 . [ reln1 ] the branch containing the real solution of @xmath93 satisfies the following , 8(1-y_1)_0 + 2y_1(2y_1 - 1 ) ( 2y_1 + 1)^2/4 , which implies @xmath94 for @xmath57 , concistent with our results for the desired ir fixed point discussed below . combining @xmath95 and @xmath96 in ( [ betafn ] ) into ( d_0+d_2)/[_0 - 8(1-y_1 ) _ 2 ^ 2-(4y_1 - 3)_2 ] = dl , and using the relation ( [ reln1 ] ) , we get 8(1-y_1)d_2/[8(1-y_1)_2 + 2y_1 - 1]=dl , for @xmath97 . integrating the above expression from @xmath28 to @xmath61 , we obtain the following expression for the bare coupling @xmath88 = -_2^ * , [ bare2 ] where @xmath98 and @xmath99 is the scaling dimension of the second tachyon harmonic . for @xmath57 , the perturbation with the second harmonic is least irrelevant and is added to improve the shape of the profile of the lump at ir fixed point . being irrelavant the perturbing operator with @xmath88 decays out . this can be seen by considering the perturbation expansion of the two - point correlation function of bosonic exponents . the integral appearing in the first nonvanishing correction to this correlation function converges at large distances for an irrelevant operator . this implies that the perturbation expansion does not contain ir singularities ; so if the bare coupling constant is small , its effect will remain small and will not be amplified in the process of renormalization . in the limit the renormalization scale , @xmath100 , _ 2 = _ 2(r ) - ( -1 ) ( ) _ 2^ * -0.125 , which is very close to the result of @xcite for the second harmonic at the stable minimum of the tachyon potential . on the other hand , the other two operators which are relevant for @xmath57 evolve to the values : @xmath101 , which are also very close to the results obtained by the level truncation technique @xcite . hence the lump profile is given by t(x_25)=0.25 - 0.35(x_25 ) -0.125(x_25 ) as a function of @xmath18 . the profile is shown in figure 1 . the profile has finite size in contrast to the result of @xcite . before returning to the calculation of boundary entropy , notice that in the rg equations ( [ betafn ] ) in the limit @xmath102 , the linear terms in @xmath103 and @xmath96 agree . the reason for these to disagree with @xmath95 is that we chose a positive sign in front of @xmath32 in the beginning . the rg equations up to linear order are @xmath104 , where the sign in front of the universal linear term depends on the sign of perturbation . rg analysis with the identity operator is trivial and it does not mix with other operators . also as we will discuss soon its contribution goes to ground state energy correction only , not to boundary entropy . next we calculate the bare @xmath86 and @xmath87 in terms of the renormalized couplings . in order to obtain the expression for bare coupling , @xmath86 in terms of the renormalized couplings we use ( [ reln1 ] ) setting the scale at uv cutoff : _ 0 ( a ) = - . inserting ( [ bare2 ] ) for the bare coupling , @xmath88 , we get the following expression for @xmath86 = - . [ bare0 ] inserting the above expression for bare coupling , @xmath86 into the pfaffian equation ( [ pfaffian ] ) we get = - , which , assuming @xmath97 , simplifies to ( ) d[8(1-y_1)_2 ] = - . integrating both sides we obtain 8(1-y_1)_2+(1-y_1 ) ln[8(1-y_1)_2 + 2y_1 - 1 ] = -ln _ 1+ln c , [ indefinite ] where the integration constant determines the correct trajectory that passes through uv and desired ir fixed points . one might wonder since many different theories flow to the same fixed point , that we can not write the bare couplings in terms of renormalized couplings . in fact , here as we are starting with the correct set of rg equations which inherently have the right uv fixed point as a trivial solution so over an infinitesimal segment from the uv fixed point we are automatically on the right trajectory . what we need to worry about is that as we integrate those equations we do nt deviate from the right trajectory and miss the ir fixed point . we reach the desired fixed point by determining the correct value of the integration constant . to determine the constant that allows us to reach the desired _ nearby _ fixed point as the end point of the flow we substitute the couplings with their fixed point values _ 1^*=- , _ 2^*=- , into ( [ indefinite ] ) resulting in the following expression for the bare coupling , @xmath87 & = & e^-(1-y_1)()^1-y_1 , [ bare1 ] where we have used ( [ bare2 ] ) . having solved the renormalization group equations for the bare couplings as functions of renormalized couplings , given by ( [ bare0 ] ) , ( [ bare1 ] ) , and ( [ bare2 ] ) we turn to the calculation for boundary entropy which measures the tension of the corresponding d - brane and test the @xmath4-_conjecture _ , which in turn is related to the minimisation of the action in the space of open string fields . the boundary has its own free energy proportional to the size of the system which diverges linearly with scale @xmath105 in thermodynamic limit . the boundary free energy gets another contribution independent of scale which counts the boundary degrees of freedom , measured by @xmath106 , f_bl =- ln z = f_bl - ln g. we first perturbatively expand the partition function @xmath107 to @xmath108 . we arrive at a uv expansion of the form which schematically looks like ln z = _ n c_n(_ir^y_i)^n = f_b^ + ln g+^-ve , where @xmath109 is the scaling dimension . we discard terms linear in negative powers of scale and nonuniversal uv divergent terms which are linear in scale as they correspond to ground state energy corrections . the remaining terms have weak dependence on the scale set by @xmath61 , which acts as an ir cut - off , as we can absorb them approximately in terms of the renormalized coupling @xmath110 using the relations ( [ bare0 ] ) , ( [ bare1 ] ) , and ( [ bare2 ] ) obtained by solving simultaneous @xmath52-function equations . thus in the @xmath100 limit ( when @xmath111 ) , these terms give the contribution to the boundary entropy . note that the weak ir cut - off dependence is consistent with the renormalization group . to cubic order in coupling the partition function is given by & & 1 + dt_1dt_2 t _ j(t_2 ) + & & + dt_1dt_2dt_3 t _ i(t_1 ) _ j(t_2 ) _ k(t_3 ) . since @xmath112 , this vanishes in the thermodynamic limit , @xmath113 . here @xmath114 is the partition function at uv fixed point . any @xmath3-point correlation function of sine - gordon perturbation , the electro - nonneutral ( @xmath115 in ( [ exp - cor1 ] ) ) part has an overall factor of @xmath116 that can not be compensated using definition of renormalized coupling @xmath110 ; therefore in above perturbative expansion contributions come from electro - neutral parts only such as @xmath75 , @xmath76 , and @xmath117 . clearly there is no contribution from the identity operator of any order as its contribution can be considered to be going to ground state energy correction . the perturbing operator has two point function given by _ i(t_2 ) = ( ) ^2(1-y_i ) . so the quadratic contribution to @xmath118 turns out to be ( considering half - cylinder geometry ) z_2 & = & _ i=1,2 a^-2y_i _ i^2 + & = & _ i=1,2 a^-2y_i _ -r/2^r/2 , [ z2a ] where @xmath119 is the relative coordinate and @xmath120 . the above integration is regularized by the cut - off @xmath121 . mapping the circle to infinite line : @xmath122 the above integration becomes z_2 = _ i=1,2 _ i^2()^2y_i _ |u|<a / r . [ z2b ] integration by parts results in z_2 & = & _ i=1,2 _ i^2()^2y_i 2 + & = & _ i=1,2 _ i^2()^2y_i 2 , [ z2c ] where @xmath123 is the euler beta function . the first term is linear in @xmath61 and does not contribute to boundary entropy . we will formally perform the computation as an expansion in @xmath124 supposing @xmath124 to be small and at the end we will set @xmath125 corresponding to @xmath57 . it is not _ a priori _ obvious that such @xmath124 expansion will converge for finite value of @xmath124 . however , as we will see , it does quite well in our case giving nice agreement with exact value of the ratio @xmath126 . the situation is similar to computation of critical exponents in three dimensions where @xmath85-expansion for @xmath127 dimensions converges for @xmath128 and gives nice agreement with the experimental data . we note that we can reach nearby fixed point at the end of the rg flow with @xmath129 indicating that @xmath124 is small enough . now in the small @xmath124 limit @xmath130 becomes z_2 -^2 , [ z2final ] where we have dropped the @xmath131 part which vanishes in the @xmath100 limit as it has an overall factor of @xmath132 with @xmath133 negative as can be seen by substituting ( [ bare2 ] ) in ( [ z2c ] ) . the cubic part is given by z_3 = dt_1dt_2dt_3e^i(t_1)e^i(t_2 ) e^-2i(t_3)[z3a ] which on half - cylinder turns out to be z_3 = a^-2y_1a^-y_2d_1d_2d_3 . [ z3b ] now introducing the relative coordinates @xmath134 and @xmath135 and center - of - mass coordinate @xmath136 the above integral simplifies to z_3 & = & a^-2y_1a^-y_2r_|u_i|>a / r du_1 du_2 ( ) ^2 + & & ( ) ^2(1-y_1 ) ( ) ^-2(1-y_2 ) + & & | ( ) /|^2(1-y_1 ) , [ z3c ] where @xmath137 with a regularization @xmath138 , @xmath139 . the above expression can be written in a more convenient form given by z_3 = ^2_|u_i|>a / r du_1 du_2 . [ z3d ] let us first concentrate on terms of @xmath140 $ ] and the ground state energy contribution from terms of @xmath141 . to achieve that we make a change of variables : @xmath142 . the above integral becomes z_3 = ^2_|u_1|>u_0(v)du_1dv . [ z3e ] where @xmath143 is the new cut - off . integration by parts with respect to @xmath144 gives z_3 & = & ^2_-^dv + & & . [ z3f ] from the surface term we will get contribution of @xmath145 $ ] which has no effect on boundary entropy . the rest is uv finite and we can remove the cut - off . however there will be a contribution of @xmath146 $ ] coming from @xmath147 $ ] term in the denominator . to extract this contribution we perform the following calculation . we first introduce step functions into the integral ( [ z3d ] ) i ( ) = _ -^du_1du_2 ( u_1 ^ 2-^2)(u_2 ^ 2-^2 ) , where @xmath148 is the cut - off . it is convenient to differentiate the integral with respect to @xmath85 : _ -^ ( u_1 ^ 2-^2 ) . the integral is identical to ( [ z2b ] ) with the same cut - off . so using previous result obtained for @xmath130 we get = 2[^-(1 - 2y_1)-y_1 ] , that gives i ( ) = ( ) ^4 - 6y_1 -()^-y_2 + [ i ] the second part gives the desired @xmath146 $ ] behaviour that has weak cut - off dependence . but the first part is of no interest to us since it is the ground state energy contribution from @xmath149 . the constant part comes from evaluating the integral ( [ z3f ] ) throwing away the surface term . writing the integral in terms of old variables @xmath150 we get & & - ^2 + & & _ -^ ( + ) + & & = - ^2_-^ ( ) . + & & we notice that in small @xmath124 limit , while performing the @xmath151 integration , the @xmath124 factor in front is cancelled by a divergence from @xmath152 : 2_a / r^ - ( ) ^4y_1 - 3 . hence in small @xmath124 limit , the constant part of ( [ i ] ) becomes & & - ^2()^4y_1 - 3_-^ ( ) + & & - ^2()^-y_2 . so ignoring higher orders in @xmath124 the final expression for @xmath118 contributing to boundary entropy is given by = 1-y_1 ^ 2 -^2()^-y_2 . having already obtained the expression for the bare couplings in terms of the renormalized couplings as given by ( [ bare2 ] ) and ( [ bare1 ] ) we can rewrite the above result in terms of renormalized couplings only . as @xmath100 , @xmath153 . hence the desired ratio between ground state entropy at ir and uv fixed points turns out to be ( in the leading order in @xmath124 ) r_p = _ r 1 - 0.50088964 y_1 ^ 2 . notice that the change in @xmath4 is negative , implying decrease of @xmath4 under flow between uv to ir point . also the ratio becomes unity for exactly marginal case , _ i.e. _ where @xmath154 , giving a line of fixed points . on the other hand the exact result is given by @xcite r_e = . for @xmath155 , perturbative result is @xmath156 compared to the exact result , @xmath157 ; so our result is within @xmath158 . on the other hand for @xmath159 , _ i.e. _ when @xmath124 is very small and the perturbation by the first tachyon harmonic is nearly marginal , the perturbative result is @xmath160 compared to the exact result , @xmath161 ; the perturbative result is now obtained to an accuracy of @xmath0 of the exact result . in the effective landau - ginzburg description of two dimensional conformal field theory , = d^2x ( ( ) ^2+v ( ) ) , [ lg ] @xmath20 being the order parameter for some physical system , the extrema of the general polynomial interaction correspond to various critical phases of the system . many systems possess a @xmath25 symmetry with an even polynomial interaction @xmath163 . for a polynomial @xmath164 of degree @xmath165 , this ensures existence of @xmath166 minima separated by @xmath167 maxima . several critical phases can coexist if the corresponding extrema coincide . hence the most critical potential is a monomial in @xmath20 and the @xmath166 critical behaviour of the theory is given by the interaction : _ int = gd^2x ^2(m-1 ) . [ mono ] by comparing the structure of the operator algebra of the above bulk critical theory with that of the unitary diagonal minimal model @xmath168 , characterized by central charge @xmath169 for @xmath170 zamolodchikov @xcite has shown that each @xmath166 multicritical behaviour of the theory ( [ lg ] ) is nothing but a minimal model , @xmath168 . generalizing this concept to boundary conformal field theory we consider the boundary potential of the form v ( ) = _ k g_k ^2k(0 ) . the bulk theory is still the @xmath21 ( _ i.e. _ @xmath171 ) minimal model throughout the flow . near the uv fixed point , as we turn on the boundary interaction ( which can be considered as even polynomial interaction of order infinity and hence an indication of underlying infinite number of critical phases of the system ) the system flows to the ir near which the system has finite multicritical behaviour . the strength of multicriticality ( _ i.e. _ how many critical phases can coexist ) depends on number of couplings involved , _ i.e. _ on dimensionality of the space of coupling constants . in other words , expanding the sine - gordon polynomial about @xmath172 near the ir fixed point , the potential has the following form : v ( ) = ( _ n _ n^ * ) -(_n _ n^*)(0)^2 + ( _ n _ n^*)(0)^4- ( -1)^k_k g_k^*(0)^2k . now using the results from previous section , @xmath173 , so one can consider @xmath174 for a large number of couplings turned on . for the perturbation with the first three tachyon harmonics ( [ double - sine - gordon ] ) , _ i.e. _ @xmath175 , we observe that for @xmath176 , only @xmath177 , and all other @xmath178 indicating @xmath179 and hence the existence of two critical phases of the system at the ir fixed point . also for @xmath19 close to one ( _ i.e. _ the self - dual radius ) , @xmath180 , @xmath181 showing the two phases again . as we turn on more and more couplings , we will see that lower order coefficients of @xmath164 become zero , but some higher order coefficient becomes nonzero producing a shift in the degree of the polynomial effective interaction indicating the presence of more and more critical phases . for @xmath182 the effective interaction is basically a monomial of very large order . we can expect that by turning on all possible couplings ( _ i.e. _ infinite number of tachyon modes ) we can probe the @xmath82-multicritical behaviour of the system . note that the values of coupling constants @xmath183 are meaningful only in the context of a particular renormalization scheme and when the composite fields @xmath184 are defined . however the multicritical behaviour observed is rg scheme independent @xcite . for the effective polynomial interaction @xmath185 , the @xmath186-critical behaviour ( _ i.e. _ presence of @xmath186 degenerate minima ) of the potential is confined to hypersurfaces @xmath187 , @xmath188 of codimension @xmath189 , _ i.e. _ of dimension @xmath190 . the @xmath191 dimensional boundary @xmath192 of this hypersurface @xmath187 is critical . whereas the form of the hypersurfaces @xmath187 and @xmath192 essentially depends on the regularization scheme of the theory , the @xmath186-critical behaviour on the hypersurface @xmath192 is universal and depends only on @xmath186 . for the sine - gordon potential ( essentially an infinite degree polynomial interaction ) , @xmath193-critical behaviours are possible and our observation is that by turning on full set of coupling constants we can fully explore the @xmath82-critical behaviour at ir . in string theory language , the @xmath186-critical behaviour in the ir is nothing but a free @xmath21 boundary conformal field theory of @xmath186-coalescing d24-branes with @xmath194 chan - paton factor . note that in order to probe the @xmath82-critical behaviour ( in other words @xmath162 symmetry of the nonperturbative bosonic vacuum reached asymptotically from the core of the d24-brane ) we need to turn on all tachyonic modes even though higher modes , being highly irrelevant , have no significant effect in perturbative rg analysis . when we perturb by the zeroth mode only ( where @xmath195 ) we end up with the closed string vacuum with no d - brane ( _ i.e. _ @xmath196 ) . at this point we can draw an interesting analogy from kondo physics . the kondo model describes the interaction between @xmath30-degenerate bands of spin-@xmath197 conduction electrons and a quantum impurity of arbitrary size @xmath198 placed at one boundary , say at origin , which essentially makes the system nontranslationally invariant . this type of problem is similar to the open string theory in the sense that boundary spin is like a dynamic degrees of freedom or chan - paton factor . the ir ( or low - temperature ) behaviour is quite different depending on the relative size of @xmath199 and @xmath30 . for the _ underscreened case _ ( _ i.e. _ @xmath200 ) the rg flow leads to ir fixed point with fermi liquid behaviour . in the course of rg flow the boundary spin fuses with internal spin of the system and the electron channel screens or swallows up a part of the impurity spin giving rise to a similar current algebra as that of the uv fixed point with a shifted kac - moody current , _ n=_n+ , leading to rearrangement in the conformal towers of the theory . , the ir fixed point is given by fusion with spin-@xmath201 priamry . the fusion rules are @xcite @xmath202 . each conformal tower is mapped into a unique conformal tower giving rise to free fermion spectrum with a @xmath203 phase shift . whereas ir fixed point in overscreened case is given by fusion with spin - s primary . ] the fact that the ir fixed point for the underscreened case is stable can be seen from the interaction between the partially screened spin and the electrons on the neighbouring lattice sites , which is ferromagnetic and hence irrelevant . the strong coupling fixed point is much the same as the uv or zero coupling fixed point . only the size of the impurity spin is reduced and there is a change in boundary condition on the otherwise free fermions ( corresponding to a @xmath203 phase shift ) . this is precisely what happens in the present context of tachyon condensation which leads to a change of boundary condition from neumann to dirichlet . the leftover impurity spin @xmath204 at ir decouples from the internal symmetry group and is similar to the @xmath205 chan - paton factor of @xmath30 d24-branes at ir . for _ exact screening _ , @xmath206 , the internal spin fully swallows up the boundary spin giving rise to a translationally invariant system which is analogous to translationally invariant nonperturbative closed string vacuum with no d - brane . for the _ overscreened _ case ( _ i.e. _ @xmath207 ) , the ir fixed point is very nontrivial and is determined by fusion with the spin-@xmath198 primary . the theory arrives at non - fermi liquid fixed point in the ir which describes completely different physics . the strong coupling ir fixed point ground state has an overscreened spin of size @xmath208 . the induced interaction is antiferromagnetic and hence unstable . the existence of such nontrivial fixed point can be proved in the large-@xmath30 limit ( keeping @xmath198 fixed ) . the ground state can not be described by a simple physical picture . in @xcite the overscreened case is described based on hamiltonian non - abelian bosonization . the idea is to represent the @xmath209 species of electrons in terms of three independent bosonic fields : a free scalar carrying @xmath210 charge , a @xmath211 wzw non - linear @xmath212-model field carring @xmath213 spin , and a @xmath214 wzw field carring @xmath215 flavour degrees of freedom . the kondo interaction involves only the spin current . the degrees of freedom at the ir fixed point are described by an @xmath216 kac - moody invariant cft . both boundary scaling dimensions and fermi - liquid bulk scaling dimensions are described by a @xmath217 kac - moody boundary cft . there are restrictions on the combinations of charge , spin , flavour degrees of freedom . unlike the underscreened case , here fermion exponents are not recovered and are generally not half - integer . they come from more general combinations of three types of bosonic fields . the fermion exponents are sums of @xmath210 , @xmath214 and @xmath211 kac - moody current algebra exponents . they are not free and in fact _ bound _ together in combinations to form fermion composites . the analogue of overscreened kondo physics in string theory is the case when one considers string theory on the @xmath213 group manifold that can be described by a world - sheet wzw action and d - branes are stabilized against shrinking by quantized @xmath210 flux as discussed by bachas , douglas and schweigert @xcite ( also see @xcite ) . they considered a static d2-brane wrapping an @xmath218 parametrized by @xmath219 breaking @xmath220 to the diagonal @xmath221 . with non - zero background @xmath222 field the gauge invariant quantity is @xmath223 . if @xmath224 , where @xmath3 is the magnetic monopole number appearing due to the world - volume flux quantization @xmath225 , the d2 brane is prevented from shrinking to one of the poles of @xmath226 @xcite . inside this range the energy of the system has a unique minimum away from the poles of @xmath226 . in the large-@xmath30 limit , that minimum energy reduces to the mass of @xmath3 d - particles . the stable configuration leads in the dual picture to a bound state of @xmath3 d - particles on @xmath218 similar to the case described in @xcite . from the above physical picture ( of the underscreening and exact screening kondo effect ) one can try to see the mechanism of generation of local @xmath194 chan - paton factors of @xmath186 coincident d24-branes as a result of breaking of inherent stringy @xmath162 symmetry due to fusion of boundary spin or dynamical degrees of freedom with internal symmetry . the local operator @xmath227 near the ir fixed point breaks @xmath162 down to @xmath228 in the core of the soliton . while there is a local @xmath194 on @xmath186 coincident d24-branes , @xmath229 is again swallowed up by the internal symmetry of the system . as @xmath230 ( the exact screening case ) , the situation is similar to perturbation by the tachyon zero mode only , which leads the theory to the nonperturbative closed string vacuum and the local @xmath210 symmetry on the d25-brane is restored to the full @xmath162 symmetry . for @xmath231 ( when perturbations with all @xmath232s are triggered ) we get full @xmath162 symmetry corresponding to infinite number of d24-branes in the core of the soliton . symmetry restoration in the same context for noncommutative tachyon solitons is discussed in @xcite . it can also be seen whether the operator @xmath227 can give rise to @xmath233 copies of identity operator at ir fixed point , which are basically @xmath233 generators of @xmath194 as all operators flow to identity operators or derivative of identity operators at ir fixed point . so from the multicritical behaviour discussed above , four copies of the identity operator are left , indicating the final configuration to be that of two d24-branes . there are many unexplored questions related to the material covered in this paper . we mention some of them below along with some remarks . it would be interesting to understand the precise relationship between boundary string field theory and chern - simons string field theory . in other words , how the action in boundary sft is related to the chern - simons action in cubic sft . it is pointed out in @xcite that the boundary sft action described in certain coordinates on the space of coupling constants must be related to the cubic action described in a particular choice of coordinates on the space of string fields . the two choices of coordinates are related by some complicated singular transformation . in @xcite the boundary conformal field theory is perturbed by a simple tachyon profile with mass parameter @xmath234 that flows from zero in the uv to infinity in the ir . the lump profile has width @xmath235 which vanishes in the ir . in contrast , the level truncation scheme in cubic sft gives solitons with finite width . on the other hand , in our scheme one gets the soliton profile with finite width situated in a nearby ir fixed point . the set of values the first three tachyon harmonics evolve to in order to hit the desired ir fixed point is very close to the one obtained in the level truncation scheme in cubic sft @xcite . also the perturbative result of the boundary entropy is in good agreement with the exact result . it might be more appropriate to choose the signs of all the couplings initially to be the same or by a symmetry to be of the same sign for the zero and second modes and of opposite sign for the first mode . the former choice does not make the analysis different - we have already mentioned that the zero mode is the identity operator which appears in the rg equations only in linear order and does not contribute to the boundary entropy ( which is the only physical quantity of interest here ) . the latter choice ( that involves opposite signs for first and second modes ) makes the calculation of the boundary entropy more complicated and it does not fit in well with the setup used in @xcite . while the relationship between the our setup and the chern - simons sft is not clear , finding the reason for this ( apparent ) agreement of our result with that of @xcite is left for future work . there is a striking similarity between the beta function equations ( [ betafn2 ] ) in the first quantized theory and the string field equations of motion given below ( in the notation of @xcite ) : 2 & = & 2t_0 - 2k^3t_0 ^ 2-k^3-t_1 ^ 2-k^3-t_2 ^ 2= 0 , + 2 & = & -(1-)t_1 + 2k^3-t_0t_1 + k^3-t_1t_2=0 , + 2 & = & -(1-)t_2 + 2k^3-t_0t_2 + k^3-t_1 ^ 2 = 0 , [ sfteqn ] where @xmath236 is the inverse of the mapping radius of the punctured disks defining the three string vertex . it is maximal for the vertex of the cubic potential @xcite . the main difference is in the zero mode dependence . in the rg case the identity operator adds a constant to the action . in the rg equation the corresponding coupling constant appears in the universal linear order only . presumably there are definite complicated nonlinear relations between the different tachyon harmonics generating term proportional to @xmath237 and the cross terms @xmath238 , @xmath239 in rg equations . on the other hand since @xmath240 is dimensionless , the mapping radius @xmath241 should be somehow related to the cut - off @xmath242 in boundary conformal field theory side . in order to get a concrete relationship between our analysis with cubic string field theory we do need to develop a concrete proposal for how we can define an off - shell string field theory using our variables ; this requires choosing an appropriate uv regulator that makes the equations of motion for all the coupling constants unambiguous and free of divergence . once this is done , one can then interprete the fixed points that we have found as ( approximate ) solutions of the equations of motion of this specific string field theory . in the absence of such a field theory it is not clear what these solutions correspond to . these are of course the rg fixed points in a specific renormalization scheme , but this is a first quantized viewpoint as opposed to a second quantized one . cubic sft clearly provides such a formulation - the equations of motion involving arbitrarily high level fields do not suffer from uv divergence . the background independent sft does not provide such a formulation for all fields , but can be solved exactly for a special class of configurations ( corresponding to operators quadratic in @xmath243 ) and hence we can consider a subsector of the theory where only these operators are switched on . in this way it avoids having to do with non - renormalizable field theories . in your formulation we have switched a finite subset of the operators in an approximation scheme , but this in higher order would induce operators of arbitrarily high dimensions and so one needs to understand how to deal with non - renormalizable field theories . one way to avoid this might be to use the techniques of integrable perturbations developed by ghoshal and zamolodchikov @xcite . we have not raised issues concerning the shape and size of the lump representing the lower dimensional d - brane . as was pointed out in @xcite , in order to get some insight into this issue one needs to find the energy density of the object . since cubic sft is nonlocal , it is presently not clear how to address this . also the size might be an artifact of the particular gauge chosen . the shape and size depends on a particular definition of off - shell string field , and can vary from one sft to another . the important fact is that in chern - simons sft everything is smooth and non - singular , so it gives a good definition of the off - shell string field . there is no invariant definition of size of the d - brane unless one looks at the coupling to metric or some closed string background . presumably one needs to follow the analysis analogous to the bulk conformal perturbation theory in the presence of a fluctuating metric ( see for example @xcite ) . it is pointed out in @xcite that the dynamics of the metric makes sense only when there is a scale in the path integral , for example by fixing the area . this is done by the so - called _ gravitational dressing _ of the microscopic matter operator by the appropriate liouville dependence and inserting it in the path integral . an analogous treatment in boundary perturbation theory might give some useful insight regarding the shape and size of the d - brane . an other important issue is to extend the scheme used in this paper to superstring theories where lower dimensional d - branes are identified with a tachyonic kink solution rather than a lump solution . in the supersymmetric setup it is more convenient and sensible to study a configuration of parallel d - branes . also the size and shape of the d - branes in the ir can be investigated considering physical picture involving closed string scattering off the d - branes @xcite . it would also be of interest to study intersecting d - brane configurations using the boundary rg analysis discussed in this paper . we are grateful to michael douglas and michael green for many interesting discussions and critical reading of the manuscript . we thank david kutasov and ashoke sen for very useful discussions , correspondence and important remarks on a preliminary draft . we thank alexander zamolodchikov for a fruitful discussion on boundary entropy . sd is grateful to damtp for its hospitality in the early stages of this work . td is grateful to rutgers new high energy theory center for its hospitality in the final stages of this work . we acknowledge partial support from the pparc spg programme , `` string theory and realistic field theory '' , ppa / g / s/1998/00613 . td is supported by a scholarship of trinity college , cambridge . in this appendix we will present the details of the rg calculation in order two and three . ( s_i_f^2-s_i^2_f ) & = & - dr dr + [ order2a ] where we have expanded the cosines in @xmath72 . the ellipsis represent the @xmath245 ( which is identical to the @xmath244 part under replacement of @xmath52 by @xmath246 ) and @xmath247 contributions . following the arguments given in section 2 , we reach to the following expression & & ( s_i_f^2-s_i^2_f ) + & & = _ 1_2 ^ 2(1-dl)dr g_f(r)dr ( + & & -)+ + & & = _ 1_2 ^ 2(1-dl)dr g_f(r)dr + , where in the last step we expanded the cosines in @xmath72 about @xmath250 . by the same argument applied to @xmath244 part we can neglect the ( gradient)@xmath74 terms . as the field @xmath20 is odd under @xmath249 reflection , the overall sine terms are not invariant under this operation and hence do not contribute . hence using ( [ fastgf1 ] ) and the nonuniversal constants resulting from the integrals ( s_i_f^2-s_i^2_f ) = 2 ^ 2dl ( 3_s(r ) -_s(r))+ , [ order2double - mixed ] neglecting the @xmath79 terms . typically the integrals in ( [ i2i3 ] ) are divergent . but we have regulated them excluding the origin which is sensible in present case . & & 1 6s_i^3_f -1 2s_i_f s_i^2_f+1 3s_i_f^3 + & = & -dt_1dt_2dt_3 \ { ( dl ln|| ) + & & + ( dl ( ln|| + 4 ) ) + & & + ( dl ( ln|| -2 ) ) + & & + ( dl ( ln|| -2 ) ) } + , + & & [ order3double1 ] neglecting the @xmath79 terms . the ellipsis represent the @xmath251 , @xmath252 , and @xmath253 contributions . the @xmath251 and @xmath253 are related to @xmath254 and @xmath252 by exchanging @xmath255 respectively . we again introduce center - of - mass coordinate @xmath256 and relative coordinates @xmath257 and @xmath258 . the relative coordinates being bounded within the uv cut - off @xmath259 , the green s functions are truly short - ranged and we can expand the cosines in @xmath260 . the result is & & 1 6s_i^3_f-1 2s_i_f s_i^2_f+1 3s_i_f^3 + & = & -_1 ^ 3dl + . + & & [ order3c ] since under reflection of @xmath261 the integrand containing sine is odd , it vanishes identically . the @xmath261 integration containing @xmath262 vanishes imposing equation of motion from the action where the bulk degrees of freedom have been integrated out . hence the expression for @xmath254 cubic contribution is & & s_i^3_f-1 2s_i_f s_i^2_f+1 3s_i_f^3 + & = & _ 1 ^ 2_2(1-dl ) dt_1dt_2dt_3 + + & & as in the previous case we introduce the center - of - mass coordinate @xmath263 and relative coordinates @xmath264 and @xmath265 . expanding the cosines in @xmath260 , the above expression can be written approximately as dldr_1dr_2 ( ln r_1-ln r_2 ) d|r ( 4_s(|r)-*1 ) which vanishes identically after integration over @xmath266 and @xmath267 . by doing similar analysis for @xmath253 , we conclude that all the contributions to @xmath252 and @xmath253 vanishes . * s. ghoshal and a. zamolodchikov , _ boundary s - matrix and boundary state in two - dimensional integrable quantum field theoory _ a9 * 3841 ( 1994 ) ; erratum - ibid . * a9 * 4353 ( 1994 ) ; hep - th/9306002 .

because of its nonperturbative nature , studying low energy hadron physics in qcd is a long standing difficult problem . for low lying pseudoscalar mesons , a widely used approach is the theory of the effective chiral lagrangian based on the consideration of the global symmetry of the system and the momentum expansion without dealing with the nonperturbative dynamics of qcd @xcite@xcite . in the chiral lagrangian approach , the coefficients in the lagrangian are all unknown phenomenological parameters which should be determined by experimental inputs . the number of unknown parameters increases rapidly with the increase of the precision in the momentum expansion . therefore studying the relation between the chiral langrangian and the fundamental principles of qcd will not only be theoretically interesting for a deeper understanding of the chiral lagrangian , but will also be helpful for reducing the number of unknown parameters and increasing the predictive power of the chiral lagrangian . in a previous paper , ref.@xcite , certain techniques were developed , with which the chiral lagrangian was formally derived from the first principles of qcd without taking approximations . the chiral lagrangian coefficients are contributed both by the anomaly part ( from the quark functional measure ) and the normal part ( from the qcd lagrangian ) . in ref.@xcite , all the chiral lagrangian coefficients contributed from the normal part of the theory are expressed in terms of certain green s functions in qcd , which can be regarded as exact qcd definitions of the chiral lagrangian coefficients . after expanding the effective action in powers of the rotated sources ( momentum expansion ) , the effective action , up to @xmath2 , contributed from the normal part is of the form @xcite @xmath3 ^ 2 -{\cal k}_2(d^{\mu}a_{\omega}^{\nu}-d^{\nu}a_{\omega}^{\mu } ) ( d_{\mu}a_{\omega,\nu}-d_{\nu}a_{\omega,\mu } ) \nonumber\\ & & \hspace{0.5 cm } + { \cal k}_3[a_{\omega}^2]^2 + { \cal k}_4a_{\omega}^{\mu}a_{\omega}^{\nu}a_{\omega,\mu}a_{\omega,\nu } + { \cal k}_5a_\omega^2{\rm tr}_f[a_{\omega}^2 ] + { \cal k}_6a_{\omega}^{\mu}a_{\omega}^{\nu } { \rm tr}_f[a_{\omega,\mu}a_{\omega,\nu}]+{\cal k}_7s_{\omega}^2 \nonumber\\ & & \hspace{0.5 cm } + { \cal k}_8s_{\omega}{\rm tr}_f[s_{\omega } ] + { \cal k}_9p_{\omega}^2 + { \cal k}_{10}p_{\omega}{\rm tr}_f[p_{\omega } ] + { \cal k}_{11}s_{\omega}a_{\omega}^2+{\cal k}_{12}s_{\omega}{\rm tr}_f [ a_{\omega}^2 ] -{\cal k}_{13}v_{\omega}^{\mu\nu}v_{\omega,\mu\nu } \nonumber\\ & & \hspace{0.5 cm } + i{\cal k}_{14}v_{\omega}^{\mu\nu}a_{\omega,\mu}a_{\omega,\nu } + { \cal k}_{15}p_{\omega}d_{\mu}a^{\mu}_{\omega}\bigg ] + o(p^6 ) , \label{p4}\end{aligned}\ ] ] where @xmath4 is related to the nonlinearly realized meson field @xmath5 by @xmath6 ; @xmath7 , and @xmath8 are , respectively , the external scalar , pseudoscalar , vector , and axial - vector sources rotated by @xmath4 ; and the @xmath9s are terms with different lorentz structures in the relevant qcd green s functions . the obtained expressions for the chiral lagrangian coefficients are @xmath10 : @xmath11[\bar{\psi}^b(x)\gamma_\mu\gamma_5\psi^a(x)]\rangle -\frac{1}{n_f}\langle [ \bar{\psi}^a(0)\gamma^\mu\gamma_5 \psi^a(0 ) ] [ \bar{\psi}^b(x)\gamma_\mu\gamma_5\psi^b(x)]\rangle\nonumber\\ & & \hspace{0.8cm}-\langle \bar{\psi}^a(0)\gamma^\mu\gamma_5\psi^b(0)\rangle \langle\bar{\psi}^b(x)\gamma_\mu\gamma_5\psi^a(x)\rangle + \frac{1}{n_f}\langle \bar{\psi}^a(0)\gamma^\mu\gamma_5\psi^a(0)\rangle \langle\bar{\psi}^b(x)\gamma_\mu\gamma_5\psi^b(x)\rangle\bigg ] , \nonumber\\ & & f_0 ^ 2b_0=-\frac{1}{n_f}\langle \bar{\psi}\psi\rangle , \label{f_0b_0}\end{aligned}\ ] ] in the literature , the anomaly part contributions are usually calculated by means of the heat kernel regularization technique @xcite . however , this technique is difficult to implement in the calculation of the normal part contributions which contain complicated functions of the momentum , say the quark self - energy @xmath1 , reflecting nonperturbative qcd dynamics ( which are even unspecified in the analytical part of the calculation ) . in order to treat the anomaly part and the normal part contributions on equal footing , a certain new regularization technique feasible for the calculations of both parts should be developed . in this paper , we use the generalized schwinger proper time regularization technique developed in ref.@xcite to regularize the system , which keeps the local chiral symmetry at every step in the calculation , and can be applied to the calculations of both the contributions from the anomaly part and from the normal part . thus , in this paper , the contributions from the anomaly part and the normal part are calculated by means of the same technique . as the first conclusion of this unified treatment , we show that _ the anomaly contributions to the chiral lagrangian coefficients given in ref.@xcite , which are independent of qcd interactions , will actually be cancelled by certain terms in the normal part contributions , and the final expressions for the coefficients concern only the remaining terms from the normal part contributions related to qcd interactions_. it should be so since the coefficients indicate meson interactions which should be residual interactions between quarks and gluons , and thus should depend on qcd interactions . these contributions have not been carefully calculated in the literature . it has been shown in ref.@xcite that in the approximations of large-@xmath0 limit , leading order in dynamical perturbation , and improved ladder approximation , the formula for @xmath14 in eqs.([f_0b_0 ] ) reduces to the well - known pagels - stokar formula @xcite in which all dynamical effects from qcd are represented by the quark self - energy @xmath1 in the formula . in this paper , we take the same approximations to calculate the chiral lagrangian coefficients ( the relevant qcd green s functions ) as an illustration of the main feature of how qcd predicts the chiral lagrangian coefficients . similar to the case of the pagels - stokar formula , the relevant qcd green s functions can all be expressed as functions of the quark self - energy @xmath1 . by solving the schwinger - dyson equation , we obtain @xmath1 , and thus the approximate qcd predicted values of the coefficients . we shall see that the obtained coefficients @xmath15 and quark condensate are consistent with the experimental values . the calculation is checked by the absence of divergences in the large-@xmath0 limit as it should be since the divergent meson - loop contributions are of next - to - the - leading order in the @xmath16 expansion . although the present approximation is rather crude , _ it reveals the main feature of qcd predictions for the chiral lagrangian coefficients_. this paper is organized as follows : in sec . ii , we calculate the anomaly part contributions to the @xmath2 coefficients using the schwinger proper time regularization technique , and the results coincide with those in ref.@xcite in the chiral limit . then , in sec . iii , we apply the same technique to the normal part , and show generally that , in the large-@xmath0 limit , the anomaly part contributions to the chiral lagrangian coefficients are exactly cancelled by the contributions from a piece in the normal part independent of the quark self - energy , and the contributions from the remaining piece in the normal part depending on the quark self - energy play the real role in the chiral lagrangian coefficients . specific approximations in the calculation of the normal part contributions and the formulae for the complete chiral lagrangian coefficients in terms of the quark self - energy are given in sec . iv . in sec . v , we present the numerical calculations of the quark self - energy and the obtained values of the chiral lagrangian coefficients . section vi is a concluding remark . in order to see the relation between the anomaly part and the normal part contributions to the chiral lagrangian coefficients , we present here the calculation of the anomaly part contributions by means of the schwinger proper time regularization . we shall see that the obtained results exactly coincide with those obtained from the heat kernel technique @xcite . our present approach is different from that in ref.@xcite in the sense that the constant constituent quark mass @xmath17 is not put in by hand as is done in ref.@xcite but is naturally included in the normal part solution through the dynamical quark mass reflecting chiral symmetry breaking . therefore our result of the anomaly part contribution is to compare with that in ref.@xcite in the chiral limit . in the schwinger proper time regularization , the anomaly part does not contribute to the coefficients of the @xmath10 terms in the case corresponding to the result with @xmath18 in ref.@xcite to regularize the divergent integrals as was done in ref.@xcite before putting in the constituent quark mass @xmath17 , the @xmath10 coefficient @xmath14 will be proportional to @xmath19 ( cf . ref.@xcite ) . as has been pointed out in ref.@xcite this term is exactly cancelled by a corresponding term in the normal part contribution [ cf . eq.(74 ) in ref.@xcite ] . ] . therefore we are only going to calculate the anomaly contribution to the coefficients of the @xmath2 terms . the anomaly term in the path - integral formalism is @xmath20\nonumber\\ & = & in_c[{\rm tr}\ln(i\partial\!\!\!\ ! /\;+j_{\omega } ) + \omega-{\rm independent~term } ] . \label{sanom}\end{aligned}\ ] ] the @xmath4-independent term is independent of the @xmath5 field , so that it is irrelevant to the chiral lagrangian coefficients . we shall only evaluate the @xmath4-dependent term in eq.([sanom ] ) . to have a unified parametrization , we can parametrize the anomaly contributed effective action similar to that in eq.([p4 ] ) , i.e. , @xmath21 ^ 2 -{\cal k}_2^{\rm ( anom)}(d^{\mu}a_{\omega}^{\nu}-d^{\nu}a_{\omega}^{\mu } ) ( d_{\mu}a_{\omega,\nu}-d_{\nu}a_{\omega,\mu } ) + { \cal k}_3^{\rm ( anom)}[a_{\omega}^2]^2 \nonumber\\ & & + { \cal k}_4^{\rm ( anom)}a_{\omega}^{\mu}a_{\omega}^{\nu } a_{\omega,\mu}a_{\omega,\nu } + { \cal k}_5^{\rm ( anom)}a_\omega^2{\rm tr}_f[a_{\omega}^2 ] + { \cal k}_6^{\rm ( anom)}a_{\omega}^{\mu}a_{\omega}^{\nu } { \rm tr}_f[a_{\omega,\mu}a_{\omega,\nu}]+{\cal k}_7^{\rm ( anom)}s_{\omega}^2 \nonumber\\ & & + { \cal k}_8^{\rm ( anom)}s_{\omega}{\rm tr}_f[s_{\omega } ] + { \cal k}_9^{\rm ( anom)}p_{\omega}^2 + { \cal k}_{10}^{\rm ( anom)}p_{\omega}{\rm tr}_f[p_{\omega } ] + { \cal k}_{11}^{\rm ( anom)}s_{\omega}a_{\omega}^2 \nonumber\\ & & + { \cal k}_{12}^{\rm ( anom)}s_{\omega}{\rm tr}_f[a_{\omega}^2 ] -{\cal k}_{13}^{\rm ( anom)}v_{\omega}^{\mu\nu}v_{\omega,\mu\nu } + i{\cal k}_{14}^{\rm ( anom)}v_{\omega}^{\mu\nu}a_{\omega,\mu}a_{\omega,\nu } + { \cal k}_{15}^{\rm ( anom)}p_{\omega}d_{\mu}a^{\mu}_{\omega}\bigg ] \nonumber\\ & & + o(p^6)+u-\mbox{independent~source~terms}. \label{p4anom}\end{aligned}\ ] ] the @xmath4-dependent term in eq.([sanom ] ) suffers from ultraviolet divergence , and we take the schwinger proper time regularization with an ultraviolet cutoff parameter @xmath22 to regularize it . to apply this regularization , we first work in the euclidean space - time , and analytically continue the results to the minkowskian space - time after the evaluation . the main procedure of evaluating the general functional determinant including the quark self - energy @xmath23 is described in appendix a. in the case of @xmath24 , there is no @xmath23-dependence in eq.([sanom ] ) . however , for regularizing the infrared divergence , we should replace the @xmath23 in eqs.([lnsigma ] ) and ( [ momexp ] ) by an infrared cutoff parameter @xmath25 . the momentum integration in eq.([momexp ] ) can be explicitly carried out with a lengthy but elementary calculation . after expanding in powers of the external sources , we can identify the expressions for @xmath26 by comparing with the form of eq.([p4anom ] ) , and we obtain @xmath27 comparing with the standard form of momentum expansion to identify the @xmath2 chiral lagrangian coefficients , we obtain the anomaly contribution to these coefficients @xmath28 these exactly coincide with the results with @xmath18 in ref.@xcite . note that the final expressions of the coefficients @xmath29 are independent of the infrared cutoff parameter @xmath25 and the ultraviolet cutoff @xmath22 although these cutoff parameters appear in @xmath30 , while @xmath31 and @xmath32 depend on the cutoff parameters . this implies that @xmath31 and @xmath32 are not measurable quantities . with @xmath33 , the values of the coefficients are ( in units of @xmath34 ) @xmath35 these are to be compared with the experimental values ( in units of @xmath34 ) @xcite @xmath36 the numbers in eqs.([anomalyl ] ) are close to the experimental results of eqs.([exptl ] ) except @xmath37 and @xmath38 are of wrong signs . this gives people an impression that the coefficients @xmath39 might mainly be contributed by the anomaly part , and the normal part might only contribute small corrections @xcite . however , we note that the results in eqs.([anomalyresults ] ) are independent of qcd interactions , i.e. , these terms remain unchanged when we switch off the qcd gauge coupling constant @xmath40 . this is somewhat confusing since these coefficients indicate meson interactions which should be residual interactions between quarks and gluons . we shall see in the next section that _ these terms will actually be completely cancelled by the terms independent of the quark self - energy in the normal part contribution , so that they do not really appear in the final form of the coefficients . what appear in the coefficients are the remaining terms in the normal part contribution which depend on the quark self - energy and hence on the qcd interactions as it should be . _ another feature of the terms in eqs.([anomalyresults ] ) indicating that they should be exactly cancelled and should not appear in the final formulae for the coefficients is the divergence of @xmath31 and @xmath32 when taking @xmath41 . we know from ref.@xcite that the ultraviolet divergences in the @xmath2 chiral lagrangian coefficients come merely from the meson - loop corrections with the @xmath10 interactions . in the @xmath16 expansion , the meson - loop corrections belonging to @xmath42 will not take place in the large-@xmath0 limit . therefore , in the large-@xmath0 limit , the final expressions for the @xmath2 coefficients should be finite when @xmath41 . now the ultraviolet divergences in @xmath31 and @xmath32 in eqs.([anomalyresults ] ) have nothing to do with the meson - loop corrections , so that they should be exactly cancelled by other terms and should not appear in the final expressions for the @xmath2 coefficients . in this section we use the same regularization technique as in sec . ii to calculate the normal part contributions to the chiral lagrangian coefficients . we start from the effective action @xmath43 given in ref.@xcite , @xmath44 } \nonumber\\ & & \hspace{0.4cm}=\int{\cal d}\xi{\cal d}\phi_{\omega}\exp\bigg\ { i\gamma_0[j_{\omega},\phi_{\omega},\pi_{\omega c}]+i\gamma_i[\phi_{\omega } ] + in_c\int d^4x{\rm tr}_{lf}\{\xi(x ) [ -i\sin\frac{\vartheta(x)}{n_f } + \gamma_5\cos\frac{\vartheta(x)}{n_f } ] \phi_{\omega}^t(x , x)\}\bigg\ } \label{seff}\end{aligned}\ ] ] which satisfies a useful relation @xcite @xmath45 the symbols are defined in ref.@xcite . in the large-@xmath0 limit , the integrations in eq.([seff ] ) can be carried out by the saddle point approximation with the saddle point equations @xmath46^{(b\zeta)(a\eta)}(y , x ) , \label{pieq}\\ & & \pi_{\omega c}^{\sigma\rho}(x , y)=-\tilde{\xi}^{\sigma\rho}(x)\delta^4(x - y ) -\sum^{\infty}_{n=1}{\int}d^{4}x_1\cdots{d^4}x_{n } d^{4}x_{1}'\cdots{d^4}x_{n}'\frac{(-i)^{n+1}(n_c g_s^2)^n}{n!}\nonumber\\ & & \hspace{0.2cm}\times\overline{g}^{\sigma\sigma_1\cdots\sigma_n}_{\rho\rho_1 \cdots\rho_n}(x , y , x_1,x'_1,\cdots , x_n , x'_n ) \phi_{\omega c}^{\sigma_1\rho_1}(x_1 , x'_1)\cdots \phi_{\omega c}^{\sigma_n\rho_n}(x_n , x'_n ) , \label{fineqnc}\\ & & { \rm tr}_l\bigg[\bigg(-i\sin\frac{\vartheta(x)}{n_f } + \gamma_5\cos\frac{\vartheta(x)}{n_f}\bigg)\phi^t_{\omega c}(x , x)\bigg]=0 , \label{constraint}\end{aligned}\ ] ] where @xmath47 is a short notation for the following quantity @xmath48\phi_{\omega c}^t(y , y)\ } \bigg|_{\xi_c\mbox { fixed}}.\label{xitilde}\end{aligned}\ ] ] then the obtained @xmath43 in this approximation is @xmath49 \nonumber\\ & = & -in_c{\rm tr}\ln[i\partial\!\!\!/+j_{\omega}-\pi_{\omega c } ] + n_c\int d^{4}xd^{4}x ' \phi^{\sigma\rho}_{\omega c}(x , x')\pi^{\sigma\rho}_{\omega c}(x , x ' ) + n_c\sum^{\infty}_{n=2}{\int}d^{4}x_1\cdots d^4x_{n}'\nonumber\\ & & \times \frac{(-i)^{n}(n_c g_s^2)^{n-1}}{n!}\bar{g}^{\sigma_1\cdots\sigma_n}_{\rho_1 \cdots\rho_n}(x_1,x'_1,\cdots , x_n , x'_n ) \phi^{\sigma_1\rho_1}_{\omega c}(x_1 , x'_1)\cdots \phi^{\sigma_n\rho_n}_{\omega c}(x_n , x'_n)\nonumber\\ & & + in_c \int d^4x~{\rm tr}_{lf}\{\xi_c(x)[-i\sin\frac{\vartheta_c(x)}{n_f } + \gamma_5\cos\frac{\vartheta_c(x)}{n_f}]\phi_{\omega , c}^t(x , x)\ } , \label{snorm}\end{aligned}\ ] ] in which the @xmath42 term @xmath50 is neglected . note that the last term in eq.([snorm ] ) actually vanishes due to eq.([constraint ] ) . we keep it here for showing the relation between the effective action @xmath43 and its stationary conditions eqs.([pieq])@xmath51 ( [ constraint ] ) . in the large-@xmath0 limit , @xmath52 on the right hand side of eq.([seffdiff ] ) . the left hand side of eq.([seffdiff ] ) can be carried out from eq.([snorm ] ) using eqs.([pieq])@xmath51([constraint ] ) . then the explicit form of eq.([seffdiff ] ) in this approximation is @xmath53^{\rho\sigma}(x , x ) = \phi^{\sigma\rho}_{\omega c}(x , x ) . \label{eq}\end{aligned}\ ] ] we see that @xmath54 and @xmath55 play the roles of the quark self - energy and the quark propagator , respectively , in the case with @xmath56 . now we decompose @xmath43 into a part _ independent of @xmath54 _ and a part _ depending on @xmath54_. the part independent of @xmath54 can be extracted from @xmath43 by setting @xmath57 , i.e. , @xmath58 + n_c\bigg[\sum^{\infty}_{n=2}{\int}d^{4}x_1\cdots d^4x_{n } ' \frac{(-i)^{n}(n_c g_s^2)^{n-1}}{n!}\bar{g}^{\sigma_1\cdots\sigma_n}_{\rho_1 \cdots\rho_n}(x_1,x'_1,\cdots , x_n , x'_n)\nonumber\\ & & \times\phi^{\sigma_1\rho_1}_{\omega c}(x_1 , x'_1)\cdots \phi^{\sigma_n\rho_n}_{\omega c}(x_n , x'_n)\bigg]_{\pi_{\omega c}=0}. \label{spi=0}\end{aligned}\ ] ] here we have ignored the last term in eq.([snorm ] ) which actually vanishes due to eq.([constraint ] ) . we show in appendix b that the last term in eq.([spi=0 ] ) is actually @xmath4-independent . therefore , eq.([spi=0 ] ) can be written as @xmath59 . \label{spi=0'}\end{aligned}\ ] ] comparing the @xmath60-dependent terms in eqs.([sanom ] ) and ( [ spi=0 ] ) , we see that they are of the same form but with an opposite sign . thus _ their contributions to the chiral lagrangian coefficients exactly cancel each other to all orders in the momentum expansion_. the cancellation in the case of the @xmath10 coefficient @xmath61 has been described in footnote 1 in sec . ii . for the @xmath2 coefficients , we have @xmath62 thus , in the large-@xmath0 limit , the anomaly part contributed chiral lagrangian coefficients in eqs.([anomalyresults ] ) do not really appear in the final results of the chiral lagrangian coefficients although their values eqs.([anomalyl ] ) are close to the experimental values . _ the chiral lagrangian coefficients are actually contributed from the @xmath63 part of @xmath43 _ , @xmath64 _ which leads to the @xmath65 part of @xmath66 _ , @xmath67 _ this is our first new conclusion in this study_. the final chiral lagrangian coefficients are then @xmath68 and @xmath69\nonumber,\\ & & l_9 = \frac{1}{8}(4{\cal k}_{13}^{({\rm norm},\pi_{\omega c}\ne 0 ) } -{\cal k}_{14}^{({\rm norm},\pi_{\omega c}\ne 0)})\nonumber,\\ & & l_{10 } = \frac{1}{2}({\cal k}_2^{({\rm norm},\pi_{\omega c}\ne 0 ) } -{\cal k}_{13}^{({\rm norm},\pi_{\omega c}\ne 0)})\nonumber,\\ & & h_1 = -\frac{1}{4}({\cal k}_2^{({\rm norm},\pi_{\omega c}\ne 0 ) } + { \cal k}_{13}^{({\rm norm},\pi_{\omega c}\ne 0)})\nonumber,\\ & & h_2 = \frac{1}{8}[-{\cal k}_1^{({\rm norm},\pi_{\omega c}\ne 0 ) } + \frac{1}{b_0 ^ 2}{\cal k}_7^{({\rm norm},\pi_{\omega c}\ne 0 ) } + \frac{1}{b_0 ^ 2}{\cal k}_9^{({\rm norm},\pi_{\omega c}\ne 0 ) } -\frac{1}{b_0}{\cal k}_{15}^{({\rm norm},\pi_{\omega c}\ne 0 ) } ] . \label{p4full}\end{aligned}\ ] ] since @xmath70 , these @xmath2 chiral lagrangian coefficients will vanish if we switch off the qcd coupling constant @xmath71 as it should be . we see that to calculate the chiral lagrangian coefficients from @xmath72 , we should mainly deal with @xmath43 given in eq.([snorm ] ) which has never been carefully calculated in the literature . ignoring the vanishing last term in eq.([snorm ] ) , there are still rather complicated terms in it . for example , the third term includes various ranks of gluon green s functions which concern very complicated calculations of qcd dynamics . as the first time of doing this kind of calculation , we shall take futher approximations to simplify the evaluation of @xmath43 . we know that the pion decay constant @xmath73 has been studied from qcd in ref.@xcite by taking the approximation of keeping only the leading order in dynamical perturbation , i.e. , taking into account only the qcd interaction in the schwinger - dyson equation leading to the nonperturbative solution of chiral symmetry breaking , and neglecting other qcd corrections in positive powers of @xmath71 ( perturbative ) . this approximation leads to the widely used pagels - stokar formula which is reasonable though not perfect . in the large-@xmath0 limit , @xmath73 is just the @xmath10 chiral lagrangian coefficient @xmath74 given in eqs.([f_0b_0 ] ) . so , as in ref.@xcite , we take the approximation of keeping only the leading order in dynamical perturbation to calculate the chiral lagrangian coefficients from @xmath43 . in this spirit , we neglect the complicated third term in eq.([snorm ] ) which contains only positive powers of @xmath71 . furthermore , we see from eq.([fineqnc ] ) that the second term in eq.([snorm ] ) is of the same order as the third term , so that we neglect the second term in eq.([snorm ] ) as well . with this approximation , @xmath43 is simplified as @xmath75 . \label{snormsimple}\end{aligned}\ ] ] now the concerned qcd dynamics resides in the @xmath54 term which is related to the quark self - energy [ cf . eq.([eq ] ) ] . we expect such a simple approximation may also lead to reasonable results of the @xmath2 chiral lagrangian coefficients since reasonable values of the @xmath2 chiral lagrangian coefficients have been obtained in a model by holdom @xcite considering only the quark self - energy contribution with certain phenomenological ansatz . we shall see in sec . v that our obtained @xmath2 chiral lagrangian coefficients are indeed reasonable . although this approximation is crude , it provides a simple illustration of the main feature of how qcd predicts the chiral lagrangian coefficients . further improved study beyond this simple approximation taking into account the second and third terms in eq.([snorm ] ) is of course needed . that will be presented in a later paper . now we need to calculate @xmath76 and carry out the explicit expression for @xmath43 in eq.([snormsimple ] ) . we have noticed that @xmath77 is related to the quark self - energy . if we find out the relation between @xmath76 and the conventional quark self - energy @xmath78 , then we can obtain @xmath78 by solving the well - known schwinger - dyson equation . as in the literature , we shall write down the schwinger - dyson equation in the landau gauge which is stable against the gauge parameter . in the same approximation of taking the leading order in dynamical perturbation theory and with the improved ladder approximation , the schwinger - dyson equation in the euclidean space - time reads @xcite @xmath79}{(p - q)^2 } \frac{\sigma(q^2)}{q^2+\sigma^2(q^2 ) } = 0 . \label{eq0}\end{aligned}\ ] ] this equation can be solved numerically , and the details will be presented in sec . v. naively , we may expect that @xmath80 . but this is not correct . under a local chiral transformation @xmath81 ( hidden symmetry transformation @xcite ) , @xmath77 transforms as @xmath82 while @xmath83 does not transform like this . the correct relation can be found by replacing the ordinary derivative @xmath84 by the covariant derivative @xmath85 since the external source @xmath86 transforms as @xmath87,\end{aligned}\ ] ] the covariant derivative @xmath88 transforms as @xmath89 thus the correct identification is @xmath90^{\sigma\rho}\delta^4(x - y ) . \end{aligned}\ ] ] then the effective action ( [ snormsimple ] ) can be written as @xmath91 . \label{sgl}\end{aligned}\ ] ] next we evaluate the effective action ( [ sgl ] ) using the schwinger proper time regularization as before [ cf . appendix a ] to obtain the expressions for the chiral lagrangian coefficients . this is not trivial since usually this regularization scheme is used in the case that @xmath23 is a constant , and thus the momentum integration can be explicitly carried out to check the local gauge invariance of the result . now we leave @xmath92 as an unspecified function in eq.([sgl ] ) , so that the momentum integration can not be carried out explicitly . organizing terms to guarantee local chiral invariance is rather tedious and the details are given in ref.@xcite . our obtained expressions in the minkowskian space - time are @xmath93 , \label{f0}\\ & & { \cal k}^{({\rm norm})}_{1}=2\int d\tilde{p}\bigg[-2a_px_p^3 + 2a_p\frac{x_p^2}{\lambda^2}-a_p\frac{x_p}{\lambda^4 } + \frac{p^2}{2}\sigma'^2_p\frac{x_p}{\lambda^2}-\frac{p^2}{2}\sigma'^2_px_p^2 , \bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{2}=\int d\tilde{p}\bigg[-2b_px_p^3 + 2b_p\frac{x_p^2}{\lambda^2}-b_p\frac{x_p}{\lambda^4 } + \frac{p^2}{2}\sigma^{\prime 2}_p\frac{x_p}{\lambda^2 } , -\frac{p^2}{2}\sigma^{\prime 2}_px_p^2\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{3}=2\int d\tilde{p}\bigg [ ( \frac{4\sigma^4_p}{3}-\frac{2p^2\sigma^2_p}{3}+\frac{p^4}{18 } ) ( 6x_p^4-\frac{6x_p^3}{\lambda^2}+\frac{3x_p^2}{\lambda^4 } -\frac{x_p}{\lambda^6 } ) , \nonumber\\ & & \hspace{0.4cm}+(-4\sigma^2_p+\frac{p^2}{2})(-2x_p^3 + \frac{2x_p^2}{\lambda^2}-\frac{x_p}{\lambda^4 } ) -\frac{x_p}{\lambda^2}+x_p^2\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{4}=\int d\tilde{p}\bigg [ ( \frac{-4\sigma^4_p}{3}+\frac{2p^2\sigma^2_p}{3}+\frac{p^4}{18 } ) ( 6x_p^4-\frac{6x_p^3}{\lambda^2}+\frac{3x_p^2}{\lambda^4 } -\frac{x_p}{\lambda^6 } ) + 4\sigma^2_p(-2x_p^3+\frac{2x_p^2}{\lambda^2 } \nonumber\\ & & \hspace{0.4cm}-\frac{x_p}{\lambda^4})+\frac{x_p}{\lambda^2}-x_p^2\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_5={\cal k}^{({\rm norm})}_6=0 , \nonumber\\ & & { \cal k}^{({\rm norm})}_7=2\int d\tilde{p}\bigg[(3\sigma^2_p+2p^2\sigma_p\sigma'_p)x_p^2 + [ -2\sigma^2_p - p^2(1 + 2\sigma_p\sigma'_p)]\frac{x_p}{\lambda^2}\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_8=0 , \nonumber\\ & & { \cal k}^{({\rm norm})}_9=2\int d\tilde{p}\bigg[(\sigma^2_p+2p^2\sigma_p\sigma'_p)x_p^2 -p^2(1 + 2\sigma_p\sigma'_p)\frac{x_p}{\lambda^2}\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{10}=0 , \nonumber\\ & & { \cal k}^{({\rm norm})}_{11}=4\int d\tilde{p } \bigg[(-4\sigma^3_p+p^2\sigma_p)x_p^3 + ( 4\sigma^3_p - p^2\sigma_p)\frac{x_p^2}{\lambda^2 } -(2\sigma^3_p-\frac{1}{2}p^2\sigma_p)\frac{x_p}{\lambda^4 } + 3\sigma_p\frac{x_p}{\lambda^2 } \nonumber\\ & & \hspace{0.4cm}-3\sigma_p x_p^2\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{12}=0 , \nonumber\\ & & { \cal k}^{({\rm norm})}_{13}=\int d\tilde{p}\bigg [ ( \frac{1}{3}p^2\sigma'_p\sigma''_p+\frac{1}{3}\sigma_p\sigma''_p)x_p + ( c_p - d_p)\frac{x_p}{\lambda^2 } -(c_p - d_p)x_p^2 - 2e_px_p^3 \nonumber\\ & & \hspace{0.4cm}+2e_p\frac{x_p^2}{\lambda^2 } -e_p\frac{x_p}{\lambda^4}\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{14}=-4\int d\tilde{p}\bigg [ -2f_px_p^3 + 2f_p\frac{x_p^2}{\lambda^2 } -f_p\frac{x_p}{\lambda^4 } + \frac{p^2}{2}\sigma_p^{\prime 2}\frac{x_p}{\lambda^2 } -\frac{p^2}{2}\sigma^{\prime 2}_px_p^2\bigg ] , \nonumber\\ & & { \cal k}^{({\rm norm})}_{15}=-4\int d\tilde{p}\bigg [ -(\sigma_p+\frac{1}{2}p^2\sigma'_p)\frac{x_p}{\lambda^2 } + ( \sigma_p+\frac{1}{2}p^2\sigma'_p)x_p^2\bigg ] , \label{kresult}\end{aligned}\ ] ] in which the short notations ( in the minkowskian space - time ) are @xmath94 for the coefficient @xmath14 , eq.([f0 ] ) is just the well - known pagels - stokar formula @xcite when taking the regularization cutoff parameter @xmath95 . it is easy to check that these @xmath96 ( @xmath97 ) do contain the @xmath54-independent ( @xmath98-independent ) piece which exactly cancel the anomaly contributions in eqs.([anomk ] ) mentioned in sec . this can be done by taking a constant @xmath98 to carry out the momentum integrations , and picking up the @xmath98-independent terms which are just the @xmath54-independent terms mentioned in sec . subtracting these @xmath54-independent terms from the obtained @xmath96 in eqs.([kresult ] ) , we get the desired @xmath99 in eq.([kpi ] ) , which is needed in the final expressions for the chiral lagrangian coefficients in eqs.([p4full ] ) . we can also check that the regularization cutoff @xmath22 does not appear in @xmath99 , so that the obtained chiral lagrangian coefficients @xmath100 are all finite as it should be since there is no divergence in the large-@xmath0 limit [ the divergent meson - loop corrections are of @xmath42 ] . @xmath125 with boundary conditions : @xmath126 where @xmath127 is a momentum cutoff regularizing the integral . we shall eventually take @xmath128 . we know that the asymptotic behavior of @xmath1 reflecting chiral symmetry breaking is @xmath129 where @xmath130 . we have found the numerical solution of eqs.([fineq1 ] ) , ( [ f01 ] ) and ( [ fp01 ] ) satisfying this asymptotic behavior . the obtained solution with @xmath128 ( a large enough number which can be regarded as infinity ) in the three cases are plotted in fig . 2 . again they are different mainly in the low momentum region . with the obtained @xmath1 , we can calculate the @xmath2 chiral lagrangian coefficients from eqs.([p4full ] ) , ( [ kpi ] ) , and ( [ kresult ] ) . the obtained values of @xmath15 are listed in table i together with the experimental values @xcite for comparison . note that there is no running of @xmath15 in this simple approach since the meson - loop effects causing the running of @xmath15 @xcite are of the order of @xmath16 , and are neglected in this approach . thus the predicted numbers of @xmath15 can be directly compared with the experimental values . we see from table i that : \(i ) these coefficients are not so sensitive to the forms of @xmath104 ; \(ii ) all the obtained @xmath15 are of the right orders of magnitude and the right signs ; \(iii ) @xmath131 and @xmath132 are consistent with the experiments at the @xmath133 level ; \(iv ) @xmath134 and @xmath38 are consistent with the experiments at the @xmath135 level ; and \(v ) only @xmath136 deviates from the experimental value by @xmath137 . considering the large theoretical uncertainty in this simple approach , _ the obtained @xmath15 are consistent with the experiments_. we see that the nonperturbative quark self - energy plays an important role in qcd contributions to the chiral lagrangian coefficients . this supports the phenomenological model of holdom @xcite . in addition to @xmath15 , we can also calculate the quark condensate @xmath138 from the @xmath10 coefficient @xmath139 in eq.([f0b0 ] ) . in the simple approach in this paper , the relation between @xmath138 and @xmath139 is @xcite @xmath140 we know that , in this simple approach , @xmath109 mev is finite . but @xmath139 in eq.([f0b0 ] ) is divergent , @xmath141 so that it needs to be renormalized . we take a simple renormalization scheme by taking the counter term as @xmath142 , in which @xmath115 is the renormalization scale ) scheme @xcite . ] . thus the renormalized quantity @xmath143 then the renormalized ( [ condensate ] ) is @xmath144 we take the renormalization scale to be @xmath145 gev to define the quark condensate . the obtained values of @xmath146 for the three forms of @xmath104 are @xmath147 these are to be compared with the experimentally determined value @xmath148 from the qcd sum rule at the scale of the typical hadronic mass @xcite . considering the large theoretical uncertainty in this calculation , the predicted quark condensate is also consistent with the experiment . the above results show that _ the present simple approach does reveal the main feature of the qcd predictions for the chiral lagrangian coefficients _ although the approximations in this approach are rather crude . of course , further improvements of the approximations beyond this simple approach are needed . this kind of study is in progress . finally , we would like to mention that , in our calculation , we have taken the ultraviolet cutoff parameters @xmath149 , i.e. , we have taken account of the qcd contributions in the whole momentum range . note that this has nothing to do with the validity range of the chiral lagrangian determined by the range in which the expansion in the meson momentum makes sense , i.e. , up to @xmath150 . to see the role of the qcd contributions from the high momentum region , say above 1 gev , we have made a check by doing the calculations with the same @xmath1 but taking @xmath151 gev instead of @xmath41 . the results are listed in table ii . comparing the nonvanishing results in table ii with the corresponding @xmath149 results in table i , we see that this change of @xmath22 does not cause much difference in @xmath152 and @xmath74 , while it causes @xmath153 and @xmath132 to reduce by at least a factor of 2 . therefore , we see that @xmath152 and @xmath74 are mainly contributed by the qcd dynamics in the low momentum region , while high momentum region contributions to @xmath154 , and @xmath132 are not negligible . in this paper , we have calculated the coefficients in the gasser - leutwyler lagrangian from the underlying theory of qcd in a simple approach with the approximations of taking the large-@xmath0 limit , the leading order in dynamical perturbation theory , and the improved ladder approximation based on the qcd formulae given in ref.@xcite to illustrate the main feature of how qcd predicts the chiral lagrangian coefficients . in the calculation , we use the same regularization technique , the generalized schwinger proper time regularization , in the calculations of the contributions from both the anomaly part and the normal part , so that the relation between the contributions from the two parts can be clearly seen . we first take the large-@xmath0 limit to evaluate the effective action in qcd . our first conclusion in this study is that , _ in the large-@xmath0 limit , to all orders in momentum expansion , the anomaly part contributions to the chiral lagrangian coefficients [ cf . eqs.([anomalyresults ] ) ] given in the literature @xcite from the effective action @xmath155 [ cf . eq.([sanom ] ) ] are exactly cancelled by the contributions from the piece of the effective action @xmath156 [ cf . eq.([spi=0 ] ) ] in the normal part contributions _ , so that _ the chiral lagrangian coefficients are eventually contributed by the remaining piece of the normal part effective action @xmath157 [ cf . eq.([spi])]_. the final qcd expressions for the @xmath2 coefficients are given in eqs.([p4full ] ) . to simplify @xmath157 , we further make the approximation of taking the leading order in dynamical perturbation theory . then @xmath157 is reduced to the simple form in eq.([sgl ] ) , and all the chiral lagrangian coefficients are approximately expressed in terms of the quark self - energy @xmath1 shown in eqs.([f0b0])@xmath51([kresult ] ) . to solve the schwinger - dyson equation for @xmath1 , we further take the improved ladder approximation . lacking of the knowledge about the running coupling constant @xmath104 in the nonperturbative region , we take certain models for it from the literature @xcite [ cf . eqs.([omega1 ] ) , ( [ omega2 ] ) and ( [ omega3 ] ) ] , and we further take the usual approximation @xmath158 to simplify the calculation . the quark self - energy reflecting chiral symmetry breaking is obtained by solving the simplified schwinger - dyson equation numerically . the obtained results of the @xmath2 coefficients are listed in table i. compared with the experimental values of @xmath159 @xcite , _ the agreement of @xmath160 , and @xmath132 is of the level of @xmath133 , and that of @xmath134 and @xmath38 is of the level of @xmath135_. only @xmath136 deviates from the experimental value by @xmath137 . considering the large theoretical uncertainty in this simple approach , _ all the obtained coefficients @xmath159 are consistent with the experiments_. we have also calculated the renormalized quark condensate @xmath146 from the obtained @xmath10 coefficient [ cf . eq.([condensaterenorm ] ) ] _ which is also consistent with the experiment_. although the approximations in this simple approach are rather crude , the above results show that _ this simple approach does reveal the main feature of qcd predictions for the chiral lagrangian coefficients_. for studying physics not requiring high precision , this simple approach may already be useful . of course further improvements of the approximations beyond this simple approach ( reflecting more about qcd dynamics ) are needed . this kind of study is in progress and will be presented in another paper . the approach can also be applied to electroweak theories to study how the coefficients in the electroweak chiral lagrangian are predicted by various kinds of underlying gauge theories of the electroweak symmetry breaking mechanism . this kind of study is also in progress , and will be presented in separate papers . this work is supported by the national natural science foundation of china , the foundation of fundamental research grant of tsinghua university , and a special grant from the ministry of education of china . in this appendix , we take the schwinger proper time regulation to regularize the one - loop functional determinant in which the quark self - energy @xmath161 reflecting chiral symmetry breaking takes place . the matrix element in eq.([lnsigma ] ) can be evaluated in the momentum representation @xmath164}|x\rangle\nonumber\\ & & = \int\frac{d^4p}{(2\pi)^4}\exp\bigg\{-\tau\bigg[\overline{e}(x ) -\nabla_x^2 - 2ip\cdot\nabla_x+p^2 + \sigma^2(-\overline{\nabla}^2 - 2ip\cdot\overline{\nabla}_x+p^2 ) \nonumber\\ & & \hspace{0.5 cm } + \hat{i}_\omega\sigma(-\overline{\nabla}_x^2 - 2ip\cdot\overline{\nabla}_x+p^2 ) + \sigma(-\overline{\nabla}_x^2 - 2ip\cdot\overline{\nabla}_x+p^2)\tilde{i}_\omega -d\!\!\ ! /\;\sigma(-\overline{\nabla}_x^2 - 2ip\cdot\overline{\nabla}_x+p^2 ) \bigg]\bigg\}. % \nonumber\\ % & & \hspace{0.5 cm } % -i[p\!\!\ ! /\;,\sigma(-\overline{\nabla}_x^2 - 2ip\cdot\overline{\nabla}_x+p^2 ) ] % \bigg]\bigg\}. \label{momexp}\end{aligned}\ ] ] then after lengthy but elementary calculations and expanding in powers of the external sources , we can identify the expressions for @xmath165 by comparing with the form of eqs.([p4 ] ) , and the obtained results in the minkowskian space - time are just those given in eqs.([f0 ] ) , ( [ f0b0 ] ) and ( [ kresult ] ) in the text . the details are given in ref.@xcite . for the evaluation of the effective action @xmath24 in eq.([sanom ] ) in the minkowskian space - time , we note that there is no @xmath166 term in eq.([lnsigma ] ) , but we still have to replace @xmath166 by an infrared cutoff parameter @xmath25 in eq.([lnsigma ] ) to regularize the infrared divergence . then the momentum integration can be explicitly carried out , and we obtain the results in eqs.([anomk ] ) in the text . here we show that the last term in eq.([spi=0 ] ) , @xmath167 with the saddle point equation [ cf . eq.([pieq ] ) ] @xmath168^{\rho\sigma}(y , x ) \label{sta1}\end{aligned}\ ] ] is @xmath4-independent . the @xmath4-rotated quantities in eqs.([3rdterm ] ) and ( [ sta1 ] ) are defined by @xcite @xmath169 ~[j(x)+i\partial\!\!\!\ ! /\;][\omega(x)p_r+\omega^{\dagger}(x)p_l],\nonumber\\ & & \phi_{\omega}^t(x , y)=[\omega^{\dagger}(x)p_r+\omega(x)p_l]~\phi^t(x , y ) [ \omega^{\dagger}(y)p_r+\omega(y)p_l ] , \label{rotation}\end{aligned}\ ] ] and the @xmath170-point green s function @xmath171 is define by @xcite @xmath172\cdots [ \overline{\psi}^{a_n}_{{\alpha}_n}(x_n ) ( \frac{\lambda_{i_n}}{2})_{\alpha_n\beta_n } \gamma^{\mu_n}{\psi}^{a_n}_{\beta_n}(x_n)]\nonumber \\ & & = { \int}d^{4}x'_1\cdots{d^{4}x'_n } g_s^{n-2}\overline{g}^{\sigma_1\cdots\sigma_n}_{\rho_1 \cdots\rho_n}(x_1,x'_1,\cdots , x_n , x'_n ) \overline{\psi}^{\sigma_1}_{\alpha_1}(x_1){\psi}^{\rho_1 } _ { \alpha_1}(x'_1)\cdots\overline{\psi}^{\sigma_n } _ { \alpha_n}(x_n){\psi}^{\rho_n}_{\alpha_n}(x'_n ) . \label{bargdef}\end{aligned}\ ] ] first we see from eqs.([rotation ] ) that eq.([sta1 ] ) can be written as @xmath173~\phi^t_c(y , x ) [ \omega^{\dagger}(x)p_r+\omega(x)p_l]\bigg ] ^{\rho\sigma}\bigg|_{\pi_c=0}\nonumber\\ & & = -i\bigg[[\omega^{\dagger}(y)p_r+\omega(y)p_l]~ [ ( i\partial\!\!\ ! /\;+j)^{-1}](y , x ) [ \omega^{\dagger}(x)p_r+\omega(x)p_l]\bigg]^{\rho\sigma}\nonumber\\ & & = -i[\omega^{\dagger}(y)p_r+\omega(y)p_l]^{\rho\rho ' } [ ( i\partial\!\!\ ! /\;+j)^{-1}]^{\rho'\sigma'}(y , x ) [ \omega^{\dagger}(x)p_r+\omega(x)p_l]^{\sigma'\sigma}\nonumber\\ & & = -i[\gamma_0v_{\omega}\gamma_0]^{\dagger\rho\rho'}(y ) [ ( i\partial\!\!\ ! /\;+j)^{-1}]^{\rho'\sigma'}(y , x ) v_{\omega}^{\sigma'\sigma}(x ) , \label{phio}\end{aligned}\ ] ] in which @xmath174 satisfies @xmath175 and @xmath176(y , x ) \label{phic}\end{aligned}\ ] ] is @xmath4-independent . next , we look at this transformed green s function @xmath181 . from the definition ( [ bargdef ] ) and the property ( [ vcondition ] ) we have @xmath182\cdots\nonumber\\ & & \cdots[(\overline{\psi}\gamma_0v^{\dagger}_\omega\gamma_0)^{a_n}_{{\alpha}_n}(x_n ) ( \frac{\lambda_{i_n}}{2})_{\alpha_n\beta_n } \gamma^{\mu_n}(v_\omega{\psi})^{a_n}_{\beta_n}(x_n)]\nonumber \\ & = & g_{\mu_1\cdots\mu_n}^{i_1\cdots{i_n } } ( x_1,\cdots , x_n)[\overline{\psi}^{a_1}_{{\alpha}_1}(x_1 ) ( \frac{\lambda_{i_1}}{2})_{\alpha_1\beta_1}\gamma^{\mu_1 } { \psi}^{a_1}_{{\beta}_1}(x_1)]\cdots [ \overline{\psi}^{a_n}_{{\alpha}_n}(x_n ) ( \frac{\lambda_{i_n}}{2})_{\alpha_n\beta_n } \gamma^{\mu_n}{\psi}^{a_n}_{\beta_n}(x_n)]\nonumber \\ & = & { \int}d^{4}x'_1\cdots{d^{4}x'_n } g_s^{n-2}\bar{g}^{\sigma_1\cdots\sigma_n}_{\rho_1 \cdots\rho_n}(x_1,x'_1,\cdots , x_n , x'_n ) \overline{\psi}^{\sigma_1}_{\alpha_1}(x_1){\psi}^{\rho_1 } _ { \alpha_1}(x'_1)\cdots\overline{\psi}^{\sigma_n } _ { \alpha_n}(x_n){\psi}^{\rho_n}_{\alpha_n}(x'_n),\nonumber\end{aligned}\ ] ] i.e. , @xmath183 thus the transformed green s function @xmath184 in eq.([sgv ] ) can be replaced by @xmath185 , and eq.([sgv ] ) becomes @xmath186 which is independent of @xmath4 . 1[biblio ] s. weinberg , physica , * 96a * , 327 ( 1979 ) . j. gasser and h. leutwyler , ann ( n.y . ) * 158 * , 142 ( 1984 ) ; nucl . phys . * b 250 * , 465 ( 1985 ) . q. wang , y .- p . kuang , x .- wang , and m. xiao , phys . d * 61 * , 054011 ( 2000 ) , hep - ph9903201 . d. espriu , e. de rafael , and j. taron , nucl . phys . * b 345 * , 22 ( 1990 ) ; * b 355 * , 278(e ) ( 1991 ) . q. lu , h. yang , and q. wang , tsinghua university preprint tuhep - th-02133 , hep - ph/0202078 , to appear in commun . h. pagels and s. stokar , phys . d * 20 * , 2947 ( 1979 ) . simic , phys . d * 34 * , 1903 ( 1986 ) . b. holdom , phys . rev . d * 45 * , 2534 ( 1992 ) ; phys . b * 294 * , 293 ( 1992 ) ; b. holdom , j. terning , and k. verbeek , _ ibid_. , * 245 * , 612 ( 1990 ) ; _ ibid_. , * 273 * , 549 ( e ) ( 1991 ) ; b. holdom and q. wang , phys . d * 55 * , 3101 ( 1997 ) . aoki , m. bando , t. kugo , m.g . mitchard , and h. nakatani , prog . . phys . * 84 * , 683(1990 ) . p. jain and h.j . munczek , phys . d * 44 * , 1873(1991 ) . b . dai , c .- s . huang , and d .- s . liu , phys . d * 43 * , 1717 ( 1991 ) ; y .- b . dai , y .- b . ding , c .- s . huang , c .- wang , and y .- l zhu , phys . rev . d * 47 * , 1256 ( 1993 ) . s .- h. tye , e. tomboulis , and e. c. poggio , phys . d * 11 * , 2839 ( 1975 ) ; v. a. miransky , sov . . phys . * 38 * , 280 ( 1983 ) ; k. higashijima , phys . d * 29 * , 1228 ( 1984 ) . for instance , h .- j . he and y .- p . kuang , z. phys . c * 47 * , 565 ( 1990 ) . shifman , a.i . vainshtein , and v.i . zakharov , nucl . * b 147 * , 385 ( 1979 ) . roberts , r.t . cahill and j. praschifka , ann . phys . * 188 * , 20 ( 1988 ) . e. witten , nucl . phys . * b 223 * , 422 ( 1983 ) . . the obtained values of the @xmath2 coefficients @xmath159 for model a [ eq.([omega1 ] ) ] , model b [ eq.([omega2 ] ) ] and model c [ eq.([omega3 ] ) ] with @xmath149 together with the experimental values [ eq.([exptl ] ) ] for comparison . @xmath106 is in mev , and the coefficients are in units of @xmath34 . the obtained @xmath1 from the schwinger - dyson equation ( [ eq0 ] ) with model a [ eq.([omega1 ] ) ] , model b [ eq.([omega2 ] ) ] and model c [ eq . ( [ omega3 ] ) ] for the running coupling constant @xmath40 . the solid , dashed , and dotted lines are for models a , b , and c , respectively .

the standard model of particle physics is known to have many problems with observations , such as the matter abundance of our universe , and many candidates of new physics have been invented so far . among them , the supersymmetry @xcite is widely noticed , due to the possibility to resolve many phenomenological problems encountered in the standard model . one of the most important experimental observables to probe the @xmath2 violation in the supersymmetric models is the electric dipole moment ( edm ) . the edm is sensitive to the violation of parity and time reversal symmetry ( or equivalently the @xmath2 symmetry ) , and is known to have very small standard model prediction @xcite . this fact gives a direct evidence of the @xmath2 violation beyond the standard model if any finite value is observed experimentally . due to the stringent experimental limits on the edm measured in many systems @xcite , many @xmath2 phases of the supersymmetric models with @xcite and without @xmath0-parity @xcite were constrained so far . in this work we focus on the @xmath2 violation of the @xmath0-parity violating ( rpv ) sector @xcite . the conservation of the @xmath0-parity is often assumed to keep the lightest supersymmetric particle stable , but this is only an _ ad hoc _ assumption , and phenomenologically , the existence of the rpv interactions is still allowed . the ( trilinear ) rpv interactions contribute to the fermion edm from the two - loop level @xcite , and the barr - zee type process ( see fig . [ fig : barr - zee ] ) is currently known to give the dominant effect @xcite . barr - zee type diagrams within @xmath0-parity violation . , width=302 ] recently , it was pointed that the two - loop level _ rainbowlike _ contribution to the fermion edm may have comparable size with the leading barr - zee type process in the @xmath0-parity conserving sector @xcite . this contribution has a fermion - sfermion inner loop , connected to the fermion external line with a higgsino and a gaugino , and the flavor structure is exactly the same as for the barr - zee type diagram , with the same coupling constants . in the rpv sector , similar rainbowlike diagrams can be drawn by replacing the higgsino with the neutrino ( see fig . [ fig : rpv_rainbow ] ) , and it is therefore of necessity to evaluate and analyze this potentially sizable contribution . example of rainbow diagram contributing to the fermion edm within @xmath0-parity violation . the neutralino is given by @xmath3 and the gluino by @xmath4 . , width=264 ] this paper is organized as follows . we first introduce the supersymmetric and rpv interactions relevant in this discussion in the next section . we then calculate in section [ sec : rainbow ] the rainbowlike contribution to the fermion edm and to the chromo - edm within @xmath0-parity violation and give the explicit formulae . in section [ sec : analysis ] , we analyze the effect of the rpv supersymmetric rainbow diagrams by comparing with the edm and the chromo - edm generated by the rpv barr - zee type diagrams . the last section is devoted to the summary . we have also collected the detailed derivation of important formulae in the appendix . let us first give the lagrangian of the particles relevant in this discussion . the rpv interactions relevant in this discussion are given by the following superpotential : @xmath5 with @xmath6 indicating the generation , @xmath7 the @xmath8 indices . the @xmath9 indices have been omitted . @xmath10 and @xmath11 denote the lepton doublet and singlet left - chiral superfields . @xmath12 , @xmath13 and @xmath14 denote respectively the quark doublet , the up - type quark singlet and the down - type quark singlet left - chiral superfields . the bilinear term has been omitted in our discussion . also baryon number violating rpv interactions ( @xmath15 ) will be omitted from now , to avoid rapid proton decay . this rpv superpotential gives the following lepton number violating yukawa interactions : @xmath16 + ( { \rm h.c.})\nonumber\\ & & -\lambda'_{ijk } \left [ \tilde \nu_i \bar d_k p_l d_j + \tilde d_{lj } \bar d_k p_l \nu_i + \tilde d_{rk}^\dagger \bar \nu_i^c p_l d_j \right . \nonumber\\ & & \hspace{3.5em } \left . -\tilde e_{li } \bar d_k p_l u_j - \tilde u_{lj } \bar d_k p_l e_i - \tilde d_{rk}^\dagger \bar e_i^c p_l u_j \right ] \nonumber\\ & & \hspace{4em } + ( { \rm h.c . } ) \ .\end{aligned}\ ] ] the soft breaking terms of the rpv sector will not be considered in this discussion . we should also give the supersymmetric interactions of the @xmath0-parity conserving sector . the neutralino mass matrix is given as follows : @xmath17 with @xmath18 where the corresponding neutralino field vector @xmath3 has as the first two components the up- and down - type higgsinos , and the last two components refer to the @xmath19 and the @xmath8 gauginos . in this work , we assume that the supersymmetry breaking scale is beyond tev , which is strongly suggested by the recent result of the lhc experiment @xcite , and we neglect the mixing between gauginos and higgsinos . in our discussion , the higgsino is thus irrelevant . the components of the mass matrix have @xmath2 phases in the general case , and they can manifest themselves as observable effect when the mass insertion occurs in the process . for the gluino , we remove its @xmath2 phase as usual . the sfermion mass matrix is given by @xmath20 where @xmath21 for up - type squarks , and @xmath22 for down - type squarks and charged sleptons . note that in this discussion , only the down - type squarks and the charged sleptons are relevant . to obtain the sfermion mass eigenbasis , we introduce the following unitary matrix with the sfermion mixing angle @xmath23 and the @xmath2 phase @xmath24 : @xmath25 with @xmath26 and @xmath27 for @xmath28 . in this approximation , the two mass eigenvalues are given by @xmath29 , and @xmath30 . the @xmath2 phase is given by @xmath31 . we should also give the gaugino - sfermion - fermion interactions . they are given by @xmath32 where @xmath33 is the gauge index of the fermion @xmath34 ( but will be omitted from now ) . the convention for the sign of the gauge coupling is @xmath35 where @xmath36 is the generator of the gauge group . the index @xmath37 gives the gauge group of the gaugino @xmath38 . we define @xmath39 as the @xmath19 gaugino , @xmath40 as the neutral @xmath8 gaugino and @xmath41 as the gluino . the fermion - sfermion - gaugino coupling constants are given as follows : for the gluino couplings , we have @xmath42 and @xmath43 . for the couplings of @xmath39 , we have @xmath44 , @xmath45 , @xmath46 and @xmath47 . we have finally for the @xmath40 couplings @xmath48 , and @xmath49 . let us now calculate the rpv rainbow diagrams . we can classify the rainbow diagrams into two types . the first type is the insertion of the effective one - loop level neutrino - gaugino - gauge boson vertex [ fig . [ fig : rainbow_type ] ( 1 ) ] . the second type is the insertion of the effective one - loop level neutrino - gaugino transition [ fig . [ fig : rainbow_type ] ( 2 ) ] . possible insertions of one - loop effective vertex for the rainbow diagram in rpv supersymmetry . the grey blob represents the one - loop effective vertex . the solid lines attached to the grey blob indicate the neutrino and gaugino . , width=302 ] the derivation of the rpv rainbow diagrams is very similar to that of the rainbow diagrams in the minimal supersymmetric standard model @xcite . the detail of the computation of the type 1 and 2 insertions is presented in appendices [ sec : type1rainbow ] and [ sec : type2rainbow ] , respectively . to obtain the edm contribution , the above rainbow diagrams are calculated to the first order in the external momentum carried by the gauge boson ( this manipulation is suitable since the edm is the first order coefficient of the multipolar expansion ) . in our calculation , we have neglected the fermion mass originating from the mass insertion of the inner loop . the formulae of the edm of the fermion @xmath50 ( @xmath51 is the flavor of the external fermion ) given by the type 1 [ fig . [ fig : rainbow_type ] ( 1 ) ] and the type 2 [ fig . [ fig : rainbow_type ] ( 2 ) ] diagrams are respectively @xmath52 \ , , \label{eq : edmnugg } \\ d^2_{f_k } & = & -\sum_{n=1,2 } { \rm i m } ( \hat \lambda_{ijj } \tilde \lambda^*_{ikk } e^{i(\theta_n -\delta_{f_j } ) } ) \frac{q_f n_c } { 64 \pi^3 } |m_{\lambda_n } | \frac{e ( g^{(n)}_{\tilde f_l } + g^{(n)}_{\tilde f_r } ) } { 4\pi } \sin \theta_{f_j } \cos \theta_{f_j } \nonumber\\ & & \hspace{10em } \times \sum_{\tilde f= \tilde f_l , \tilde f_r } g^{(n)}_{\tilde f } m_{\tilde f_{k}}^2 \left [ f''(|m_{\lambda_n}|^2 , 0 , m_{\tilde f_{k}}^2 , m_{\tilde f_{1j}}^2 \ , ) -f''(|m_{\lambda_n}|^2 , 0 , m_{\tilde f_{k}}^2 , m_{\tilde f_{2j}}^2 \ , ) \right ] \ , , \label{eq : edmnug}\end{aligned}\ ] ] where @xmath53 for inner quark - squark loop and @xmath54 for inner lepton - slepton loop . here @xmath55 and @xmath56 are the flavor indices of the intermediate neutrino and the fermion - sfermion of the inner loop , respectively . the rpv couplings are given by @xmath57 for the charged lepton - slepton inner loop , @xmath58 in the case of the quark - squark inner loop , @xmath59 for the external lepton , and @xmath60 for the external quark . the constant @xmath61 is @xmath62 for the left - handed sfermion @xmath63 and @xmath64 for the right - handed sfermion @xmath65 . the electric charge of the inner loop fermion ( sfermion ) and the external fermion in the unit of @xmath66 is denoted , respectively , by @xmath67 and @xmath68 . the functions @xmath69 and @xmath70 are defined by @xmath71 -\frac{f}{ac } \frac{\pi^2}{6 } \ , , \label{eq : f ' } \\ % % % % % % % % % % % % % % f '' ( a,0,c , f ) % % % % % % % % % % % % % % f''(a,0,c , f ) & \equiv & \frac{(a - f)^2 } { a(a - c)^3 } \left [ { \rm li}_2 ( 1-a / f ) -{\rm li}_2 ( 1-c / f ) \right ] - \frac{f^2}{ac^3 } \left [ { \rm li}_2 ( 1-c / f ) - \frac{\pi^2}{6 } \right ] \nonumber\\ & & + \frac{1}{a(c - f ) } \left [ \frac{(a - f)^2}{(a - c)^2 } -\frac{f^2}{c^2 } \right ] \ln \frac{c}{f } - \frac{c \ln c - c+2f } { 2c^2 ( a - c ) } + \frac{f+a \ln f}{(a - c)^2 } \left [ \frac{1}{c } -\frac{1}{a - c } \ln \frac{a}{c } \right ] \ , , \label{eq : f''}\end{aligned}\ ] ] with @xmath72 denoting the dilogarithm function . the original functions @xmath73 and @xmath74 were defined in ref . similarly , the chromo - edm is given as follows : @xmath75 \ , . \label{eq : cedm}\ ] ] note that the gluino mass parameter @xmath76 can be negative , and the overall sign of @xmath77 may change . we see that the formulae ( [ eq : edmnugg ] ) , ( [ eq : edmnug ] ) , and ( [ eq : cedm ] ) are very similar to the supersymmetric rainbow contribution with @xmath0-parity conservation @xcite ( this is just a replacement of the matter - higgs yukawa couplings by the rpv couplings , and the higgsino mass @xmath78 by the negligible mass of the neutrino ) . the rpv rainbow diagrams vanish if the mass eigenvalues of the inner loop sfermion @xmath79 and @xmath80 are degenerate . this property can also be seen for the supersymmetric rainbow process with @xmath0-parity conservation . we now analyze the contribution of the rpv rainbow diagrams to the edm and to the chromo - edm of the fermions . in this discussion , we only consider processes with the third generation fermion - sfermion inner loop , since they give the dominant effect due to @xmath81 . we choose the following values for the sparticle masses in the function of the supersymmetry breaking scale @xmath82 : @xmath83 , @xmath84 , @xmath85 , @xmath86 , and @xmath87 . the down squark masses for both gauge eigenstates are given by @xmath88 , and @xmath89 for the selectron . the @xmath19 and @xmath8 gaugino mass is given by @xmath90 , and the sneutrino mass by @xmath91 . in our discussion , we fix the @xmath92 parameter to @xmath93 tev . for @xmath1 , we assign @xmath94 . the large value of @xmath1 is important , since the rainbow diagram contribution will be enhanced for large @xmath1 , due to the factor @xmath95 . the rainbow diagram will be compared with the leading rpv barr - zee type diagram @xcite . the fermion edm and the quark chromo - edm generated from the barr - zee type diagram is given respectively as : @xmath96 the first case to consider is the rainbow diagram with external down - quark and quark - squark inner loop . in this case , the @xmath2 violating combinations of parameters im(@xmath97 ) , im(@xmath98 ) , and im(@xmath99 ) ( @xmath100 ) contribute to the neutron edm via the edm and the chromo - edm . the dependence of the quark edm and the chromo - edm on the neutron edm is given by the following relation , obtained from the analysis using qcd sum rules @xcite : @xmath101 where we have assumed the peccei - quinn symmetry @xcite . the rainbow diagram effect ( for both edm and chromo - edm ) with external down - quark and bottom quark - bottom squark inner loop , together with that of rpv barr - zee diagram , is plotted in fig . [ fig : rpv_q - qrbnedm ] . contribution of the rpv rainbow diagram with external down quark and @xmath102 inner loop ( absolute value ) to the neutron edm plotted in the function of the supersymmetry breaking scale [ we have set @xmath103 . the barr - zee type contribution is also plotted for comparison . , width=302 ] we see that the chromo - edm generated by the rpv barr - zee diagram dominates over all other contributions . the dominance of the chromo - edm over the quark edm can be explained by the large value of the strong coupling @xmath104 over the electromagnetic coupling @xmath105 . we remark also that the rainbow and barr - zee type contributions have very close size near @xmath106 tev for the quark edm , even though the two effects have a difference by about one order of magnitude for the chromo - edm . this is due to the suppression of the barr - zee type diagram of the quark edm , due to the double factor of the fractional charge of the down - type quarks ( @xmath107 ) . we can see also the fast decrease of the rainbow diagrams against the barr - zee type effect in high energy region of @xmath82 . this can be explained by simple dimensional analysis of the scaling in @xmath82 . from eqs . ( [ eq : barr - zee ] ) and ( [ eq : chromo - barr - zee ] ) , we see that the rpv barr - zee type diagram is scaled as @xmath108 . the rpv rainbow diagram is however scaled as @xmath109 , due to @xmath110 , where we have fixed @xmath92 . if the @xmath92-parameter have a supersymmetric dynamical origin @xcite and the @xmath92-parameter is scaled as @xmath82 , the rainbow diagram contribution will be scaled as @xmath108 . we must note that the rainbow diagram and the barr - zee type diagram can interfere constructively or destructively , depending on the phases @xmath111 , @xmath112 and the signs of the gluino mass parameter @xmath76 and of @xmath113 . the rpv barr - zee type contribution to the quark chromo - edm however dominates over the rainbow chromo - edm with external down quark and @xmath114 inner loop by one order of magnitude , so it is difficult to enhance or cancel the total contribution , and the result will not be much changed from the analysis taking into account the barr - zee type diagrams alone . in fig . [ fig : rpv_q - qrbnedm ] , we can see that the rpv contribution to the neutron edm ( plotted with @xmath115 ) largely exceed the currently known experimental data of the neutron edm ( @xmath116 ) @xcite . it is thus possible to constrain this combination of rpv couplings . if @xmath117 and the interference is constructive , we obtain the following limit @xmath118 for @xmath106 tev . contribution of the rpv rainbow diagram with external electron and @xmath119 inner loop ( absolute value ) to the electron edm plotted in the function of the supersymmetry breaking scale [ we have set @xmath120 . the barr - zee type contribution is also plotted for comparison . , width=302 ] the next case to consider is the rainbow diagram with external electron and charged lepton - slepton inner loop . in this case , the @xmath2 violating combinations of parameters im(@xmath121 ) and im(@xmath122 ) contribute to the electron edm . the rainbow diagram effect with external electron and @xmath123-@xmath124 inner loop , together with that of the rpv barr - zee type diagram , is plotted in fig . [ fig : rpv_l - lrbeedm ] . here we find that the rainbow and the barr - zee type contributions have similar size . this means that important enhancement or cancellation may occur , depending on the phase of @xmath125 and the sign of @xmath126 . in this case we do not have the suppression of the barr - zee type process by fractional charge . the comparable size of these two contributions is natural , since both involve no colored sparticles which were assumed to be heavier than uncolored sparticles . here we find again that the rainbow diagram decreases faster than the barr - zee type diagram in heavier supersymmetry breaking region . the reason is the same as for the down - quark edm and chromo - edm with quark - squark inner loop . as we can see in fig . [ fig : rpv_l - lrbeedm ] , this rpv contribution to the electron edm ( for @xmath127 ) exceeds largely the limit given by the experimental limit of the electron edm ( @xmath128 ) @xcite . if the interference is constructive , the limit of the rpv couplings @xmath129 is improved by about factor 1.6 from the analysis of the barr - zee type diagram alone ( for @xmath106 tev ) : @xmath130 if the interference is destructive , the limit to the rpv couplings will of course be loosened . contribution of the rpv rainbow diagram with external electron and @xmath102 inner loop ( absolute value ) to the electron edm plotted in the function of the supersymmetry breaking scale [ we have set @xmath131 . the barr - zee type contribution is also plotted for comparison . , width=302 ] the third case to consider is the rainbow diagram with external electron and quark - squark inner loop . in this case , the @xmath2 violating combinations of parameters im(@xmath132 ) and im(@xmath133 ) ( @xmath134 ) contribute to the electron edm . the rainbow diagram effect with external electron and bottom quark - squark inner loop is plotted in fig . [ fig : rpv_l - qrbeedm ] , with also the barr - zee type contribution for comparison . here again we find that the rainbow and barr - zee type contributions have similar size . this is due to the suppression of the barr - zee type diagram with a factor of the fractional charge of the bottom quark . as for the electron edm with lepton - slepton loop , the interference can enhance or suppress the total contribution . in this case , we can constrain the rpv couplings im(@xmath133 ) with the experimental data of the electron edm . if the interference is constructive and maximal , we obtain the following limit to the rpv couplings for @xmath135 tev : @xmath136 this limit is tighter than the constraint given by the experimental data of the @xmath137hg atom @xcite via the @xmath138 , @xmath2-odd electron - nucleon interaction @xcite @xmath139 this result is important since , the barr - zee type contribution alone could not give the tightest constraint . if the interference is destructive , the limit to the rpv couplings is loosened , and may even disappear . contribution of the rpv rainbow diagram with external down quark and @xmath140 inner loop ( absolute value ) to the neutron edm plotted in the function of the supersymmetry breaking scale [ we have set @xmath141 . the barr - zee type contribution is also plotted for comparison . , width=302 ] the final case to consider is the rainbow diagram with external down - quark and charged lepton - slepton inner loop . in this case , im(@xmath142 ) and im(@xmath143 ) ( @xmath134 ) contribute to the neutron edm . by again using the relation ( [ eq : qcdsumrule ] ) , we obtain the plot of fig . [ fig : rpv_q - lrbnedm ] . we see that although being comparable , the rpv rainbow diagram is smaller than the barr - zee type contribution . this is because the effect of the type 2 rainbow diagram @xmath144 is suppressed by the fractional charge of the quark [ see eq . ( [ eq : edmnug ] ) ] . as for the other cases , if the interference is constructive , the limit on the rpv couplings is tightened as @xmath145 for @xmath106 tev . on the contrary , the limit will be loosened for the destructive interference . in this paper , we have calculated a new sizable rpv contribution to the fermion edm , the rainbow diagram . the calculation was similar to that of the rainbow diagram of the minimal supersymmetric standard model , and we have given the formulae for it . it was found that the rpv rainbow diagrams have sizable effects compared with the currently known rpv barr - zee type diagrams , for the electron edm with @xmath102 or @xmath119 inner loops , and for the neutron edm with @xmath119 loop . we have also given new limits on the related combinations of rpv couplings when the interference is maximally constructive . the rpv rainbow diagrams have also many interesting properties , such as the large enhancement due to large @xmath1 , the cancellation for the equal mass eigenvalues of the inner loop sfermions , the @xmath109 scaling , etc . the interference between the barr - zee type and rainbow diagrams is dependent on the phases @xmath146 , @xmath147 and the signs of the gluino mass parameter and of @xmath148 . in this work , we did not considered the rpv rainbow diagrams with chargino propagation . as the matter - chargino interactions have comparable coupling with the matter - neutralino ones , it is highly probable that the rainbow diagrams with chargino have sizable effects . we should study this topic in the next work . the author thanks h. kamano for useful discussion and comments . in this section , we calculate the type 1 rainbow diagram with the insertion of the one - loop effective neutrino - gaugino - gauge boson vertex [ see fig . [ fig : rainbow_type ] ( 1 ) ] . the inner one - loop diagrams are depicted in fig . [ fig : nugg ] . one - loop contribution to the effective neutrino - gaugino - gauge boson vertex . , width=680 ] this contribution can generate both edm and chromo - edm . the amplitude of the effective neutrino - gaugino - gauge boson vertex is given by the sum of the diagrams shown in fig . [ fig : nugg ] : @xmath149 the off - shell amplitudes @xmath150 , @xmath151 , @xmath152 , and @xmath153 are given as : @xmath154\nonumber\\ & & \hspace{1.5em}+ 2g^{(n)}_{\tilde f_r } \bar \lambda_n ( k\hspace{-.5em}/\ , + q\hspace{-.45em}/_2 ) p_l \nu_i \cdot k^\mu \sin \theta_{f_j } \cos \theta_{f_j } e^{i\frac{\theta_n}{2 } - i\delta_{f_j } } \left [ d ( m_{\tilde f_{2j}}^2 , m_{\tilde f_{2j}}^2 , m _ { f_j}^2 ) - d ( m_{\tilde f_{1j}}^2 , m_{\tilde f_{1j}}^2 , m _ { f_j}^2 ) \right ] \biggl\ } \ , \label{eq : m(a)}\end{aligned}\ ] ] @xmath155\nonumber\\ & & \hspace{1.5em } -2g^{(n)}_{\tilde f_l } \bar \lambda_n ( k\hspace{-.5em}/\ , + q\hspace{-.5em}/_2 ) p_l \nu_i \cdot k^\mu \sin \theta_{f_j } \cos \theta_{f_j } e^{i\frac{\theta_n}{2 } - i\delta_{f_j } } \left [ d ( m_{\tilde f_{2j}}^2 , m_{\tilde f_{2j}}^2 , m _ { f_j}^2 ) - d ( m_{\tilde f_{1j}}^2 , m_{\tilde f_{1j}}^2 , m _ { f_j}^2 ) \right ] \biggl\ } \ , \label{eq : m(a')}\end{aligned}\ ] ] @xmath156 \nonumber\\ & & \hspace{1.5em } - g^{(n)}_{\tilde f_l } \bar \lambda_n ( k\hspace{-.5em}/\ , \gamma^\mu ( k\hspace{-.5em}/\ , - q\hspace{-.5em}/_1 ) + m_{f_j}^2 \gamma^\mu ) p_l \nu_i \cos \theta_{f_j } \sin \theta_{f_j } e^{i\frac{\theta_n}{2 } - i\delta_{f_j } } \left [ d ( m _ { f_j}^2 , m _ { f_j}^2 , m_{\tilde f_{2j}}^2 ) - d ( m _ { f_j}^2 , m _ { f_j}^2 , m_{\tilde f_{1j}}^2 ) \right ] \biggr\ } \ , , \nonumber\\ \label{eq : m(b ) } \\ % % % % % % % % % % % % % % % % m_(b ' ) % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % i{\cal m}_{\rm ( b ' ) } & = & \hat \lambda_{ijj } \sqrt{2 } g_f t_n \epsilon_\mu^ * ( q_1 ) \int \frac{d^4 k}{(2\pi ) ^4 } \nonumber\\ & & \times \biggl\ { -g^{(n)}_{\tilde f_l } m_{f_j } \bar \lambda_n ( 2k^\mu - \gamma^\mu q\hspace{-.5em}/_1 ) p_l \nu_i \ , e^{-i\frac{\theta_n}{2 } } \left [ \cos^2 \theta_{f_j } d ( m _ { f_j}^2 , m _ { f_j}^2 , m_{\tilde f_{1j}}^2 ) + \sin^2 \theta_{f_j } d ( m _ { f_j}^2 , m _ { f_j}^2 , m_{\tilde f_{2j}}^2 ) \right ] \nonumber\\ & & \hspace{1.5em}+ g^{(n)}_{\tilde f_r } \bar \lambda_n ( k\hspace{-.5em}/\ , \gamma^\mu ( k\hspace{-.5em}/\ , - q\hspace{-.5em}/_1 ) + m_{f_j}^2 \gamma^\mu ) p_l \nu_i \cos \theta_{f_j } \sin \theta_{f_j } e^{i\frac{\theta_n}{2 } - i\delta_{f_j } } \left [ d ( m _ { f_j}^2 , m _ { f_j}^2 , m_{\tilde f_{2j}}^2 ) - d ( m _ { f_j}^2 , m _ { f_j}^2 , m_{\tilde f_{1j}}^2 ) \right ] \biggr\ } \ , , \nonumber\\ \label{eq : m(b')}\end{aligned}\ ] ] where @xmath157 is the electromagnetic gauge coupling of the fermion @xmath34 or sfermion @xmath158 of the inner loop ( the strong gauge coupling @xmath159 for the chromo - edm ) , @xmath160 the gauge coupling to the fermion - sfermion pair [ see eq . ( [ eq : gaugino - int ] ) for the lagrangian ] . here @xmath161 is the rpv coupling , with @xmath162 when charged lepton - slepton runs in the loop , and @xmath58 for the case of down - type quark - squark loop . the factor @xmath163 is given by @xmath164 for the contribution with external photon and @xmath165 for external gluon . the momenta carried by the exiting gauge boson and gaugino are denoted by @xmath166 and @xmath167 , respectively . the gaugino and the neutrino spinors are given by @xmath38 and @xmath168 , respectively . note that the gaugino @xmath38 is majorana . the polarization vector of the external gauge boson is given by @xmath169 . here we have used the following notation to simplify the above equations : @xmath170 \left [ ( k - q_1)^2 - y \right ] \left [ ( k+q_2)^2 - z \right ] } \ .\ ] ] terms with @xmath171 and @xmath172 are the contribution from the mass insertion in the fermion propagator , and terms with @xmath173 are given by the mass insertion in the sfermion propagator . by integrating eqs . ( [ eq : m(a ) ] ) , ( [ eq : m(a ) ] ) , ( [ eq : m(b ) ] ) , and ( [ eq : m(b ) ] ) , we obtain the following one - loop level effective neutrino - gaugino - gauge boson vertex ( expanded up to the first order in the external momentum carried by the gauge boson ) : @xmath174 in writing the above off - shell amplitude , we have omitted several other dirac bilinears with different lorentz structure , such as @xmath175 , @xmath176 . these terms do not contribute to the edm operator @xmath177 after insertion into the second loop ( they cancel with their complex conjugates ) , so we do not consider them further . we have also neglected the quark and lepton masses . similarly , the off - shell effective antineutrino - gaugino - gauge boson amplitude is given by @xmath178 this amplitude is given by the complex conjugated rpv coupling @xmath179 . note that we also have to consider the amplitudes generated by transposed fermion interactions . they are given by @xmath180 here the momentum @xmath167 is of course that carried by the exiting neutrino or antineutrino . we should note that the one - loop level effective vertices ( [ eq : nulamg ] ) , ( [ eq : antinulamg ] ) , ( [ eq : lamnug ] ) , and ( [ eq : lamantinug ] ) are all first order in @xmath166 . we now insert the above effective vertices ( [ eq : nulamg ] ) , ( [ eq : antinulamg ] ) , ( [ eq : lamnug ] ) , and ( [ eq : lamantinug ] ) to the second loop , as shown in fig . [ fig : rainbow_type ] ( 1 ) . we obtain then the following contribution for the type 1 insertion ( expanded in the first order of @xmath166 ) : @xmath181 \left [ k^2 - m_{\tilde f_k}^2 \right ] } \int_0 ^ 1 dz \,\frac{z } { z k^2 - m_{\tilde f_{2j}}^2 } - ( m_{\tilde f_{2j}}^2 \leftrightarrow m_{\tilde f_{1j}}^2 ) \nonumber\\ & \approx & \frac{i}{64 \pi^3 } { \rm i m } ( \hat \lambda_{ijj } \tilde \lambda^*_{ikk } e^{i(\theta_n -\delta_{f_j } ) } ) \nonumber\\ & & \hspace{2em } \times \epsilon_\mu^ * ( q_1 ) \bar f_k \sigma^{\mu \nu } ( q_1)_\nu \gamma_5 f_k \sum_{\tilde f= \tilde f_l , \tilde f_r } s\ , g^{(n)}_{\tilde f } \left [ f'(|m_{\lambda_n}|^2 , 0 , m_{\tilde f_{k}}^2 , m_{\tilde f_{1j}}^2 \ , ) -f'(|m_{\lambda_n}|^2 , 0 , m_{\tilde f_{k}}^2 , m_{\tilde f_{2j}}^2 \ , ) \right ] \ , , \end{aligned}\ ] ] where we have omitted terms which do not contribute to the edm operator @xmath177 . the rpv coupling @xmath182 is given by @xmath59 for external lepton and @xmath183 for external quark . the constant @xmath61 is @xmath62 for left - handed sfermion @xmath184 and @xmath64 for right - handed sfermion @xmath185 . the functions @xmath69 and @xmath70 are defined by eqs . ( [ eq : f ] ) and ( [ eq : f ] ) , respectively . from the above equation , we obtain the formulae for the edm ( [ eq : edmnugg ] ) and for the chromo - edm ( [ eq : cedm ] ) . in this section , we calculate the second type rainbow diagram with the insertion of the one - loop effective neutrino - gaugino transition ( see fig . [ fig : rainbow_type ] ( 2 ) ) . the contributing one - loop diagrams are shown in fig . [ fig : nug ] . the one - loop level effective neutrino - gaugino transition within @xmath0-parity violation is given by @xmath186 \left [ ( k+q_2)^2 - m_{f_j}^2 \right ] } + \frac{g^{(n)}_{\tilde f_l } \sin^2 \theta_{f_j } + g^{(n)}_{\tilde f_r } \cos^2 \theta_{f_j } } { \left [ k^2 - m_{\tilde f_{2i}}^2 \right ] \left [ ( k+q_2)^2 - m_{f_j}^2 \right ] } \right\ } \nonumber\\ & & - \sqrt{2 } \hat \lambda_{ijj } n_c ( g^{(n)}_{\tilde f_l } + g^{(n)}_{\tilde f_r } ) \sin \theta_{f_j } \cos \theta_{f_j } e^{i\frac{\theta_n}{2 } - i\delta_{f_j } } \int \frac{d^4 k}{(2\pi ) ^4 } \left\ { \frac{\bar \lambda_n ( k\hspace{-.5em}/\ , + q\hspace{-.5em}/_2 ) p_l \nu_i } { \left [ k^2 - m_{\tilde f_{2i}}^2 \right ] \left [ ( k+q_2)^2 - m_{f_j}^2 \right ] } - ( m_{\tilde f_{2i}}^2 \leftrightarrow m_{\tilde f_{1i}}^2 ) \right\ } , \nonumber\\\end{aligned}\ ] ] where @xmath167 is the momentum carried by the incident fermion . the other notations are the same as for appendix [ sec : type1rainbow ] . it is important to note that the first line of the above equation is divergent . this divergence occurs because the first term is a one - loop correction to the tree - level rpv higgs - slepton transition ( note the mass insertion @xmath187 ) . as we have assumed that the bilinear rpv interaction can be rotated away , the divergent term is renormalized to zero . the second line is however finite and can not be renormalized away . this contribution does not give the same effective interaction as the renormalizable first term ( note the lorentz structure @xmath188 of the effective operator ) . it is only generated at the loop level through the trilinear rpv interactions . after renormalization , we obtain @xmath189 where we have also written the off - shell amplitudes given by the transposed and hermitian conjugated interactions . here again we have neglected the quark and lepton masses . by inserting the above effective vertices to the second loop as shown in fig . [ fig : rainbow_type ] ( 2 ) , we obtain the following contribution for the type 2 insertion ( expanded in the first order of the momentum @xmath190 carried by the external photon ) : @xmath191 \left [ ( k+q - p)^2 -m_{\tilde f_k}^2 \right ] \left [ ( k - p)^2 -m_{\tilde f_k}^2 \right ] } \int_0 ^ 1 dx \ , ( 1-x ) \ln \left ( \frac{xk^2 -m_{\tilde f_{1i}}^2 } { xk^2 -m_{\tilde f_{2i}}^2 } \right ) \nonumber\\ % % % % % % % % % % % % % % & \approx & \frac{2i}{(4\pi ) ^2 } { \rm i m } ( \hat \lambda_{ijj } \tilde \lambda^*_{ikk } e^{i(\theta_n -\delta_{f_j } ) } ) n_c q_f e ( g^{(n)}_{\tilde f_l } + g^{(n)}_{\tilde f_r } ) \sin \theta_{f_j } \cos \theta_{f_j } |m_{\lambda_n}| \epsilon^*_\mu ( q ) \nonumber\\ & & \hspace{4em}\times \sum_{\tilde f = \tilde f_l , \tilde f_r } \int \frac{d^4 k}{(2\pi)^4 } \frac{-g^{(n)}_{\tilde f } m_{\tilde f_k}^2 \bar f_k \gamma_5 f_k \cdot ( 2p^\mu -q^\mu ) } { \left [ k^2 - |m_{\lambda_n}|^2 \right ] \left [ k^2 -m_{\tilde f_k}^2 \right]^3 } \int_0 ^ 1 dx \ , ( 1-x ) \ln \left ( \frac{xk^2 - m_{\tilde f_{1i}}^2 } { xk^2 - m_{\tilde f_{2i}}^2 } \right ) \nonumber\\ % % % % % % % % % % % % % % % & \approx & \frac{i } { ( 4\pi ) ^4 } { \rm i m } ( \hat \lambda_{ijj } \tilde \lambda^*_{ikk } e^{i(\theta_n -\delta_{f_j } ) } ) n_c q_f e ( g^{(n)}_{\tilde f_l } + g^{(n)}_{\tilde f_r } ) \sin \theta_{f_j } \cos \theta_{f_j } |m_{\lambda_n}| \epsilon^*_\mu ( q ) \bar f_k \sigma^{\mu \nu } q_\nu \gamma_5 f_k \nonumber\\ & & \hspace{8em } \times \sum_{\tilde f= \tilde f_l,\tilde f_r } g^{(n)}_{\tilde f } m_{\tilde f_k}^2 \left [ f''(|m_{\lambda_n}|^2 , 0 , m_{\tilde f_k}^2 , m_{\tilde f_{1i}}^2 ) -f''(|m_{\lambda_n}|^2 , 0 , m_{\tilde f_k}^2 , m_{\tilde f_{2i}}^2 ) \right ] \ , \end{aligned}\ ] ] where the function @xmath70 is defined in eq . ( [ eq : f ] ) , and the external momentum brought by the fermion @xmath50 is given by @xmath192 . this leads to the formula ( [ eq : edmnug ] ) . h. e. haber and g. l. kane , phys . rept . * 117 * , 75 ( 1985 ) ; j. f. gunion and h. e. haber , nucl . b * 272 * , 1 ( 1986 ) ; s. p. martin , in _ perspectives on supersymmetry ii _ , edited by g. l. kane ( world scientific , singapore , 2010 ) , p. 1 [ arxiv : hep - ph/9709356 ] . j. ellis and m. k. gaillard , nucl . b * 150 * , 141 ( 1979 ) ; e. p. shabalin , sov . . phys . * 31 * , 864 ( 1980 ) ; i. b. khriplovich and a. r. zhitnitsky , phys . b * 109 * , 490 ( 1982 ) . i. b. khriplovich , sov . j. nucl * 44 * , 659 ( 1986 ) ; phys . b * 173 * , 193 ( 1986 ) ; m. e. pospelov and i. b. khriplovich , sov . j. nucl . phys . * 53 * , 638 ( 1991 ) [ yad . fiz . * 53 * , 1030 ( 1991 ) ] ; x .- g . he and b. mckellar , phys . d * 46 * , 2131 ( 1992 ) ; x .- g . he , b. h. j. mckellar , s. pakvasa , phys . b * 283 * , 348 ( 1992 ) ; a. czarnecki and b. krause , phys . lett . * 78 * , 4339 ( 1997 ) ; t. mannel and n. uraltsev , phys . d * 85 * , 096002 ( 2012 ) . j. r. ellis , s. ferrera and d. v. nanopoulos , phys . b * 114 * , 231 ( 1982 ) ; w. buchm@xmath193ller and d. wyler , phys . b * 121 * , 321 ( 1983 ) ; j. polchinski and m. b. wise , phys . b * 125 * , 393 ( 1983 ) ; f. del aguila , m. b. gavela , j. a. grifols and a. mendez , phys . b * 126 * , 71 ( 1983 ) ; d. v. nanopoulos and m. srednicki , phys . lett . b * 128 * , 61 ( 1983 ) ; m. dugan , b. grinstein and l. j. hall , nucl . b * 255 * , 413 ( 1985 ) ; p. nath , phys . * 66 * , 2565 ( 1991 ) ; y. kizukuri and n. oshimo , phys . d * 45 * , 1806 ( 1992 ) ; phys . d * 46 * , 3025 ( 1992 ) ; w. fischler , s. paban and s. d. thomas , phys . b * 289 * , 373 ( 1992 ) ; t. inui , y. mimura , n. sakai and t. sasaki , nucl . b * 449 * , 49 ( 1995 ) ; t. ibrahim and p. nath , phys . d * 57 * , 478 ( 1998 ) ; phys . b * 418 * , 98 ( 1998 ) ; phys . d * 58 * , 111301 ( 1998 ) ; s. pokorski , j. rosiek and c. a. savoy , nucl . b * 570 * , 81 ( 2000 ) ; s. y. ayazi and y. farzan , phys . d * 74 * , 055008 ( 2006 ) ; jhep * 0706 * , 013 ( 2007 ) ; t. h. west , phys . d * 50 * , 7025 ( 1994 ) ; t. kadoyoshi and n. oshimo , phys . rev . d * 55 * , 1481 ( 1997 ) ; d. chang , w .- y . keung and a. pilaftsis , phys . lett . * 82 * , 900 ( 1999 ) ; a. pilaftsis , phys . b * 471 * , 174 ( 1999 ) ; d. chang , w .- f . chang and w .- y . keung , phys . b * 478 * , 239 ( 2000 ) ; d. chang , w .- f . chang and w keung , phys . d * 66 * , 116008 ( 2002 ) ; t .- f . feng , x .- q . li , j. maalampi and x. zhang , phys . d * 71 * , 056005 ( 2005 ) ; n. arkani - hamed , s. dimopoulos , g.f . giudice and a. romanino , nucl . b * 709 * , 3 ( 2005 ) ; d. chang , w .- f . chang and w .- y . keung , phys . d * 71 * , 076006 ( 2005 ) ; g. f. giudice and a. romanino , phys . lett . b * 634 * , 307 ( 2006 ) ; t .- f . feng , x .- q . li , l. lin , j. maalampi and h .- s . song , phys . d * 73 * , 116001 ( 2006 ) ; y. li , s. profumo and m. j. ramsey - musolf , phys . d * 78 * , 075009 ( 2008 ) . j. dai , h. dykstra , r. g. leigh , s. paban and d. dicus , phys . b * 237 * , 216 ( 1990 ) ; t. falk , k. a. olive , m. pospelov and r. roiban , nucl . b * 560 * , 3 ( 1999 ) ; m. brhlik , g. j. good and g. l. kane , phys . rev . d * 59 * , 115004 ( 1999 ) ; s. abel , s. khalil and o. lebedev , nucl . b * 606 * , 151 ( 2001 ) ; a. pilaftsis , nucl . b * 644 * , 263 ( 2002 ) ; o. lebedev and m. pospelov , phys . lett . * 89 * , 101801 ( 2002 ) ; d. demir , o. lebedev , k. a. olive , m. pospelov and a. ritz , nucl . phys . b * 680 * , 339 ( 2004 ) ; o. lebedev , k. a. olive , m. pospelov and a. ritz , phys . rev . d * 70 * , 016003 ( 2004 ) ; k. a. olive , m. pospelov , a. ritz and y. santoso , phys . d * 72 * , 075001 ( 2005 ) ; g. degrassi , e. franco , s. marchetti and l. silvestrini , jhep * 0511 * , 044 ( 2005 ) ; s. abel and o. lebedev , jhep * 0601 * , 133 ( 2006 ) ; j. ellis , j. s. lee and a. pilaftsis , jhep * 0810 * , 049 ( 2008 ) ; jhep * 1010 * , 049 ( 2010 ) ; jhep * 1102 * , 045 ( 2011 ) ; y. li , s. profumo and m. j. ramsey - musolf , jhep * 1008 * , 062 ( 2010 ) ; t. fukuyama , int . j. mod a * 27 * , 1230015 ( 2012 ) . j. hisano and y. shimizu , phys . d * 70 * , 093001 ( 2004 ) ; m. endo , m. kakizaki and m. yamaguchi , phys . b * 583 * , 186 ( 2004 ) ; g .- c . cho , n. haba and m. honda , mod . a * 20 * , 2969 ( 2005 ) ; j. hisano , m. nagai and p. paradisi , phys . lett . b * 642 * , 510 ( 2006 ) ; phys . d * 78 * , 075019 ( 2008 ) ; phys . d * 80 * , 095014 ( 2009 ) ; w. altmannshofer , a. j. buras and p. paradisi , phys . b * 688 * , 202 ( 2010 ) . r. m. godbole , s. pakvasa , s. d. rindani and x. tata , phys . d * 61 * , 113003 ( 2000 ) ; s. a. abel , a. dedes and h. k. dreiner , jhep * 0005 * , 13 ( 2000 ) ; d. chang , w .- f . chang , m. frank and w .- y . keung , phys . d * 62 * , 095002 ( 2000 ) . k. choi , e. j. chun and k. hwang , phys . d * 63 * , 013002 ( 2000 ) ; y.y . keum and otto c.w . kong , phys . rev . d * 63 * , 113012 ( 2001 ) ; c .- c . chiou , o. c. w. kong and r. d. vaidya , phys . d * 76 * , 013003 ( 2007 ) . g. bhattacharyya , arxiv : hep - ph/9709395 ; h. k. dreiner , _ in perspectives on supersymmetry ii _ , edited by g. l. kane ( world scientific , singapore , 1997 ) , p. 565 ; r. barbier _ et al . _ , . rept . * 420 * , 1 ( 2005 ) ; m. chemtob , prog . part . phys . * 54 * , 71 ( 2005 ) . atlas collaboration ( georges aad _ et al . _ ) , phys . rev . * 108 * , 261804 ( 2012 ) ; arxiv:1208.4688 [ hep - ex ] ; arxiv:1210.1314 [ hep - ex ] ; cms collaboration ( vardan khachatryan _ _ ) , phys . rev . lett . * 107 * , 221804 ( 2011 ) ; arxiv:1210.2052 [ hep - ex ] ; arxiv:1210.8115 [ hep - ex ] .

one of the most important problems today in cosmology is the nature of dark matter and dark energy that must dominate the matter content of the universe @xcite . the existence of dark matter is suggested by the anomalies in the dynamics of galaxies and clusters of galaxies @xcite . dark energy seems to be an inevitable consequence of the present acceleration of the universe as indicated by the type ia supernovae ( sne ia ) data , which asks for a fluid of negative pressure which does not agglomerate at small scales @xcite . while in general the nature of dark matter can be connected with relics of fundamental theories , like axions @xcite , there are two main candidates to represent dark energy : cosmological constant @xcite and quintessence which is a self interacting scalar field @xcite@xcite . both models suffers of many drawbacks , mainly linked with fine tuning of parameters , as in the quintessence model @xcite , or disagreement with the theoretically predicted values , like in the cosmological constant case @xcite . the traditional chaplygin gas model ( cgm ) and the generalized chaplygin gas model ( gcgm ) have been widely considered as alternatives to the cosmological constant and to quintessence as the dark energy that drives the present acceleration of the universe @xcite@xcite . the appealing of the gcgm comes from , among other reasons , the fact that it can unify the description of dark matter and dark energy . the generalized chaplygin gas ( gcg ) is characterized by the equation of state @xmath14 where @xmath15 and @xmath16 are constants that in general are considered as positive numbers . the conservation equation leads to the relation @xmath17 some of the main interesting features of eq . ( [ densidade ] ) are the following . initially , the gcgm behaves as a dust fluid , with @xmath18 , while at late times the gcgm behaves as a cosmological constant term , @xmath19 . hence , the gcgm interpolates a matter dominated phase ( where the formation of structure can happen ) and a de sitter phase . at the same time , the sound velocity of the gcgm is positive which assures that , in spite of exhibiting a negative pressure , it is stable against perturbations of the background configuration @xcite . a lot of effort has been done in order to constrain the free parameters of the gcgm . type ia supernovae data @xcite@xcite , the spectra of anisotropy of cosmic microwave background radiation @xcite , the mass power spectrum @xcite , gravitational lenses @xcite , x - ray data @xcite and age estimates of high-@xmath20 objects @xcite have extensively been used in this sense . this is essential for many obvious reasons . one of them is that if the parameter @xmath16 is close to zero , the gcgm reduces essentially to the cosmological constant model . if observations indicate that this is the preferred value , the appealing for the cosmological constant as dark energy is reinforced . in a previous work @xcite , we have analysed the case of the traditional chaplygin gas model , where @xmath21 , using the sne ia data . one important aspect of this analysis is the use of the bayesian statistics , for one side , and also the fact that all free parameters ( hubble constant , the curvature term , the value of the sound velocity and the proportion of ordinary matter and chaplygin gas ) of the model were taken into account in the treatment of the problem . this contrasts with most of the analyses previously found in the literature in the sense that they employ a more simplified statistical analysis or some input parameters were fixed a priori @xcite@xcite . in that work @xcite , it was concluded that the unification scenario ( i.e. , without pressureless matter ) , and a sound velocity near ( but not equal ) to the velocity of light were preferred . moreover , a closed universe was clearly favoured . the aim of the present work is to apply the same analysis , using the bayesian statistics and taking into account all free parameters , to the gcgm . in order to keep contact with the previous work , we will use the same supernovae sample . this sample is much smaller than those available today ( see , for example , ref . @xcite@xcite ) . but , the quality of the data is extremely good , which leads to a smaller value of @xmath22 ( the quantity that measures the quality of the fitting ) : with this restricted sample the best - fitting gives @xmath22 around @xmath23 , while for the complete `` gold '' sample @xcite , the value mounts to unity . in the gcgm there are five free parameters instead of the four free parameters of the traditional chaplygin gas model ( @xmath21 ) : the hubble constant @xmath24 , the parameter @xmath0 related to the sound velocity of the chaplygin gas , the curvature density parameter @xmath25 , the ordinary matter parameter @xmath26 ( or alternatively , the chaplygin gas parameter @xmath27 ) and the equation of state parameter @xmath16 . one important point in performing this generalized analysis is the range of validity of the parameter @xmath16 . usually , it has been assumed that @xmath28 . the reason is twofold : @xmath16 greater than one could lead , even if in the future , to a sound velocity greater than the velocity of the light ; @xmath16 negative should lead to an imaginary sound velocity , that is , instability . however , an analysis of the supernovae data may taken into account values of @xmath16 greater than one or negative . in particular , in reference @xcite , using a different statistical approach with respect to the method employed here , a value of @xmath29 near @xmath30 was favoured . in fact , values of @xmath16 outside that limited range can be considered . the reason lies in the fact that the gcgm with the equation of state ( [ eos ] ) may be a phenomenological manifestation of a theory expressed in terms , for example , of scalar field with a potential , or even of tachyon condensate model @xcite . in this case , the notion of sound velocity can not be identified with the ordinary expression used in the perfect fluid approach , and no violation of causality or instability may occur if @xmath16 is greater than one or negative . hence , in our analysis the parameter @xmath16 will vary from negative to highly positive values . the prediction for each parameter ( @xmath16 , @xmath24 , @xmath25 , @xmath26 , @xmath27 and @xmath0 ) is taken marginalizing , i.e. , integrating over the other parameters . we stress that such marginalization procedure gives predictions that can be quite different from a two dimensional analysis , for example , where one searches for the simultaneous preferred values for any two parameters . this marginalization is in fact essential in order to have the correct prediction for a given free parameter . we will consider the following cases : all five parameters free ; spatial curvature zero ( four free parameters ) ; pressureless matter given only by the barionic component ( four free parameters ) ; pressureless matter absent ( four free parameters ) ; spatial curvature zero and no pressureless matter component ( three free parameters ) ; spatial curvature zero and pressureless matter component given by the barionic component ( three free parameteres ) . for the more general case with five free parameters the predictions we obtain are the following : @xmath31 , @xmath32 , @xmath5 , @xmath33 , @xmath34 , @xmath35 . moreover the predicted age of the universe and the value of de deceleration parameter today are @xmath36 and @xmath37 , respectively . the results are consistent with , for example , the age of globular clusters @xcite . two crucial parameters are @xmath16 and @xmath38 , since if @xmath39 and/or @xmath40 are favoured , the gcgm becomes equivalent to the cosmological constant model ( @xmath41cdm ) . the value of the parameter @xmath0 is in fact peaked at unity . but for reasons discussed later , this does not imply that this is the most favoured value ; indeed it is just expected to be near unity . on the other hand , the parameter @xmath16 is peaked in a negative value , but the spread is very large . in particular , the traditional chaplygin gas with @xmath42 can not be excluded or even disfavoured : in most of the cases this value of @xmath16 is favoured with a probability around @xmath43 . hence , our analysis does not exclude the traditional chaplygin gas model . this paper is organized as follows . in next section , the basic equations and quantities are exhibited . section @xmath44 explains the type ia supernovae fitting . in section @xmath45 , we display the results of the bayesian analysis of type ia supernovae data with the gcgm . we present our conclusions in section @xmath46 . regardless the possibility of unification of dark matter and dark energy using the gcgm , we will consider a model for the universe today with two fluids : pressureless matter and the gcg . in doing so , our aim is to verify whether the unified model is the case favoured by observations . moreover , even if in the gcgm there is no need of a dark matter component , there is baryonic matter in the universe which is represented also by a pressureless matter . hence , the dynamics of the universe is driven by the friedmann s equation @xmath47 where @xmath48 and @xmath49 stand for the pressureless matter and generalized chaplygin gas component . dot means derivative with respect to the cosmic time @xmath50 . as usual , @xmath51 indicates a flat , closed and open spatial section . in the conservation equation for the generalized chaplygin gas component ( [ be5 ] ) , the equation of state ( [ eos ] ) has been introduced . the equations expressing the conservation law for each fluid ( [ be4])-([be5 ] ) lead to @xmath52 henceforth , we will parametrize the scale factor such that its value today is equal to unity , @xmath53 . hence , @xmath54 and @xmath55 are the pressureless matter and gcg densities today . eliminating from the last relation the parameter @xmath56 , the gcg density at any time can be reexpressed as @xmath57 where @xmath58 . this parameter @xmath0 is connected with the sound velocity for the gcg today by the relation @xmath59 . the luminosity distance is given by the expression @xmath60 @xmath61 being the co - moving coordinate of the source . using the expression for the propagation of light @xmath62 and the friedmann s equation ( [ be3 ] ) , we can reexpress the luminosity distance as @xmath63\quad , \ ] ] where @xmath64 and @xmath65^{1/(1+\alpha ) } -\omega _ { k0}(z^{\prime } + 1)^{2}\}^{1/2}}\quad , \ ] ] with the definitions @xmath66 where @xmath67 . the final equations have been also expressed in terms of the redshift @xmath68 . the age of the universe today can be expressed as @xmath69^{1/(1+\alpha ) } -\omega _ { k0}(z^{\prime } + 1)^{2}\}^{1/2}}\quad , \ ] ] where @xmath70 , so @xmath71 has units of years . the value of the decelerated parameter today , @xmath72 , is given by @xmath73 the value @xmath74 of the scale factor that shows the beginning of the recent accelerating phase of the universe is given by the roots of the equation @xmath75 and is related to the redshift value @xmath76 such that @xmath77 or @xmath78 . we proceed by fitting the sne ia data using the gcg model described above . essentially , we compute the quantity distance moduli , @xmath79 and compare the same distance moduli as obtained from observations . the quality of the fitting is characterized by the @xmath80 parameter of the least - squares statistic , as defined in ref . @xcite , @xmath81 in this expression , @xmath82 is the measured value , @xmath83 is the value calculated through the model described above , @xmath84 is the measurement error , @xmath85 is the dispersion in the distance modulus due to the dispersion in galaxy redshift due to peculiar velocities . this quantity we will taken as @xmath86 where , following ref . @xcite , @xmath87 . in this article , we have used the same @xmath88 type ia supernovae in table @xmath2 of ref . @xcite , so we do not need to analyse and compare again the cgm with the @xmath41cdm ( this work has been done in ref . @xcite ) , but instead we can concentrate our efforts on the gcgm . although nowadays there are approximately @xmath89 sne ia available , the choice of these @xmath88 are confirmed to be of excellent quality by analysing the low values of @xmath90 ( the estimated errors for degree of freedom , i.e. , @xmath80 divided by @xmath88 , the number of sne ia ) , shown in table [ tablebestfitgcg ] . @xmath91 values range between @xmath92 to @xmath93 , while other larger sne ia samples have been used in many cosmological models @xcite , typically resulting @xmath91 near the unity . .the best - fitting parameters , i.e. , when @xmath94 is minimum , for each type of spatial section and matter content of the generalized chaplygin gas model . @xmath24 is given in @xmath95 , @xmath96 in units of @xmath97 , @xmath71 in @xmath98 and @xmath74 in units of @xmath99 . @xmath100 and @xmath101 are the constants of the relation @xmath102 . [ cols="^,^,^,^,^,^,^ " , ] table [ tableparestgcg ] is very comprehensive and resumes our results , listing all the bayesian parameter estimations for six different cases of spatial section and matter content within the context of generalized chaplygin gas model . the following sub - sections will analyse in detail this table and accompanying figures . more emphasis is spent into the analyses of the generalized chaplygin gas parameter @xmath16 , because it is the subject of constant debate . in figures [ figsalpha ] , [ figsa ] and [ figsomegas ] , the one - dimensional pdf shapes are matched against half - gaussians to estimate the probabilities outside the left and right edges , in this way increasing the precision of the parameter estimations without excessively enlarging the parameter space . the parameter estimation of @xmath16 is strongly related to the @xmath0 parameter , i.e. , the joint pdfs for the parameter space of @xmath104 do not have a gaussian shape , so the marginalization process changes the peak values and credible regions . table [ tableparestgcg ] , figures [ figsalphaa ] and [ figsalpha ] are carefully analysed below with respect to @xmath16 . in brief , the estimations for @xmath16 are quite spread ( i.e. , they have large credible regions ) and they differ a lot depending on if two or one - dimensional parameter space is used . for the generalized chaplygin gas model . the joint pdf peak is shown by the large dot , the credible regions of @xmath105 ( @xmath106 ) by the red dotted line , the @xmath107 ( @xmath108 ) in blue dashed line and the @xmath109 ( @xmath110 ) in green dashed - dotted line . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the joint pdf peak is shown by the large dot , the credible regions of @xmath105 ( @xmath106 ) by the red dotted line , the @xmath107 ( @xmath108 ) in blue dashed line and the @xmath109 ( @xmath110 ) in green dashed - dotted line . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the joint pdf peak is shown by the large dot , the credible regions of @xmath105 ( @xmath106 ) by the red dotted line , the @xmath107 ( @xmath108 ) in blue dashed line and the @xmath109 ( @xmath110 ) in green dashed - dotted line . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the joint pdf peak is shown by the large dot , the credible regions of @xmath105 ( @xmath106 ) by the red dotted line , the @xmath107 ( @xmath108 ) in blue dashed line and the @xmath109 ( @xmath110 ) in green dashed - dotted line . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] in figures [ figsalphaa ] , the gcgm with @xmath46 free parameters has a joint pdf peak of @xmath113 at @xmath114 with @xmath115 ( @xmath116 ) , @xmath117 ( @xmath118 ) and @xmath119 ( @xmath120 ) credible region contours with pdf values of @xmath121 , @xmath122 and @xmath123 , respectively . the two - dimensional credible regions are not gaussian - shaped , and their extremes give the estimations @xmath124 at @xmath115 , @xmath125 at @xmath117 , and @xmath126 at @xmath127 confidence levels . the symbol ( @xmath128 ) in the upper credible value for @xmath16 is due to the fact that the graphics is limited to @xmath129 . the marginalization ( integration ) of @xmath130 over the @xmath0 space yields @xmath131 , shown in figure [ figsalpha ] , where the credible intervals are @xmath132 at @xmath115 and @xmath133 at @xmath117 ( the @xmath127 pdf level can not be seen with @xmath134 ) . the pdf peak of @xmath135 at @xmath136 is greater than the pdf value of @xmath137 for @xmath21 ( ordinary chaplygin gas model ) , so the region with pdf greater than @xmath137 has a cdf of @xmath138 , i.e. , we can say that @xmath21 is ruled out at a @xmath139 ( @xmath140 ) confidence level ( or it is preferred at @xmath141 ) . table [ tableparestgcg ] shows that positive values of @xmath16 are more likely at a @xmath142 ( @xmath143)confidence level . the @xmath45-parameter case with @xmath144 has a joint pdf peak of @xmath145 at @xmath146 with @xmath115 , @xmath117 and @xmath119 credible region contour levels of @xmath147 , @xmath148 and @xmath149 , respectively . the two - dimensional credible regions give the estimations @xmath150 at @xmath151 , @xmath152 at @xmath117 , and @xmath153 at @xmath154 confidence levels . the estimation based on @xmath131 yields @xmath155 at @xmath115 and @xmath156 at @xmath117 confidence levels . the pdf peak of @xmath157 at @xmath158 is greater than the pdf value of @xmath159 for @xmath21 such that @xmath160 is ruled out at a @xmath161 ( @xmath162 ) confidence level . positive values of @xmath16 are preferred at a @xmath163 ( @xmath164 ) confidence level . for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines and the @xmath167 ( @xmath168 ) credible regions are given by blue dashed lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines and the @xmath167 ( @xmath168 ) credible regions are given by blue dashed lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines and the @xmath167 ( @xmath168 ) credible regions are given by blue dashed lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines and the @xmath167 ( @xmath168 ) credible regions are given by blue dashed lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] the joint pdf has a peak value of @xmath169 at @xmath170 with @xmath115 , @xmath117 and @xmath119 credible region contour levels of @xmath171 , @xmath172 and @xmath149 , respectively . these levels indicate that the joint two - dimensional pdf is strongly peaked . the two - dimensional credible regions give the estimations @xmath173 at @xmath115 , @xmath174 at @xmath117 , and @xmath175 at @xmath127 confidence levels . the estimations based on one - dimensional pdf yield the same peak value ( due to the highly peaked joint pdf ) , with @xmath176 at @xmath151 and @xmath177 at @xmath117 confidence levels . the pdf peak of @xmath178 at @xmath179 is greater than the pdf value of @xmath180 for @xmath21 , and we can say that @xmath21 is preferred at a @xmath181 ( @xmath182 ) confidence level . positive values of @xmath16 are preferred at a @xmath183 ( @xmath184)confidence level . the joint pdf has a peak value of @xmath185 at @xmath186 with @xmath115 , @xmath117 and @xmath119 credible region contour levels of @xmath187 , @xmath188 and @xmath189 , respectively , which show a highly peaked behaviour . the two - dimensional credible regions give the estimations @xmath190 at @xmath151 , @xmath191 at @xmath117 , and @xmath192 at @xmath127 confidence levels . the estimations based on one - dimensional pdf give @xmath193 at @xmath115 and @xmath194 at @xmath117 confidence levels . the pdf peak of @xmath195 at @xmath196 is greater than the pdf value of @xmath197 for @xmath21 , and we can say that @xmath21 is preferred at a @xmath198 ( @xmath199 ) confidence level . positive values of @xmath16 are preferred at a @xmath200 ( @xmath201 ) confidence level . the joint pdf has a peak value of @xmath202 at @xmath203 with @xmath115 , @xmath117 and @xmath119 credible region contour levels of @xmath204 , @xmath205 and @xmath149 , respectively . the two - dimensional credible regions give the estimations @xmath206 at @xmath115 , @xmath207 at @xmath117 , and @xmath208 at @xmath127 confidence levels . the estimations based on one - dimensional pdf give @xmath209 at @xmath115 and @xmath210 at @xmath117 confidence levels . the pdf peak of @xmath211 at @xmath212 is greater than the pdf value of @xmath213 for @xmath21 , and we can say that @xmath214 is preferred at a @xmath215 ( @xmath216 ) confidence level . positive values of @xmath16 are preferred at a @xmath217 ( @xmath218 ) confidence level . the joint pdf has a peak value of @xmath219 at @xmath220 with @xmath115 , @xmath117 and @xmath119 credible region contour levels of @xmath221 , @xmath222 and @xmath149 , respectively . the two - dimensional credible regions give the estimations @xmath223 at @xmath115 , @xmath224 at @xmath117 , and @xmath225 at @xmath127 confidence levels . the estimations based on one - dimensional pdf give @xmath226 at @xmath115 and @xmath227 at @xmath117 confidence levels . the pdf peak of @xmath228 at @xmath229 is greater than the pdf value of @xmath230 for @xmath21 , and we can say that @xmath21 is preferred at a @xmath161 ( @xmath162 ) confidence level . positive values of @xmath16 are preferred at a @xmath231 ( @xmath232 ) confidence level . the non - gaussian shape of the credible region contours in the @xmath233 two - dimensional parameter space is the culprit of this difference for the estimations of @xmath16 , so it is crucial to emphasize how the @xmath16 parameter is estimated : from a joint two - dimensional pdf or an one - dimensional pdf ( after marginalization of the joint two - dimension pdf ) . the most robust parameter estimation is based on the one - dimensional pdf , therefore giving small importance to regions in the joint two - dimensions which have a small joint pdf volume ( cdf ) . comparing the estimations , the ones based on two - dimensional parameter space always give larger values for @xmath29 , after marginalization the pdf is peaked at lower values of @xmath234 . we also expect that a large sample of sne ia will narrow the credible regions of @xmath16 , so the issue of a ordinary chaplygin gas model ( cgm ) being preferred or ruled out will be probably better addressed . the generalized chaplygin gas model ( gcgm ) estimations for @xmath0 are given by table [ tableparestgcg ] , which if compared with the results of the cgm ( table @xmath46 of ref . @xcite ) clearly indicate that the pdf peak is closer to @xmath40 and the dispersion is increased when the gcgm is considered . this behaviour is due to the shape of the joint pdf in the parameter space @xmath235 , figures [ figsalphaa ] explicitly show that as @xmath16 increases there is a narrower region of high pdf values closer to higher values of @xmath0 . but the estimations for @xmath0 shown in table [ tableparestgcg ] and figures [ figsa ] have to be interpreted . they do not indicate that the @xmath236 limit ( @xmath40 ) is favoured because the resolution for the @xmath38 sampling is @xmath237 , i.e , a pdf peak at @xmath238 means the peak happens for @xmath239 indeed . therefore , in figures [ figsa ] , the graphics for the pdf of @xmath0 which are peaked at @xmath240 have the shape of the graphics for the case @xmath112 , with the pdf peak closer to @xmath40 . @xcite for a detailed comparison between the cgm and @xmath241 cases . for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath242 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath242 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath242 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath165 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath242 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . the cases for @xmath111 are not shown here because they are similar to the ones with @xmath112 . ] the analyses of @xmath16 presented above have considered the joint pdfs for the two - dimensional parameter space of @xmath235 , and the extremes of the credible regions have given estimations for @xmath235 at @xmath151 , @xmath117 , and @xmath154 confidence levels . these estimations for @xmath0 are almost always different from the estimations based on the one - dimensional pdf of @xmath0 , obtained by marginalizing @xmath245 over the @xmath16 space . we stress again that the most robust estimations are based on the one - dimensional pdfs . the general behaviour of table [ tableparestgcg ] is that the pdf peak is at @xmath246 and the @xmath117 credible interval begins at @xmath247 and finishes at @xmath248 . the parameter estimation of @xmath24 shown in table [ tableparestgcg ] is almost the same for all cases , around @xmath249 with a narrow credible region of @xmath250 . this estimation , like the one for the cgm in table @xmath46 of ref . @xcite , is compatible with other sne ia analyses . table [ tableparestgcg ] and figure [ figsomegas ] show that the curvature density parameter @xmath25 estimation for the gcgm is almost like for the cgm of ref . @xcite , i.e. , it also strongly depends on the cold dark matter content of the gcgm . in the @xmath46-parameter case a closed universe ( @xmath251 ) is preferred at @xmath252 ( @xmath253 ) and a spatially flat universe is ruled out at @xmath254 ( @xmath255 ) confidence level , i.e. , the pdf of @xmath256 is smaller than the pdf of @xmath257 and this region has a cdf of @xmath254 ( @xmath255 ) . we can also see the figure [ figomegam0c0 ] that the integral of the joint pdf over the upper region gives the same high likelihood of a closed universe , although the joint pdf peak is located in the open universe region . for the generalized chaplygin gas model , where @xmath258 is a integral of @xmath259 over the @xmath260 parameter space . it shows a different shape for the confidence regions with respect to the @xmath241 but similar to the cgm of ref . the joint pdf peak has the value @xmath261 for @xmath262 ( shown by the large dot ) , the credible regions of @xmath263 ( @xmath106 , shown in red dotted line ) , @xmath264 ( @xmath108 , in blue dashed line ) and @xmath109 ( @xmath265 , in green dashed - dotted line ) have pdf levels of @xmath266 , @xmath267and @xmath268 , respectively . as @xmath269 , the probability for a spatially flat universe is on the line @xmath270 , above it we have the region for a closed universe ( @xmath271 , @xmath251 ) , and below , the region for an open universe ( @xmath272 , @xmath273 ) . ] but assuming the hypothesis that @xmath112 , a @xmath45-parameter case , the behaviour is opposite , a closed universe ( @xmath251 ) is preferred at @xmath274 ( @xmath275 ) and a spatially flat universe is preferred at @xmath276 ( @xmath277 ) confidence level , i.e. , the pdf of @xmath256 is smaller than the pdf of @xmath278 and this region has a small cdf of @xmath279 . therefore , if we impose a low value for @xmath280 then the spatially flat universe is favoured . , @xmath280 and @xmath27 for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath281 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath282 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . as @xmath283 , for @xmath111 we have @xmath284 , for @xmath112 then @xmath285 and for @xmath286 we also have @xmath287 . ] , @xmath280 and @xmath27 for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath281 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath282 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . as @xmath283 , for @xmath111 we have @xmath284 , for @xmath112 then @xmath285 and for @xmath286 we also have @xmath287 . ] , @xmath280 and @xmath27 for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath281 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath282 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . as @xmath283 , for @xmath111 we have @xmath284 , for @xmath112 then @xmath285 and for @xmath286 we also have @xmath287 . ] , @xmath280 and @xmath27 for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath281 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath282 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . as @xmath283 , for @xmath111 we have @xmath284 , for @xmath112 then @xmath285 and for @xmath286 we also have @xmath287 . ] , @xmath280 and @xmath27 for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath281 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath282 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . as @xmath283 , for @xmath111 we have @xmath284 , for @xmath112 then @xmath285 and for @xmath286 we also have @xmath287 . ] , @xmath280 and @xmath27 for the generalized chaplygin gas model . the solid lines are the pdf , the @xmath281 ( @xmath166 ) regions are delimited by red dotted lines , the @xmath282 ( @xmath168 ) credible regions are given by blue dashed lines and the @xmath243 ( @xmath244 ) regions are delimited by green dashed - dotted lines . as @xmath283 , for @xmath111 we have @xmath284 , for @xmath112 then @xmath285 and for @xmath286 we also have @xmath287 . ] the generalized chaplygin gas model favours small values of @xmath26 like the ordinary cgm , see table [ tableparestgcg ] and table @xmath46 of ref . the graphics for @xmath26 in figure [ figsomegas ] , for the @xmath46-parameter case and @xmath45-parameter case with @xmath144 , shows a estimation for @xmath26 peaked at @xmath288 at a high level of confidence , with @xmath289 ( for the @xmath46-parameter case ) and @xmath290 ( for @xmath45-parameter case with @xmath144 ) at @xmath115 ( @xmath166 ) confidence level . as @xmath26 was sampled with bin size @xmath291 , then the peaks can be somewhere between @xmath292 and @xmath237 . the joint pdf for the parameter space of @xmath293 , i.e , @xmath294 , is shown in figure [ figomegam0c0 ] . the pdf peak is now located at @xmath295 , with the same the central value of @xmath26 after bayesian marginalization . but , in this two - parameter space with simultaneously values of @xmath296 , the credible regions would be given by @xmath297 at @xmath115 ( @xmath106 ) and @xmath298 at @xmath117 ( @xmath299 ) confidence levels . of course , the integral ( marginalization ) of @xmath300 over the @xmath27 space yields @xmath301 , shown in figure [ figsomegas ] . the conclusion is that the marginalization decreases the confidence regions , even if it does not change the peak value of probability . the generalized chaplygin gas density parameter @xmath27 estimation ( table [ tableparestgcg ] ) is like the estimation in the case of cgm shown by table @xmath46 of ref . @xcite , and is quite spread , i.e. , it has a large credible region , for example @xmath302 in the case of @xmath46-parameter space . figure [ figsomegas ] show the pdf of @xmath303 for the @xmath46-parameter space ( showing @xmath304 at @xmath115 confidence level ) and the @xmath45-parameter space with @xmath144 ( when @xmath305 ) . as discussed above for the gcgm , the joint pdf for the parameter space of @xmath306 in figure [ figomegam0c0 ] shows that the credible regions would be given by @xmath297 at @xmath115 ( @xmath106 ) and @xmath298 at @xmath117 ( @xmath299 ) confidence level . so , the marginalization of @xmath307 over the @xmath26 space gives @xmath308 , yielding a different peak value as well as different ( narrower ) confidence regions . the estimates for the dynamical age of the universe today , @xmath71 , given in table [ tableparestgcg ] , show that the generalized chaplygin gas model increases the @xmath117 credible regions ( e.g. , @xmath309 ) , when compared to the estimations of the cgm in table @xmath46 of ref . @xcite ( e.g. , @xmath310 ) . the peak values of @xmath311 range between @xmath312 to @xmath313 , while the cgm features @xmath71 peaked between @xmath314 to @xmath315 . we expect that a large sample of sne ia will narrow the credible regions and allow , by means of independent estimations of @xmath71 from different observations ( globular agglomerates , etc ) , to verify which cosmological model is preferred : gcgm , cgm and @xmath316cdm . table [ tablebestfitgcg ] shows the @xmath71 for the best - fitting parameters without any marginalization ( i.e. , the @xmath101-parameter model has its @xmath101 parameters taken simultaneously yielding a minimum @xmath80 ) : the @xmath71 ranges from @xmath317 to @xmath318 , well inside the estimated credible regions for @xmath71 . all the estimations based on the generalized chaplygin gas model ( table [ tableparestgcg ] ) suggest an accelerating universe today , i.e. , @xmath320 , with a high confidence level ranging from @xmath277 ( @xmath321 ) to @xmath322 ( @xmath323 ) , depending on the curvature and cold dark matter densities of the universe . the peak values of @xmath319 range between @xmath324 to @xmath325 . the data of cgm in table @xmath46 of ref . @xcite show a similar behaviour . although the @xmath319 values of table [ tablebestfitgcg ] , from @xmath326 to @xmath327 , are inside the estimated credible regions for @xmath319 , the estimated peak values of @xmath319 differs from the @xmath319 for the best - fitting parameters . we have also estimated the value of the scale factor that shows the beginning of the recent accelerating phase of the universe . table [ tableparestgcg ] shows that the most probable value for @xmath74 ranges between @xmath328 to @xmath329 , where @xmath99 is the value of scale factor today , with usually an small upper credible region and a larger lower credible region . the probability of @xmath330 is theoretically the same having an accelerating universe today ( @xmath320 ) , and these independent probability calculations agree almost exactly as table [ tableparestgcg ] shows , indicating that the bayesian probability analyses of this work are accurate and reliable for the given sne ia data . we can obviously express the scale factor value @xmath74 in terms of redshift value @xmath76 , @xmath331 or @xmath332 . for example , @xmath333 corresponds to @xmath334 . we expect that a large sample of sne ia with large redshift values will help improve this type of estimation of @xmath76 or equivalently @xmath74 . the @xmath74 values are almost equal in table [ tablebestfitgcg ] for the best - fitting parameters , between @xmath335 and @xmath336 , and this range fits well inside the the estimated credible regions for @xmath74 of table [ tableparestgcg ] . the probability of having a eternally expanding universe , given here by demanding @xmath337 during the future evolution of the universe , is estimated to be exactly one or almost one ( at almost a @xmath338 level ) , see table [ tableparestgcg ] . the estimations for the cgm and @xmath339cdm cases ( tables @xmath45 and @xmath46 , ref . @xcite ) were miscalculated , they are also exactly one or almost one . in this paper we investigated some constraints on the generalized chaplygin gas model ( gcgm ) using type ia supernovae data and bayesian statistics . the model is built with the chaplygin gas and a pressureless fluid as the matter content of the universe , whose density parameters today are @xmath27 and @xmath26 respectively , such that the curvature term is @xmath340 . the parameter @xmath0 related to the sound velocity of the chaplygin gas , the hubble parameter @xmath24 and the @xmath29 parameter are also independent free variables . we allowed the parameter @xmath16 to vary between negative and large positive values . in principle , this could lead to instability problems at perturbative level ( @xmath29 negative ) , or superluminal sound velocity ( @xmath16 greater than one ) , but both potential problems can be resolved using a fundamental description for the gcg . when the five independent parameters are free , the bayesian estimation yields @xmath341 , @xmath342 , @xmath343 , @xmath344 , @xmath302 , @xmath345 , @xmath36 , @xmath346 and @xmath333 where @xmath311 is the age of the universe , @xmath319 is the value of the deceleration parameter today and @xmath74 is the value of the scale factor that marks the start of the recent accelerating phase of the universe . we have also estimated the cosmological parameters fixing the pressureless matter ( through the constraints of nucleosynthesis ) or the curvature ( imposing the flat condition ) , or even both . the estimations of the remaining free parameters do not change dramatically in each case , with some exceptions . for example , the prediction for the curvature parameter depends on whenever the pressureless matter is fixed or not : a spatially flat universe is favoured only if @xmath26 is fixed at low values . in particular , we verify that the @xmath16 parameter is strongly related to the @xmath0 parameter . in this manner , the best value of @xmath16 have the peak values and the credible regions changed in function of the marginalization process used . the @xmath16 estimations are very spread and differ a lot if one or two - dimensional parameter space is used . many authors @xcite@xcite , prefer to estimate @xmath16 and @xmath96 simultaneously using two - dimensional credible regions and they do not marginalize the pdf to obtain the more statistically robust one - dimensional pdfs for @xmath16 and @xmath0 , so any comparison should be taken carefully as the credible region estimations on two - dimensional parameter space are usually different from the one - dimensional estimations . here we take the marginalization approach to obtain one - dimensional pdfs , and one important point arises in this analysis : despite the dependency on the @xmath0 parameter , the value @xmath21 , that represents the ordinary cgm , is not ruled out since this particular case is preferred at @xmath347 to @xmath348 , depending on the spatial section and matter content . we calculated the value of the scale factor when the dark energy starts to prevail , @xmath349 , for the case of five free parameters . in general , the value for @xmath74 ranges between @xmath350 and @xmath351 , where @xmath99 is the value of the scale factor today . this results are coherent with the suposition that the recent acceleration of the universe could be started with the redshift parameter @xmath20 between @xmath288 and @xmath2 obtained by the type ia supernovae data . in the case of the ordinary matter parameter @xmath26 our analyses show that the preferred values are small , around zero . this situation is the same of the case of the ordinary chaplygin gas . the result reinforces the idea of the quartessence model @xcite , that unifies dark matter and dark energy into a single fluid , the chaplygin gas . the predicted age of the universe is well above the age of globular clusters . at same time , the deceleration parameter is consistent with the estimations from the wmap data . further developments of the present work include the use of a large sample of type ia supernovae . however , the sample of @xmath88 sne ia used here has an excellent quality with respect to the low values of @xmath91 . moreover , the dispersion of the parameter estimations obtained here has the same order of estimations using larger samples . in some cases our estimation are even less spread , as our preliminary investigation on larger sne ia indicates . we also plan to apply the bayesian analysis to the gcgm using large scale structures data ( 2dfrgs ) and to the anisotropy of the cosmic microwave background . in this way , we may consistently cross the estimations of different observational data and constrain more sharply the predictions to each free parameter of the model . * acknowledgments * we would like to thank orfeu bertolami , ioav waga and martn makler for many discussions about the generalized chaplygin gas model . the authors are also grateful by the received financial support during this work , from cnpq ( j. c. f. and s. v. b. g. ) and facitec / pmv ( r. c. jr . ) .

growing body of experimental information on fixed - target nucleus - nucleus collisions from ags and cern - sps accelerators provides a reliable basis for the systematic investigation of strongly interacting hadronic matter under extreme conditions . particularly , production of particles containing strange quarks is expected to reflect the reaction dynamics at the early stage of collisions . in this context , @xmath0 mesons , particles consisting mainly of @xmath4 pairs , are of a special interest . since the interaction of @xmath5 mesons with non - strange hadronic matter is suppressed according to ozi rule , @xmath5 mesons are expected to decouple easily from the hadronic fireball . several calculations have been done for @xmath0 meson properties in dense hadronic environment ( mainly at higher baryon densities ) @xcite . it has been found that modifications of the @xmath0 width and mass are sensitive to the strangeness content of the surrounding medium . recently , new experimental data from cern - sps on @xmath0 meson production in central pb+pb collisions at 158gev / a beam energy became available . advantage of the rich experimental program at cern is that it allows to study the @xmath0 production via different @xmath0 decay channels . results of na49 collaboration @xcite are based on the @xmath0 meson identification via its hadronic decay @xmath6 , while na50 collaboration has recently reported @xcite about preliminary analysis of @xmath5 mesons identified via the dileptonic decay channel @xmath7 . the purpose of this paper is to show , that the @xmath5 meson production spectra reconstructed via @xmath2 and @xmath1 decay channels can be significantly different up to the level of present experimental observations . after introductory considerations in section ii we derive in section iii expressions for momentum distributions of @xmath0 mesons detected via @xmath8 and @xmath9 channels . we take into account rescattering of decay kaons and in - medium modification of @xmath0 meson properties discussed in section iv . numerical results are presented in sections v and vi and conclusions are drawn in section vii . for our estimates of the observed @xmath5 meson spectra we assume a simple two - stage picture of central heavy ion collisions , which is close to those considered within cascade - transport @xcite , hydrodynamical @xcite and thermodynamical approaches @xcite . initial stage of a collision is characterized by temperature close to the qcd phase transition @xmath10 mev . then the system expands up to the point , when numbers of different kinds of particles freeze in - a chemical freeze - out . thermodynamical parameters of this stage could be obtained by fitting the final total hadron multiplicities . typical temperature is found to be @xmath11 mev . during the second stage of the expansion , elastic scatterings change momentum distributions of hadrons until they cease and distributions freeze in . the freeze - out temperature can be extracted from a simultaneous fit to the single - particle @xmath12-spectra of different particles supplemented by the analysis of particle correlation data . according to refs . @xcite one has @xmath13 mev . in our considerations we assume that the fireball created in heavy - ion collision consists , mainly , of pions , kaons and excited mesonic resonances in central rapidity region . mean free path @xmath14 of @xmath0 mesons in hadron gas is estimated in ref . comparison with mean free paths of pions and kaons , @xmath15 , from ref . @xcite gives @xmath16 for temperatures @xmath17 . hence , one can expect that @xmath0 mesons decouple from the pion - kaon subsystem at some earlier stage between the chemical and thermal freeze - out . upon this stage @xmath0 mesons stream freely out from the fireball . pions and kaons , on the other hand , may still participate in mutual secondary interactions up to the stage of total thermal freeze - out . therefore , if a @xmath0 meson decays inside a fireball via hadronic channel we have to take into account possible interaction of its decay products with surrounding hadronic environment . as a result of secondary interactions , the reconstructed invariant mass of a given pair of @xmath0 daughter kaons falls out from the original @xmath0 meson peak into the region identified as a combinatorical background . therefore , the @xmath0 mesons decaying in medium , can be partially unrecognized in experimental analysis of @xmath2 pairs . together with negligible final state interaction of secondary dimuons originating from @xmath18 decays this behaviour results in the relative suppression of @xmath0 meson yield observed via hadronic @xmath8 channel . such mechanism has been quantitatively studied in ref . @xcite where suppression at the level 4060% has been obtained from simulation using rqmd code . in this paper we discuss effects which may enhance the suppression of observed @xmath5 mesons identified via kaon channel . possible increase of a @xmath0 meson width in medium will enlarge the probability of @xmath5 decay inside a fireball enhancing thus consequences of mechanism studied in ref . alternatively , we consider a possibility that the kaon decay channel of a @xmath0 meson becomes kinematically quenched in the medium . additionally we argue that substantial relative difference in properties of @xmath19 and @xmath20 meson in hadronic environment caused by the isospin asymmetry and/or by the large baryonic admixture , would also prevent the reconstruction of @xmath0 mesons decaying into kaon pairs inside the medium . let us denote the phase - space distribution of @xmath0 mesons in the center - of - mass system of two colliding nuclei at freeze - out as @xmath21 . then the primary momentum distribution of @xmath0 mesons is given by [ integr ] _ 0(p ) = _ d^3 ^p _ f _ ( _ , ) , where integration goes over the fireball volume within a freeze - out hyper - surface @xmath22 ( surface normal vector @xmath23 contracted with a @xmath0 meson momentum @xmath24 ) @xcite . here we assume the case of a time - like freeze - out hyper - surface , @xmath25 , which is relevant for applications below ( for discussions of alternative cases cf . @xcite ) . in the absence of any in - medium modifications of decay products and final state rescattering the shape of _ observed _ momentum distributions @xmath26 of muon pairs @xmath27 and @xmath28 of kaon pairs @xmath29 would be the same . here @xmath30 is @xmath0 meson total width and @xmath31 , @xmath32 are the @xmath5 partial decay widths in kaon and muon decay channels . accordingly , in the experimental analysis , @xmath0 meson distribution is reconstructed from momentum distribution of decay products multiplied by the corresponding inverse branching ratio . in this section we derive expressions for the apparent momentum distributions of @xmath0 mesons reconstructed via kaon and dimuon decay channels taking into account possible modifications of meson properties in medium and consequences of kaon rescattering . let us consider @xmath5 meson suffered the last interaction at position @xmath33 inside the hadronic fireball . the probability , that @xmath0 meson lives for a time @xmath34 is [ pbx ] d_(t)&= & , where @xmath35 is total width of a moving meson with energy @xmath36 . asterisk * denotes in - medium values of the quantities . after traveling for a given time @xmath34 with velocity @xmath37 the @xmath5 meson decays at the position @xmath38 with a probability @xmath39 into two kaons . in - medium values of widths @xmath40 are determined by the current local temperature and density of the system . daughter kaons from @xmath5 meson decay have momenta @xmath41 in the rest frame of the @xmath0 meson . value of @xmath42 follows from equation @xmath43 where @xmath44 and @xmath45 are in - medium spectra of kaons and anti - kaons . in the center - of - mass system of two colliding nuclei ( cms ) the kaon momenta are equal to [ kmoment ] _ k^&= & , + & = & p_kk \ { _ k+ _ ( _ -1 ) ( _ k _ ) } , where @xmath46 . the unit vector @xmath47 is uniformly distributed in the @xmath0 rest frame , direction of @xmath48 is determined by @xmath0 meson momentum @xmath49 in cms . for a successful identification of @xmath5 mesons in invariant mass spectrum of observed @xmath2 pairs it is essential that momenta of daughter kaons do not change while leaving the hadronic fireball . here we investigate two mechanisms which may change momenta of daughter kaons : rescattering in surrounding hadronic environment and change of momentum due to in - medium @xmath50 meson potential . probability that secondary kaon and anti - kaon leave a fireball without rescattering is determined by their mean free paths @xmath51 , the time which kaons need to reach the fireball border @xmath52 , and their velocities @xmath53 ( in cms ) as follows : [ pl]p_(t ) & = & .thus , the probability to register @xmath0 meson in the kaon channel when it decays in medium can be expressed as : [ probx ] p_i(1-p_(t))+p_ii_(t ) , where @xmath54 and @xmath55 are probabilities to identify a @xmath0 meson from rescattered ( @xmath54 ) and non - rescattered ( @xmath55 ) kaons . single rescattering of one of the kaons changes momentum of the kaon pair by vector @xmath56 , with @xmath57 being a average thermal momentum of pions . correspondingly the invariant mass changes by @xmath58 . at high temperatures ( @xmath59 ) average thermal momentum of pions is @xmath60 mev and hence a single rescattering may shift @xmath61 far from the @xmath5 meson mass . in this case we put @xmath62 . neglecting in - medium effects , one takes @xmath63 , assuming that without a hard rescattering all kaon pairs from @xmath0 decays can be identified . however the modification of particle properties can provide another mechanism preventing @xmath0 identification . in medium kaons feel an effective mean - field potential inducing a change of their spectra @xmath64 . leaving the fireball , kaons have to come back to their vacuum mass shell . this changes the invariant mass and momentum of the pair . going out of medium , kaon stays on the same energy level , and its momentum outside the fireball becomes @xmath65 ( @xmath66 ) , where the kaon energy @xmath67 is evaluated at the moment of @xmath0 decay . as a result , momentum of the kaon pair is changed by @xmath68 , where we put @xmath69 and @xmath70 . the invariant mass shift is then given by @xmath71 . in the nucleon free , isospin symmetrical meson gas , in - medium spectra of kaons and anti - kaons are identical @xmath72 and the invariant mass shift reduces to @xmath73 $ ] . this expression can be interpolated between the limit cases @xmath74 and @xmath75 as @xmath76 . therefore in the symmetrical case @xmath77 and @xmath78 vanish for small @xmath79 . moreover in the isospin symmetrical case kaons suffer only a small mass modification ( @xmath8030 mev for @xmath59 ) . hence , the release of kaons from a small potential well formed inside a fireball does not affect the momentum and invariant mass of a pair strongly enough to prevent a @xmath0 meson reconstruction . for this case we take @xmath81 . in the case of rather strong isospin asymmetry and/or significant baryonic admixture @xmath19 and @xmath20 mesons can have quite different spectra in medium @xcite . this may lead to a wide spread of the kaon pair invariant masses , making the reconstruction of @xmath0 mesons decaying inside fireball impossible ( @xmath82 ) . note , that in this case our results are insensitive to the details of kaon propagation in medium . they depend only on the total @xmath0 meson width via function @xmath83 . finally the probability that a @xmath0 meson created at position @xmath84 will be detected via the kaon channel is : & & _ 0^ t^*_kk(t ) d_(t ) p_(t ) p_ii + [ kaonprob ] & & = _ ( _ ) + _ 0^ _ d t^*_kk(t ) d_(t ) p_(t ) p_ii , where @xmath85 is the time of flight of a @xmath0 meson through the fireball , for @xmath86 we have @xmath87 and @xmath88)$ ] . momentum distribution of the kaon pairs from @xmath0 decays is then obtained by averaging expression ( [ kaonprob ] ) over all @xmath47 directions in the @xmath0 meson rest frame , and integrating over the fireball volume with a primary @xmath0 distribution - see eq.([integr ] ) . then , multiplying by the inverse branching ratio we express `` reconstructed '' @xmath0 distribution in kaon channel as : _ k(p ) & = & _ ( _ ) + _ 0^_t d_(t ) _ kk^*(t ) p_ii , + [ ratrescat ] & & = _ ( t ) . here and the brackets denote integration : = _ d^3 ^p _ f _ ( _ , ) ( ) . note that according to this definition @xmath89 . consideration of the di - muon pair momentum distribution is more straightforward . with a change of total @xmath0 width in medium the branching ratio of the @xmath90 decay @xmath91 changes too . therefore , the apparent @xmath0 distribution reconstructed in the muonic channel is : [ ratmu ] _ ( p ) & = & _ ( _ ) + _ tot^0 _ 0^ _ dt d_(t ) . note that for @xmath92 we have @xmath93 , for @xmath94 we have @xmath95 and if @xmath96 then @xmath97 . in our numerical estimates we will consider ratio of the reconstructed @xmath5 meson distributions eq.([ratrescat],[ratmu ] ) which does not depend on the overall normalization : [ eta ] ( p)=_k(p)/_(p ) . for a direct comparison with experimental results on @xmath12 distributions of identified @xmath5 mesons , one has to integrate over the rapidity interval accessible to the experiments , utilizing @xmath98 . dependence of the ratio ( [ eta ] ) on @xmath12 reads [ etamt ] ( m_t)=<_k(p)>_y/<_(p)>_y . we also define ratio of the apparent and primary @xmath0 momentum distributions for @xmath2 and @xmath1 decay channels as a function of @xmath12 : _ k,(m_t)=<_k,(p)>_y/<_0(p ) > _ y main hadronic decay channels of @xmath0 meson are @xmath100 and @xmath101 . let us now consider the change of a @xmath0 decay width in a hot meson gas due to the modification of kaon , pion and @xmath102-meson properties . in - medium properties of pions can be effectively incorporated by a small mass shift @xmath103 with @xmath104 . similar behaviour is expected for kaons . spectral function of @xmath102 meson in medium was extensively investigated in the context of di - lepton production in heavy - ion collisions at sps energies @xcite . in baryonic matter , coupling of @xmath102 mesons to resonance nucleon - hole modes @xcite together with the modification of pions @xcite play dominant role . the @xmath102 meson becomes very broad in medium and its spectral density strength is driven effectively to lower energies in analogy to brown - rho - scaling picture @xcite . in purely mesonic systems the @xmath102 mass is found to be almost independent of the temperature due to cancellation of @xmath105- and @xmath106-loop contributions @xcite . the @xmath102 width , on the other hand , is expected to increase in meson gas considerably e.g. by 80 mev at @xmath107 mev and by 160 mev at @xmath108 mev @xcite . the partial width of the @xmath109 decay in medium depends on the kaon mass @xmath110 as : [ gkk]_kk^*= _ kk^0 p_cm^3(m_^2,m_k^*,m_k^*)/ p_cm^3(m_^2,m_k , m_k ) , where @xmath111 is a kaon momentum in the rest frame of @xmath5 meson decay , obeying the equation : @xmath112 . assuming isospin symmetry we have @xmath113 mev . then vacuum width is equal to @xmath114 mev . note that for @xmath115 the in - medium width @xmath116 decreases fast and vanishes for @xmath117 mev . the second hadronic decay channel @xmath101 is effected by the decrease of the @xmath102 meson mass , increase of the @xmath102 meson width and the bose - einstein enhancement factor for pions . for the @xmath0 meson at rest we write [ grho]_^ * = ^0__(m_^*,_^*,t)/ _ ( m_^0,_^0,0 ) , and [ grho ] _ ( m_,,t ) & = & _ 2m_^m_-m _ [ ( -m_)^2-m_^2]^32 + & & , where @xmath118 and @xmath119 are in - medium @xmath102 meson mass and width , respectively . vacuum values are @xmath120 mev , @xmath121 mev , and @xmath122 mev . we use the constant width approximation for the @xmath102 meson spectral density and denote @xmath123 . the bose - einstein distribution of pions with temperature @xmath124 is @xmath125 . in the zero - width limit we obviously have @xmath126 . numerical evaluation of eq . ( [ grho ] ) leads to the approximated relation @xmath127\big ) $ ] valid for @xmath128 mev , @xmath129 mev , and @xmath130 . the shift of a pion mass produces a minor effect and it is , therefore , neglected here . finally , the total width of @xmath0 is given by @xmath131 . here and below we do not consider the @xmath0 meson mass shift , which is small and cancels as soon as we consider the ratio of the momentum spectra , cf . eq.([eta],[etamt ] ) . for completeness we remark that the di - lepton decay channel @xmath132 also suffers a modification in medium , because the vector - meson photon coupling is suppressed by meson fluctuations @xcite . for the @xmath133 coupling the suppression factor is determined only by kaon fluctuations @xmath134 . here @xmath135^{-1}\,$ ] , @xmath136 and @xmath137 mev is the pion decay constant . because of the large kaon mass this correction is negligibly small : @xmath138 . finally we , specify , how the in - medium properties of kaons and @xmath102 mesons relax to their vacuum values during the fireball expansion . assuming that @xmath139 and @xmath140 are proportional to the density of the system , we have m_k,(t ) = m_k,^0 , _ ( t ) = _ ^0 , where @xmath141 and @xmath142 are input parameters . before finishing this section we would like to point out that the estimations done here are valid only for an almost baryon free fireball . in the presence of baryons the calculation of a phi self - energy becomes more elaborated , cf . . however , in this case the spectra of kaons and anti - kaons differ dramatically , inhibiting thereby the @xmath0 reconstruction even more . it corresponds to the case @xmath143 when the reconstructed @xmath0 distributions ( [ ratrescat],[ratmu ] ) are the functions of a total width only . to simulate effectively the in - medium modification of @xmath144 we will use eq . ( [ gkk ] ) and vary the kaon mass . after the considerations above let us now specify the model of the fireball expansion , which we will use in our numerical calculations . assume a simple homogeneous spherical fireball with the constant density and temperature profiles . the @xmath0 meson momentum distribution [ dist ] f_(,)=,is determined by the temperature @xmath145 , flow velocity profile @xmath146 and radius @xmath147 . time of flight of a @xmath0 meson and its daughter kaons through the medium contained in eqs . ( [ pl ] ) and ( [ ratrescat ] ) can be expressed as @xmath148 and @xmath149 , where @xmath150 stands for a time , during which a particle with velocity @xmath151 passes a distance from position @xmath152 to the border of a sphere with the radius @xmath153 . since the fireball is expanding with radial velocity @xmath154 , this time satisfies the equation @xmath155 . this implies : _ r ( , ) & = & ( + [ tau ] & -&(-v_fr ) ) ( ^2-v_f^2)^-1 . solution ( [ tau ] ) is valid for @xmath156 and @xmath157 . in the case @xmath158 we put @xmath159 . during the expansion @xmath160 , fireball density drops as @xmath161 and the temperature decreases as @xmath162 as expected for relativistic pion gas , cf . the kaon mean free path is @xmath163 and therefore _ k(t ) & = & _ k^0 r^3(t)/r_0 ^ 3 .to incorporate the freeze - out effect we will assume that as soon as @xmath164 kaons become free and @xmath165 as well as @xmath166 . the freeze - out time is then given by [ taofo ] _ f.o.= ( -1 ) . parameters @xmath167 , @xmath168 , @xmath154 , and @xmath169 serve as input for numerical evaluations below . the parameter @xmath170 is used for an effective parameterization of the freeze - out time @xmath171 . it should not be considered as a true freeze - out temperature , since our estimation is based on the simplified hydrodynamical description of a fireball . the model set up here is a rather crude approximation . however , the final results are found to be rather insensitive to the details of hydrodynamical evolution of a fireball , being determined mainly by the values of @xmath172 , @xmath173 , and @xmath154 . we shall vary the input parameters within a broad range to illustrate different possibilities . in this section we perform numerical evaluation of our expressions for @xmath5 meson yields reconstructed via @xmath2 and @xmath1 channels in central pb+pb collisions at 158gev / n sps energy . first we investigate to what extend rescattering of secondary kaons enhanced by the in - medium modification of a @xmath0 meson width can suppress experimentally observed yield of @xmath5 mesons identified via @xmath2 channel . we remind that results of ref . @xcite give maximal suppression factor 40% for the @xmath5 meson observation in the kaon decay channel . in our evaluation we use several combinations of input parameters . freeze - out temperatures @xmath145 of @xmath5 mesons distributed according to eq.([dist ] ) vary between @xmath174 and @xmath170 : ( @xmath175 ) @xmath176 mev , ( @xmath177 ) @xmath178 mev , ( @xmath179 ) @xmath180 mev . size of the fireball @xmath147 at the stage of the @xmath0 freeze - out has to be comparable with @xmath0 meson mean free path @xmath181 at given temperature @xmath182 with @xmath183 . for different temperature parameters above we take according to ref . @xcite : @xmath184 fm , @xmath185 fm , and @xmath186 fm . first we evaluate suppression factor ( [ etamt ] ) without any modifications of particle properties in medium , i.e. , @xmath187 . in this case we have @xmath188 and @xmath189 . flow velocities corresponding to selected temperatures @xmath145 are adjusted to reproduce the slope of @xmath5 meson @xmath12 distribution measured by the na50 collaboration : @xmath190 mev : @xmath191 , @xmath192 , @xmath193 . we take @xmath194 and vary the mean free path of kaons @xmath195 within the interval @xmath196 , where @xmath197 follows from estimations of ref . @xcite : @xmath198 fm , @xmath199 fm , and @xmath200 fm . we vary also @xmath170 between 100 mev and 80 mev in agreement with analysis @xcite . this translates into the interval of freeze - out time values @xmath201 . all these variations produce the upper grey areas shown in fig . [ fig : nomed ] ) as a function of @xmath12 calculated for three sets of parameters ( @xmath202 without inclusion of in - medium modifications of kaons and @xmath102 mesons . the upper grey area corresponds to variations of kaon mean free paths @xmath203 and freeze - out temperature @xmath170 as described in the text . the lower grey areas are obtained for the same set of parameters but with the common expansion velocity @xmath204.,width=188 ] for all three sets of parameters @xmath205 we observe that @xmath206 does not fall below @xmath207 significantly . this is related to the large expansion velocity of the fireball . in this case @xmath208 and @xmath5 mesons decay after the thermal freeze - out . to illustrate this effect we recalculate ratio @xmath206 for the same three cases fixing @xmath204 what corresponds to @xmath209 . results obtained ( @xmath210 ) are shown as lower gray areas in fig . [ fig : nomed ] . to reproduce results of rqmd calculations described in ref . @xcite we take somewhat larger size of a fireball with @xmath211 , freeze - out temperature @xmath212 mev and @xmath213 fm . this corresponds to the lowest limit allowed by the analysis @xcite . the results are shown in fig . [ fig : nomedr15 ] by solid lines . the limiting scenario considered in @xcite , when the freeze - out volume is determined by the last kaon interactions , can be reproduced with @xmath214 mev . this case is shown by dash lines in fig . [ fig : nomedr15 ] . ) calculated for cases @xmath215@xmath216 with @xmath211 , @xmath217 fm and @xmath213 fm . solid lines are calculated with @xmath212 mev and dash lines correspond to @xmath214 mev . no medium effects are included.,width=188 ] we take solid lines in fig . [ fig : nomedr15 ] as a reference point for our further investigation of in - medium effects . first we consider modifications of @xmath102 meson properties . we choose @xmath218 mass shift to be @xmath219 mev for our three parameter sets . the @xmath218 meson width depends on the temperature . relying on ref . @xcite we take : @xmath220 mev , @xmath221 mev , and @xmath222 mev . comparison with fig . [ fig : nomedr15 ] ( solid lines ) shows slight decrease of ratio @xmath206 which corresponds to small increase of the total @xmath0 meson width by 40% due to @xmath223 channel . [ fig : med ] shows results obtained taking into account modification of @xmath50 meson properties in - medium . calculated for in - medium modification of @xmath0 meson properties . results for different parameters sets @xmath215@xmath216 are depicted by solid , dash and dotted lines , respectively . the left plane corresponds to the case when the kaon mass increases in medium @xmath224 mev , whereas the right plane shows results for a decreasing kaon mass @xmath225 mev . thin lines show suppression factors in the muon @xmath226 ( b ) and kaon @xmath227 ( c ) channels calculated for parameter set ( @xmath175 ) . upper plots are calculated for the case @xmath228 , lower plots correspond to @xmath143 . in all cases modification of @xmath102 meson properties in medium is taken into account.,width=302 ] first we investigate the case when @xmath100 channel is closed initially ( @xmath224 mev ) and it opens only during the fireball expansion . results are shown in the left part of fig . [ fig : med ] , where the upper plot is calculated for @xmath228 and the lower one corresponds to a strong suppression of the kaon channel in medium with @xmath143 . comparing ratios @xmath206 in fig . [ fig : nomedr15 ] and in fig . [ fig : med ] shown by thick solid line ( a ) calculated for parameter set ( @xmath175 ) we observe that kinematical quenching of @xmath229 channel by increasing kaon mass decreases ratio @xmath206 very slightly for both values of @xmath55 . this happens because @xmath0 meson width becomes very small in this case and therefore probability of @xmath5 meson decay inside the expanding fireball and consequently also probability for rescattering of daughter kaons are small . compare lines ( c ) , calculated for parameter set ( @xmath175 ) , with a corresponding line in fig . [ fig : nomedr15 ] . lines ( b ) in fig . [ fig : med ] ( left side ) show that @xmath226 becomes larger than one for increasing kaon mass . since increase of @xmath230 by 10%20% does not change considerably the effective @xmath12 slope of @xmath231 distribution , we do not need to readjust the flow velocity parameter . this increase of @xmath230 leads at the end to a small decrease of @xmath232 . let us now consider the case when the @xmath0 meson width increases strongly in hadronic medium due to the increase of @xmath116 . we simulate this effect by decrease of the kaon mass in medium , which can result e.g. from rescattering of kaons on pions through @xmath233 and heavier kaonic resonances @xcite . here we restrict ourselves to rather conservative modification of kaon masses @xmath234 mev@xmath235 which corresponds to the @xmath0 width @xmath236 mev . in this case @xmath237 and at small @xmath238 region @xmath230 can be suppressed up to 4060% . thus , for a given freeze - out temperature @xmath145 and total width @xmath144 we readjust flow velocity @xmath239 to reproduce the slope of the @xmath12 distribution measured in di - muon channel by na50 @xcite . for our three sets of parameters ( @xmath240 ) we obtain new flow velocities : ( @xmath175 ) @xmath241 , ( @xmath177 ) @xmath242 , ( @xmath179 ) @xmath243 . right part of fig . [ fig : med ] shows our results obtained for @xmath225 mev . corresponding partial width is @xmath244 mev , and the total width is @xmath245 mev . this leads to @xmath246 , which provides a strong suppression of @xmath247 as it is shown for the parameter set ( @xmath175 ) in fig . [ fig : med ] ( right side , lines c ) . we find @xmath248 for @xmath228 and @xmath249 for @xmath143 . however , since the ratio @xmath250 also shown in fig . [ fig : med ] ( lines b ) is also suppressed the resulting ratio @xmath206 remains on the level @xmath251 for small @xmath238 , provided we put @xmath143 and @xmath252 for @xmath228 . taking even larger values of the total decay width in the most preferable case ( @xmath175 ) we obtain @xmath253 for @xmath254 mev , @xmath242 , and @xmath255 for @xmath256 mev , @xmath257 ( both for @xmath258 ) . it is instructive to investigate ratio ( [ eta ] ) as a function of the @xmath0 momentum . averaging over rapidity mixes momenta within a broad interval , e.g. , for @xmath259 momenta @xmath260 mev what partially washes out the final suppression effect . to illustrate this point we show ratio @xmath261 as a function of a @xmath0 momentum in fig . [ fig : pdep ] . ) as a function of @xmath0 meson momentum . curve styles correspond to those in fig . [ fig : med ] with readjusted flow velocities . curves a are calculated for @xmath262 , whereas sets b are calculated with @xmath62 . , height=188 ] for increasing kaon mass and correspondingly vanishing @xmath31 we obtain a decrease of @xmath261 at small momenta by 4050% ( left plane ) . this is again a direct consequence of a rapid fireball expansion which brings @xmath263 in the integral ( [ ratrescat ] ) fast to its vacuum value . in the extreme case for @xmath264 and @xmath265 we get @xmath266 . for decreasing kaon mass the reduction is even stronger @xmath26770% . momentum dependence of @xmath232 at small momenta @xmath268 is completely washed out by rapidity averaging . in fig . [ fig : ydist ] ( full circles ) , @xmath269 ( open circles ) and @xmath270 ( open diamonds ) calculated for the parameter set ( @xmath175 ) with @xmath225 mev , the flow velocity @xmath241 and @xmath62 . , width=188 ] we compare the original rapidity distribution of @xmath0 mesons @xmath271 , with the distributions which can be reconstructed via kaonic @xmath269 , and muonic @xmath270 decay channels . calculations are done for the parameter set ( @xmath175 ) with @xmath272 mev , flow velocity @xmath241 and @xmath143 . distributions are normalized to give @xmath273 . we observe a considerable broadening of the rapidity distribution measured in the kaon channel . we have studied distributions of @xmath5 mesons in heavy - ion collisions at sps energies reconstructed via hadronic @xmath2 and dilepton @xmath274 decay channels . the analysis of @xmath5 meson mean free path allows to suppose that @xmath0 mesons decouple from the hadronic system at somewhat earlier stage before the common breakup of the hadronic fireball . therefore , kaon pairs originated from the @xmath0 decays inside a fireball can be rescattered or absorbed . such kaon pairs will not contribute to a @xmath0 meson reconstruction , whereas the leptonic probes can leave a fireball freely . we derive the expressions ( [ ratrescat ] ) and ( [ ratmu ] ) for the apparent momentum distribution of @xmath0 mesons in kaonic and muonic channels respectively . within a simple model of spherically expanding fireball we investigate dependence of a relative suppression factor of the hadronic channel with respect to the dileptonic one on parameters of the system and on the @xmath0 meson in - medium properties . for a vacuum @xmath0-meson width @xmath275 mev the maximal suppression 0.60.8 is obtained for the fireball size and expansion time @xmath276 fm . these values are in agreement with results of rqmd simulations @xcite . the crucial parameter is the fast expansion of a fireball with @xmath2770.5 corresponding to the @xmath0 freeze - out temperature range @xmath278170 mev . width of hadronic @xmath0 meson decay channels ( @xmath100 and @xmath279 ) can be modified in medium due to changes of the meson properties . we have found that increase of the @xmath280 channel width due to the broadening of @xmath102 meson and decrease of @xmath218 meson mass leads alone to a tiny increase of the suppression . other possibility is kinematical quenching of the kaon decay channel , which we simulate by simultaneous increase of @xmath19 and @xmath20 masses . since total width @xmath281 of @xmath5 meson in medium becomes small ( increase of the @xmath280 channel width is not strong enough ) @xmath0 mesons decay mainly outside the fireball , where vacuum properties of @xmath5 mesons are restored and rescattering of daughter kaons is negligible . together with relative amplification of the muon decay channel by 20% the resulting suppression factor found for a quenched kaon decay channel was @xmath252 . the increase of the @xmath0 meson width in medium provides , on the other hand , a mechanism for strong suppression ( @xmath249 ) of the kaonic detection channel due to the enhancement of @xmath5 decay probability inside a fireball increasing thus rescattering of daughter kaons . however , increase of the @xmath0 total width reduces simultaneously the branching ratio of @xmath282 decay and suppresses the spectrum of @xmath5 mesons reconstructed via @xmath1 decay channel . obtained suppression of the muonic decay channel at the level @xmath28360% requires readjustment of the flow velocity to be compatible with experimental slope of na50 . adjusted flow velocity @xmath2840.4 for @xmath285 mev and @xmath278170 mev ( compare to @xmath2770.5 obtained for a vacuum @xmath0 width ) gives final net relative suppression factor of kaon channel to muon channel @xmath251 this value is close to experimental observations at cern sps @xcite . strong increase of @xmath5 meson in - medium width can take place if the kaon mass decreases in medium by 30 mev . the mechanism for such kaon mass modification can be similar to that studied in ref . @xcite . finally we suppose that to improve understanding of experimental results on @xmath5 meson production in heavy ion collisions at cern sps @xcite further detailed investigations taking into account in - medium effects within transport or hydrodynamical models are necessary . the authors would like to thank yu.b . ivanov , k. redlich , v.d . toneev , and d.n . voskresensky for discussions . work of e.k . was supported by the grant of ect@xmath286 , p.f . was partially supported by slovak grant agency for sciences grant no . 2/5085/00 . 99 c.m . ko et al . , phys . rev . * c45 * , 1400 ( 1992 ) . m. asakawa and c.m . ko , nucl . * a572 * , 732 ( 1994 ) . ko and d. seibert , phys . * c49 * , 2198 ( 1994 ) ; phys . rev . * 50 * , r559 ( 1994 ) . k. haglin , nucl . * a584 * , 719 ( 1995 ) . c. song , phys . * b388 * , 141 ( 1996 ) . w. smith and k.l . haglin , phys . * c57 * , 1449 ( 1998 ) . f. klingl , t. waas , and w. weise , phys . v. friese for na49 , j. phys . * g23 * , 1837 ( 1997 ) ; f. sikler for na49 , nucl . phys . * a661 * , 45c ( 1999 ) ; s.v . afanasiev et al . , to be published . n. willis for na50 , nucl a661 * , 534c ( 1999 ) ; l. villatte for na50 , in proc . xxxviii international winter meeting on nuclear physics , bormio , italy , january 24 - 29 ( 2000 ) . bass et al . , prog . part . * 42 * , 313 ( 1999 ) . bratkovskaya , w. cassing , nucl . * a619 * , 413 ( 1997 ) . bravina et al . , phys . rev . * c60 * , 044905 ( 1999 ) . j. sollfrank et al . , j. phys . * g25 * , 363 ( 1999 ) . p. filip , e - print : hep - ex/9605001 , nucl - th/9801058 ; p. filip and j. pisut , phys . lett . * b411 * , 187 ( 1997 ) . j. cleymans , k. redlich , and d.k . srivastava , phys . rev . * c55 * , 1431 ( 1997 ) . yu.b . ivanov , e.g. nikonov , and v.d . toneev , privat communications . peressounko and yu.e . pokrovsky , e - print : hep - ph/0009025 . p. braun - munzinger , i. heppe , and j. stachel , phys . * b465 * , 15 ( 1999 ) ; i. heppe , ( diploma thesis , uni . heidelberg ) . b. kmpfer , e - print : hep - ph/9612336 ; b. kmpfer et al . , j. phys . * g23 * , 2001 ( 1997 ) . b. tomik , u.a . wiedemann and u. heinz , e - print : nucl - th/9907096 . t. peitzmann , e - print : nucl - th/0006025 . t. alber et al . ( na49 coll . ) j. * c2 * , 643 ( 1998 ) . f. cooper and g. frye , phys . * d10 * , 186 ( 1974 ) . bugaev , nucl . * a606 * , 559 ( 1996 ) ; d. molnar and m. gyulassy , phys . rev . * c62 * , 054907 ( 2000 ) . haglin and s. pratt , phys . lett . * b328 * , 255 ( 1994 ) . h.a . weldon , z. phys . * c54 * , 431 ( 1992 ) . s.c . johnson , b.v . jacak , and a. drees , eur . j. * c18 * , 645 ( 2001 ) . g.e . brown et al . , phys . c * 43 * , 1881 ( 1991 ) ; g.e . brown , c.m . ko and k. kubodera , z. phys . * a341 * , 301 ( 1992 ) ; m. lutz , a. steiner and w. weise , nucl . * a574 * , 755 ( 1994 ) ; e.e . kolomeitsev , b. kmpfer and d.n . voskresensky , nucl . * a588 * , 889 ( 1995 ) ; e.e . kolomeitsev , b. kmpfer and d.n . voskresensky , int . j. mod . * e5 * , 313 ( 1996 ) ; c.m . ko , j. phys . * g27 * , 327 ( 2001 ) ; m.f.m . lutz and c.l . korpa , eprint : nucl - th/0105067 ; b. friman and h.j . pirner , nucl.phys . * a617 * , 496 ( 1997 ) ; m. lutz , b. friman , and g. wolf , nucl . phys . * a661 * , 526 ( 1999 ) ; m. post , s. leupold , and u. mosel , eprint : nucl - th/0008027 . m. asakawa et al . , phys . rev . * c46 * , 1159 ( 1992 ) . g. chanfray and p. schuck , nucl . * a555 * , 329 ( 1993 ) ; m. herrmann , b. friman , and w. nrenberg , nucl.phys . * a560 * ( 1993 ) 411 ; m. urban et al . * a641 * , 433 ( 1998 ) . brown and m. rho , phys . 66 * , 2720 ( 1991 ) ; g.q . li , c.m . ko , and g.e . brown , nucl . a606 * , 568 ( 1996 ) ; y. kim et al . , phys . * c62 * , 015202 ( 2000 ) . titov , t.i . gulamov , and b. kmpfer , phys . * d53 * , 3770 ( 1996 ) ; c. song , phys . * d53 * , 3962 ( 1996 ) ; h. van hees and j. knoll , eprint : hep - ph/0007070 . r. rapp and c. gale , phys . rev . * c60 * , 024903 . lee , c. song , and h. yabu , phys . lett . , 407 ( 1995 ) . c. song , s.h . lee and c.m . ko , phys . * c52 * , 476 ( 1995 ) . voskresensky , jetp * 78 * , 793 ( 1994 ) . shuryak , phys . rev . * d42 * , 1764 ( 1990 ) ; nucl . phys . * a533 * , 761 ( 1991 ) ; e.v . shuryak and v. thorsson , nucl . * a536 * , 739 ( 1992 ) .

this work is part of a series of papers devoted to the study of gas adsorption and capillary condensation in silica aerogels@xcite . our goal is to provide a theoretical interpretation of @xmath0he adsorption experiments that have been performed over the last 15 years in several low - temperature physics groups@xcite . more generally , we wish to propose a theoretical framework for describing the _ nonequilibrium _ ( hysteretic ) behavior of fluids in disordered porous solids@xcite . it is worth noting that the experimental studies of helium in aerogel were originally aimed at understanding the effect of quenched random impurities on _ equilibrium _ critical phenomena and phase separation . in particular , the study of the liquid - vapor transition of @xmath0he followed from the suggestion that the critical behavior of mixtures or pure fluids confined in porous media or in contact with a gel network could be interpreted in terms of the random field ising model ( rfim)@xcite . in this perspective , the use of aerogel was motivated by its very tenuous structure ( which minimizes the effects of confinement ) and by the fact that the porosity ( and thus the amount of disorder ) can be varied in a controlled way . however , in spite of the early claim that mesoscopic equilibrium phase separation of @xmath0he could be observed in a @xmath1 porosity aerogel@xcite , leading to an estimation of the order parameter critical exponent @xmath2 , it now seems more and more questionable that true equilibrium behavior is observed in experiments . in particular , recent measurements of sorption isotherms using a high - precision capacitive technique@xcite show that hysteresis between filling and emptying persists to temperatures very close to the critical temperature of pure helium ( @xmath3 , and gradual adsorption is always observed above @xmath4 , instead of the sharp vertical step expected for two - phase coexistence . similar observations have been made by another group@xcite , using a low - frequency mechanical pendulum technique and , more recently , coupling pressure and optical measurements . these results are reminiscent of the behavior of fluids confined in dense porous glasses which is usually interpreted in terms of capillary condensation . although this moves away the exciting perspective of studying the putative equilibrium vapor - liquid critical point of @xmath0he inside aerogel , the remarkable changes in the hysteretic behavior of the adsorbed fluid with temperature and porosity remain to be explained . in particular , the hysteresis loops in very light aerogels have a rectangular ( instead of triangular ) shape at very low temperature ( for instance , at @xmath5 in @xmath6 porosity aerogel@xcite ) with a very steep adsorption branch that suggests the presence of a genuine , but _ nonequilibrium _ , first - order phase transition . this is in agreement with our theoretical description@xcite which predicts the change from a continuous to a discontinuous adsorption isotherm as one decreases the temperature ( at constant porosity ) or increases the porosity ( at constant temperature ) , the jump in the fluid density corresponding to a macroscopic avalanche in the system . ( note that we only focus here on adsorption . the behavior on desorption is also interesting but is related to other physical processes as discussed in refs.@xcite . ) this scenario , which resembles the one predicted for the field - driven rfim at zero temperature@xcite and observed in some disordered magnetic systems@xcite , suggests the existence of a line of disorder - induced critical points in the temperature - porosity diagram with an associated scaling behavior . in the present work we further elaborate on this issue by presenting new results concerning the avalanche properties as a function of temperature and porosity and by studying the scaling behavior of the adsorption isotherms in the vicinity of the nonequilibrium critical points . we are also interested in characterizing the avalanches geometrically and studying their possible relation with the aerogel structure . the paper is arranged as follows . in section 2 we review the model and the theory . in section 3 we present the results for the statistical properties of avalanches . in section 4 we discuss the scaling behavior of the adsorption isotherms and present the phase diagram of the model in the temperature - porosity plane . we conclude , in section 5 , with a discussion of the relevance of our results to experiments . since both the model and the theory were described in detail in ealier work ( see in particular ref.@xcite ) , we will restrict the presentation to some key aspects that are relevant to our present goal . the gel - fluid system is modelled by a _ coarse - grained _ lattice - gas where each cell of the size of a silica particle ( typically a few nanometers@xcite ) is occupied either by the solid or the fluid . the distribution of silica particles on the lattice is obtained by a diffusion - limited cluster - cluster aggregation ( dlca ) algorithm@xcite , so as to model the intricate microstructure of actual base - catalized silica aerogels . on the other hand , fluid particles can equilibrate with a reservoir that fixes their chemical potential @xmath7 and the temperature @xmath8 . multiple occupancy of a site is forbidden and only nearest - neighbors ( nn ) attractive interactions are considered . for reasons that are explained in ref.@xcite , we use a body - centered cubic ( bcc ) lattice and periodic boundary conditions ( @xmath9 where @xmath10 is the linear size ) . for a given configuration of the solid matrix , fluid configurations are obtained by minimizing a grand - potential functional in the local mean field approximation@xcite , @xmath11&=k_bt \sum_{i } [ \rho_i\ln \rho_i + ( \eta_i-\rho_i)\ln ( \eta_i-\rho_i ) ] -w_{ff } \sum_{<ij > } \rho_i\rho_j -w_{sf } \sum_{<ij > } [ \rho_i(1-\eta_j ) + \rho_j(1-\eta_i)]-\mu \sum_{i } \rho_i\end{aligned}\ ] ] where @xmath12 ( @xmath13 ) is the thermally averaged fluid density at site @xmath14 and @xmath15 is the quenched occupancy variable for the gel particles ( @xmath16 if the site is occupied by the gel and @xmath17 otherwise ) . @xmath18 and @xmath19 denote the fluid - fluid and solid - fluid interactions , respectively , and the double summations run over all distinct pairs of nn sites . the gel porosity is given by @xmath20 and the ratio @xmath21 controls the wettability of the solid surface . @xmath22 $ ] thus defines a grand - potential landscape whose extrema are obtained by solving the corresponding euler - lagrange equations @xmath23 . at fixed @xmath8 and @xmath24 , and for a given realization of the solid , this yields a set of @xmath25 coupled nonlinear equations @xmath26 ) } } \ , \end{aligned}\ ] ] where @xmath27 and the sum runs over the nearest neighbors of site @xmath14 . a crucial feature of these equations is that they may have a large number of solutions at low temperature ( and in a certain range of @xmath24 ) as a consequence of the randomness of the solid matrix@xcite . we believe that this feature is the key for understanding the physics of adsorption in disordered porous solids . especially important are the local _ minima _ of @xmath22 $ ] which are obtained by solving the set of eqs . ( 2 ) iteratively@xcite and which correspond to metastable states ( to be distinguished from the global minimum of @xmath22 $ ] that defines the true equilibrium state ) . note that despite its mean - field character ( by which thermal fluctuations are only crudely described ) , the present approach fully accounts for the disorder - induced fluctuations . in contrast with the standard global mean - field approximation , our treatment still allows to describe strongly inhomogenous configurations and to predict non - classical critical behavior . computing the metastable states is however not sufficient for describing the nonequilibrium behavior of the system , and one also needs to specify a dynamics . our theoretical description@xcite is based on the use of a non - ergodic , deterministic , _ zero - temperature_-like dynamics which neglects all thermally activated processes that would allow the system to cross the ( grand - potential ) barriers : a change in the fluid configurations is thus only due to a change in the landscape that comes itself from a change in the temperature or from a variation of the external field ( here the chemical potential @xmath24 ) . as @xmath24 varies ( for instance , when the pressure in the reservoir is slowly increased at constant @xmath8 ) , the system either follows the minimum in which it was trapped as this minimum deforms gradually , or it falls instantaneoulsy into another minimum when the former reaches its stability limit . this latter move is a discontinuous and irreversible process , an _ avalanche _ , which is at the origin of the history - dependent behavior of the system . the avalanche corresponds to some _ collective _ condensation event inside the gel network which manifests itself by a jump in the adsorption isotherm ( whereas the gradual deformation of a minimum only corresponds to a slight swelling of the liquid domains ) . the size of the avalanche is thus given by the size of the jump , i.e. the discontinuous change in the average fluid density @xmath28 . as was shown qualitatively in ref.@xcite , these condensation events occur at different length scales inside the gel , depending on chemical potential , temperature , and porosity . in particular , when @xmath29 is large enough , for instance @xmath30 , there exists a critical point separating a high - temperature regime where all condensation events are of microscopic or mesoscopic size ( and the adsorption isotherm is smooth in the thermodynamic limit ) , and a low - temperature regime where a macroscopic fraction of the gas condenses , which yields a finite jump in the isotherm . it is one goal of the present work to put these observations on a more quantitative basis . we shall discuss below in more detail the relevance of this theoretical description to the actual experimental situation . let us just recall that this picture is expected to be valid when there are widely separated time scales in the problem . specifically , the observation time must be much larger than the local equilibration time ( i.e. the time needed to relax from a marginally stable state to a nearby metastable state ) but much smaller than the time associated to thermal activation ( i.e. the time that it takes to escape from a local minimum via a nearby saddle point)@xcite . this requires that energy barriers are much larger than temperature , which is indeed the case in many experimental situations dealing with random systems . it may also be noted that our definition of avalanches generalizes the one used at @xmath31 ( see e.g. ref.@xcite ) . we indeed introduce an explicit temperature - dependence in the problem by allowing the ( mean - field ) free - energy landscape itself to change with @xmath8 ( for a similar approach in the context of random magnets , see e.g. ref . as shown in our previous works , this dependence explains quite convincingly the variations observed in the hysteretic behavior of gases adsorbed in disordered solids at low enough temperature . as discussed in the preceding papers of this series@xcite , the number , size and shape of the condensation events along an adsorption isotherm depend on porosity and temperature . they also depend on the value of the chemical potential at which they occur . the random character of these events ( each finite - size sample has a different isotherm ) requires a statistical study of their properties , an issue that has been investigated in great detail in the context of the field - driven rfim@xcite . the first problem is to identify each avalanche properly . at @xmath31 , when the microscopic variables can only take discrete values ( such as the ising spins in the rfim ) , this is an easy task because the energy landscape only changes through the disappearance of minima , which induces the avalanches . one can then vary the external field until a spin becomes unstable ( this spin is unique when the random field distribution is continuous ) and then count the number of spins that flip during the avalanche to determine its size . at finite temperature , on the other hand , one has to deal with continuous variables ( the local fluid densities @xmath12 in the present case ) , and since @xmath32 , the increment in the chemical potential along the adsorption isotherm , is always finite , one has to distinguish between the gradual deformation of a minimum and the jump associated to the passage to another local minimum , both leading to a change in the total density . moreover , if @xmath32 is not small enough , several avalanches may overlap , which biases the statistics of their number or size distribution . although , in principle , one could precisely locate the beginning of each avalanche by following the evolution of the eigenvalues of the hessian matrix @xmath33 with @xmath24 or @xmath8 ( an avalanche starts when a local spinodal corresponding to the vanishing of a certain eigenvalue is reached@xcite ) , this is a daunting numerical task for a large system ( with @xmath34 ) and it is more advisable to use another procedure . let us first note that the determination of the location and size of the avalanches depends on the precision in the computation of the metastable states . since these latter are obtained by solving the set of eqs . ( 2 ) iteratively , it must be checked that the results are not significantly modified when the convergence criterium of the iterative procedure is changed . all the results presented here were computed with a precision @xmath35 ( i.e. , the calculation was stopped at iteration @xmath36 when @xmath37 ) . it was found that decreasing @xmath38 to @xmath39 did not change the results by more than @xmath40 ( but in turn the computational time was increased by a considerable factor ) . porosity aerogel at @xmath41 ( @xmath42 is the average fluid density inside the aerogel and @xmath24 the gas chemical potential ) . calculations were done in a sample of size @xmath43 using the dichotomy procedure explained in the text.,width=302 ] consider now the variations @xmath44 in the local fluid densities when increasing the chemical potential by a very small ( but finite ) increment @xmath32 ( for instance @xmath45 , with @xmath18 taken as the energy unit ) . we show in fig . 1 a small portion of a typical low - temperature isotherm computed in a @xmath46 porosity sample of size @xmath43 at @xmath41 ( @xmath47 ) . the random character of the condensation events results in the fine staircase structure of the curve , with the vertical steps corresponding to the avalanches separated by nearly horizontal portions corresponding to the smooth deformation of the grand - potential local minima ( note the small scale on the vertical axis ) . the slope of these near - plateaus depends on temperature and its order of magnitude can be estimated from the isothermal compressibility of the pure lattice gas . it turns out that _ all _ the @xmath12 s vary by a very small amount in these flat portions of the isotherms ( less than @xmath48 ) whereas the vertical steps are due to a significant variation of _ some _ of the @xmath12 s ( a low temperature , this typically corresponds to a gazeous region becoming liquid ) . we can use this feature to identify the avalanches unambiguously . specifically , we consider that the change @xmath49 in the isotherm is only due to a smooth deformation of a grand - potential minimum when @xmath50 is smaller that some threshold , which we take equal to @xmath51 . otherwise there is an avalanche , and we associate to it all the sites for which @xmath52 . the size of the avalanche is then @xmath53 , where the sum is restricted to these sites@xcite ( by taking @xmath54 one would neglect the small increase of @xmath42 which is due to the smooth " variation of the @xmath12 s on the other sites of the lattice ) . this criterium thus implies that avalanches for which @xmath55 can not be detected . by choosing a smaller threshold , e.g. @xmath56 , one may find new avalanches , but their size @xmath57 is very small ( always less than @xmath58 ) . as a consequence , avalanches with @xmath59 will not be considered in the following . in any experiment there is also a threshold below which signals can not be detected . since one must collect the results of many samples to get a good statistics of the avalanches properties , it is impossible in practice to study a whole isotherm with an increment @xmath32 as small as @xmath60 . it is much more efficient to proceed by dichotomy , comparing the configurations obtained for two values of @xmath24 , say @xmath61 and @xmath62 ( initially , @xmath61 is a large negative value and @xmath63 , the value at saturation ) . if @xmath64 between the two configurations , one then compares the configurations for @xmath61 and @xmath65 on the one hand , and @xmath65 and @xmath62 on the other hand . one then looks for an avalanche in the two intervals , and the same procedure is repeated until no avalanche is found between two configurations , or the difference in chemical potential becomes smaller than @xmath60 ( this is the method that was used for obtaining the isotherm shown in fig . 1 , which explains that the points are irregularly spaced ) . finally , to characterize geometrically an avalanche , which is a random object , one may define an effective radius of gyration " @xmath66 by attributing to each site @xmath14 a mass " that corresponds to the change in the local density , @xmath67 where the sum is restricted to the sites contributing to the avalanche ( i.e. with @xmath68 ) and @xmath69 is the center of mass " of the avalanche@xcite . let us first study how the number of avalanches with a given size varies with temperature and porosity . to illustrate the general trend , we consider isotherms obtained in @xmath46 and @xmath1 porosity aerogels , as shown in figs . 2(a ) and 3(a ) . the sequence of avalanches was recorded in @xmath70 samples of linear size @xmath43 and @xmath71 samples of linear size @xmath72 , respectively ( this approximately corresponds to the same ratio @xmath73 , where @xmath74 is the correlation length of the aerogel@xcite ) . figs . 2(b ) and 3(b ) show @xmath75 , the ( unnormalized ) avalanche size distribution ( @xmath76 , the total number of avalanches of size larger than @xmath58 ) . as @xmath8 increases , @xmath77 decreases from @xmath78 to @xmath79 and from @xmath80 to @xmath81 , respectively . the size of the largest possible avalanche corresponding to a simultaneous change of all the @xmath12 s from @xmath82 to @xmath17 is @xmath83 . one thus has @xmath84 and @xmath85 in the @xmath46 and @xmath1 aerogels , respectively . it is readily seen that there exist much larger avalanches in the lighter aerogel . this is a direct consequence of the larger void space available , as discussed in our previous papers@xcite . in particular , at @xmath86 , some avalanches span the whole system , signaling that a macroscopic avalanche possibly occurs in the thermodynamic limit . note however that the statistics for very large avalanches is not good , which explains the erratic behavior seen in fig . 3(b ) for large @xmath57 . moreover , finite - size effects play an important role in the large - porosity / low - temperature regime and they should be taken into account to properly describe the evolution towards the thermodynamic limit ( see ref.@xcite for a careful study in the case of the rfim ) . unfortunately , this would require a considerable amount of numerical work , and we shall therefore only concentrate our discussion on the non - spanning avalanches . porosity aerogel at @xmath87 , and @xmath88 ( from top to bottom ) ; data correspond to an average over @xmath70 gel samples of size @xmath43 . ( b)corresponding ( unnormalized ) avalanche size distributions @xmath75 . the solid lines represent the fit @xmath89 ( color on line).,width=377 ] porosity aerogel at @xmath90 , and @xmath91 ; data correspond to an average over @xmath71 gel samples of size @xmath72 ( color on line).,width=377 ] figs . 2(b ) and 3(b ) show that the evolution of @xmath75 with temperature is quite similar in the two aerogels : as @xmath8 increases , both large _ and _ small avalanches are suppressed , and , as a consequence , the total number @xmath77 decreases . only the number of avalanches of intermediate size stays roughly constant with @xmath8 . if one refers to what happens in the field - driven rfim at @xmath31 , this result may appear surprising . indeed , the isotherms get smoother as one increases the temperature , like the magnetization curves @xmath92 in the rfim as one increases the width of the random field distribution@xcite . this seems to imply that the temperature plays the role of an effective disorder in the system . therefore , one would expect that large avalanches become rare as @xmath8 increases ( which is indeed what happens ) , but also that small avalanches proliferate , so that @xmath77 increases on average@xcite . it turns out , however , that temperature also plays another role in our system , a role which is well documented in the literature devoted to capillary condensation@xcite : adsorption in a small cavity becomes more gradual as @xmath8 increases . specifically , mean - field calculations for a single pore of simple geometry ( for instance a slit of width @xmath93 ) predict that adsorption is continuous and reversible above some critical temperature @xmath94 which decreases as @xmath93 decreases . since small avalanches are associated with gas condensation in small cavities or crevices defined by the neighboring gel strands , this phenomenon explains the decrease of the number of small avalanches ( when @xmath95 , the temperature at which the hysteresis disappears , there are no avalanches at all and the isotherms become reversible@xcite ) . it is clear that the random - field picture can only be applied at a scale much larger than the correlation length of the solid : schematically , capillary condensation prevails for @xmath96 whereas the disordered character of the aerogel structure plays a major role for @xmath97 . the two mechanisms compete at the scale @xmath98 which explains that the number of avalanches of intermediate size does not vary significantly with @xmath8 . if we now concentrate on the disorder - dominated regime , the most remarkable feature , quite visible in the @xmath99 gel at @xmath86 , is that @xmath75 follows an approximate power - law behavior over several decades . this is the defining signature of criticality , an issue that has been at the focus of intense research activity in recent years , for instance to explain the scaling properties of the barkausen noise in magnetic materials@xcite . in the rfim picture , the power - law behavior is related to the existence of a disorder - induced critical point@xcite . in this context , the avalanche size distribution should be analyzed using the form @xmath100 where @xmath101 is a critical exponent@xcite , @xmath102 is a scaling function , and @xmath103 a cut - off that depends on the distance to the critical point ( @xmath104 diverges as the critical disorder is approached ) . our results are however too limited to obtain sensible scaling collapses ( although a crude attempt shows that the scaling function is non - monotonic , in agreement with the rfim behavior@xcite ) . we simply replace the function @xmath102 by an exponential form only describing the cut - off of the power - law at the scale @xmath104 . figs . 2(b ) and 3(b ) show the fits of @xmath75 in the two aerogels according to this approximate treatment . the power - law regime is very limited in the @xmath46 aerogel even at the lowest temperature , indicating that the system is clearly outside the critical region . on the other hand , in the @xmath1 gel , the curves show a linear portion on a log - log scale that increases as @xmath8 decreases : the cutoff @xmath104 is pushed to larger values of @xmath57 ( from @xmath105 to @xmath106 as @xmath107 varies from @xmath91 to @xmath88 ) and reaches the box size for @xmath86 ( our set of data is however too limited to estimate the critical exponent associated to the divergence of @xmath104 ) . the exponent @xmath101 increases and becomes approximately equal to @xmath108 in the fully developed power - law regime at @xmath86 . taking into account the fact that our systems are small and that @xmath101 in a finite system depends on disorder@xcite , this value does not appear inconsistent with that predicted for the rfim , @xmath109@xcite . in any case , the gel - fluid system is clearly close to criticality at this temperature . recorded in the @xmath46 porosity samples as a function of chemical potential @xmath24 at @xmath86 ( a ) and @xmath110 ( b),title="fig:",height=264 ] recorded in the @xmath46 porosity samples as a function of chemical potential @xmath24 at @xmath86 ( a ) and @xmath110 ( b),title="fig:",height=264 ] porosity samples at @xmath86 ( a ) and @xmath111 ( b).,title="fig:",height=264 ] porosity samples at @xmath86 ( a ) and @xmath111 ( b).,title="fig:",height=264 ] the number and the size of the avalanches along an adsorption isotherm strongly depend on the value of the chemical potential . we thus also recorded the value of @xmath24 at which each avalanche of size @xmath57 occured . figs . 4 and 5 show the corresponding point clouds in the @xmath46 and @xmath1 gels at @xmath86 and @xmath88 , and @xmath86 and @xmath91 , respectively . as a general rule , the size of the largest avalanches increases with @xmath24 , especially in the last stage of the adsorption ( note the logarithmic scale on the vertical axis ) . but the most striking feature is that avalanches of very different sizes occur in a small interval of @xmath24 . for instance , in the @xmath46 gel at @xmath86 and @xmath112 , @xmath57 varies from @xmath58 to more than @xmath113 . in the @xmath1 gel at the same temperature and @xmath114 , the avalanches span almost all possible sizes between @xmath58 and @xmath115 . these observations are in complete contradiction with the picture based on the independent pore model which is often used to interpret capillary condensation phenomena@xcite . in this model , only pores of a certain characterisic size @xmath93 fill at a given value of the chemical potential , and the adsorbed fluid density is directly related to the number of pores of this size that can be found in the material . ( one then try to extract from the isotherm the pore size distribution . ) this picture is clearly wrong in the present case because of the very open and intricate character of the gel structure that allows condensation events of very different sizes . the geometrical characterization of the avalanches is rather difficult . using @xmath116 as an estimate of the extension of an avalanche is misleading because events with very different shapes are observed ( see for instance figs . 6 and 7 in ref.@xcite ) . this variety in the geometry is reflected in the values of the radius of gyration @xmath66 , as defined by eq . the value of @xmath66 is indeed proportional to the maximal extension of the avalanche , but the prefactor depends on its shape . figs . 6 and 7 show how @xmath66 varies with @xmath57 in the @xmath46 and @xmath1 aerogels . one can see that avalanches with a similar size may have quite different radii of gyration . for instance , in the @xmath46 aerogel at @xmath86 , avalanches of size @xmath117 have a radius that varies between @xmath118 and @xmath58 ( note that for a compact object , a factor of @xmath119 in @xmath66 would correspond to a factor @xmath120 in @xmath57 ) . similarly , in the @xmath1 aerogel at the same temperature , @xmath66 varies from @xmath58 to @xmath121 for avalanches of size @xmath122 . it is therefore impossible to associate a unique characteristic length to an avalanche of a given size . moreover , by examining the avalanches that occur in a small interval of @xmath24 , we find that they also have very different radii of gyration . as the temperature increases , the values of @xmath66 for a given size @xmath57 are less scattered . indeed , avalanches become more compact as they correspond to condensation events that occur in a single cavity of the aerogel . this evolution is very clear in the @xmath46 aerogel : at @xmath110 , the radius of gyration of most avalanches is between @xmath119 and @xmath118 and there are almost no avalanches with @xmath66 larger than 6 . on the other hand , there is still a large domain of variation of @xmath66 in the @xmath1 aerogel at @xmath111 , and one should go to even higher temperatures to observe a significant reduction . there is in fact a remarkable resemblance between figs . 6(a ) and 7(b ) which suggests that the avalanche properties are similar apart from a scale factor of about @xmath119 . the adsorption isotherms have also similar shapes , as can be seen in figs . 2(a ) and 3(a ) . this confirms the observation made in ref.@xcite that the adsorption processes in a low porosity gel at low temperature and in high porosity gel at a higher temperature have similar properties . and size @xmath57 of the avalanches in the @xmath46 porosity samples at @xmath86 ( a ) and @xmath110 ( b ) . the inset in ( a ) is a magnification of the point cloud of the spherical avalanches ( see text ) in log - log scale ; the solid line represents the equation @xmath123 ( color on line).,title="fig:",height=264 ] and size @xmath57 of the avalanches in the @xmath46 porosity samples at @xmath86 ( a ) and @xmath110 ( b ) . the inset in ( a ) is a magnification of the point cloud of the spherical avalanches ( see text ) in log - log scale ; the solid line represents the equation @xmath123 ( color on line).,title="fig:",height=264 ] porosity samples at @xmath86 and @xmath91 . in the inset of ( b ) the solid line represents the equation @xmath124 ( color on line).,title="fig:",height=264 ] porosity samples at @xmath86 and @xmath91 . in the inset of ( b ) the solid line represents the equation @xmath124 ( color on line).,title="fig:",height=264 ] in both fig . 6(a ) and fig . 7(b ) , one may notice that a small set of avalanches stands out . in the @xmath46 gel , at @xmath86 , these avalanches have a size @xmath57 in the range @xmath125 and a smaller radius of gyration than the other ones . they are therefore more compact , and a direct observation of some of them shows that their shape is almost perfectly spherical . this is confirmed by a systematic study of the asphericity parameter which is always smaller than @xmath126@xcite . as shown in the inset of fig . 6(a ) on a log - log scale , the radius of gyration of these avalanches varies approximately like @xmath127 which is indeed close to the expected value for spherical compact objects ( @xmath128^{1/3}\approx 0.48 s^{1/3}$ ] for a sphere of radius @xmath129 ) . the radius @xmath129 varies between @xmath130 and @xmath118 , this latter value corresponding approximately to @xmath74 which is also more or less the size of the largest voids in the aerogel @xcite . moreover , as shown in fig . 8 , most of these avalanches occur at the end of the adsorption process : they thus correspond to the condensation of the last gas bubbles that remain in the large cavities . of spherical avalanches along the adsorption isotherm in the @xmath46 porosity aerogels at @xmath86 . the average isotherm is shown as a continuous line.,width=226 ] a similar behavior is observed in the @xmath1 aerogel at @xmath111 ( fig . the avalanches are again approximately spherical and occur in the last stage of the adsorption process . however , since the temperature is higher , the boundaries of the gas bubbles are less well defined , and the radius of gyration of varies approximately like @xmath131 . the largest observed value of @xmath132 is again comparable to the size of the largest cavities in the aerogel ( @xmath133 ) . although there are also some other remaining bubbles of more complicated shape ( see figs . 7 and 9 in ref.@xcite ) , these spherical avalanches actually represent the only case where one can relate directly the characteristic size of the condensation events to the structure of the aerogel . we stress again that , in general , the avalanches have a complicated shape so that one can not describe the condensation of the adsorbed gas in terms of constant curvature bubbles"@xcite . as a function of the inverse radius of the spherical avalanches in the @xmath46 and @xmath1 aerogels at @xmath86 and @xmath111 , respectively . ( b ) is a magnification of the small point cloud in ( a ) near the origin corresponding to the avalanches in the @xmath1 aerogel . the solid line with a slope @xmath17 corresponds to the kelvin equation ( see text ) ( color on line).,width=359 ] the fact that spherical avalanches are clearly identified gives us the opportunity for testing a macroscopic description of the gas condensation in terms of a competition between volume and surface contributions . such a description underlies the classical interpretation of capillary condensation based on the kelvin equation@xcite that is commonly used to extract from the isotherms information on the characteristic pore size in the solid . we thus write the change in the grand potential associated to the condensation of a spherical gas bubble of radius @xmath129 at the chemical potential @xmath24 as@xcite @xmath134 \frac{4\pi}{3}r^3-\gamma 4\pi r^2\ ] ] where @xmath135 is the corresponding bulk pressure , @xmath136 is the pressure of the metastable liquid at the same chemical potential ( both pressures are computed from the equation of state of the bulk lattice gas ) , and @xmath137 is the ( planar ) liquid - gas surface tension . all these terms except @xmath129 are temperature dependent . assuming that condensation occurs at equilibrium , the kelvin equation ( in its laplace form ) corresponds to setting @xmath138 in the equation . expanding as usual @xmath139 and @xmath140 about @xmath141 and to first - order in @xmath142 ( which is valid for small undersaturations ) , this yields @xmath143 $ ] , where @xmath144 and @xmath145 are the densities of the coexisting liquid and gas at this temperature . since the surface tension on a lattice depends on the orientation of the surface with respect to the lattice vectors , we estimate @xmath137 by taking the average of the three values calculated in the principle lattice planes @xmath146 , @xmath147 and @xmath148@xcite . the corresponding kelvin plot " is shown in fig . we see that the macroscopic " kelvin equation provides the correct trend for the variation of @xmath149 as a function of @xmath150 , in spite of the fact that actual condensation does not occur at equilibrium ( @xmath151@xcite ) and that the adsorbent potential and the curvature corrections are neglected in eq . ( 4 ) ( note that @xmath149 is ten times smaller in the lighter gel , which is due to the effect of temperature , correctly described by the term @xmath152 , and to the change in the average radius @xmath129 by a factor of about @xmath119 ) . the scatter in the data , however , shows that the kelvin equation is only an approximation . the existence of a critical point in our system , at which an infinite avalanche occurs for the first time in the thermodynamic limit and the avalanche size distribution has a power - law behavior , suggests the possibility of performing also a scaling collapse of the adsorption isotherms . in the case of the nonequilibrium rfim at @xmath31 where the disorder is controlled by the width @xmath153 of the distribution of the random fields , standard renormalization group arguments suggest that the magnetization curve @xmath92 for different values of @xmath153 may be collapsed using the scaling form@xcite @xmath154 where @xmath155 is the critical magnetization ( i.e. the magnetization at the critical field @xmath156 for the critical disorder @xmath157 ) , @xmath158 and @xmath159 are the reduced disorder and reduced field respectively and @xmath160 is a universal scaling function ( @xmath161 refers to the sign of @xmath162 ) . the estimated values of the critical exponents in three dimensions are @xmath163 and @xmath164@xcite . is rescaled by the density at @xmath165 ) . from left to right : @xmath166 at @xmath167 . the curves result from an average over @xmath70 samples of size @xmath72 ( color on line).,width=340 ] the situation in our system is more complicated than in the rfim because the hamiltonian ( or , more correctly , the grand - potential functional @xmath168 ) contains three parameters that can be tuned independently ( the porosity @xmath29 , the interaction ratio @xmath169 , and the chemical potential @xmath24 ) , and , moreover , the temperature controls the ruggedness of the free - energy landscape . although a rigorous proof is lacking , it seems however reasonable to assume that the critical behavior is governed by a unique fixed point , which may or may not be the same as for the @xmath31 nonequilibrium rfim . in other words , there are only two relevant variables from a renormalization - group standpoint : the chemical potential ( that plays the role of the magnetic field @xmath170 in the rfim ) and an effective disorder @xmath153 which is some smooth funtion of @xmath169 , @xmath29 and @xmath8 . ( in addition , one could also have to take into account the microstructure of the aerogel , on top of the dependence on the porosity : in particular , one may suspect that the fractal character of very light aerogels , e.g. @xmath171 , modifies the values of the critical exponents or , at least , induces a different scaling behavior in some intermediate regime ; this possibility is not taken into account in the present study which does not consider aerogels of porosity larger then @xmath1 . ) for fixed @xmath169 and @xmath8 , there is a critical value @xmath172 of the porosity , and a critical value @xmath173 of the chemical potential . alternatively , for fixed @xmath169 and @xmath29 , there is a temperature @xmath174 and a chemical potential @xmath175 at which the system becomes critical . if this picture is correct , one expects to observe the same critical behavior whether the critical line @xmath176 ( or @xmath177 ) is approached along a constant-@xmath178 or a constant-@xmath8 trajectory in the temperature - porosity diagram ( for simplicity , we forget the dependence on @xmath169 which is kept constant and equal to @xmath119 in all our computations@xcite ) . consider for instance the collection of isotherms displayed in fig . 10 which covers a significant range of temperatures and porosities . this would correspond to @xmath179 for the real @xmath0he - aerogel system , but this must be considered only as a crude estimation : we indeed recall that our treatment is not aimed at a quantitative description of actual systems because of the coarse - grained character of the model and the use of a ( local ) mean - field approximation . note that the density of the adsorbed fluid is rescaled by its value at @xmath165 so that all the curves go to @xmath17 as @xmath180 ( one has @xmath181 where @xmath182 is the density of the bulk liquid at coexistence ) . isotherms for @xmath183 and @xmath184 according to eq . 8 where @xmath185/\mid \phi- \phi_c(t)\mid^{\beta \delta}$ ] and @xmath186 with @xmath187 , @xmath188 , and @xmath189 ( color on line).,width=302 ] let us first study the scaling behavior when varying the porosity at fixed temperature . by analogy with the rfim , our basic scaling hypothesis is the existence , in the limit @xmath190 , of only two scaling variables @xmath191 and @xmath192 that measure the distance to the critical point @xmath193 . accordingly , the adsorption isotherms in the vicinity of the critical point @xmath194 should scale as @xmath195 where @xmath196 are the corresponding scaling functions and @xmath197 and @xmath198 are two nonuniversal scale factors that are temperature - dependent ( they actually depend on all the specific features of the model@xcite : @xmath169 , the lattice structure , the range of interaction , etc ... ) . note that the scaling variables @xmath199 and @xmath200 need not be exactly @xmath201 and @xmath202 , but their dependence on @xmath29 and @xmath24 should be analytic@xcite . unless one uses the proper variables , the curves do not collapse . in the case of the rfim@xcite , it was shown that the proper scaling must be done with replacing @xmath159 in eq . ( 6 ) by the rotated " variable @xmath203 , where @xmath204 is some non - universal constant ( on the other hand , there is no need to rotate @xmath162 because the corresponding correction is irrelevant when @xmath205 ) . in practice , the rotation of the x - axis can be avoided by simply lining up the peaks in the derivative of the curves , with no correction terms , as is often done in analyzing experimental data ( by considering the slope @xmath206 instead of @xmath207 we also get rid of the dependence on @xmath208 ) . what to do on the y - axis is a matter of choice and , for simplicity , we just take @xmath209 . ( note that in ref.@xcite the curves @xmath210 in the rfim are plotted using @xmath211 instead of the more standard choice @xmath212 ; alternatively , it has been proposed to expand @xmath162 to second - order in @xmath213@xcite ; these different choices are equivalent asymptotically if no rotation is required , but they may change significantly the amplitude of the critical regions . ) therefore , to summarize , we look for a scaling of the form @xmath214 where @xmath209 , @xmath215 is the value of the chemical potential at the maximum of @xmath206 , @xmath216 is the derivative of @xmath196 , and @xmath217 and @xmath218 account for the dependence on @xmath197 and @xmath198 . since it is important to have smooth isotherms so that one can take the derivative numerically , the average must be performed over a significant number of samples : the curves shown in fig . 10 result from an average over @xmath70 gel realizations of size @xmath72 ( however , there are still some annoying irregularities in the average slopes , especially around @xmath219 ) . calculations are of course done in finite systems so that one should also take into account finite - size effects@xcite . these are especially important in the low - disorder regime ( @xmath220 in which there is a discontinuity in the isotherms in the thermodynamic limit . this introduces an additional difficulty in the analysis and it is a very demanding numerical task to perform a systematic finite - size scaling study of all the curves ( see ref.@xcite for such a study of the @xmath221 isotherm in the @xmath1 porosity aerogel ) . as a consequence , we shall only consider here the isotherms that are clearly in the strong disorder regime , which leads to reasonably size - independent results : this eliminates in particular the @xmath222 and @xmath223 isotherms in the @xmath1 porosity aerogel . fig . 11 shows the scaling plot of the slope of the @xmath224 isotherms for @xmath225 and @xmath184 obtained with the choice @xmath226 , @xmath187 , and @xmath189 ( the scale factors @xmath227 and @xmath228 are arbitrarily fixed equal to @xmath17 ) . it is actually very difficult to adjust independently the three unknown parameters @xmath229 , @xmath2 and @xmath176 ; the values of the exponents used here are the ones that allow for a collapse of a whole set of isotherms , as discussed below . it can be seen that the @xmath230 and @xmath231 isotherms collapse onto a single curve , which is not the case of the @xmath232 isotherm , especially on the left - hand side of the peak . this part of the isotherm is indeed dominated by the presence of a liquid layer that coats the aerogel strands@xcite , and in the low - porosity gels the remaining available void space is too small for allowing large avalanches to spread out . this contribution to the adsorption has nothing to do with a critical phenomenon and it is reasonable to eliminate from the scaling plot all the isotherms obtained in the @xmath46 porosity aerogel . more generally , at each temperature , we choose to discard the isotherms that do not belong to the scaling region because the porosity is obviously too far from @xmath176 ( i.e. , the distance from the effective critical disorder is too large ) . this finally leads to the global scaling plot shown in fig . 12 , obtained by adjusting also the nonuniversal scale factors @xmath217 and @xmath218 ( which , however , always remain of order @xmath17 ) . the overall quality of the collapse is not perfect , but similar to that found in the rfim@xcite ( it is actually better because we only keep two isotherms at each temperature@xcite ) . in fact , as discussed in ref.@xcite , the magnetization curves ( or , here , the isotherms ) are quantities that do not collapse well , which makes it difficult to obtain from the scaling plots a precise determination of the critical exponents or of the location of the critical point . although our estimated values of the exponents @xmath229 and @xmath2 are compatible with those of the rfim , the uncertainty is much too large to conclude that our model belongs to the same universality class . on the other hand , it is noteworthy that the value @xmath233 for the exponent @xmath229 is consistent with the value @xmath234 for the finite - size scaling exponent @xmath235 obtained in ref.@xcite , if one uses for @xmath236 the rfim value , @xmath237@xcite . ) . @xmath238 and @xmath169 are the corresponding scaling variables and scaling functions ( color on line).,width=302 ] ( or @xmath176 ) below which there is a first - order nonequilibrium transition in the adsorption isotherm . the squares indicate the isotherms that are included in the scaling plot of fig . 12 ( color on line).,width=340 ] and @xmath239 isotherms in the @xmath1 aerogel according to eq . 9 where @xmath185/\mid t / t_c(\phi)-1\mid^{\beta \delta}$ ] and @xmath240 with @xmath187 , @xmath188 , and @xmath241 ( color on line).,width=302 ] from the values of @xmath176 extracted from the scaling plots at each temperature , we can map out an approximate phase diagram of the model in the temperature - porosity plane . this is shown in fig . 13 , where the critical line has been forced to extrapolate towards the zero - disorder limit @xmath242 . ( let us recall however that , in a real system , there can not be a clear - cut transition close to @xmath243 because thermal fluctuations become larger and can allow the system to cross the free - energy barriers on the experimental time - scale ; moreover , as noted before , we expect the fractal character of the microstructure in very light aerogels to come into play . ) the present calculations indicate that the critical temperature drops very quickly with the porosity , and the extrapolation to @xmath31 predicts that there is no transition at finite temperature in gels of porosity smaller than @xmath244 . this is in agreement with finite - size scaling calculations performed in the @xmath46 porosity aerogel that show that the isotherms are continuous at all temperatures in the thermodynamic limit . on the other hand , in the @xmath1 porosity aerogel , the study of ref.@xcite predicts a transition at @xmath221 , which is indeed below the critical line of fig . 13 . morever , as shown in the preceding section , the avalanche size - distribution has a power - law behavior at @xmath86 , which is the critical temperature extracted from the scaling plot ( @xmath245 ) . it turns out that the same critical temperature is also obtained by studying the scaling behavior when varying the temperature at fixed porosity . the scaling is now performed according to the expression @xmath246 where @xmath247 and @xmath248 and @xmath249 are porosity - dependent scale factors . it is clear from fig . 13 that this procedure can only be used with the @xmath1 porosity aerogel . the scaling plot of the @xmath110 and @xmath239 isotherms is shown in fig . 14 ( again , the @xmath250 isotherm is discarded because of the presence of important finite - size effects ) . it can be seen that a good scaling is obtained with the same values of the critical exponents as used before . remarkably , the resulting scaling function is also the same and it can be included in fig . 12 by adjusting the scale factors @xmath248 and @xmath249 . this is a clear indication that the critical behavior of the system is controlled by the same fixed point for all temperatures and porosities . we now discuss the comparison of all these results with experiments . in this work and our earlier work@xcite , we have presented a detailed numerical study of the adsorption of @xmath0he in silica aerogel that explains the modification in the shape of the isotherms as the temperature or the porosity is changed . we have found that the adsorption occurs via a sequence of irreversible condensation events ( avalanches ) of varying size and shape . the traditional picture of capillary condensation in mesoporous solids in which pores of increasing size fill up successively as the chemical potential or the gas pressure increases does not hold in aerogel , except , approximately , for the special category of spherical avalanches that occur at the end of the adsorption process . our calculations predict that there exists a line of critical points in the temperature - porosity diagram separating two regimes of continuous and discontinuous adsorption and characterized by a power - law behavior of the avalanche size distribution . there is a rather large scaling region where adsorption isotherms at different temperatures and different porosities collapse onto a single universal curve . although the system sizes considered in this work are too small to reach a definite conclusion , the estimated values of the critical exponents seem to be consistent with those of the three - dimensional field - driven rfim at @xmath31@xcite . the qualitative agreement between the simulated and experimental loops@xcite is a strong argument in favor of the validity of the above scenario . however , a complete validation still awaits the experimental description of the scaling behavior of the isotherms and the observation of a power - law behavior in the avalanche size distribution . this latter issue requires some additional comments . it may be indeed distressing that avalanches have not been observed so far with @xmath0he in aerogel . in fact , the only exemple of avalanche behavior in adsorption / desorption experiments concerns the draining of superfluid @xmath0he from nuclepore@xcite , a porous material that contains a random spatial distribution of interconnected cylindrical pores . in this case , superfluidity appears to be a critical factor for the existence and propagation of avalanches via the presence of a superfluid film on the nuclepore substrate that provides a coupling mechanism even between distant pores . it is likely however that such a mechanism is not necessary for avalanches to occur in aerogels because of very open character of the microstructure . there are actually several requirements for observing avalanches in sorption experiments , which are all satisfied in the case of the experiments with nuclepore@xcite , but not in the case of the existing aerogel studies@xcite . let us first recall that , by definition , avalanches correspond to some collective reorganization of the system _ in response _ to a variation of the external driving force . if one is changing the density rather than the chemical potential or the pressure ( by introducing a controlled amount of gas into the sample and then measuring the pressure change ) , the system may follow a different path among the metastable states , and the vertical jumps in the isotherms that are the signature of the avalanches may not be observed@xcite ( this could explain that there are many data points along the adsorption isotherm at @xmath5 in @xmath6 aerogel displayed in ref.@xcite ) . secondly , as already noted , the observation time must be larger than the time the system takes to respond to the change in the external control variable . this can be done by continuously changing the chemical potential at a very slow rate , and measuring the amount of gas in the solid regularly ( every @xmath238 seconds ) . in practice , this procedure is only possible if the ( local ) equilibration time is short ( for instance , it is smaller than one second with superfluid helium in nuclepore ) . unfortunately , the relaxation times in aerogel are much longer ( of the order of two hours with @xmath0he at low temperature in the experiments of ref.@xcite ) , and one has to change @xmath24 or @xmath251 by a finite step and keep it fixed until a new local equilibrium ( i.e. , a new metastable state ) is reached . with this procedure , the issue of having a system with a long equilibration time boils down to stability of the experimental conditions and of course patience ; in principle , one could then observe avalanches . yet , if @xmath149 or @xmath252 is not small enough , the avalanches may overlap and the resulting isotherm may appear continuous . this is probably the situation with the experiments in ref.@xcite where @xmath253 ( to be compared with @xmath254 in the nuclepore experiments@xcite ) . finally , a last requirement is that the sensitivity of the measurements is large enough to discriminate the jumps in the adsorbed fluid density . this condition is more easily satisfied when these jumps are large enough , which implies that one must be close to the critical line displayed in fig . although one must be very cautious in comparing our temperature scale with the true experimental one , it seems that this last condition is also not satisfied in the existing experiments . f. detcheverry , e. kierlik , m. l. rosinberg , and g. tarjus , phys . e * 68 * , 61504 ( 2003 ) . f. detcheverry , e. kierlik , m. l. rosinberg , and g. tarjus , langmuir * 20 * , 8006 ( 2004 ) . a. p. y. wong and m. h. w. chan , phys . 65 * , 2567 ( 1990 ) . d. j. tulimieri , j. yoon , and m. h. w. chan , phys . lett . * 82 * , 121 ( 1999 ) . j. beamish and t. herman , physica b * 340 * , 329 ( 2003 ) ; j. beamish and t. herman , j. low temp . phys . * 134 * , 339 ( 2004 ) ; t. herman , ph.d . thesis , university of alberta ( 2005 ) ; t. herman , j. day , and j. beamish , preprint cond - mat/0505430 . c. gabay , p.e . wolf , and l. puech , physica b * 284 * , 97 ( 2000 ) ; c. gabay , f. despetis , p.e . wolf , and l. puech , j. low temp . 121 * , 585 ( 2000 ) ; c. gabay , ph.d . thesis , universit joseph fourier , grenoble i ( 2001 ) ; t. lambert , c. gabay , l. puech , and p.e . wolf , j. low temp . phys . * 134 * , 293 ( 2004 ) ; t. lambert , ph.d . thesis , universit joseph fourier , grenoble i ( 2004 ) . e. kierlik , p. a. monson , m. l. rosinberg , l. sarkisov , and g. tarjus , phys . lett . * 87 * , 055701 ( 2001 ) ; e. kierlik , p. a. monson , m. l. rosinberg , and g. tarjus , j. phys . : condens . matter . * 14 * , 9295 ( 2002 ) ; m.l . rosinberg , e. kierlik , and g. tarjus , europhys . lett . * 62 * , 377 ( 2003 ) ; f. detcheverry , e. kierlik , m. l. rosinberg , and g. tarjus , physica b * 343 * , 303 ( 2004 ) ; f. detcheverry , e. kierlik , m. l. rosinberg , and g. tarjus , adsorption * 11 * , 115 ( 2005 ) . f. brochard and p.g . de gennes , j. phys . lett . * 44 * , l785 ( 1983 ) ; p. g. de gennes , j. phys . chem . * 88 * , 6469 ( 1984 ) . j. p. sethna , k. dahmen , s. kartha , j. a. krumhansl , b. w. roberts , and j. d. shore , phys . * 70 * , 3347 ( 1993 ) ; o. perkovic , k. dahnen , and j. p. sethna , phys . 75 * , 4528 ( 1995 ) . o. perkovic , k. dahnen , and j. p. sethna , cond - mat/9609072 ( unpublished ) ; k. dahmen and j. p. sethna , phys . b * 53 * , 14872 ( 1996 ) ; o. perkovic , k. dahnen , and j. p. sethna , phys . rev . b * 59 * , 6106 ( 1999 ) . a. berger , a. inomata , j. s. jiang , j. e. pearson , and s. d. bader , phys . rev . lett . * 85 * , 4176 ( 2000 ) . j. marcos , e. vives , l. maosa , m. acet , e. duman , m. morin , v. novak , and a. planes , phys . b * 67 * , 224406 ( 2003 ) . p. meakin , phys . lett . * 51 * , 1119 ( 1983 ) ; m. kolb , r. botet , and r. jullien , phys . * 51 * , 1123 ( 1983 ) . more precisely , we expect the number of solutions to scale exponentially with the system size . the fact that only minima are obtained by solving iteratively nonlinear equations such as eqs . ( 2 ) is shown by d. ling , d. r. bowman , and k. levin , phys . b * 28 * , 262 ( 1983 ) . in practice , as explained in ref.@xcite , we use a more efficient algorithm than simple iteration . however , we checked numerically that it yields the same solutions as the simple iteration procedure . in actual sorption experiments , there are also kinetic effects associated to mass and/or heat transfer which may significantly slow down the relaxation to a new metastable state . in particular , silica aerogels are exceptional thermal insulators and convection in the adsorbed fluid is probably suppressed by the aerogel strands@xcite . these effects are neglected in the present treatment in which the fluid responds instantaneously to a change in chemical potential or temperature . mass transfert can be incorporated in an effective grand - canonical treatment by rescaling the relaxation time so as to take into account the constraint of mass conservation ( woo and monson , phys . e * 67 * , 041207 ( 2003 ) ) . in any case , at low enough temperature , the ( local ) relaxation time remains much smaller than the time associated to thermally activated processes between metastable states , which justifies the use of a zero - temperature dynamics . j. p. sethna , k. dahmen , and c. r. myers , nature * 410 * , 242 ( 2001 ) . c. m. soukoulis , k. levin , and g. s. grest , phys . b * 28 * , 1495 ( 1983 ) . f. j. perez - reche and e. vives , phys . rev . b * 67 * , 134421 ( 2003 ) ; f. j. perez - reche and e. vives , phys . rev . b * 70 * , 214422 ( 2004 ) . it is easy to see from eq . ( 1 ) that the elements of the hessian matrix @xmath255 are given by @xmath256 -w_{ff}n_{ij}$ ] where @xmath257 is the kronecker symbol and @xmath258 when @xmath14 and @xmath259 are nearest neighbors and zero otherwise . @xmath255 can thus be diagonalized by a unitary transformation . the threshold @xmath260 ensures that the avalanche size is computed with enough precision . the calculation of @xmath66 is restricted to avalanches that do not span the whole lattice . r. evans , j. phys . : condensed matter * 2 * , 8989 ( 1990 ) . for a recent review on the barkhausen effect , see g. durin and s. zapperi in _ the science of hysteresis _ , edited by g. bertotti and i. mayergoyz , elsevier ( 2004 ) . since we consider the avalanche size distribution integrated over a whole isotherm , our exponent @xmath101 identifies with the exponent @xmath261 in the notation of ref.@xcite . s. j. gregg and k. s. w. sing , _ adsorption , surface area , and porosity _ , academic press , new york ( 1982 ) . we measure the asphericity via the parameter @xmath262/2r_g^2 $ ] . the normalisation is such that @xmath263 for spherically symmetric objects and @xmath17 for cigar " shaped objects . note that @xmath264 is not an invariant quantity with this definition since the components of the radius of gyration tensor are not taken along the principal axes ( see j. rudnick and g. gaspari , j. phys . a. * 19 * , l191 ( 1986 ) ) . it is likely that , on average , we underestimate the actual deviation from sphericity . r. salazar and l. d. gelb , mol . phys . * 102 * , 1015 ( 2004 ) . note however that @xmath265 in eq . 5 is a small quantity compared to the bulk and surface contributions . v. privman and m. e. fisher , phys . b * 30 * , 322 ( 1984 ) . see e.g. j. cardy , _ scaling and renormalization in statistical physics _ , cambridge university press , cambridge ( 1996 ) . except at @xmath110 where we only take into account the isotherm in the @xmath30 porosity aerogel .. r. evans , u. marini bettolo marconi , and p. tarazona , j. chem . , farady trans . * 2 * , 82 ( 1986 ) . lilly , a. h. wooters , and r. b. hallock , phys . lett . * 77 * , 4222 ( 1996 ) ; m. p. lilly and r. b. hallock , phys . b * 64 * , 024516 ( 2001 ) ; m. p. lilly , a. h. wooters , and r. b. hallock , phys . rev . b * 65 * , 104503 ( 2002 ) . moreover , as emphasized in ref.@xcite , adding volumetric bursts to the system induces an unknown degree of local heating .

low - dimensional quantum heisenberg antiferromagnets are known to exhibit non - trivial properties due to quantum fluctuations . a most striking example comes from spin ladders : interpolating between the chain geometry and the square lattice geometry , they have provided an interesting new playground to explore the physics of low - dimensional strongly correlated electron systems @xcite . specifically , it has been shown that af heisenberg spin-@xmath1 ladders with an even number of legs are quantum spin liquids with purely short - range spin correlations . their spin correlations decay exponentially due to a finite singlet - triplet gap in the energy spectrum . in contrast , heisenberg ladders with an odd number of legs display properties similar to those of single chains at low temperature , namely gapless spin excitations , and power - law spin correlations . this difference between even - leg and odd - leg ladders was predicted by theory and confirmed by experiments in a variety of systems @xcite . in this paper we focus on disordered spin ladders . for the case of doping with static non - magnetic impurities , one distinguishes between site and bond disorder . site dilution occurs upon doping the magnetic ions with non - magnetic ones ( as for instance substitution of cu@xmath2 with zn@xmath2 ) . this leads to the formation of local moments ( lms ) close to the dopant site @xcite . effective residual interactions between the lms lead to long - range antiferromagnetic ordering@xcite , experimentally detected in systems such as @xmath3 ( ref . ) and bi(cu@xmath4zn@xmath5po@xmath6 ( ref . ) . bond disorder occurs instead when the dopant ions replace the ions which act as bridges between the magnetic ions . for instance , bond disorder can be introduced in ipa-@xmath7 by a partial substitution of non - magnetic @xmath8 for the likewise non - magnetic @xmath9 , affecting the bond angles in the @xmath10-halogen - halogen-@xmath10 superexchange pathways @xcite . the reduced strength of magnetic interactions on the affected bonds leads to bose - glass behavior when a strong magnetic field is applied . from the theory side , a number of studies have addressed the effects of site and bond disorder on heisenberg ladders . ref . examined the correlation length of randomly site - diluted spin-1/2 heisenberg 2-leg ladders at weak and intermediate interchain couplings , showing an apparent divergence of the spin correlations at low temperatures , due to the presence of the impurities . a related divergence of the staggered susceptibility has been reported in ref . . this behavior has been related to that of the random - exchange heisenberg model ( rehm ) @xcite , describing the effective random interactions between lms @xcite . in a more recent study , three of us have investigated bond - diluted 2-leg ladders @xcite and observed enhancement of spin correlations due to bond dilution . however , a general understanding of the mechanisms for enhancement of correlations in site- and bond - diluted heisenberg ladders is still lacking . it is also unclear whether such an enhancement of spin correlations generally exists for even - leg ladders with a number of legs greater than two . in this paper , we present a study of site- and bond - diluted spin-1/2 af heisenberg ladders with n=2 , 4 and 6 legs . making use of quantum monte carlo simulations , we can address the correlation length of the system down to extremely small temperatures at which the asymptotic @xmath11 behavior sets in . our main findings can be summarized as follows : 1 . for site - diluted even - leg ladders , we observe a strong enhancement of correlations upon doping up to @xmath12 legs . in the specific case of 2-leg ladders , we find that the system realizes the physics of the rehm with a _ discrete _ distribution of the effective couplings , leading to a _ non - universal _ behavior at low temperatures ; hence the known predictions for the universal regime of the rehm with a continuous distribution of couplings are expected not to apply to realistic models of doped spin ladders . 2 . for bond - diluted ladders , we find that correlations are _ suppressed _ by a low level of dilution , due to dilution - induced dimerization . hence the system is first driven to a gapless phase with short - range correlations . the correlation length becomes logarithmically divergent for vanishing temperatures only beyond a critical dilution . we study the heisenberg model on @xmath13-leg ladders @xmath14 where @xmath15 is the quantum spin operator at site @xmath16 along the @xmath17-th leg . the first term in the hamiltonian describes interactions along the legs with coupling @xmath18 , whereas the second term represents the rung interaction with coupling @xmath19 . the random variables @xmath20 , @xmath21 express the site or bond dilution : in the case of site dilution , @xmath22 and @xmath23 , where @xmath24 takes value 1 if the site @xmath25 is occupied ( with probability @xmath26 ) or 0 if it is empty ( with probability @xmath27 ) . in the case of bond dilution , @xmath28 take values 0 with probability @xmath29 if the bond is not occupied , and take values 1 otherwise . the simulations are performed using the stochastic series expansion ( sse ) quantum monte carlo ( qmc ) method based on the directed loop algorithm @xcite . periodic boundary conditions ( pbc ) along the leg direction are used . to access the regime of very low temperatures , at which the asymptotic low-@xmath30 behavior of the correlation length sets in , we have made use a @xmath31-doubling scheme @xcite , allowing us to efficiently access inverse temperatures up to @xmath32 . the main objective of this paper is to study the correlation length along the leg direction , which is calculated via the disorder - averaged second - moment estimator : @xcite : @xmath33_{\rm av}}{[s(\pi+2\pi / l,\pi)]_{\rm av}}-1 } . \label{e.2ndmoment}\ ] ] here @xmath34 is the time - averaged structure factor @xmath35 with @xmath36 corresponding to the total number of sites . @xmath37_{\rm av}$ ] denotes the disorder average , which is performed over 300 - 600 disorder realizations . antiferromagnetic even - leg ladders without doping display a rung - singlet ground state @xcite with a finite gap @xmath38 to triplet excitations . in this state , the @xmath13 spins on the same rung preferentially form a singlet state , and therefore effectively decouple from the rest of the ladder . this leads to exponentially decaying correlations in the direction of the legs , characterized by a finite correlation length @xmath39 . at low enough concentration of dopants , the main effect of site dilution on an even - leg ladder is that of turning the number of spins on a rung from even to odd : in this situation , the state on the rung turns from a singlet into a _ doublet _ , which corresponds to an effective @xmath0 localized moment ( lm ) . this @xmath0 moment remains exponentially localized ( over a characteristic length @xmath39 ) close to the impurity site @xcite , but its finite overlap with other lms leads to an effective interaction which generically decreases exponentially with the distance . in the case of dominant rung interactions , @xmath40 , the interaction between lms is appropriately described within second - order perturbation theory ( in @xmath41 ) as resulting from the exchange of virtual massive triplets between two lms @xcite . in the case of a 2-leg ladder , in which we can identify the location of a lm with a spin site next to a missing rung partner ( dangling spin ) , the effective coupling between lms has the form of a su(2)-invariant heisenberg interaction , and at large separation between lms , @xmath42 , the coupling strength takes the form @xcite @xmath43 where the staggering factor takes value @xmath44 if the two lms belong to the same sublattice and @xmath45 otherwise . eq . holds strictly speaking only in the case in which @xmath46 , but a similar exponential decay of the effective lm coupling ( _ without _ the square - root denominator ) has been observed numerically in the case of much stronger leg interaction @xcite , as _ e.g. _ for @xmath47 , for which @xmath48 . moreover the above expression is numerically found to account for the decay of the effective couplings already for moderate distances , @xmath49 . hence , towards an effective model of the site - diluted ladders , we will assume in the following that the exponential decay of lm couplings with decay rate @xmath39 remains valid even when perturbation theory is no longer applicable which is the case of interest . in this paper , we deliberately choose @xmath50 to have a sizable correlation length @xmath39 , for reasons which will become clear in the following . our assumption then implies that non - perturbative effects only affect the prefactor to the exponential in eq . . in addition , we will assume that the exponential decay present in eq . sets in for distances between lms of the order of @xmath39 . in the case of @xmath13-leg ladders with @xmath51 a detailed theory of effective lm couplings is not available to our knowledge , but for widely spaced vacancies , the extended structure of the localized doublet becomes irrelevant , and for @xmath52 a perturbation approach analogous to that of 2-leg ladders should be applicable , leading to effective staggering couplings between lms with exponential dependence on the distance . ( color online ) . probability distribution ( @xmath53 ) and cumulative distribution ( @xmath54 ) of the effective couplings @xmath55 between nearest - neighbor localized moments in site - diluted 2-leg ladders , with parameters @xmath56 , @xmath57 ( upper panels ) , and @xmath58 , @xmath59 ( lower panels ) . the red dashed lines show the probability distributions in the continuum approximation according to eq . .,title="fig:",width=359 ] ( color online ) . probability distribution ( @xmath53 ) and cumulative distribution ( @xmath54 ) of the effective couplings @xmath55 between nearest - neighbor localized moments in site - diluted 2-leg ladders , with parameters @xmath56 , @xmath57 ( upper panels ) , and @xmath58 , @xmath59 ( lower panels ) . the red dashed lines show the probability distributions in the continuum approximation according to eq . .,title="fig:",width=359 ] in the case of low doping , @xmath60 , the probability distribution of having two nearest - neighboring lms at a distance @xmath61 along the leg direction in a 2-leg ladder is given by @xmath62 with a corresponding average @xmath63 . as found numerically in ref . , the effective coupling has the form @xmath64 for @xmath65 . as an approximation , one may assume that the same exponential behavior survives for @xmath66 . then the probability distribution for the strength of the couplings between nearest neighboring lms reads @xmath67.\end{aligned}\ ] ] this result suggests the possibility of extracting analytically the probability distribution for @xmath68 , as done in ref . . to this end , one may further take the continuum approximation , namely , approximate the summation in eq . by an integral : @xmath69~. \label{e.pdpj}\end{aligned}\ ] ] this allows then to write @xmath70 where @xmath71 . hence , within the continuum approximation the n.n . couplings obey a simple power - law distribution.@xcite the average absolute value of the coupling between neighboring lms takes then the form @xcite : @xmath72_{\rm av } = \frac{1-\gamma}{2-\gamma } ~j_0 = \frac{2z\xi_0}{1 + 2z\xi_0 } ~j_0 . \label{e.jav}\ ] ] however , a critical analysis of the above derivation shows that the continuum approximation is problematic . in fact , it requires that the distances @xmath61 giving a significant contribution to the sum in eq . be @xmath73 , which is in contradiction with the fact that @xmath74 decreases exponentially with @xmath61 ; the characteristic decay length is @xmath75 , which for @xmath76 takes values @xmath77 . [ f.pjeff-numerical ] shows the distribution @xmath78 determined numerically according to eq . , _ i.e. _ , by sampling the _ discrete _ distribution lengths @xmath74 , for @xmath79 , @xmath57 ( corresponding to @xmath80 , and giving @xmath81 ) , and @xmath82 , @xmath59 ( corresponding to @xmath83 , and giving @xmath84 ) these parameter sets will be relevant for our study of correlations in the following . we notice that the distribution shown in fig . [ f.pjeff-numerical ] is only quantitatively correct when @xmath85 , namely for @xmath86 , while it is only an approximation otherwise . it is clear that for @xmath85 , the distribution of n.n . couplings obtained numerically deviates strongly from the prediction of eq . . over most of its support , the distribution of n.n . couplings has a fundamentally _ discrete _ structure , due to the fact that the largest values of @xmath68 , which also take the largest probabilities , are associated with short separations @xmath61 among n.n . these probabilities _ grow _ as a power law of the absolute value of the coupling , opposed to what predicted by eq . for @xmath87 . obviously a coarse graining on the @xmath88-peaks of the exact distribution would gradually reconstruct the shape of eq . , but given the finite separation between the @xmath88-peaks even in the thermodynamic limit , this coarse graining is not justified over a large range of @xmath68 . in fact the @xmath88-peaks become dense only in the limit @xmath89 , which corresponds to larger and larger separations among the lms , and hence to lesser and lesser probabilities . only in that limit the continuum approximation appears legitimate ( in fact numerically we recover the analytical prediction for @xmath90 due to the finite width of our bins , @xmath91 ) . in conclusion , for @xmath92 and @xmath93 , @xmath78 has a remarkable structure : it assigns the largest probability to the range in which @xmath68 is a _ discrete _ variable , preventing a straightforward approximation of the distribution with a continuous function . this will have significant consequences on the analysis of doped ladder systems in relation with models of randomly coupled spins . ( color online ) . correlation length @xmath94 of site - diluted ladders with @xmath95 and @xmath96 . upper panel : comparison between 2-leg , 4-leg and 6-leg ladders at inverse temperature @xmath97 . lower panel : correlation length of the 2-leg ladder at different temperatures.,width=283 ] in both cases of @xmath98 and @xmath51 the unfrustrated effective couplings between lms can lead to the formation of a _ _ ground state with algebraically decaying correlations between the sites , and hence an infinite correlation length . when looking at a fixed finite temperature @xmath30 the correlation length is necessarily finite for one - dimensional su(2 ) invariant systems with short - range interactions , but it is expected to increase because the average coupling between lms increases with @xmath27 , as in eq . , so that the effective temperature @xmath99_{\rm av}$ ] decreases . despite the increase of correlations due to disorder , the range of correlations is fatally bounded by the fact that any finite amount of site dilution will break an infinite @xmath13-leg ladders into finite segments , of characteristic length @xmath100 ( corresponding to the inverse linear density of rungs which are fully removed by doping ) when considering doped ladders at finite temperatures no geometric bound on the correlation length ( neither coming from finite - size effects nor from ladder fragmentation ) is present as long as the correlation length satisfies the condition @xmath101 it is clear that such a condition can only be satisfied when the doping is sufficiently weak . hence , for any fixed temperature , upon increasing the doping concentration the correlation length @xmath102 will cross over from a growing behavior at low doping @xmath103 to a decreasing behavior for @xmath104 , hence going through a maximum for a given optimal doping value @xmath105 , which is quite stable at sufficiently low temperatures . this is indeed seen in fig . [ f.sitedil](a ) , in which we observe @xcite that , for @xmath106 , @xmath107 for @xmath98 , 4 , and 6 . for any finite value of doping and at arbitrarily low temperatures , the correlation length is upper bounded by the average length of the ladder segments , @xmath108 . yet , as shown in fig . [ f.sitedil](b ) , at low enough doping the upper bound is reached extremely slowly in temperature - in fact , for the 2-leg ladder at @xmath109 and @xmath47 , the correlation length is still not saturated to its finite @xmath110 value for @xmath111 . it is easy to understand that this occurs because the average effective coupling @xmath112_{\rm av}$ ] decreases as @xmath27 decreases , leading to exceedingly weak correlations even at extremely low temperatures . this means that for sufficiently low doping , the condition @xmath113 is actually satisfied over all numerically accessible low temperatures ( and quite possibly also over all experimentally accessible temperatures ) , so that the segmented nature of the doped ladders become in fact _ irrelevant_. in the following we will focus our discussion on this regime . as discussed in section [ s.pjeff ] , from the statistical point of view the system of interacting lms randomly distributed on the bipartite 2-leg ladder can be approximated by a system of @xmath0 spins interacting with randomly distributed couplings , following an intrinsically _ discrete _ distribution . in general terms , a doped 2-leg ladder represents a physical realization of the so - called random - exchange heisenberg model ( rehm ) @xcite , but with a special structure of the coupling distribution . ( color online ) . temperature dependence of the correlation length of site - diluted 2-leg ladders with @xmath114 , and power - law fits at low temperature.,width=283 ] numerical real - space renormalization group studies @xcite show that the rehm colorred for a _ continuous _ distribution of couplings @xmath115 has two regimes : * _ universal regime _ : if the distribution is not singular , or has a power - law singularity @xmath116 with @xmath117 ( where @xmath118 ) , the low - temperature thermodynamics of the rehm is dominated by a unique fixed point ; the correlation length diverges as @xmath119 with a universal scaling exponent , @xmath120 with @xmath121 . this result has been confirmed by quantum monte carlo @xcite for a non - singular distribution ( square box , @xmath122 ) . * _ non - universal regime _ : for a more strongly singular distribution @xmath123 the renormalization group flows to a non - universal fixed point , which depends on the initial coupling distribution . studies on doped coupled spin - peierls chains using real space renormalization group and sse qmc in this regime also confirm that the exponent @xmath124 depends on the doping concentration@xcite . as discussed in sec . [ s.pjeff ] , the distribution of couplings between lms in a doped 2-leg ladder is intrinsically discrete , approximating a continuous distribution only for vanishing couplings . it is therefore interesting to ask whether the doped 2-leg ladder reproduces the known physics of the rehm model with a continuous distribution of couplings . if a continuous approximation to the coupling distribution is in order , as in eq . , the system should exhibit a _ universal rehm behavior _ for @xmath125 , and a non - universal behavior for @xmath126 . in other words , ladders with sufficiently strong doping @xmath27 or sufficiently large correlation length @xmath39 in the undoped limit should exhibit universal scaling properties at low temperature . to our knowledge the only numerical study of correlations in site - diluted ladders is ref . , which , making use of quantum monte carlo , investigates ladders with @xmath127 and dopings @xmath128 , 0.04 and 0.1 . given that @xmath129 in this case @xcite , for @xmath130 and 0.1 one obtains @xmath131 and @xmath132 respectively , which appear to be both safely on the universal side of the rehm with continuous couplings . indeed a fit to the low - temperature behavior of the correlation length to the scaling law @xmath133 gives an estimate of @xmath134 ( independent of @xmath27 ) which appears consistent with the predictions of refs . . we have performed qmc simulations to extensively study the low - temperature behavior of the correlations in site - diluted 2-leg ladders . our qmc results , however , do not support the conclusion in ref . about a universal rehm behavior of diluted 2-leg ladders . we investigate doped ladders at similar concentrations with respect to those of ref . , down to significantly lower temperatures - ref . stops at @xmath135 , while we descend down to @xmath136 . a collection of our results is shown in fig . [ f.xit ] . in qualitative agreement with what expected from the theory of the rehm model , we also observe the clear onset of a low - temperature power law scaling of the type @xmath137 . yet our central observation is that this power - law scaling exhibits a _ non - universal , doping dependent _ exponent @xmath138 . for all the parameter sets we considered , a fit of the low - temperature correlation length provides values of @xmath124 which are systematically below the one exhibited by the universal regime of the rehm with continuous couplings . the data shown in fig . [ f.xit ] refer to 2-leg ladders with @xmath139 and @xmath140 and @xmath141 , with corresponding @xmath142 and @xmath143 respectively ( here we use @xmath59 , which we estimate independently with qmc ) ; and 2-leg ladders with @xmath127 ( as in ref . ) and @xmath144 , @xmath145 , and @xmath146 , corresponding to @xmath147 , @xmath148 , and @xmath149 respectively . although all the values of @xmath150 fall below the critical value @xmath151 , as shown in fig . [ f.xit ] , in fact , all the corresponding data sets for the correlation length display a _ non - universal _ power - law scaling . in particular , we observe that data sets with close values of the parameter @xmath150 ( _ e.g. _ @xmath139 , @xmath140 with @xmath152 , and @xmath127 , @xmath153 , with @xmath154 ) show a clearly different scaling . we also critically revisit the parameter sets explored in ref . . a comparison to our data , as done in fig . [ f.greven ] , shows that in the data of ref . the significant scattering due to the statistical uncertainty is masking the correct asymptotic scaling regime at low temperature . a comprehensive summary of our results on the low - temperature scaling of the correlation length is provided in fig . [ ph.diagram ] , where we report the values of the @xmath124 exponent obtained for different doping values and @xmath155 ratios . the values of doping concentrations and ratios were chosen such that the average distance between lms @xmath156 is at least larger than @xmath39 . the lms , therefore , are not overlapping on average and can be described by an effective rehm with couplings obeying eq . . furthermore , for all the points shown in fig . [ ph.diagram ] , @xmath117 , we do not observe a universal @xmath124 value . indeed , it is found that @xmath124 depends not only on the doping concentration but also on the ratio of couplings @xmath155 . and an @xmath124 exponent close to that of the universal rehm regime is only observed in a narrow region around @xmath157 and @xmath76 . ( color online ) . values of the @xmath124 exponent extracted from the low - temperature behavior of site - diluted 2-leg ladders.,width=283 ] the above results lead us to conclude that doped 2-leg ladders do not realize in general the universal physics of the rehm with a continuous coupling distribution . in fact , the coupling distribution for the lms in doped 2-leg ladders has a fundamentally discrete structure , approximating a continuous one only for vanishing couplings . the weight assigned by the distribution to strong discrete couplings is dominant over that of weak , quasi - continuous couplings in the case of sizable doping @xmath27 ( suppressing the probability of large separation between the lms ) and of large correlation length @xmath158 ( increasing the strength of the couplings between lms , the distance being fixed ) . nonetheless , this is the regime in which the parameter @xmath159 is small , and in which one would expect the universal rehm physics to be manifested in the continuum approximation . on the other hand , the quasi - continuous part of the distribution ( corresponding to very small couplings ) acquires a bigger weight for small @xmath27 and @xmath39 , which determine a small average coupling according to eq . and hence increase the probability of the small values of the coupling . in this regime @xmath160 is closer to 1 , which leads to too strong a singularity in the probability distribution for the universal regime of the rehm with continuous couplings to be manifested . as a consequence , we generally observe a low - temperature power - law scaling of the correlation length with doping - dependent exponents . as far as the numerical results are concerned , this scaling appears to be the asymptotic one for @xmath11 , given that it sets in for temperatures well below the average lm coupling . on the other hand , according to a real - space renormalization group ( rsrg ) approach @xcite , the system at a lower temperature is increasingly sensitive to the part of the distribution related to the weaker couplings ; given that this part is the one which best approximates a continuous distribution , one could naturally expect that at very low temperatures the physics of the rehm with continuous couplings be reproduced by the doped 2-leg ladders . this argument would then suggest that , when lowering the temperature , the system will sooner or later attain the universal regime of the rehm with continuous couplings if @xmath117 . nonetheless one can argue that in a doped 2-leg ladder the rsrg flow has to be necessarily stopped at a finite length , corresponding roughly to the average segment length @xmath108 . in practice this imposes an upper bound to the correlation length and a lower bound to the temperature through the condition eq . ; these bounds might prevent one from attaining the low - temperature regime at which universal rehm physics would manifest itself . finally , our study leaves one open question concerning the role of discreteness of the initial coupling distribution for the physics of the rehm . the numerical rg study of ref . only addresses initial continuous distributions ; a systematic rg study of the flow starting from realistic distributions stemming from the doped - ladder physics would be highly desirable . in the following , we consider the doping- and temperature dependence of correlations in bond - diluted ladders , which show a marked difference with respect to site dilution . this is due to a fundamental geometric aspect which distinguishes site and bond dilution : diluting a site leaves a single unpaired spin , giving rise to a lm , while eliminating a bond leaves always _ two _ unpaired spins , located on different sublattices . the corresponding lms are therefore interacting with an effective antiferromagnetic coupling , mediated by the the shortest path of bonds connecting them . in the following , we will focus for simplicity on the case @xmath96 . [ f.bonddil-fixt ] shows the evolution of the correlation length at a fixed , low temperature @xmath161 . a striking difference with respect to the case of site dilution , shown in fig . [ f.bonddil-fixt ] , is that the correlation length does not increase immediately with doping : it remains essentially constant for @xmath98 , while it even _ decreases _ for @xmath51 . in the case of a 2-leg ladder , we can understand qualitatively the behavior of correlations by considering that bond dilution has two fundamentally different effects ( sketched in fig . [ f.bonddil-cartoon](b ) ) : * if a rung bond is diluted , two lms are liberated but still interact antiferromagnetically , so that they can screen each other to form a rung singlet at a lower energy @xmath162 ( corresponding to the effective coupling between lms which are rung neighbors ) . this screening is very effective at low dilution because in that case the @xmath162 interaction is largely dominant over the interaction between two lms belonging to different rungs , due to the large spacing of two diluted rung bonds this spacing is @xmath163 ( given that only one bond in three is a rung bond ) ; * if a leg bond is diluted , the rung - singlet state on the two adjacent rungs is further reinforced because its connectivity to the other rungs is lowered . ( color online ) . correlation length of bond - diluted 2- , 4- , and 6-leg ladders with @xmath164 , @xmath96 , and temperature @xmath165,width=283 ] the dilution of rung bonds has therefore the effect of lowering locally the gap over the ground state , whereas the dilution of leg bonds has the effect of increasing it @xcite . the fact that there are twice as many leg bonds as rung bonds suggests that the second effect dominates over the first . therefore the spin gap of the undoped ladder is locally preserved or even enhanced , leading to short - range correlations even in presence of bond dilution . nonetheless , standard consideration on rare - event physics lead to conclude that even a very weak bond dilution leads the system to a _ gapless , griffiths - like phase_. in fact , the gap can close locally if rare regions appear in which only rung bonds are diluted the limiting case being that of the formation of local strands made of two uncoupled @xmath0 chains . these exponentially rare regions lead to the closing of the spin gap in the thermodynamic limit ; yet global correlations remain short ranged , due to the localized nature of the rare , locally gapless regions . this griffiths phase is reminiscent of what has been observed by some of us in the heisenberg model on the inhomogeneously bond - diluted square lattice @xcite . in that case , the predominance of correlation - suppressing dilution was guaranteed by the inhomogeneity of doping probabilities ( favoring the appearance of ladder - like or dimer - like structures ) . ( color online ) . sketch of the bond dilution effects on 2-leg ladders . a ) pure ladder ; b ) ladder at low dilution : enhancement of rung singlets by leg - bond dilution , and low - energy singlets formed between lms after rung - bond dilution ; c ) ladder at stronger dilution : couplings between lms belonging to different diluted rungs.,width=302 ] ( color online ) . correlation length for 2-leg and 4-leg ladders with dilution of ( center-)rung bonds . the average length of ladders segments @xmath108 is obtained upon homogeneous dilution . , width=283 ] for a sufficiently large bond dilution @xmath166 ( @xmath167 ) , the correlation length starts increasing with increasing @xmath27 . this implies necessarily that the liberated lms belonging to different rungs begin to correlate with each other ( fig . [ f.bonddil-cartoon](c ) , see also refs . ) . such correlations appear because the spacing between diluted rung bonds decreases , and hence the couplings between lms belonging to different rungs become of the same order of @xmath162 . we have isolated the effect of enhancement of correlations through rung - bond dilution by considering uniquely this form of dilution , namely by taking @xmath168 with probability @xmath27 , while @xmath169 with probability 1 . the resulting correlations @xmath170 ( r = rung ) as a function of @xmath27 for 2- and 4-leg ladders are shown in fig . [ f.rungdilution ] , and they are seen to increase quite fast with dilution ( following the approximate form @xmath171 for sufficiently low dilution ) . yet in the original system the correlations induced by rung - bond dilution are upper - bounded by the characteristic length of the ladder segments @xmath108 , which we have estimated by generalizing an efficient algorithm recently developed for homogeneous percolation @xcite . it is found that at finite @xmath27 , @xmath172 . it follows the same asymptotic behavior as from the naive estimate @xmath173 , but with a renormalized factor @xmath174 . hence the enhancement of correlations due to rung - bond dilution competes with this geometrical restriction on correlations for sufficiently strong doping . quite remarkably , we observe that the optimal doping for the enhancement of correlations in the homogeneously doped 2-leg ladder coincides with the value of dilution ( @xmath175 ) at which the correlation length of the rung - only diluted ladder equals the average length of ladder segments @xmath108 . therefore we can conclude that the non - monotonic behavior of the correlation length at intermediate dilution values is completely determined by the competition between the correlation enhancement through rung - bond dilution and the correlation suppression due to ladder fragmentation . ( color online ) . ( a ) correlation length of bond - diluted 2-leg ladders with @xmath95 and @xmath176 at various temperatures . ( b ) temperature dependence of the correlation length of above system . the curves with @xmath177 are omitted due to @xmath178 is limited by the geometric length . ( c ) and ( d ) fitting parameters of correlation length @xmath179 , as a function of bond dilution.,width=321 ] as discussed in the previous subsection , for @xmath180 correlations start to increase as a function of bond dilution . [ f.bonddilution.t](a ) shows the doping dependence of the correlation length of 2-leg ladders for higher temperatures than those considered in fig . [ f.bonddil-fixt ] : here we observe that the correlation length does not appear to evolve significantly with temperature for doping values @xmath181 , whereas it becomes significantly dependent on temperature for @xmath166 . moreover , the low - temperature dependence of the correlation length , see fig . [ f.bonddilution.t](b ) , indicates that in this regime the correlation length grows logarithmically with decreasing temperature as @xmath182 ( color online ) . ( a ) temperature dependence of correlation lengths for four cases of rung - bond and random dilution . , width=283 ] the dependence of the coefficients @xmath183 and @xmath184 on bond dilution is shown in fig . [ f.bonddilution.t](c ) and ( d ) . while @xmath183 decreases with increasing bond dilution , @xmath184 increases . in particular @xmath184 appears to vanish for @xmath181 , and it seems to suggest that a phase transition occurs at finite @xmath180 from a phase with correlations converging to a finite value at low @xmath30 to a phase with correlations diverging logarithmically with decreasing temperature . obviously the divergent behavior of correlation length only persists up to lengths of the order @xmath108 , corresponding to the characteristic length of ladder segments , but the exceedingly slow growth of @xmath178 with decreasing @xmath30 keeps it safely below @xmath108 for all the temperatures and doping values we explored . to gain a deeper understanding on the above results , we study the temperature dependence of rung - bond dilution and random dilution separately . we consider rung - bond dilution with concentrations @xmath185 such that the fraction of diluted rung bonds is the same as for the cases of random dilution that we have investigated . this allows us to quantitatively ascertain the separate effect of leg - bond and rung - bond dilution . in presence of rung - bond dilution only , the correlation length is seen to converge to a finite value as @xmath186 , even at relatively high concentration . two examples with concentration @xmath187 and @xmath188 are shown in fig . [ para.z ] ( data for higher concentration , not shown here , exhibit a similar behavior ) . this implies that the system with rung - bond dilution only is gapless ( because of the unbounded size of rare regions with _ e.g. _ diluted adjacent rung bonds ) and it has a finite correlation length , namely it is in a griffiths - like phase . adding leg - bond dilution ( with concentrations @xmath189 and @xmath190 respectively ) , we recover a randomly bond - diluted ladder ( with concentrations @xmath191 and @xmath192 respectively ) . for large enough dilution the correlation is found to grow logarithmically with decreasing temperature , as in eq . , which seems atypical for a griffiths - like phase . on the other hand , we observe that , for dilution @xmath185 , the correlation length for the rung - bond diluted ladder with concentration @xmath193 is an upper bound to that of the randomly diluted ladder with concentration @xmath27 , a fact which is somewhat intuitive , given that rung - bond dilution is the main mechanism leading to the enhancement of correlations . therefore it is a priori unclear whether the logarithmic growth of the randomly diluted ladder persists to even lower temperatures than the ones explored here , because this would eventually lead the correlation length of the randomly diluted ladder to exceed that of the rung - bond diluted one . we can then conclude that the correlation length results are _ a priori _ consistent with two different scenarios for the randomly diluted 2-leg ladders . in a first scenario the system transitions from a griffiths - like phase with a finite correlations for @xmath194 to a new phase with logarithmically diverging correlations , which is reminiscent of the behavior of a system controlled by an infinite - randomness fixed point ( irfp ) see the discussion below . in a second scenario the correlation length exhibits different temperature dependences at intermediate temperatures for increasing bond dilution . yet the correlation length of the randomly diluted ladders converges always to a finite value , because it is expected to be upper - bounded by that of rung - bond diluted ladders , which is seen to converge to a finite value . in this second scenario the system is therefore in a griffiths - like phase for all the values of bond dilution we explored . in the following we find that the analysis of the temperature dependence of the uniform susceptibility helps clarifying which scenario is the most appropriate . [ xu.z](a ) shows the uniform susceptibility of a 2-leg ladder for various values of bond dilution . in the undoped system , the susceptibility of 2-leg ladders vanishes exponentially at low temperatures , in agreement with well - known previous results , and showing the gapped nature of the spectrum . in contrast , the susceptibility of the doped system diverges following a curie law for low temperatures . furthermore , fig . [ xu.z](b ) shows that the curie coefficient increases as a function of bond dilution concentration following a power law , @xmath195 . one might expect that curie paramagnetism comes trivially from spins which have remained isolated after bond dilution . yet the concentration of such spins would scale as @xmath196 with dilution , given that one needs to dilute at least 3 bonds to decouple one spin from the rest of the ladder . therefore the paramagnetism observed in the system has a collective nature . in fact , two contributions add to the one of free spins : ( color online ) . ( a ) low - temperature uniform susceptibility of bond - diluted ladders with @xmath197 and @xmath198 . ( b ) the curie coefficient increases as a power - law function of the bond dilution concentration @xmath195.,width=302 ] * odd - numbered clusters , obtained after fragmentation of the ladder into two ( or more ) pieces , have a doublet ground state , behaving as a collective spin 1/2 . in our simulations on systems with an even total number of sites , three contiguous diluted bonds lead to an isolated site plus an odd - numbered cluster , so that the latter clusters have a probability @xmath199 . alternatively odd - numbered clusters can be obtained by cutting the ladder at two distinct locations into two odd - numbered extended clusters ( bigger than one single site ) . yet one can easily see that this would require at least 5 diluted bonds at specific locations , giving rise to a probability @xmath200 , and consequently to a very small contribution to the uniform susceptibility . therefore the contribution to collective paramagnetism can not come from such clusters . * clusters with an even number of spins , but a non - equal number of spins on a and b sublattices , can also contribute to a diverging susceptibility . they represent therefore a further candidate for collective paramagnetism . ( color online ) . ( a ) uniform susceptibility of even - numbered spin clusters of randomly bond - diluted 2-leg ladders . the solid lines are ( a ) logarithmic fits of the form @xmath201 and ( b ) power - law fits @xmath202.,width=302 ] to isolate the second contribution , we calculate the uniform susceptibility coming from _ even - numbered _ spin clusters only , @xmath203 . this quantity is shown in fig . [ xuc.t ] . we find that even - numbered uniform susceptibility at low temperatures can be equally well described by two fitting laws : a logarithmically - corrected curie behavior : @xmath204 and a power - law corrected curie behavior : @xmath205 in both cases we can conclude that the contribution to the uniform susceptibility coming from even - numbered clusters diverges more weakly than for a curie law , and therefore that our system exhibits a highly non - trivial collective paramagnetism . the fitting coefficients @xmath206 , and @xmath207 for both cases are shown in fig . [ fit.para1 ] . we observe that all these fitting coefficients evolve smoothly with @xmath27 , apparently contradicting the picture of a possible phase transition suggested by the behavior of the correlation length . in particular the non - universal power - law dependence of eq . is typical of griffiths behavior , as observed _ e.g. _ in anisotropically bond - diluted square lattices @xcite . therefore the uniform susceptibility seems to favor the scenario for which the system is in a griffiths phase for all values of the bond dilution we explored . for completeness we should also mention a third , less likely scenario , for which the correlation length is diverging logarithmically as @xmath11 even for infinitesimal doping ( but the prefactor @xmath184 of the logarithmic divergence becomes lower than our resolution ) . therefore the system would be controlled by an irfp for all doping values explored here ; this scenario is somewhat consistent with the fact that a logarithmically corrected curie law is consistent with the susceptibility at all doping values considered in this section , and that it could be another sign of irfp behavior . ( color online ) . fitting coefficients as a function of bond dilution . left panels are fitting coefficients of the logarithmic function @xmath201 . right panels are fitting coefficients of the power - law function @xmath202.,width=302 ] as sketched in fig . [ f.bonddil-cartoon ] , the rung lms in a bond - diluted ladder realize a ladder with weaker random couplings which are at once rung , leg and diagonal ones but without frustration . the rung couplings are randomized by the variety of different local environments which mediate the effective interaction between the rung lms ; while the leg couplings and the diagonal couplings are obviously randomized ( in a correlated way ) by the positional disorder of the diluted rung bonds . for ladders with @xmath47 we can not easily rely on perturbation theory to extract accurate expressions for the effective couplings , and an explicit calculation of the disorder statistics goes beyond the purpose of this paper . in any instance it is clear that the rung couplings obey a fundamentally different probability distribution with respect to the leg and the diagonal couplings : the rung couplings are always antiferromagnetic and weakly randomized , while the leg and diagonal couplings always take opposite signs and can be either ferromagnetic or antiferromagnetic - they are hence strongly disordered . to the best of our knowledge , such a peculiar model of a random ladder has not been investigated before . in a critical quantum spin system the relationship between energy scales ( @xmath208 ) and length scales ( @xmath209 at finite temperature ) is governed by the so - called dynamical critical exponent @xmath210 , @xmath211 . the logarithmic dependence of the correlation length on temperature , eq . , implies that @xmath212 : as mentioned before , this behavior , together with the logarithmically corrected curie law in the susceptibility , could be suggestive of the fact that the bond - diluted ladder at sufficiently strong disorder is governed by an irfp , similarly to what happens to _ e.g. _ random antiferromagnetic @xmath0 chains @xcite . nonetheless , it is not the same irfp as in random antiferromagnetic chains , given that the correlation length and the susceptibility diverge differently ( as @xmath213 and as @xmath214 respectively ) in the latter . existing rg studies of disordered ladder systems focus primarily on the case of antiferromagnetic ladders , either with nearest neighbor couplings only or with frustrated ones @xcite . the occurrence of extended parameter regions with irfp is only observed in the case of frustrated ladders . therefore it is quite remarkable to observe an irfp - like behavior in our unfrustrated system for a large interval of doping values . on the other hand , the absence of a documented irfp for unfrustrated random ladders suggest that the logarithmic divergence observed in our system might not be the true asymptotic behavior for @xmath11 . nonetheless , a definite conclusion on such an asymptotic behavior would require a precise determination of the distribution of the effective couplings between lms , which goes beyond the scopes of this paper . [ f.bonddil-fixt ] shows that the dependence of the correlation length on bond dilution in 4-leg and 6-leg ladders exhibits a qualitatively different behavior compared to 2-leg ladders . indeed for 4-leg and 6-leg ladders @xmath178 is initially _ decreasing _ with increasing bond dilution . in the case @xmath176 the correlation length reaches a minimum for @xmath215 , and then starts growing with doping , up to a maximum attained for @xmath175 . in the following we will analyze this phenomenon in details . ( color online ) . bond - dilution effects on 4-leg ladders : ( 1 ) rung - singlet enhancement upon leg - bond dilution ; ( 2 ) center - rung dilution leading to the formation of two 2-spin rung singlets ; ( 3 ) outer - rung dilution with formation of a low - energy singlets between a lms and a 3-spin doublet.,width=302 ] similarly to what done for 2-leg ladders , we can distinguish among different effects of doping depending on which bond is diluted ; we will focus here on the case of a 4-leg ladder ( which is sketched in fig . [ f.bonddil-cartoon-4leg ] ) : 1 . the dilution of a _ leg bond _ leads to the the enhancement of the 4-spin rung singlet on the two adjacent rungs ; 2 . the dilution of a _ center rung bond _ leads to the formation of two 2-spin rung singlets on the intact outer rungs ; 3 . the dilution of an _ outer rung bond _ separates an outer lm from three rung spins , which tend to form locally a spin - doublet . this doublet should have then tendency to form again a low - energy singlet with the lm through the effective antiferromagnetic coupling which binds them . hence we observe that bond dilution of type 1 ) and 2 ) have the tendency to enhance locally the spin gap of the ladder , while dilution of type 3 ) has the tendency to suppress it . the case of 6-leg ladders is analogous , with the only difference that dilution of the central rung bond leads to the formation of two rung doublets . therefore we will hereafter focus on the 4-leg case for simplicity . we have disentangled the competing effects of dilution on 4-leg ladders by considering only one at a time . first of all we consider the effect of enhancement of correlations due to outer rung - bond dilution only , whose results are shown in fig . [ f.rungdilution ] . analogously to what has been seen for rung - bond dilution in the 2-leg ladders , outer - rung - bond dilution in 4-leg ladders leads to a strong enhancement of correlations ( with a similar exponential scaling @xmath216 , or = outer rung ) . in homogeneously diluted ladders this effect of enhancement is fundamentally limited by the finiteness of ladder segments ( due to leg - bond dilution ) : indeed a numerical estimate of the average length of ladder segments , @xmath108 , shows that @xmath108 crosses the correlation length @xmath217 for a value of @xmath27 ( @xmath218 ) which nicely corresponds to the optimal doping in the homogeneously bond - diluted ladder . hence , similarly to 2-leg ladders , we can conclude that the optimal doping is completely dictated by the competing effect of outer - rung - bond dilution and ladder fragmentation . ( color online ) . correlation length with selective bond dilution . bond dilution is distinguished as : ( a ) rung - bond and leg - bond dilution in 2-leg ladders ; ( b ) center - rung bond , outer - rung bond and leg bond dilution in 4-leg ladders . the dashed line is the average correlation length from above selectively bond - diluted mechanisms , width=302 ] finally , we study the effect of selective dilution on the correlation length . in fig . [ f.selective-dilution](a ) , we investigate the effect of rung - bond and leg - bond dilution in 2-leg ladders . in fig . [ f.selective-dilution](b ) , we distinguish dilution of outer - rung bonds , center - rung bonds , and leg bonds in 4-leg ladders . as anticipated , latter forms of dilution in above cases lead to a significant suppression of correlations , again with approximate exponential scaling , @xmath219 ( l = leg , cr = center rung ) . we observe that @xmath220 ( @xmath221 for 2-led ladder ) , so that the simultaneous suppression and enhancement of correlations due to dilution of different types of bonds are in strong competition . the fact that at low doping the suppression of correlation wins is simply due to combinatorics : in 2-leg ladders ( a ) , rung bonds only hold a fraction @xmath222 while leg bonds contain @xmath223 of all bonds ; in 4-leg ladders ( b ) , outer - rung bonds only represent a fraction @xmath224 of all the bonds , whereas leg bonds and center - rung bonds are respectively @xmath225 and @xmath226 . hence correlation - enhancing bond dilution has a global lower probability than correlation - suppressing one . in the limit of very low dilution ( @xmath227 ) we can easily imagine to separate a ladder into correlated regions ( of characteristic length @xmath39 ) which are statistically independent from each other . upon dilution , each correlated region will only be affected by a single missing bond on average , with a probability @xmath228 for the @xmath229-th bond type ( @xmath229 = r , l in 2-leg ladders ; and @xmath229 = or , l , cr in 4-leg ladders ) . therefore the local correlations will be enhanced or suppressed in the same way as for the case of selective bond doping of type @xmath229 . for @xmath227 it is then tempting to write the correlation length of the homogeneously diluted ladder as a spatial average of different local correlation lengths . assuming that the ratio of structure factors entering in eq . can be written as @xmath230_{\rm av}}{[s(\pi+2\pi / l,\pi)]_{\rm av } } \approx \sum_i f_i \frac{[s(\pi,\pi)]_{{\rm av},i}}{[s(\pi+2\pi / l,\pi)]_{{\rm av},i}}\ ] ] where @xmath37_{{\rm av},i}$ ] denotes averaging over disorder realizations with dilution of bond of @xmath229-th type only ( and with dilution concentration @xmath231 ) , we obtain that @xmath232^{1/2}~.\ ] ] this average correlation length is presented by the dashed lines in fig . [ f.selective-dilution ] . it clearly shows a quantitative agreement with random bond dilution results at small dilution . in this paper we have discussed the effects of site and bond dilution on the low - temperature correlations of even - leg @xmath0 ladders with antiferromagnetic heisenberg interactions . site dilution is found to prune rung singlets and thus create localized moments which interact through unfrustrated , distance dependent couplings . the hamiltonian describing the effective interaction between these moments is a random exchange heisenberg model ( rehm ) , with a gapless spectrum and power - law diverging correlations as temperature decreases to zero . we find that the distribution of the effective couplings has an intrinsically discrete structure , which prevents the system from realizing the universal regime of the rehm with continuous couplings . further studies would hence be desired to clarify the role of a discrete distribution of effective couplings in determining the fixed point which governs the low temperature physics of this system . bond dilution , on the other hand , can either enhance or suppress locally the spin gap in even - leg ladders , depending on the location of the bond which is diluted . as a result of these competing effects , weak bond dilution is not enhancing the correlations of pure ladders . in fact , correlations can even be suppressed by disorder , especially in 4-leg and 6-leg ladders . the resulting short - range correlated phase has a gapless , griffiths nature , due to the appearance of exponentially rare , but locally gapless regions . beyond a critical concentration , the correlation - enhancing effect of bond dilution leads to a phase in 2-leg ladders , with a logarithmic scaling of the correlation length with decreasing temperature . this behavior may suggests that the low - temperature behavior of the system is governed by an infinite randomness fixed point over an extended range of parameters . nonetheless , based on the investigation of the uniform susceptibility of the system , and by comparison with existing results on disordered ladders , we argue that this might not be the true asymptotic behavior for @xmath11 , and that the system is more likely to remain in a griffiths phase for all values of bond dilution . in order to connect our results to experiments on site - diluted and bond - disordered ladders , a fundamental question to address is the role of finite inter - ladder couplings , which are unavoidable in real materials . if @xmath233 is their energy scale , such couplings drive the system towards a three - dimensional magnetically ordered phase below a temperature @xmath234 , which can be estimated via a mean - field approach as @xmath235 , where @xmath178 is the correlation length of the uncoupled ladders . for a gapless critical phase , as that induced by site dilution ( at any concentration ) and by bond dilution ( beyond a critical concentration ) , @xmath102 diverges for @xmath11 , so that the system orders at finite temperature . on the other hand , the divergence of @xmath178 with decreasing temperature appears to be relatively weak ( power law with a small power @xmath236 for site dilution , or logarithmic for bond dilution ) . hence a difference of several orders of magnitude between the intra - ladder and the interladder couplings , which is common to various ladder materials , allows the asymptotic low - temperature behavior of @xmath178 for the uncoupled doped ladders to be manifested at temperatures which lie well above the critical temperature @xmath234 for magnetic ordering . for bond - diluted ladders , moreover , short - range correlations proper of the pure ladder persist up to a critical doping , so that three - dimensional ordering in coupled ladders should be absent at weak dilution . increasing the bond dilution leads to an increase of correlations at low temperature , so that the system is expected to undergo a disorder - induced quantum phase transition from a griffiths phase to a magnetically ordered phase ( at @xmath110 ) , but with exponentially small transition temperatures , @xmath237 , which might be hardly detectable in experiments . as a result the temperature scaling of correlations investigated here for uncoupled ladders is expected to be relevant to the behavior of real ladder materials with non - magnetic doping . we would like to thank n. bray - ali , s. garnerone , and b. normand for valuable discussions . we also acknowledge financial support by the department of energy , grant number de - fg03 - 01er45908 . t. r. acknowledges the hospitality of the mpipks - dresden , where part of this work was completed . r. y. was supported by nsf grant no . dmr-1006985 and robert a. welch foundation grant no . the numerical computations were carried out on the university of southern california high - performance supercomputer cluster . 200 e. dagotto , t. m. rice , science * 271 * , 618 ( 1996 ) . m. sigrist and a. furusaki , j. phys . . jpn . * 65 * , 2385 ( 1996 ) . sandvik , e. dagotto , and d. j. scalapino , phys . rev . b * 56 * , 11701 ( 1997 ) . g.b . martins , m. laukamp , j. riera , and e. dagotto , phys . * 78 * , 3563 ( 1997 ) . s. wessel , b. normand , m. sigrist , and s. haas , phys . * 86 * , 1086 ( 2001 ) . n. laflorencie and d. poilblanc , phys . lett . * 90 * , 157202 ( 2003 ) . c. yasuda , s. todo , m. matsumoto , and h. takayama , phys b * 64 * , 092405 ( 2001 ) . m. imada and y. iino , j. phys . 66 * , 568 ( 1997 ) . h. weber and m. vojta , eur . j. b * 53 * , 185 ( 2006 ) . m. azuma , y. fujishiro , m. takano , m. nohara and h. takagi , phys . rev . b. * 55 * , r8658 ( 1997 ) . j. bobroff , n. laflorencie , l. k. alexander , a. v. mahajan , b. koteswararao , and p. mendels , phys . lett . * 103 * , 047201 ( 2009 ) . h. manaka , a. v. kolomiets , and t. goto , phys . * 101 * 077204 ( 2008 ) . h. manaka , h. a. katori , o. v. kolomiets , and t. goto , phys . b * 79 * 092401 ( 2009 ) . tao hong , a. zheludev , h. manaka , and l .- p . regnault , phys . b * 81 * 060410 ( 2010 ) . m. greven and r. j. birgeneau , phys . lett . * 81 * , 1945 ( 1998 ) . t. miyazaki , m. troyer , m. ogata , k. ueda , and d. yoshioka , j. phys . * 66 * , 2580 ( 1997 ) . a. furusaki , m. sigrist , p. a. lee , k. tanaka , and n. nagaosa , phys . lett . * 73 * , 2622 ( 1994 ) . e. westerberg , a. furusaki , m. sigrist , and p. a. lee , phys . lett . * 75 * , 4302 ( 1995 ) . n. nagaosa , a. furusaki , m. sigrist , and h. fukuyama , j. phys . . jpn . * 67 * , 3724 ( 1996 ) . r. yu , t. roscilde , and s. haas , phys . lett . * 94 * , 197204 ( 2005 ) . r. yu , t. roscilde , and s. haas , phys . b * 73 * , 064406 ( 2006 ) . c. yasuda , s. todo , and h takayama , j. phys . soc . jpn . * 75 * , 124704 ( 2006 ) . o. f. syljusen and a. w. sandvik , phys . e * 66 * , 046701 ( 2002 ) . a. w. sandvik , phys . b * 66 * 024418 ( 2002 ) . f. cooper , b. freedman , and d. preston , nucl . b * 210 * , 210 ( 1982 ) . mikeska , u. neugebauer , and u. schollwck , phys . b * 55 * , 2955 ( 1997 ) . e. westerberg , a. furusaki , m. sigrist , and p. a. lee , phys . b * 55 * , 12578 ( 1997 ) . b. frischmuth , m. sigrist , b. ammon , and m. troyer , phys . b * 60 * , 3388 ( 1999 ) . n. laflorencie , d. poilblanc , and m. sigrist , phys . b. * 71 * , 212403 ( 2005 ) . the notion of local gap is associated to the local susceptibility at site @xmath229 , @xmath238 . assuming that @xmath239 one obtains that @xmath240 , where @xmath241 is the local gap . d. stauffer , `` introduction to percolation theory '' , taylor and francis , london 1985 . m. e. j. newman and r. m. ziff , phys . 85 * , 4104 ( 2000 ) ; m. e. j. newman and r. m. ziff , phys . e * 64 * , 016706 ( 2001 ) . d. s. fisher , phys . b * 50 * , 3799 ( 1994 ) . e. yusuf and k. yang , phys . b * 65 * , 224428 ( 2002 ) . r. mlin , y .- c . lin , p. lajk , h. rieger , and f. igli , phys . b * 65 * , 104415 ( 2002 ) . j. a. hoyos and e. miranda , phys . rev . b * 69 * , 214411 ( 2004 ) . f. igli and c. monthus , phys . rep . * 412 * , 277 ( 2005 ) . it is to be said that , at the low temperature to which the figure refers ( @xmath242 ) , the correlation length on the doped 4-leg and 6-leg ladders grows to values which are no longer satisfying the condition @xmath243 , and hence it is limited by the system size ( @xmath164 ) explored there .

computer simulation methods play an increasingly important role in the study of complex fluids . historically , many simulations have concentrated on fluids modeled as an assembly of monodisperse spherical particles interacting , e.g. , via a bare excluded - volume potential ( hard - sphere fluid ) or a lennard - jones interaction @xcite . while such systems can already display a wealth of interesting features , such as a solid liquid transition and in the presence of attractive interactions a critical point , it is also clear that real fluids and solutions can exhibit far richer behavior . various factors contribute to the additional features found in these systems , including internal degrees of freedom of the constituents ( such as in polymeric systems ) , interactions induced by electrical charges , and the presence of various species with a strong size asymmetry . modeling any of these properties can greatly increase the required numerical efforts and in certain situations can make the simulations prohibitively expensive . although available computational power continues to increase steadily , further progress in simulating such systems will critically depend on algorithmic advances . recently , we have introduced a simulation method that addresses one of the above - mentioned complicating factors , namely the slow - down arising in simulations of solutions containing species of largely different sizes @xcite ( see also ref . this method , which generalizes an original idea due to dress and krauth @xcite to identify clusters based upon geometric symmetry operations and accordingly is called the ( generalized ) _ geometric cluster algorithm _ ( gca ) , exhibits two noteworthy features . firstly , it employs a nonlocal monte carlo ( mc ) scheme to move the constituent particles in a nonphysical way , thus introducing artificial dynamics while preserving all thermodynamic equilibrium properties . this greatly accelerates the generation of uncorrelated configurations of particles . we emphasize that our scheme does not involve any approximations ; the molecular configurations produced by the gca are generated according to the same boltzmann distribution that would govern a conventional simulation of the system under consideration , but the gca follows a different trajectory through phase space . secondly , the nonlocal moves involve clusters of particles that are constructed in such a way that _ all proposed moves are accepted_. this is not only an additional factor contributing to the efficiency of the method , but it also has a more profound significance . namely , in order to satisfy detailed balance @xcite without imposing the usual metropolis @xcite acceptance criterion , the ratio of the probability of transforming a particle configuration @xmath0 into a new configuration @xmath1 and the probability of the reverse transformation must be identical to the ratio of the boltzmann factors of the configuration @xmath1 and the original configuration @xmath0 . clearly , this can only be realized if the transformations ( `` moves '' ) are proposed with a probability that involves knowledge about the physical properties of a system . for the vast majority of mc algorithms , this is not the case . the most well - known exceptions are the swendsen wang ( sw ) algorithm @xcite for lattice spin models and its variant due to wolff @xcite . among continuum systems , hard - sphere fluids constitute a special case , since all configurations without particle overlap have the same energy and thus the same boltzmann factor . however , it is generally accepted that these _ rejection - free _ algorithms are an exception ( cf . , e.g. , ref . @xcite , sec . indeed , ever since the invention of the sw cluster algorithm for ising and potts models , its extension to fluids has been a widely - pursued goal . for lattice gases the extension can be accomplished in a straightforward manner , since they are isomorphic to potts models . however , for off - lattice ( continuum ) fluids , this mapping can not be applied , owing to the absence of particle hole symmetry . this symmetry is a critical ingredient for the sw algorithm for ising and potts models , as clusters of spins ( or variables , in the case of the potts model ) are identified using a symmetry operation that , if applied globally , would leave the hamiltonian of the system invariant . the only known extensions to continuum systems apply to the widom rowlinson model for fluid mixtures @xcite , in which identical particles do not interact and unlike species experience a hard - core repulsion , and the closely - related stillinger helfand model , in which the hard - core is replaced by a soft repulsion @xcite . however , no generalization has been found in which identical particles do not behave as an ideal gas . in a separate development , dress and krauth @xcite observed that , for hard - core fluids , an overlap - free configuration can be transformed into a new configuration of nonoverlapping particles by means of a _ geometric _ symmetry operation . the particular advantage of this scheme is its ability to efficiently relax size - asymmetric mixtures @xcite . in ref . @xcite we showed how this advantage can be retained for fluids with _ arbitrary _ pair potentials , while simultaneously exploiting the invariance of the hamiltonian under the symmetry operation to create particle clusters in a manner that is fully analogous to the sw or wolff method . an application to a concrete physical problem has been presented in ref . @xcite , where colloid nanoparticle mixtures featuring attractive , repulsive and hard - core interactions were simulated for size asymmetries up to 100 . we remark that geometric cluster algorithms have also been applied successfully to lattice - based models . heringa and blte have employed it to investigate lattice gases with nearest - neighbor exclusion @xcite and developed a version for simulation of the ising model in the constant - magnetization ensemble @xcite . the `` pocket algorithm '' of krauth and moessner @xcite , which is essentially a geometric cluster algorithm for dimer models , was used to investigate hard - core dimers on three - dimensional lattices @xcite . furthermore , we note that there is a profound difference between the geometric cluster algorithm and other monte carlo schemes that move groups of particles simultaneously @xcite . such algorithms can work efficiently in specific cases , and are sometimes viewed as counterparts of the sw algorithm merely because they also invoke the notion of a `` cluster , '' but they typically involve a tunable parameter and are neither rejection free nor do they exploit a symmetry property of the hamiltonian . in the current article we provide a detailed description of the generalized geometric cluster algorithm for continuum fluids and study basic properties such as the dependence of cluster - size distribution on temperature and density . we also describe a multiple - cluster variant of the generalized gca in which a particle configuration is fully decomposed into clusters that can be moved independently . for reference in future sections that highlight the similarities between the generalized gca @xcite for off - lattice fluids and the sw @xcite and wolff @xcite algorithms for lattice spin models , we briefly summarize the lattice cluster algorithms for the case of a @xmath2-dimensional ising model with nearest - neighbor interactions , described by the hamiltonian @xmath3 the spins @xmath4 are placed on the vertices of a square ( @xmath5 ) or simple cubic ( @xmath6 ) lattice and take values @xmath7 . the sum runs over all pairs of nearest neighbors , which are coupled via a ferromagnetic coupling with strength @xmath8 . starting from a given configuration of spins , the sw algorithm now proceeds as follows : 1 . a `` bond '' is formed between every pair of nearest neighbors that are aligned , with a probability @xmath9 , where @xmath10 . all spins that are connected , directly or indirectly , via bonds belong to a single cluster . thus , the bond assignment procedure divides the system into clusters of parallel spins . note how the bond probability ( and hence the typical cluster size ) grows with increasing coupling strength @xmath11 ( decreasing temperature ) . for finite @xmath11 , @xmath12 and hence a cluster is generally a _ subset _ of all spins of a given sign . all spins in each cluster are flipped _ collectively _ with a probability @xmath13 . i.e. , for each cluster of spins a spin value @xmath7 is chosen and this value is assigned to all spins that belong to the cluster . the last , and crucial , step is made possible by the fortuin kasteleyn mapping @xcite of the potts model on the random - cluster model , which implies that the partition function of the former can be written as a whitney polynomial that represents the partition function of the latter @xcite . accordingly , all spins in a cluster ( a connected component in the random - cluster model ) are uncorrelated with all other spins , and can be assigned a new spin value . this algorithm possesses several noteworthy properties . first , it strongly suppresses dynamic slowing down near a critical point @xcite by efficiently destroying nonlocal correlations ( see also ref . @xcite for a pedagogical introduction ) . secondly , this algorithm is _ rejection free_. indeed , the assignment of bonds involves random numbers , but once the clusters have been formed each of them can be flipped independently without imposing an acceptance criterion involving the energy change induced by such a collective spin - reversal operation . thirdly , the fortuin kasteleyn mapping can also be applied to systems in which each spin interacts not only with its nearest neighbors , but also with other spins @xcite . in particular , the coupling strength can be different for different spin pairs , leading to a probability @xmath14 that is , e.g. , dependent on the separation between @xmath15 and @xmath16 . thus , cluster algorithms can be designed for spin systems with medium- @xcite and long - range interactions @xcite . critical slowing down is suppressed even more strongly in a variant of this algorithm due to wolff @xcite . in this implementation , no decomposition of the entire spin configuration into clusters takes place . instead , a single cluster is formed , which is then always flipped : 1 . a spin @xmath15 is selected at random . all nearest neighbors @xmath16 of this spin are added to the cluster with a probability @xmath9 , provided spins @xmath15 and @xmath16 are parallel and the bond between @xmath15 and @xmath16 has not been considered before . each spin @xmath16 that is indeed added to the cluster is also placed on the stack . once all neighbors of @xmath15 have been considered for inclusion in the cluster , a spin is retrieved from the stack and all its neighbors are considered in turn for inclusion in the cluster as well , following step 2 . steps 2 and 3 are repeated iteratively until the stack is empty . once the cluster has been completed , all spins in the cluster are inverted . again , this is a rejection - free algorithm , in the sense that the cluster is always flipped . just as in the sw algorithm , the cluster - construction process involves random numbers , but the individual probabilities @xmath14 involve single - particle energies rather than an acceptance criterion that involves the total energy change induced by a cluster flip . since suppression of critical slowing is a highly attractive feature for fluid simulations as well , the generalization of the sw and wolff algorithms to fluid systems has been a widely - pursued goal . in the lattice - gas interpretation , where a spin @xmath17 corresponds to a particle and a spin @xmath18 corresponds to an empty site , a spin - inversion operation corresponds to a particle being inserted in or removed from the system . this `` particle hole symmetry '' is absent in off - lattice ( continuum ) systems . while a particle in a fluid configuration can straightforwardly be deleted , there is no unambiguous prescription on how to transform empty space into a particle . more precisely , in the lattice cluster algorithms the operation performed on every spin is self - inverse . this requirement is not fulfilled for off - lattice fluids . , @xmath19 , @xmath20 ) . each pair is denoted a _ cluster_. the particles in any one of these clusters can be point - reflected with respect to the pivot without affecting the other two clusters . this can be used to carry out the point reflection for every cluster with a pre - set probability . c ) final configuration that results if , starting from the original configuration , only the particles in the third cluster @xmath20 are point - reflected . this approach guarantees that every generated configuration will be free of overlaps . note that the pivot will generally not be placed in the center of the cell , and that the periodic boundary conditions indeed permit any position.,scaledwidth=96.0% ] dress and krauth @xcite proposed a method to efficiently generate particle configurations for a hard - sphere liquid . geometric cluster algorithm _ proceeds as follows ( cf . fig . [ fig : geomc - hs ] ) . 1 . in a given configuration @xmath0 , a `` pivot '' is chosen at random . a configuration @xmath21 is now generated by carrying out a point reflection for all particles in @xmath0 with respect to the pivot . the configuration @xmath0 and its transformed counterpart @xmath21 are superimposed , which leads to groups of overlapping particles . the groups generally come in pairs , except possibly for a single group that is symmetric with respect to the pivot . each pair is denoted a `` cluster '' @xcite . 4 . for each cluster , all particles can be exchanged between @xmath0 and @xmath21 without affecting particles belonging to other clusters . this exchange is performed for each cluster independently with a probability @xmath22 . thus , if the superposition of @xmath0 and @xmath21 is decomposed into @xmath23 clusters , there are @xmath24 possible new configurations . the configurations that are actually realized are denoted @xmath1 and @xmath25 , i.e. , the original configuration @xmath0 is transformed into @xmath1 and its point - reflected counterpart @xmath21 is transformed into @xmath25 . the configuration @xmath25 is discarded and @xmath1 is the new configuration , serving as the starting point for the next iteration of the algorithm . note that a new pivot is chosen in every iteration . observe that periodic boundary conditions must be employed , such that an arbitrary placement of the pivot is possible . other self - inverse operations are permissible , such as a reflection in a plane @xcite , in which case various orientations of the plane must be chosen in order to satisfy ergodicity . while operating in the canonical rather than in the grand - canonical ensemble , this prescription clearly bears great resemblance to the original sw algorithm . the original configuration is decomposed into clusters by exploiting a symmetry operation that leaves the hamiltonian invariant if applied to the entire configuration ; in the sw algorithm this is the spin - inversion operation and in the geometric cluster algorithm it is a geometric symmetry operation . subsequently , a new configuration is created by moving each cluster independently with a certain probability . we note that this method represents an approach of great generality . for example , it is not restricted to monodisperse systems , and has indeed been applied successfully to binary @xcite and polydisperse @xcite mixtures . indeed , the nonlocal character of the particle moves makes them exquisitely suitable to overcome the jamming problems that slow down the simulation of size - asymmetric mixtures . an important limitation of the algorithm is the fact that the average cluster size increases very rapidly beyond a certain density , corresponding to the percolation threshold of the combined system containing the superposition of the configurations @xmath0 and @xmath21 . once this cluster spans the entire system , the algorithm is clearly no longer ergodic . we can take the analogy with the lattice cluster algorithms one step further , by showing that a single - cluster ( wolff ) variant can be formulated as well @xcite . 1 . in a given configuration @xmath0 , a `` pivot '' is chosen at random . a particle @xmath15 is selected as the first particle that belongs to the cluster . this particle is moved via a point reflection with respect to the pivot . in its new position , the particle is referred to as @xmath26 . step 2 is repeated _ iteratively _ for each particle @xmath16 that overlaps with @xmath26 . thus , if the ( moved ) particle @xmath27 overlaps with another particle @xmath28 , particle @xmath28 is moved as well . note that all translations involve the same pivot . once all overlaps have been resolved , the cluster move is completed . as in the sw - like prescription , a new pivot is chosen for each cluster that is constructed . the geometric cluster algorithm described in the previous section is formulated for hard - core interactions . for application to general pair potentials , it was suggested in ref . @xcite to impose a metropolis - type acceptance criterion based upon the energy difference induced by the cluster move . indeed , in the case of potentials that consist of a hard - core ( excluded - volume ) contribution supplemented by an attractive or repulsive tail , such as a yukawa potential , the cluster - construction procedure guarantees that no overlaps are generated and the acceptance criterion takes into account the tail of the interactions . for `` soft - core '' potentials , such as a lennard - jones interaction , the situation becomes already somewhat more complicated , since an arbitrary excluded - volume distance must be chosen for the cluster construction . as the algorithm will not generate configurations in which the separation between a pair of particles is less than this distance ( i.e. , the particle `` diameter , '' in the case of monodisperse systems ) , it must be set to a value that is smaller than any separation that would typically occur , as already noted in ref . @xcite . in either case , the clusters that are generated have only limited physical relevance , and the evaluation of a considerable part of the energy change resulting from a cluster move is deferred until the acceptance step . rejection is not only likely , but also costly , given the computational efforts of both cluster construction and energy evaluation . we note that this approach was nevertheless applied to yukawa mixtures with moderate size asymmetry ( diameter ratio @xmath29 ) @xcite . on the other hand , heringa and blte @xcite devised an geometric cluster algorithm for the ising model in which the nearest - neighbor interactions between spins are taken into account already during the cluster construction . while this lattice model can obviously be simulated by the sw and wolff algorithms , their approach permits simulation in the constant - magnetization ensemble . the geometric operations employed map the spin lattice onto itself , such that excluded - volume conditions are satisfied automatically : every spin move amounts to an _ exchange _ of spins . for every spin pair @xmath30 that is exchanged , each of its nearest - neighbor pairs @xmath31 are exchanged with a probability that depends on the change in pair energy , @xmath32 . this procedure is then again performed iteratively for the neighbors of all spin pairs that are exchanged . , open and shaded disks denote the particles before and after the geometric operation , respectively , and the small disk denotes the pivot . however , in the _ generalized _ gca a single cluster is constructed , to which particles are added with an interaction - dependent probability . a ) original configuration . b ) a cluster is constructed as follows . particle 1 is point - reflected with respect to the pivot . if , in its new position , it has a repulsive interaction with particle 2 , the latter has a certain probability to be point - reflected as well , with respect to the same pivot . assuming an attractive interaction between particles 2 and 3 , particle 3 is translated as well , but again only with a certain probability . if particles 46 are not affected by these point reflections , the cluster construction terminates . c ) the new configuration consists of particles 13 in their new positions and particles 46 in the original positions . a new pivot is chosen and the procedure is repeated.,scaledwidth=96.0% ] in ref . @xcite we introduced a generalization of the gca for off - lattice fluids , in which particles undergo a geometric operation in a stochastic manner , akin to the approach of ref . the differences arise from the fact that no underlying lattice is present , so that particles are added to the cluster on an individual basis , rather than in pairs . all interactions are treated in a unified manner , so there is no technical distinction between attractive and repulsive interactions or between hard - core and soft - core potentials . generalized gca _ is most easily described as a combination of the single - cluster methods formulated in sec . [ sec : sw - wolff ] and [ sec : hardcore - gca ] . we assume a general pair potential @xmath33 that does not have to be identical for all pairs @xmath34 ( see fig . [ fig : geomc - wolff ] ) . 1 . in a given configuration @xmath0 , a `` pivot '' is chosen at random . a particle @xmath15 at position @xmath35 is selected as the first particle that belongs to the cluster . this particle is moved via a point reflection with respect to the pivot . in its new position , the particle is referred to as @xmath26 , at position @xmath36 . each particle @xmath16 that interacts with @xmath15 or @xmath26 is now considered for addition to the cluster . unlike the first particle , particle @xmath16 is point - reflected with respect to the pivot only with a probability @xmath37 $ ] , where @xmath38 . a particle @xmath16 that interacts with @xmath15 both in its old and in its new position is nevertheless treated only once . each particle @xmath16 that is indeed added to the cluster ( i.e. , moved ) is also placed on the stack . once all particles interacting with @xmath15 or @xmath26 have been considered , a particle is retrieved from the stack and all its neighbors that are not yet part of the cluster are considered in turn for inclusion in the cluster as well , following step 3 . steps 3 and 4 are repeated iteratively until the stack is empty . the cluster move is now complete . if a particle interacts with multiple other particles that have been added to the cluster , it can thus be considered multiple times for inclusion . however , once it has been added to the cluster , it can not be removed . this is an important point in practice , since particles undergo a point reflection already during the cluster construction process ( and thus need to be tagged , in order to prevent them from being returned to their original position by a second point reflection ) . a crucial aspect is that the probability @xmath14 _ only _ depends on the change in _ pair energy _ between @xmath15 and @xmath16 . a change in the relative position of @xmath15 and @xmath16 occurs if particle @xmath16 is _ not _ added to the cluster . this happens with a probability @xmath39 $ ] . the similarity with the metropolis acceptance criterion is deceptive ( and merely reflects the fact that both algorithms aim to generate configurations according to the boltzmann distribution ) , since @xmath40 does not represent the _ total _ energy change resulting from the translation of particle @xmath15 . instead , other energy changes are taken into account via the iterative nature of the algorithm . it is instructive to note that the expression for @xmath14 bears close resemblance to the probability employed in the sw algorithm ( sec . [ sec : sw - wolff ] ) . in the latter , two different situations can be discerned that lead to a change in the relative energy @xmath41 between a spin @xmath15 that belongs to the cluster and a spin @xmath16 that does not yet belong to the cluster . if @xmath15 and @xmath16 are initially _ antiparallel _ , @xmath16 will never be added to the cluster and only spin @xmath15 will be inverted , yielding an energy change @xmath42 that occurs with probability unity . if @xmath15 and @xmath16 are initially _ parallel _ and @xmath16 is not added to the cluster , the resulting change in the pair energy equals @xmath43 . this occurs with a probability @xmath44 . these two situations can indeed be summarized as @xmath45 $ ] , just as in the generalized gca . the ergodicity of this algorithm follows from the fact that there is a nonvanishing probability that a cluster consists of only one particle , which can be moved over an arbitrarily small distance , since the location of the pivot is chosen at random . this obviously requires that not all particles are part of the cluster , a condition that is violated at high packing fractions . while a compact proof of detailed balance of the generalized gca has already been given in ref . @xcite , we include it here for the sake of completeness . we consider a configuration @xmath46 that is transformed into a configuration @xmath47 by means of a cluster move . all particles included in the cluster maintain their relative separation ; as noted above , an energy change arises if a particle is _ not _ included in the cluster , but interacts with a particle that does belong to the cluster . following wolff @xcite we denote each of these interactions as a `` broken bond . '' a broken bond @xmath28 that corresponds to an energy change @xmath48 occurs with a probability @xmath49 if @xmath50 and a probability @xmath51 if @xmath52 . the formation of an entire cluster corresponds to the breaking of a set @xmath53 of bonds , which has a probability @xmath54 . this set is comprised of the subset @xmath55 of broken bonds @xmath56 that lead to an increase in pair energy and the subset @xmath57 of broken bonds that lead to a decrease in pair energy , such that @xmath58 \;.\ ] ] the transition probability from configuration @xmath46 to configuration @xmath47 is proportional to the cluster formation probability , @xmath59 \;,\ ] ] where the factor @xmath0 accounts for the fact that various arrangements of bonds within the cluster ( `` internal bonds '' ) correspond to the same set of broken bonds . in addition , it incorporates the probability of choosing a particular pivot and a specific particle as the starting point for the cluster . if we now consider the reverse transition @xmath60 , we observe that this again involves the set @xmath53 , but all the energy differences change sign compared to the forward move . consequently , the subset @xmath55 in eq . ( [ eq : trans - fw ] ) is replaced by its complement @xmath57 and the transition probability is given by @xmath61 \;,\ ] ] where the factor @xmath0 is identical to the prefactor in eq . ( [ eq : trans - fw ] ) . since we require the geometric operation to be self - inverse we thus find that the cluster move satisfies detailed balance at an acceptance ratio of unity , @xmath62 } { \exp\left[+ \beta \sum_{m } \delta_m\right ] } = \exp\left [ -\beta \sum_{k } \delta_k\right ] = \exp\left [ -\beta ( e_y - e_x)\right ] = \frac{\exp(-\beta e_{y})}{\exp(-\beta e_{x } ) } \;,\ ] ] where @xmath63 and @xmath64 are the internal energies of configurations @xmath46 and @xmath47 , respectively . i.e. , the ratio of the forward and reverse transition probabilities is equal to the inverse ratio of the boltzmann factors , so that we indeed have created a rejection - free algorithm . this is obscured to some extent by the fact that in our prescription the cluster is moved while it is being constructed , similar to the wolff algorithm in sec . [ sec : sw - wolff ] . the central point , however , is that the construction solely involves single - particle energies , whereas a metropolis - type approach only evaluates the total energy change induced by a multi - particle move and then frequently rejects this move . by contrast , the gca avoids large energy differences by incorporating `` offending '' particles into the cluster with a high probability . we now introduce a sw implementation of the generalized gca . the merit of this formulation , which builds upon the wolff version described in the previous section , is twofold : first , it demonstrates that the algorithm constitutes the true off - lattice counterpart of the sw and wolff cluster algorithms for spin models outlined in sec . [ sec : sw - wolff ] . secondly , the sw formulation produces a full decomposition of an off - lattice fluid configuration into _ stochastically independent clusters_. this implies an interesting and remarkable analogy with the ising model . as observed by coniglio and klein @xcite for the two - dimensional ising model at its critical point , the clusters created according to the prescription in sec . [ sec : sw - wolff ] are just the so - called `` fisher droplets '' @xcite . while ref . @xcite makes no reference to the work by fortuin and kasteleyn , these `` coniglio klein clusters '' are implied by the fortuin kasteleyn mapping of the potts model onto the random - cluster model @xcite , which in turn constitutes the basis for the swendsen wang approach @xcite . the clusters generated by the gca do not have an immediate physical interpretation , as they typically consist of two spatially disconnected parts . however , just like the ising clusters can be inverted at random , each cluster of fluid particles can be moved independently with respect to the remainder of the system . as such , the generalized gca can be viewed as a continuum version of the fortuin kasteleyn mapping . the cluster decomposition of a configuration proceeds as follows . first , a cluster is constructed according to the wolff version of sec . [ sec : gca ] , with the exception that the cluster is only _ identified _ ; particles belonging to the cluster are marked but not actually moved . the pivot employed will also be used for the construction of all subsequent clusters in this decomposition . these subsequent clusters are built just like the first cluster , except that particles that are already part of an earlier cluster will never be considered for a new cluster . once each particle is part of exactly one cluster the decomposition is completed . like in the sw algorithm , every cluster can then be moved ( i.e. , all particles belonging to it are translated via a point reflection ) independently , e.g. , with a probability @xmath65 . despite the fact that all clusters except the first are built in a restricted fashion , each individual cluster is constructed according to the rules of the wolff formulation of sec . [ sec : gca ] . the exclusion of particles that are already part of another cluster simply corresponds to the fact that every bond should be considered only once . if a bond is broken during the construction of an earlier cluster it should not be re - established during the construction of a subsequent cluster . the cluster decomposition thus obtained is not unique , as it depends on the placement of the pivot and the choice of the first particle . evidently , this also holds for the sw algorithm . in order to establish that this prescription is a true equivalent of the sw algorithm , we prove that each cluster can be moved ( reflected ) independently while preserving detailed balance . if only a single cluster is actually moved , this essentially corresponds to the wolff version of the gca , since each cluster is built according to the gca prescription . the same holds true if several clusters are moved and no interactions are present between particles that belong to different clusters ( the hard - sphere algorithm is a particular realization of this situation ) . if two or more clusters are moved and _ broken _ bonds exist between these clusters , i.e. , a nonvanishing interaction exists between particles that belong to disparate ( moving ) clusters , then the shared broken bonds are actually preserved and the proof of detailed balance provided in the previous section no longer applies in its original form . however , since these bonds are identical in the forward and the reverse move , the corresponding factors cancel out . this is illustrated for the situation of two clusters whose construction involves , respectively , two sets of broken bonds @xmath66 and @xmath67 . each set comprises bonds @xmath56 ( @xmath68 and @xmath69 , respectively ) that lead to an _ increase _ in pair energy and bonds @xmath70 ( @xmath71 and @xmath72 , respectively ) that lead to a _ decrease _ in pair energy . we further subdivide these sets into _ external _ bonds that connect cluster 1 or 2 with the remainder of the system and _ joint _ bonds that connect cluster 1 with cluster 2 . accordingly , the probability of creating cluster 1 is given by @xmath73 upon construction of the first cluster , the creation of the second cluster has a probability @xmath74 since all joint bonds in @xmath75 already have been broken . the factors @xmath76 and @xmath77 refer to the probability of realizing a particular arrangement of internal bonds in clusters 1 and 2 , respectively ( cf . [ sec : gca ] ) . hence , the total transition probability of moving _ both _ clusters is given by @xmath78 \;.\ ] ] in the reverse move , the energy differences for all external broken bonds have changed sign , but the energy differences for the joint bonds connecting cluster 1 and 2 are the same as in the forward move . thus , cluster 1 is created with probability @xmath79 \prod_{j \in \left\{l_1^{\rm joint}\right\ } } \exp[-\beta \delta_j ] \;,\ ] ] where the @xmath80 reflects the sign change of the energy differences compared to the forward move and the product over the external bonds involves the complement of the set @xmath81 . the creation probability for the second cluster is @xmath82\ ] ] and the total transition probability for the reverse move is @xmath83 \;.\ ] ] accordingly , detailed balance is still fulfilled with an acceptance ratio of unity , @xmath84 = \exp\left [ - \beta ( e_y - e_x ) \right ] \;,\ ] ] in which @xmath85 and @xmath86 and @xmath63 and @xmath64 refer to the internal energy of the system before and after the move , respectively . this treatment applies to any simultaneous move of clusters , so that _ each cluster in the decomposition indeed can be moved independently _ without violating detailed balance . this completes the proof of the multiple - cluster version of the gca . it is noteworthy that the probabilities for breaking joint bonds in the forward and reverse moves cancel only because the probability in the cluster construction factorizes into individual probabilities . in order to illustrate the validity of this approach , we have applied it to the binary lennard - jones mixture employed in ref . this system consists of @xmath87 small particles ( diameter @xmath88 ; reduced coupling @xmath89 ) and @xmath90 large particles ( diameter @xmath91 ; reduced coupling @xmath92 ) at a total packing fraction @xmath93 . following the lorentz berthelot mixing rules @xcite we set the parameters for the large small lennard - jones interaction to @xmath94 and @xmath95 . the particles are contained in a cubic cell with linear size @xmath96 . periodic boundary conditions are applied and the cut - off for all interactions is set to @xmath97 . we perform a full cluster decomposition for every configuration and carry out a reflection for every cluster with a probability @xmath98 . as illustrated in fig . [ fig : wolff - sw ] , the correlation functions for pairs of large particles and for pairs of large and small particles agree perfectly with the results obtained in ref . @xcite by means of the single - cluster version . in sec . [ sec : efficiency ] we address the relative efficiency of both approaches . ; solid lines ) and the multiple - cluster version ( sec . [ sec : gca - sw ] ; symbols ) of the generalized geometric cluster algorithm . the figure shows pair correlation functions for the size - asymmetric lennard - jones mixture described in the text . @xmath99 and @xmath100 represent the large large and large small correlation functions , respectively . there is excellent agreement between both algorithms . ] the actual implementation of the generalized gca involves a variety of issues . the point reflection with respect to the pivot requires careful consideration of the periodic boundary conditions . furthermore , as mentioned above , particles that have been translated via a point reflection must not be translated again in the same cluster move , and particles that interact with a given cluster particle both before and after the translation of that cluster particle must be considered only once , on the basis of the difference in pair potential . in order to account for all interacting pairs in an efficient manner , we employ the cell index method @xcite . for mixtures with large size asymmetries ( the situation where the generalized gca excels ) , it is natural to set up different cell structures , with cell lengths based upon the cut - offs of the various particle interactions . for example , in the case of a binary mixture of two species with very different sizes and cutoff radii ( @xmath101 and @xmath102 , respectively ) , the use of an identical cell structure with a cell size that is determined by the large particles would be highly inefficient for the smaller particles . thus , two cell structures are constructed in this case ( with cell sizes @xmath103 and @xmath104 , respectively ) and each particle is stored in the appropriate cell of the structure belonging to its species , and incorporated in the corresponding linked list , following the standard approach @xcite . however , in order to efficiently deal with interactions between unlike species ( which have a cut - off @xmath105 ) , a mapping between the two cell structures is required . if all small particles must be located that interact with a given large particle , we proceed as follows . first , the small cell @xmath106 is identified in which the center of the large particle resides . subsequently , the interacting particles are located by scanning over all small cells within a cubic box with linear size @xmath107 , centered around @xmath106 . this set of cells is predetermined at the beginning of a run and their indices are stored in an array . each set contains approximately @xmath108 members . in an efficient implementation , @xmath109 is not much larger than @xmath110 , which for short - range interactions is of the order of the size of a small particle . likewise , @xmath111 is typically of the order of the size of the large particle , so that @xmath112 , where @xmath113 denotes the size asymmetry between the two species . since @xmath114 indices must be stored for each large cell , the memory requirements become very large for cases with large size asymmetry , cf . ref . @xcite for @xmath115 . whereas the autocorrelation time per particle ( expressed in @xmath116s of cpu time per particle move ) rapidly increases with size ratio , the gca features only a weak dependence on @xmath113 . ] the most notable feature of the generalized gca , as emphasized in ref . @xcite , is the efficiency with which it generates uncorrelated configurations for size - asymmetric mixtures . this performance directly derives from the nonlocal character of the point reflection employed . in general , the translation of a single particle over large distances is impossible in all but the most dilute situations . on the other hand , multiple - particle moves typically entail an energy difference that strongly suppresses the likelihood of acceptance . by enabling collective moves while maintaining a high ( and , in fact , maximal ) acceptance probability , fluid mixtures can be simulated efficiently over a wide parameter range ( volume fraction , size asymmetry and temperature ) . following ref . @xcite , we illustrate this for a simple binary mixture in which the autocorrelation time is determined as a function of size asymmetry . this system contains 150 large particles of size @xmath117 , at fixed volume fraction @xmath118 . furthermore , @xmath119 small particles are present , also at fixed volume fraction @xmath120 . thus , as the size @xmath121 of these small particles is varied from @xmath122 to @xmath123 ( i.e. , the size ratio @xmath124 is increased from 2 to 15 ) , their number increases from @xmath125 to @xmath126 . pairs of small particles and pairs involving a large and a small particle act like hard spheres . however , in order to prevent depletion - driven aggregation of the large particles @xcite , we introduce a short - ranged yukawa repulsion , @xmath127 & \quad r > \sigma_{22 } \end{array } \right . \;,\ ] ] where @xmath128 and the screening length @xmath129 . in the simulation , the exponential tail is cut off at @xmath130 . the additional yukawa interactions also provides a fluctuating internal energy @xmath131 that permits us to determine the rate at which the system decorrelates . we consider the integrated autocorrelation time @xmath132 obtained from the energy autocorrelation function @xcite , @xmath133 and compare @xmath132 for a conventional ( metropolis ) mc algorithm and the generalized gca . in order to avoid arbitrariness resulting from the computational cost involved with a single sweep or the construction of a cluster , we express @xmath132 in actual cpu time ( assuming that both methodologies have been programmed in an efficient manner ) . furthermore , @xmath132 is normalized by the total number of particles in the system , to account for the variation in @xmath119 as the size ratio @xmath113 is increased . for conventional mc calculations , @xmath134 rapidly increases with increasing @xmath113 , because the large particles tend to get trapped by the small particles . indeed , already for @xmath135 it is not feasible to obtain an accurate estimate for @xmath134 . by contrast , @xmath136 exhibits a very different dependence on @xmath113 . at @xmath137 both algorithms require virtually identical simulation time , which establishes that the gca does not involve considerable overhead compared to standard algorithms ( if any , it is mitigated by the fact that all moves are accepted ) . upon increase of @xmath113 , @xmath138 initially _ decreases _ until it starts to increase weakly . the nonmonotonic variation of @xmath136 results from the changing ratio @xmath139 which causes the cluster composition to vary with @xmath113 . the main points to note are : ( i ) the gca greatly suppresses the autocorrelation time , @xmath140 for @xmath141 , with an efficiency increase that amounts to more than three orders of magnitude already for @xmath142 ; ( ii ) the increase of the autocorrelation time with @xmath113 is much slower for the gca than for a local - move mc algorithm , making the gca increasingly advantageous with increasing size asymmetry . this second observation is confirmed by the large size asymmetries that could be attained in ref . @xcite . , for a monodisperse lennard - jones fluid . ( a ) critical density , @xmath143 . the distribution is strongly bimodal in the vicinity of the critical temperature and remains bimodal up to relatively high temperatures . ( b ) twice smaller density @xmath144 . the cluster - size distribution only becomes bimodal for temperatures relatively close to the critical temperature and decreases monotonously for higher temperatures . identical symbols refer to identical temperatures in both panels . all temperatures are indicated in terms of @xmath145.,title="fig : " ] , for a monodisperse lennard - jones fluid . ( a ) critical density , @xmath143 . the distribution is strongly bimodal in the vicinity of the critical temperature and remains bimodal up to relatively high temperatures . ( b ) twice smaller density @xmath144 . the cluster - size distribution only becomes bimodal for temperatures relatively close to the critical temperature and decreases monotonously for higher temperatures . identical symbols refer to identical temperatures in both panels . all temperatures are indicated in terms of @xmath145.,title="fig : " ] the cluster size clearly has a crucial influence on the performance of the gca . if a cluster contains more than 50% of all particles , an equivalent change to the system could have been made by moving its complement ; unfortunately it is unclear how to determine this complement without constructing the cluster . nevertheless , it is found that the algorithm operates in a comparatively efficient manner for average relative cluster sizes as large as 90% or more . once the total packing fraction of the system exceeds a certain value , the original hard - core gca breaks down because each cluster occupies the entire system . the same phenomenon occurs in the generalized gca , but in addition the cluster size can saturate because of strong interactions . thus , the maximum accessible volume fraction depends on a considerable number of parameters , including the range of the potentials and the temperature . for multi - component mixtures , size asymmetry and relative abundance of the components are of importance as well , and the situation can be complicated further by the presence of competing interactions . as an illustration , we consider the cluster - size distribution for a monodisperse lennard - jones fluid ( particle diameter @xmath146 ) . for an interaction cut - off of @xmath147 , the critical temperature @xmath148 lies just below @xmath149 @xcite , where @xmath150 denotes the coupling strength . figure [ fig : clusterdistr](a ) shows this distribution at the critical density @xmath151 ( @xmath152 ) for a range of supercritical temperatures . already at temperatures that are far above the critical temperature , the cluster - size distribution tends toward a bimodal form , indicative of the formation of large clusters . the gap between the two peaks widens with decreasing temperature and in the vicinity of the critical temperature the average cluster size becomes very large . this is greatly different from the sw and wolff algorithms , which operate at the percolation threshold when applied to a critical system . remarkably , when applied to a lattice gas at its critical density , the geometric cluster algorithm was also found to yield the power - law distribution that is characteristic for a percolating system @xcite . this can be understood from the fact that excluded - volume effects play no role in geometric operations applied to lattice - based systems . in continuum systems , the superposition of a system and its point - reflected counterpart percolates already when the original system has a density that is considerably below the percolation threshold . motivated by this , we investigate the cluster - size distribution in the same lennard - jones fluid at a twice lower density , @xmath153 , see fig . [ fig : clusterdistr](b ) . for the highest temperature ( which is already twice lower than the highest temperature in panel ( a ) ) , this distribution now is a monotonously decreasing function of cluster size , and the bimodal character only appears for temperatures around 25% above the critical temperature . it turns out to be possible to influence the cluster - size distribution by placing the pivot in a biased manner . rather than first choosing the pivot location , a particle is selected that will become the first member of the cluster . subsequently , the pivot is placed at random within a cubic box of linear size @xmath154 , centered around the position of this particle . by decreasing @xmath155 , the displacement of the first particle is decreased , as well as the number of other particles affected by this displacement . as a consequence , the average cluster size decreases , and higher volume fractions can be reached . ultimately , the cluster size will still occupy the entire system , but we found that the maximum accessible volume fraction can be increased from approximately @xmath156 to a value close to @xmath157 . this value indeed corresponds to the percolation threshold for hard spheres , @xmath158 @xcite . note that the proof of detailed balance is not affected by this modification . as a function of linear system size for the critical lennard - jones fluid , in units of particle sweeps , for three different monte carlo algorithms : local moves ( `` metropolis mc '' ) ; gca with swendsen wang type cluster decomposition and probability 0.50 ( `` sw 50% '' ) and 0.75 ( `` sw 75% '' ) of moving each cluster ; single - cluster gca ( `` wolff '' ) . ] the cluster - size distributions obtained in the previous section suggest that the generalized gca will not suppress critical slowing down for the lennard - jones fluid . as emphasized in ref . @xcite , this does not have to be viewed as a great shortcoming of the algorithm , because of the efficiency improvement it delivers for the simulation of size - asymmetric fluids over a wide range of temperatures and packing fractions ( cf . fig . [ fig : efficiency ] ) . nevertheless , since suppression of critical slowing down plays such an important role for lattice cluster algorithms and since it is a feature that has not been realized by any fluid simulation algorithm , we have investigated the integrated autocorrelation time for the energy at the critical point , as a function of linear system size . in fig . [ fig : critdyn ] these times are collected for three algorithms . ( 1 ) conventional local - update metropolis algorithm ; ( 2 ) wolff version of the gca ; ( 3a ) sw version of the gca , in which each cluster is point - reflected with a probability @xmath159 ; ( 3b ) sw version of the gca , in which each cluster is point - reflected with a probability @xmath160 . this also serves as a performance comparison between the single - cluster gca and the multiple - cluster variant . just as for spin models , the single - cluster version is more efficient than the sw - like approach . however , all variants of the gca exhibit the same power - law behavior , which outperforms the metropolis algorithm by a factor @xmath161 . it is important to emphasize that this acceleration may be due to the suppression of the hydrodynamic slowing down @xcite caused by the conservation of the density ( which may couple to the energy correlations @xcite ) . remarkably , already for moderate system sizes the generalized gca outperforms the metropolis algorithm , despite the time - consuming construction of large clusters [ cf . [ fig : clusterdistr](a ) ] which lead to only small configurational changes . as an illustration of the capabilities of the generalized gca we apply it to a binary mixture first studied in ref . it contains dilute colloidal particles in a solvent of smaller particles . both species are modeled as hard spheres , but unlike pairs experience a yukawa attraction which promotes the accumulation of the solvent particles around the colloids . systems like this are relevant for an improved understanding of depletion effects in the presence of additional nonadditive interactions @xcite , but have hitherto been studied only to a limited extent , because of computational limitations . specifically , we set up a simulation cell containing 29000 small particles ( species 1 ) at volume fraction @xmath162 and two large colloids ( species 2 ) at volume fraction @xmath163 . the pair potentials are chosen as @xmath164 and @xmath165 & \quad r > \sigma_{12 } \end{array } \right . \;,\ ] ] where @xmath94 and @xmath166 . the coupling strength is set to @xmath167 , corresponding to a contact energy @xmath168 , and the decay parameter ( inverse screening length ) is set to @xmath169 @xcite . and volume fraction @xmath163 ) in an environment of smaller particles ( diameter @xmath170 and volume fraction @xmath162 ) that experience a yukawa - type attractive interaction with the colloids . the symbols represent data obtained by means of the hard - core gca supplemented by an acceptance criterion @xcite . the solid line is obtained via the generalized gca . ] in ref . @xcite this system was investigated by means of the hard - core gca supplemented by an acceptance criterion in order to take into account the yukawa attractions . in view of the size asymmetry between the two species , this already yields a considerable efficiency improvement over conventional mc simulations that only employ local moves . however , the acceptance criterion potentially greatly deteriorates performance , as clusters will be constructed that are subsequently rejected in their entirety . indeed , the authors report @xcite that accurate direct sampling of the colloidal pair correlation function @xmath171 was prohibitively expensive , so that they instead obtained it via numerical differentiation of the _ integrated _ pair correlation function ( see symbols in fig . [ fig : amokrane ] ) . this differentiation involves a polynomial fit and the result was found to be sensitive to the degree of this polynomial . in the generalized gca the yukawa attractions are directly incorporated in the cluster construction , so that all clusters are accepted . as demonstrated in fig . [ fig : amokrane ] ( solid line ) , accurate results for @xmath171 can now be obtained through direct sampling . note that our choice for the colloid concentration @xmath172 is slightly smaller than in ref . @xcite ( leading to a somewhat larger number of small particles in our calculation ) , which is however irrelevant for the results , as they have already converged to the dilute colloid limit . thus , the generalized gca opens possibilities for a systematic investigation of the effect of interaction strength and range on the potential of mean force , as a function of size asymmetry and solvent concentration . the addition of small , nonadsorbing additives , e.g. , polymers , to a colloidal suspension can lead to the well - known depletion interaction between colloids . this interaction has been modeled successfully by the asakura oosawa ( ao ) model @xcite which treats the polymer chains as ideal , noninteracting spheres that have an excluded - volume interaction with the colloids . alternatively , the polymers can be treated as hard spheres , leading to an additive binary hard - sphere mixture @xcite . although on a qualitative level both theories agree with the experimentally observed trends for depletion interactions , it recently has been suggested that a more accurate description can be obtained by means of a model in which the polymer pair potential is described by a gaussian @xcite , @xmath173 \;,\ ] ] where @xmath150 is the strength of the repulsive potential and @xmath174 its width . for dilute and semidilute polymer solutions , @xmath175 and @xmath176 , respectively , were found to be appropriate parameter values @xcite . this potential can be viewed as a model that interpolates between the ao model for ideal polymers and the binary hard - sphere mixture . for the gaussian polymer model , eq . ( [ eq : gauss ] ) . panel ( a ) pertains to a fixed interaction strength @xmath177 for various interaction widths @xmath178 , @xmath179 , @xmath180 ( triangles , circles and squares , respectively ) . in panel ( b ) , @xmath150 is varied ( @xmath181 , @xmath182 and @xmath183 , indicated by squares , circles and triangles , respectively ) for fixed @xmath184 . the solid line in both figures represents the hard - sphere result obtained by density - functional theory @xcite and the dotted line pertains to the expression by vrij @xcite for the ao model @xcite.,title="fig : " ] for the gaussian polymer model , eq . ( [ eq : gauss ] ) . panel ( a ) pertains to a fixed interaction strength @xmath177 for various interaction widths @xmath178 , @xmath179 , @xmath180 ( triangles , circles and squares , respectively ) . in panel ( b ) , @xmath150 is varied ( @xmath181 , @xmath182 and @xmath183 , indicated by squares , circles and triangles , respectively ) for fixed @xmath184 . the solid line in both figures represents the hard - sphere result obtained by density - functional theory @xcite and the dotted line pertains to the expression by vrij @xcite for the ao model @xcite.,title="fig : " ] in order to demonstrate that these system can be accurately and efficiently simulated by means of the generalized gca , we study the depletion interactions between colloidal particles as a function of the width @xmath113 and the strength @xmath150 . the colloid and polymer volume fractions are fixed at @xmath185 and @xmath186 , respectively , and their size ratio is set to @xmath187 . we employ the polymer coil diameter ( @xmath188 ) as unit length scale . the simulation involves @xmath189 `` polymers '' and 20 colloidal particles . figure [ fig : gauss ] shows that the ao model generally overestimates the depletion attraction in both strength and range . indeed , the soft - core polymer model yields an effective colloidal pair potential that decreases in strength and range with increasing @xmath113 and @xmath150 . in addition , this potential exhibits a repulsive barrier for a separation @xmath190 , owing to many - body effects that arise from the mutual repulsion between polymers . by contrast , there is good agreement between the additive hard - sphere mixture and the gaussian polymer model , in particular for @xmath191 and @xmath192 . interestingly , these are precisely the values that were found to reasonably represent dilute and semidilute self - avoiding random walk polymers @xcite . we have presented a detailed description of the generalized geometric cluster algorithm for continuum fluids introduced in ref . @xcite , which is a generalization of the work by dress and krauth @xcite . in order to emphasize the connection with the swendsen wang algorithm for lattice models we have derived a multiple - cluster variant of the gca , in which a particle configuration is decomposed into clusters that can be independently point - reflected with respect to a given pivot point , without affecting the remainder of the system . this algorithm establishes that it is generally possible to identify such clusters and to devise a rejection - free monte carlo scheme , independent of the detailed nature of the pair potentials between particles . no restrictions are imposed upon the number of species involved , but ergodicity can only be maintained if not every cluster contains all particles of a given species . for monodisperse hard - sphere fluids this implies a threshold packing fraction around @xmath156 , which can be increased to approximately @xmath157 if the pivot is placed in a biased manner . for interactions with a longer range , the clusters will typically be larger and hence the maximum accessible volume fraction decreases . the nonlocal character of the particle translations permits the efficient decorrelation of fluid mixtures that involve a strong size asymmetry between the components . unavoidably , the resulting dynamics have no physical interpretation , but all thermodynamic equilibrium properties are identical to those sampled by conventional algorithms . while cluster - size distributions for the lennard - jones fluid indicate that the percolation threshold and the critical point do not coincide , the algorithm significantly accelerates canonical simulations at the critical point . two specific example applications have been discussed , both involving depletion interactions in size - asymmetric binary mixtures . in the first illustration , we assess the effect of attractive interactions between unlike species on the pair correlation function of the larger species . this system has been investigated earlier by means of the hard - sphere gca supplemented by an acceptance criterion , and is included here to illustrate that it is possible to perform the calculation in a rejection - free manner . in the second illustration , we calculate the depletion interaction induced by a depletant that acts as a hard sphere for the larger species but has a gaussian interaction potential with other depletant particles . it is demonstrated how such calculations can be performed efficiently for relatively large size ratios and lead to depletion potentials that interpolate between the well - known ao potential and the depletion potential for binary hard - sphere mixtures . a variety of extensions to the gca can be devised . in particular , the application to nonspherical particles is straightforward , but may require an additional ( rotational ) monte carlo move that permits an efficient relaxation of the rotational degrees of freedom . periodic boundary conditions are an essential ingredient for the point - reflection moves employed , but application to layered systems ( i.e. , with only a periodicity in the @xmath193 and @xmath194 directions ) can be realized if the point reflection is performed in a horizontal plane and the relaxation along the @xmath195-coordinate is performed via local monte carlo moves . for strongly aspherical particles , the overlap threshold and hence the range of accessible volume fractions decreases significantly . in summary , the generalized gca offers a wide range of opportunities to efficiently simulate fluid systems that were hitherto inaccessible to computer simulations . in addition , it may well be possible to generalize this algorithm to other situations in which cluster algorithms are employed , such as quantum - mechanical systems . this material is based upon work supported by the national science foundation under career award no . dmr-0346914 and grant no . cts-0120978 and by the u.s . department of energy , division of materials sciences under award no . defg02 - 91er45439 , through the frederick seitz materials research laboratory at the university of illinois at urbana - champaign .

in recent years , scaling limits of nonlinear and/or singular functionals of stochastic lattice models have attracted a lot of attention . just to give a couple of examples , we mention the extensive studies of the kpz universality class ( see @xcite for a review ) and gaussian multiplicative chaos associated to liouiville quantum gravity ( see @xcite for a review ) . in @xcite the author has proposed a general framework ( the so - called theory of _ regularity structures _ ) in order to deal with ill - posed stochastic pde s on which the trouble comes from a nonlinear term ( like in the kpz or stochastic allen - cahn equations ) or from a singular linear term ( like in the parabolic anderson model ) . the theory of regularity structures allows to make sense of troublesome equations in a meaningful way . moreover , various scaling limits of stochastic lattice models on which these singular and/or nonlinear observables play an important role should be given in terms of solutions to these equations . however , aside from models on which a great deal of integrability is present ( the term _ stochastic integrability _ was coined in @xcite ) , the question of convergence of nonlinear fluctuations of stochastic lattice models is basically open ; see however @xcite , @xcite . one of the main ingredients of the theory of regularity structures consists in making sense _ a priori _ of enough nonlinear and/or singular observables of solutions of _ linear _ stochastic pde s . in the case of the kpz equation @xcite , one starts with the solution of the ornstein - uhlenbeck equation @xmath0 where @xmath1 is a space - time white noise and tries to make sense of various nonlinear functionals of it ( the twelve tree - labeled processes in @xcite ) . to avoid uncomfortable issues arising from the lack of compactness , @xcite restricts himself to the circle @xmath2 . the simplest of these processes corresponds to @xmath3 . since @xmath4 turns out to be a distribution , it is far from clear how to define @xmath3 . let us restrict ourselves to the stationary situation , on which for any fixed time @xmath5 , @xmath4 is a spatial white noise of variance @xmath6 . the simplest choice should be to take an approximation of the identity @xmath7 centered at @xmath8 , and to consider @xmath3 as the limit of @xmath9 as @xmath10 in some sense . it turns out that this plan can be formalized after a wick renormalization : when integrated in time and space against a smooth test function , @xmath11 has a meaningful limit as @xmath12 . of course , if one wants to use @xcite in order to obtain a unique solution of the kpz equation , there are still 11 processes to go , but on this paper we just focus on this one process ( we don _ not _ claim we can treat the other 11 processes ! ) . up to our knowledge , this _ squared field _ was first considered in @xcite , where the convergence of space - time fluctuations of two - point functions to the squared field of the ornstein - uhlenbeck process was obtained . a more general , different proof was implicitly obtained in @xcite and explicitly stated in @xcite , as a part of a program towards the derivation of the kpz equation from general stochastic lattice models . in this paper we propose a new , different approach in order to define the field @xmath3 . the _ quadratic field _ associated to @xmath4 is the two - dimensional process formally defined as @xmath13 . in order to define this object in a weak sense , some care needs to be taken at the diagonal ; the simplest choice is to take @xmath14 ( anyway @xmath15 is just a formal object ) . blindly applying it s formula we see that @xmath15 satisfies an equation of the form @xmath16 where @xmath17 is a distribution - valued martingale that can be computed in terms of the ornstein - uhlenbeck process @xmath4 and the noise @xmath18 . it turns out that this equation is well - posed and solutions of it can be constructed straightforwardly . let @xmath19 be a two - dimensional approximation of the identity . let @xmath20 be a mean - zero , regular function and consider the process @xmath21 it turns out that the process @xmath22 has a non - trivial limit as @xmath12 and this limit coincides with the squared process @xmath3 constructed in @xcite . the main difference is that now we obtain the squared process as a singular _ linear _ observable of the solution of a reasonably well - behaved stochastic pde . the main advantage of this representation is that solving a simple laplace problem , we can obtain various properties of the squared process in a more or less straightforward way . in order keep the paper at a reasonable length , we focus on two issues about the quadratic field @xmath15 and we keep the model as simple as possible . first we show that the quadratic fluctuations of the simple exclusion process converge to the quadratic field @xmath15 . it is well known that the simple exclusion process is integrable in the sense that various quantities of interest , among them @xmath23-point correlation functions , can be computed almost explicitly . we do _ not _ take full advantage of this feature . given the technical boltzmann - gibbs principle for granted , a careful reading of our proof shows that this convergence result can be extended for the speed - change exclusion processes considered in @xcite . this technical principle has been proved in @xcite . and then we obtain short - time and long - time properties of the process @xmath24 , using the construction outlined above . these properties have not been obtained before , and they are good examples of the advantages of our construction with respect to previous ones . this paper is organized as follows . in section [ s1 ] we define the exclusion process and we define the various fluctuation fields we want to study on this article . on the way , we provide various topological definitions which are needed to handle distribution - valued processes . we also state the main results of the article . in section [ s2 ] we show that the discrete quadratic fields form a relatively compact sequence of distributions with respect to the @xmath25-skorohod topology on the space of distribution - valued , _ cdlg _ paths . in section [ s3 ] we show that the discrete quadratic fields converge to the unique stationary solution of equation . the proof is divided in three parts . recall that limit points exists due to tightness . in section [ s3.1 ] we show that any limit point is continuous ; this will be important later when characterizing some martingales in terms of quadratic variations . in section [ s3.2 ] we show that various martingales associated to the discrete quadratic fields have limit points which are martingales . this shows that any limit point of the discrete quadratic fields satisfies a martingale formulation of . in section [ s3.3 ] we show that this martingale problem has a unique solution , which closes the proof of the convergence result . in section [ s4 ] let @xmath26 denote the discrete circle with @xmath23 points . let @xmath27 be the state space of a continuous - time markov chain which we describe as follows . we denote by @xmath28 the elements of @xmath29 and we call them _ configurations_. we consider the set @xmath30 as embedded in the continuous circle @xmath31 . we call the elements of @xmath30 _ sites_. we say that two sites @xmath32 are _ neighbors _ if @xmath33 . in this case we write @xmath34 . to each pair @xmath35 of neighbors we attach a poisson clock of rate @xmath36 . each poisson clock is independent of the other clocks . each time a poisson clock rings , we exchange the occupation numbers of the corresponding pair of neighbors . for each @xmath37 and each @xmath32 , let @xmath38 denote the configuration obtained from @xmath39 by exchanging the occupation numbers at @xmath40 and @xmath41 , that is , @xmath42 this informal description corresponds to the markov chain @xmath43 generated by the operator @xmath44 , given by @xmath45 for any @xmath46 , where @xmath47 is given by @xmath48 for any @xmath32 and any @xmath37 . the sum is over _ unordered pairs _ @xmath35 of neighbors in @xmath30 . we say that a site @xmath40 is _ occupied _ by a particle at time @xmath49 if @xmath50 . if @xmath51 , we say that the site @xmath40 is _ empty _ at time @xmath49 . with this convention about particles and empty sites ( or _ holes _ ) , the dynamics of @xmath43 has the following interpretation . each particle tries to jump to each of its two neighbours with exponential rate @xmath36 . at each attempt , it verifies whether the destination site is empty , on which case it jumps to it . otherwise the particle stays where it is . this particle interpretation gives the name _ simple exclusion process _ to the family of processes @xmath43 . notice that particles are neither created nor annihilated by this dynamics . by reversibility , it is easy to check that the uniform measures on the spaces @xmath52 are invariant with respect to the dynamics of @xmath53 for any @xmath54 . let us call @xmath55 these measures . checking the irreducibility of the sets @xmath56 with respect to the dynamics , it can be concluded that @xmath55 is actually ergodic under the evolution of @xmath43 for any @xmath57 . notice that the product measures @xmath58 given by @xmath59 are invariant and reversible under @xmath43 for any @xmath60 $ ] . however , these measures are not ergodic due to the conservation of the number of particles . in fact , these measures are obtained as proper convex combinations of the measures @xmath61 . from now on we start the process @xmath43 from the invariant measure @xmath62 for @xmath63 . to avoid uninteresting topological issues , we fix @xmath64 and we restrict the process @xmath53 to the interval @xmath65 $ ] . we denote by @xmath66 the distribution of @xmath67\}$ ] on the space @xmath68 ; \omega_n)$ ] of cdlg trajectories from @xmath65 $ ] to @xmath29 , and we denote by @xmath69 the expectation with respect to @xmath66 . the expectation with respect to @xmath55 will be denoted by @xmath70 , and the expectation with respect to @xmath58 will be denoted by @xmath71 . in order to define the fluctuation fields associated to the process @xmath67\}$ ] in a proper way , we need to introduce some notations and some topologies . for a given compact set @xmath72 with a differentiable structure , let @xmath73 denote the set of infinitely differentiable functions @xmath74 . we will only consider @xmath75 for @xmath76 , that is , the circle @xmath2 and the two - dimensional torus @xmath77 . the space @xmath73 is a polish space with respect to the topology generated by the metric @xmath78 given by @xmath79 here @xmath80 denotes the @xmath57-derivative of @xmath81 , which is a function in the case @xmath82 and an @xmath57-dimensional , symmetric tensor in the case @xmath83 . the space @xmath84 ( do not confuse with @xmath68 ; \omega_n)$ ] ) will denote the set of linear , continuous functions @xmath85 . in other words , @xmath84 is the topological dual of @xmath73 , which is known in the literature as the _ space of distributions _ in @xmath72 . the functions @xmath86 are called _ test functions_. we will define two distribution - valued processes associated to @xmath67\}$ ] ; one with values in @xmath87 and another one with values in @xmath87 . it turns out that the spaces @xmath87 , @xmath88 equipped with the weak topology are polish spaces . for an arbitrary polish space @xmath89 , we denote by @xmath68 ; { { \mathscr e}})$ ] the space of cdlg trajectories equipped with the @xmath25-skorohod topology . notice that with respect to the @xmath25-skorohod topology @xmath68 ; { { \mathscr e}})$ ] is also a polish space . for a given compact space @xmath72 and @xmath90 , we denote by @xmath91 the @xmath92 of dirac at @xmath40 , that is , the atomic probability measure supported at @xmath93 . let @xmath94\}$ ] denote the @xmath87-valued process given by @xmath95 for any @xmath96 and any @xmath97 $ ] . the process @xmath94\}$ ] is known in the literature as the _ density fluctuation field _ associated to the process @xmath67\}$ ] . this process has been extensively studied , and in particular a scaling limit for it is available . the process @xmath94\}$ ] converges in distribution with respect to the @xmath25-topology of @xmath68 ; { { \mathscr d}}({{\mathbb t}}))$ ] to the stationary solution of the infinite - dimensional ornstein - uhlenbeck equation @xmath98 where @xmath99\}$ ] is a space - time white noise . as far as we understand , this result was proved by the first time in the lecture notes by de masi , ianiro , pellegrinotti and presutti @xcite . let us introduce a second fluctuation field , this time with values in @xmath88 . let @xmath100\}$ ] be the process with values in @xmath88 defined as @xmath101 for any test function @xmath102 and any @xmath97 $ ] . without loss of generality , from now on and up to the end of the article we will assume that any test function @xmath102 is symmetric . in fact , for any antisymmetric function @xmath102 , @xmath103 . we call the process @xmath100\}$ ] the _ quadratic _ fluctuation field associated to the process @xmath67\}$ ] . our aim will be to obtain the scaling limit of the process @xmath100\}$ ] as @xmath104 . [ t1 ] let @xmath105\}$ ] be the martingale process defined as @xmath106 for any test function @xmath102 . the process @xmath100\}$ ] converges in distribution , as @xmath104 , to the stationary solution of the equation @xmath107 for a given function @xmath108 , it will be useful to introduce the notation @xmath109 ( without the index @xmath5 ) for the function from @xmath29 to @xmath110 given by @xmath111 let us write @xmath112 . in the same spirit , for a ( symmetric ) function @xmath113 and for @xmath37 , we define @xmath114 the main purpose of the introduction of the fluctuation field @xmath115\}$ ] , is the study of the _ quadratic field _ @xmath116\}$ ] formally defined as follows . for each function @xmath96 , we denote by @xmath117 the distribution in @xmath77 given by @xmath118 in other words , @xmath119 , where @xmath120 represents the uniform measure on the diagonal @xmath121 . then we formally define @xmath122 as @xmath123 of course , it is not clear at all whether this definition makes any sense . the simplest idea one can use is to approximate the singular object @xmath124 by a sequence of more regular functions . let us consider the approximation of the identity @xmath125 given by @xmath126 this approximation of the identity is not the smoothest one we can use , but it is very convenient computationally . the following theorem explains how to define the singular process @xmath116\}$ ] . [ t2 ] let @xmath127 be the approximation of the identity in @xmath77 defined above . let @xmath128\}$ ] be the @xmath87-valued process defined as @xmath129 for any @xmath96 and any @xmath97 $ ] . then , @xmath128\}$ ] converges in distribution with respect to the uniform topology to a well - defined , @xmath87-valued process @xmath116\}$ ] . this theorem was proved in @xcite , although the proof there is very different from the proof we will present here . moreover , the proof we present here has one important advantage : as we will see , it gives a more explicit construction of the process @xmath116\}$ ] , which allows to obtain various properties of it . for completeness , we present the following convergence result , obtained in @xcite : let @xmath130\}$ ] the @xmath87-valued process defined by @xmath131 for any @xmath96 . the process @xmath130\}$ ] converges in distribution as @xmath104 , to the process @xmath116\}$ ] defined above . notice that concatenating intervals of size @xmath132 , we can assume that @xmath133 is a well - defined process . as we mentioned in the introduction , our construction allows to obtain some properties of the field @xmath133 . the short - time properties of @xmath134 are given by the following theorem : [ t4.5 ] as @xmath12 , for each @xmath96 the field @xmath135 converges in the uniform topology to a stationary gaussian process @xmath136 of covariances @xmath137 = \tfrac{\kappa}{8}{\langle}f,-\delta f{\rangle}\big\ { t^{3/2}+s^{3/2}-|t - s|^{3/2}\big\},\ ] ] where @xmath138 notice that this result is very similar in spirit to theorem 2.5 of @xcite . as far as we understand , this result has not been predicted in the literature . for large times @xmath5 , we only know the limiting variance of @xmath122 : [ t2.6 ] there exists a compact operator @xmath139 such that @xmath140 = \tfrac{1}{4 } { \langle}f , -\tfrac{1}{\sqrt{2 \pi } } ( -\delta)^{\frac{1}{4 } } f + { { \mathscr k } } f{\rangle}.\ ] ] we will prove theorem [ t1 ] using the standard three - steps method to get convergence in distribution of stochastic processes , namely , we first prove tightness of the sequence of processes in a suitable topology , then we deduce some properties of the possible limit points using the approximating processes , and then we show that these aforementioned properties characterize the limit point in a unique way . in this section we perform the first step , that is , we prove tightness , and we prepare the ground for the other steps by introducing a bunch of martingales . we start recalling the following well - known fact about continuous - time markov chains . let @xmath141 \to { { \mathbb r}}$ ] be a smooth function on the time variable . then , the process @xmath142 is a positive martingale with unit expectation . we will apply this formula for the function @xmath143 , where @xmath144 and @xmath145 is defined above . for each @xmath144 and each @xmath146 , let @xmath147\}$ ] denote the martingale given by @xmath148 for any @xmath97 $ ] . another simple observation is that for any @xmath149 , the process @xmath150 is a martingale . let us compute @xmath151 in a more explicit way . notice that for any function @xmath46 , @xmath152 notice as well that @xmath153 this term will appear repeatedly in what follows , so we will give it a name . let @xmath154 denote the function @xmath155 if @xmath156 , we will write @xmath157 instead of @xmath158 . we will write @xmath159 , @xmath160 as well . with this notation , we can write in the more compact form @xmath161 when @xmath162 , the difference @xmath163 is of order @xmath164 . in fact , since @xmath146 , this difference is approximated by @xmath165 , with an error term of order @xmath166 . in particular , we have the _ a priori _ bound @xmath167 for any @xmath168 , any @xmath169 and any @xmath97 $ ] . since the measure @xmath170 is of product form , we have a better bound for the second moment of @xmath171 : @xmath172 \leq \frac{c_2(f)}{n}\ ] ] for any @xmath168 , any @xmath169 and any @xmath97 $ ] . we point out here that the _ a priori _ bound will be very useful , since the term @xmath173 will actually appear in a double exponential , and therefore moment bounds will not be as useful . going back to the computation of the martingale , we see that @xmath174 recall that taking derivatives of @xmath175 with respect to @xmath176 we can find other martingales associated to the process @xmath67\}$ ] . taking @xmath177 in , we see that the process @xmath178\}$ ] given by @xmath179 is a martingale . for @xmath102 , let us define @xmath180 as @xmath181 in other words , @xmath182 is a discrete version of the laplacian @xmath183 . for functions @xmath96 , we define @xmath184 which corresponds to a discrete version of the laplacian ( actually the second derivative ! ) @xmath185 . for functions @xmath81 defined only in @xmath30 , we adopt the same notation and definition for @xmath182 . after some computations , we see that @xmath186 where the function @xmath187 is defined as @xmath188 for any @xmath8 , that is , @xmath189 is the value of @xmath81 on the _ diagonal _ @xmath190 . notice that the first term on the right - hand side of can be written as @xmath191 . with respect to the measure @xmath170 , the expectation of the second term on the right - hand side of is equal to @xmath192 , since the function @xmath193 has mean @xmath192 with respect to the counting measure on @xmath30 . it is also uniformly bounded in @xmath23 and @xmath39 , thanks to the smoothness of @xmath81 . moreover , its variance with respect to @xmath170 vanishes as @xmath104 . in other words , only the first term on the right - hand side of will be relevant when considering scaling limits . the term @xmath194 will appear repeatedly in what follows , so it is convenient to give it a name . let us define @xmath195 and notice that @xmath196 for any @xmath168 , any @xmath169 and any @xmath37 . taking @xmath197 in , we see that the process @xmath198 is also a martingale . in other words , quadratic variation _ of the martingale @xmath178\}$ ] is given by @xmath199 notice that the moment bound implies a bound of the form @xmath200 for the quadratic variation process , and therefore a moment bound of the form @xmath201 \leq c(f ) t$ ] for the martingale process @xmath178\}$ ] . this observation will be relevant when showing tightness of the processes @xmath100\}$ ] . in this section we prove tightness of the sequence of processes @xmath202\}_{n \in { { \mathbb n}}}$ ] . the first step is to reduce the problem from distribution - valued processes to real - valued processes . this is done throught the so - called _ criterion_. the sequence of @xmath88-valued processes @xmath202\}_{n \in { { \mathbb n}}}$ ] is tight in @xmath68 ; { { \mathscr d}}({{\mathbb t}}^2))$ ] if and only if the sequence @xmath203\}_{n \in { { \mathbb n}}}$ ] is tight in @xmath68 ; { { \mathbb r}})$ ] for any @xmath102 . moreover , if any limit point of @xmath203\}_{n \in { { \mathbb n}}}$ ] is supported on continuous , real - valued trajectories for any @xmath102 , then any limit point of @xmath100\}_{n \in { { \mathbb n}}}$ ] is supported on continuous trajectories on @xmath88 . applying this criterion , we see that it is enough to prove tightness for the process @xmath204 for each @xmath102 . the decomposition is also a declaration of intentions : we will not prove tightness directly for the process @xmath203\}$ ] , but for each one of the process appearing on the right - hand side of . the simplest one is the sequence @xmath205 of real - valued random variables . in fact , taking characteristic functions , it is easy to see that @xmath206 converges to a gaussian random variable of mean @xmath192 and variance @xmath207 . notice that this is also true for the sequence @xmath208 for any time @xmath97 $ ] . since any convergent sequence is tight , we are done with this term . next in line is the integral term @xmath209 there is a very simple criterion that applies in this situation ( see theorem 2.3 of @xcite for example ) . [ p2.2 ] the sequence of processes @xmath210\}_{n \in { { \mathbb n}}}$ ] is tight with respect to the uniform topology of @xmath211 ; { { \mathbb r}})$ ] if @xmath212 < \infty.\ ] ] by the stationarity of the process @xmath67\}$ ] , it is enough to get a bound for @xmath213.\ ] ] recall decomposition and the comments thereafter . since @xmath81 is smooth , we see that there exists a constant @xmath214 which does not depend on @xmath23 such that @xmath215 \leq c(f)\ ] ] for any @xmath168 . in the other hand , @xmath216 \leq \frac{1}{n^2 } \sum_{x , z \in { { \mathbb t}}_n } \big ( \delta_n f(x , z)\big)^2,\ ] ] and due to the smoothness of @xmath81 , the right - hand side of this inequality can also be bounded by a ( maybe different ) constant @xmath214 which does not depend on @xmath23 . we conclude that the sequence of processes @xmath210\}_{n \in { { \mathbb n}}}$ ] is tight with respect to the uniform topology of @xmath211 ; { { \mathbb r}})$ ] . now it is the turn of the martingale term @xmath178\}$ ] . we use the following well - known criterion . the sequence of real - valued processes @xmath178\}_{n \in { { \mathbb n}}}$ ] is tight in @xmath68 ; { { \mathbb r}})$ ] if @xmath217 dt < \infty.\ ] ] let us show that this supremum is finite . first we recall a simple version of burkholder s inequality . for any @xmath218 there exists a universal constant @xmath219 such that @xmath220 \leq \kappa_p \big(\sum_{z \in a } f_z^2\big)^p\ ] ] for any @xmath221 and any @xmath222 . notice that the inequality above is actually an identity when @xmath223 and @xmath224 . by stationarity , we just need to show that @xmath225 < \infty.\ ] ] since there are no cancellations between the @xmath157 terms , using the crude estimate @xmath226 should not be really bad . therefore , we need to show that @xmath227 < \infty.\ ] ] bounding above by @xmath228 the term @xmath229 in @xmath230 , we are exactly on the setup of burkholder s inequality for @xmath231 . therefore , we have the bound @xmath232 \leq \kappa_2 n \sum_{x \in { { \mathbb t}}_n } \big(\sum_{z \in { { \mathbb t}}_n } \big(f(z , x+\tfrac{1}{n})-f(z , x)\big)^2\big)^2.\ ] ] due to the smoothness of @xmath81 , the right - hand side of this inequality converges , as @xmath104 , to @xmath233 which shows that the supremum in is finite . we conclude that the martingale sequence @xmath178\}_{n \in { { \mathbb n}}}$ ] is tight in @xmath68 ; { { \mathbb r}})$ ] . recapitulating what we have accomplished up to here in this section , we have proved tightness of @xmath100\}_{n \in { { \mathbb n}}}$ ] by combining mitoma s criterion with the tightness of the processes appearing in the decomposition . here we just remark that we do not know whether the limit points of @xmath178\}_{n \in { { \mathbb n}}}$ ] are continuous , and in particular , we do not know whether the limit points of @xmath100\}$ ] are concentrated on continuous , @xmath88-valued trajectories . we will address this question in the following section , when discussing the characterization of such limit points . in the previous section , we have showed tightness of the sequence @xmath100\}_{n \in { { \mathbb n}}}$ ] . let @xmath115\}$ ] be one of such limit points . for ease of notation , the superscript @xmath23 will denote along this section , the subsequence for which the process @xmath100\}$ ] converges to @xmath115\}$ ] . without loss of generality , we can assume that the process @xmath178\}$ ] converges to a process @xmath234\}$ ] , along the subsequence @xmath23 , for any @xmath102 . interpreting @xmath235\}$ ] as an integral process , we see that the estimates used to prove tightness of @xmath178\}$ ] are exactly the ones needed to apply the tightness criterion [ p2.2 ] to @xmath235\}$ ] . we leave the details to the reader . taking a further sub - subsequence if necessary , we can assume that along @xmath23 , the process @xmath235\}$ ] converges in distribution to some process @xmath236\}$ ] without further comments , taking sub - subsequences if necessary , we will assume that any process we need to define is convergent along the subsequence @xmath23 . notice that based solely on convergence in distribution , we can not argue that @xmath236\}$ ] is the quadratic variation of @xmath234\}$ ] . in fact , we do not even know whether @xmath234\}$ ] is a martingale . in order to address the questions raised in the previous section , it will be more convenient to work with the exponential martingales @xmath237\}$ ] . for the moment we do not know whether these martingales are tight , and therefore we can not say anything about convergence . the following simple taylor estimate will be very useful . for any @xmath238 and any @xmath149 , @xmath239 fix @xmath102 and @xmath144 . we will use this estimate for @xmath240 and @xmath241 . recall that @xmath242 . we have that @xmath243 therefore , we can write @xmath244 where the error term @xmath245 satisfies @xmath246 and in particular this error term goes to @xmath192 uniformly in @xmath23 . notice that @xmath247 @xmath248 \leq \frac{c_1(f ) c_2(f)}{n}\ ] ] and we see that the cubic term in is uniformly bounded in @xmath23 and it converges to @xmath192 in @xmath66-probability as @xmath104 . the uniform bound is relevant , because this cubic term is on top of an exponential in . looking into , we have just showed that along the subsequence @xmath23 , for any fixed @xmath97 $ ] and any @xmath146 , the _ random variable _ @xmath249 converges in distribution to @xmath250 notice that although we have not proved that the convergence holds at the level of processes , we do know that @xmath251\}$ ] is a well - defined process with trajectories in @xmath68 ; { { \mathbb r}})$ ] . in order to prove that @xmath251\}$ ] is a martingale , it is enough to show that the sequence @xmath252 is uniformly integrable . it is here where we will take full advantage of the decomposition . in fact , the error term and the cubic term in are uniformly bounded in @xmath23 by a deterministic constant . therefore , we can neglect them . moreover , the term @xmath253 is non - negative and it appears with a minus sign . therefore , we can also neglect it . we are left to prove the uniform integrability of the sequence @xmath254 . the simplest criterion for uniform integrability is a uniform @xmath255-bound for some @xmath256 . in other words , we want to estimate @xmath257 $ ] for some @xmath256 . recall that @xmath258=1 $ ] . we will use the elementary estimate @xmath259 \leq e[e^{2(x - y)}]^{1/2}e[e^{2y}]^{1/2}\ ] ] for @xmath260 and @xmath261 with these choices , we get the bound @xmath262 \leq { { \mathbb e}}_n[\exp\ { 2 y\}]^{\frac{1}{2}}.\ ] ] we can again neglect the cubic term and the error term in , since they are uniformly bounded in @xmath23 by a deterministic constant . therefore , uniform integrability of @xmath263 will be proved if we can show that @xmath264 < \infty.\ ] ] recall the definition of @xmath253 and rewrite it as @xmath265 by the convexity of the exponential function , @xmath266 \leq \frac{1}{t } \int_0^t { { \mathbb e}}_n\big[\exp\{2p^2\theta^2 t \sum_{x \in { { \mathbb t}}_n } \xi^n_t(f;x)^2 \}\big ] dt.\ ] ] therefore , we just need to show that @xmath267 < \infty.\ ] ] at this point , we need something stronger than burkholder s inequality in order to bound this expectation . we will use _ hoeffding s inequality_. there exist constants @xmath268 , @xmath269 such that for any @xmath168 , for any @xmath221 and for any @xmath222 , @xmath270 let us go back to . as we pointed out before , there are no cancellations we can take advantage when adding up the terms @xmath230 . therefore , we use the crude exponential hlder estimate @xmath271 \leq \prod_i e[\exp\{nx_i\}]^{1/n}$ ] to get the bound @xmath272 \leq \prod_{x \in { { \mathbb t}}_n } e_{\frac{1}{2 } } \big [ \exp\{2p^2\theta^2 t n \xi^n(f;x)^2\}\big]^{\frac{1}{n}}.\ ] ] to simplify the notation , let us write @xmath273 . bounding @xmath229 by one , we get the estimate @xmath274 \leq e_{\frac{1}{2 } } \big [ \exp\big\{\beta n \big(\sum_{z \in a_x } \big(\eta(z)-\tfrac{1}{2}\big)g(z , x)\big)^2\big\}\big],\ ] ] where @xmath275 and @xmath276 . we are almost at the setting of hoeffding s inequality . we just need the following simple observation . for any non - negative random variable @xmath277 and any regular function @xmath81 , @xmath278 = -f(0)+ \int_0^\infty f'(t ) p(x \geq t ) dt.\ ] ] therefore , the right - hand side of is bounded above by @xmath279 finally we are in position to use hoeffding s inequality . taking @xmath280 in , the integral above is bounded by @xmath281 by the smoothness of @xmath81 , there is a constant @xmath282 such that @xmath283 for any @xmath168 and any @xmath169 . therefore , for any @xmath284 , @xmath285 and we conclude that @xmath272 \leq c_h\big\{\frac{c_h}{c_h-2p^2\theta^2 t c_2(f)}-1\big\}\ ] ] for any @xmath144 small enough . we conclude that the sequence @xmath263 is uniformly integrable for any @xmath144 satisfying @xmath286 we summarize what we have shown above in the following theorem . [ t3.3 ] for any function @xmath146 , there is a ( strictly positive ) constant @xmath287 such that the process @xmath288\}$ ] given by @xmath289 is a martingale for any @xmath144 such that @xmath290 . notice that this theorem implies that the processes @xmath291\big\}\ ] ] are martingales for any @xmath292 . in particular , considering the cases @xmath76 we get the following result . [ c3.4 ] for any function @xmath146 , the process @xmath234\}$ ] is a martingale of quadratic variation @xmath236\}$ ] . considering the case @xmath241 , it is possible to obtain the following bound : @xmath293 \leq 125 { { \mathbb e}}_n\big[{\langle}{{\mathscr w}}_t(f){\rangle}^2\big].\ ] ] using a crude cauchy - schwarz estimate to bound @xmath294 $ ] , we get a bound of the form @xmath295 \leq c(f ) t^2 $ ] for some constant depending on @xmath81 . by stationary and the classical kolmogorov - centsov continuity criterion , we conclude that the process @xmath234\}$ ] has continuous trajectories , which answers the question raised in the previous section about the continuity of @xmath115\}$ ] . we state this as a corollary . [ c3.5 ] any limit point of the sequence of @xmath88-valued processes @xmath100\}_{n \in { { \mathbb n}}}$ ] is concentrated on continuous , @xmath88-valued tractories . our next goal will be to characterize the martingale @xmath234\}$ ] and its quadratic variation in terms of the processes @xmath115\}$ ] and @xmath296\}$ ] . recall the definition of the process @xmath296\}$ ] as the scaling limit of the density fluctuation fields @xmath94\}_{n \in { { \mathbb n}}}$ ] . by , @xmath297 the second sum on the right - hand side of this identity converges to @xmath192 in @xmath298 . the first sum is equal to @xmath299 plus an error term which also goes to @xmath192 in @xmath298 . taking the limit in through the subsequence @xmath23 , we get the identity @xmath300 valid for any @xmath146 and any @xmath97 $ ] . for @xmath8 , @xmath168 and @xmath146 , let @xmath301 be defined as @xmath302 notice that @xmath303 is a discrete approximation of the partial derivative @xmath304 . the function @xmath303 provides a very compact way of identify @xmath171 in terms of @xmath305 . in fact , looking back to definition , we see that @xmath306 in particular , there exists a constant @xmath307 such that @xmath308 for any @xmath168 , any @xmath169 and any @xmath97 $ ] . using the simple identity @xmath309 , we see that @xmath310 we conclude that @xmath311 in the sense that this difference is uniformly bounded by a deterministic sequence which goes to @xmath192 as @xmath104 . now we explain how to change @xmath312 by @xmath313 in the sum above . notice that @xmath314 & \leq { { \mathbb e}}_n\big[\big({{\mathscr y}}_t^n(f_1)-{{\mathscr y}}_t^n(f_2)\big)^2\big]^{\frac{1}{2 } } \times\\ & \quad \times \big(2\big\{{{\mathbb e}}_n\big[{{\mathscr y}}_t^n(f_1)^2\big]+{{\mathbb e}}_n\big[{{\mathscr y}}_t^n(f_2)^2\big]\big)^{\frac{1}{2}}.\\ \end{split}\ ] ] using the product structure of @xmath170 , the right - hand side of this inequality is equal to @xmath315 therefore , the application @xmath316 from @xmath317 to @xmath318 is uniformly continuous in @xmath23 and @xmath5 . let @xmath319 be defined as @xmath320 . we see that @xmath303 converges uniformly to @xmath321 , the convergence being uniform in @xmath23 and @xmath40 . we conclude that @xmath322 in @xmath318 . now we need to get rid of the term @xmath323 , which is not a function of @xmath324 . the idea is to take advantage of the continuity of @xmath325 with respect to @xmath40 , in order to introduce an average of @xmath326 over some finite interval . notice that for any @xmath327 , @xmath328 using this limit in conjunction with , we see that for any @xmath149 , @xmath329 in @xmath318 . passing the sum over @xmath330 to the @xmath326-term , we see that @xmath331 in @xmath318 . the expectation of @xmath229 is equal to @xmath332 . moreover , @xmath326 and @xmath333 are independent under @xmath170 as soon as @xmath334 . if we split the sum @xmath335 into two sums , one running over even values of @xmath330 and another running over odd values of @xmath330 , we can use the elementary inequality @xmath336 plus burkholder s inequality on each sum to get the bound @xmath337 \leq \frac{c_p}{\ell^p}\ ] ] for any @xmath218 and any @xmath149 , where @xmath338 is a constant that depends only on @xmath339 . in particular , using hlder s inequality ( to split the product ) and burkholder s inequality ( for @xmath340 ) , we get the bound @xmath341 \leq \frac{c_3(f)}{\sqrt \ell}\ ] ] for some constant @xmath307 which does not depend on @xmath342 or @xmath343 $ ] . we conclude that @xmath344 in @xmath318 . in particular , @xmath345 in @xmath318 . both terms in this limit are convergent in distribution when @xmath104 . we summarize what we have proved up to here in the following theorem . [ t3.5 ] for any @xmath146 and any @xmath97 $ ] , @xmath346 where @xmath319 is defined as @xmath320 . recall that by definition @xmath347 , where @xmath348 is the symmetric part of @xmath81 . the expression for @xmath349 does not look very symmetric in the coordinates @xmath350 . notice that for any smooth , symmetric function @xmath351 , @xmath352 this relation allows us to write @xmath349 in the more symmetric way @xmath353 putting theorems [ t3.3 ] ( actually corollaries [ c3.4 ] and [ c3.5 ] ) and [ t3.5 ] together , we see that the process @xmath115\}$ ] satisfies the following martingale problem . * for any function @xmath102 , the process @xmath354 is a continuous martingale of quadratic variation @xmath355 very formally speaking , this is the martingale problem associated to the spde @xmath356 where @xmath357\}$ ] is a space - time white noise . the problem comes from the fact that this martingale problem does not tell us anything about the relation between the processes @xmath296\}$ ] and @xmath358\}$ ] . we would like to say that @xmath357\}$ ] is the driving process of @xmath296\}$ ] , that is , the white noise appearing in . let us consider a test function of the form @xmath359 for some functions @xmath360 . then , by we have @xmath361 in the other hand , by definition of @xmath362\}$ ] , we see that @xmath363 taking the limit along the subsequence @xmath23 , we conclude that @xmath364 using it s formula , we can obtain the martingale decomposition of the process @xmath365\}$ ] . in particular , @xmath366 let us rewrite this identity in terms of the process @xmath115\}$ ] . first notice that @xmath367 . then notice that the term @xmath368 is exactly the normalization term missing in the integral involving the laplacians . we conclude that @xmath369 for any function of the form @xmath359 . therefore , at least for functions of this product form , we are able to identify @xmath370 with @xmath371 . actually , since the set of linear combinations of functions of the form @xmath372 is dense in @xmath373 , this relation allows us to identify the martingale appearing in * ( mp ) * in terms of @xmath371 . let us explain this in a more rigorous way . let us define the @xmath87-valued process @xmath374\}$ ] as @xmath375 for any @xmath96 . by the definition of @xmath296\}$ ] , the process @xmath376\}$ ] is a continuous martingale of quadratic variation @xmath377 . the relation can be rewritten as @xmath378 notice that a by - product of our proof of tightness for @xmath94\}_{n \in { { \mathbb n}}}$ ] is a proof of tightness for @xmath379\}$ ] as a @xmath88-valued martingale . therefore , @xmath105\}$ ] is a well - defined , @xmath88-valued martingale process . in particular , its distribution is determined by the values of @xmath234\}$ ] for @xmath81 of the form @xmath380 . in there is no mention to the process @xmath115\}$ ] . we are finally ready to establish a uniqueness result for the martingale problem * ( mp)*. we state it as a theorem . [ t3.8 ] let @xmath381\}$ ] be a @xmath382-valued , continuous processes . assume that * for any @xmath96 , the process @xmath376\}$ ] is a continuous martingale of quadratic variation @xmath383 , * for any @xmath96 , the process @xmath384\}$ ] satisfies the relation @xmath385 * there exists a @xmath88-valued process @xmath105\}$ ] such that for any @xmath386 , @xmath387 * for any @xmath102 , we have @xmath388 * for any @xmath102 and any @xmath97 $ ] , the real - valued random variable @xmath389 has a gaussian distribution of mean zero and variance @xmath390 . then the distribution of @xmath381\}$ ] is uniquely determined . the proof was basically accomplished above . notice that the tightness result shows existence of the triple @xmath391\}$ ] . condition e ) implies _ stationarity _ of the process @xmath115\}$ ] , and could be relaxed to some moment bound , but since we are only interested on the stationary case , we will not discuss these generalizations here . using this stationarity condition , it is not hard to show that for any @xmath97 $ ] and any smooth path @xmath392 \to { { \mathscr c}}^\infty({{\mathbb t}}^2)$ ] , @xmath393 this relation allows us to use duhamel s formula to compute @xmath389 in terms of the process @xmath105\}$ ] . fix @xmath102 and @xmath97 $ ] , and consider the test function @xmath394 , where @xmath395\}$ ] is the semigroup generated by @xmath396 . then , @xmath397 and @xmath398 therefore , given the process @xmath105\}$ ] and the initial distribution of @xmath399 , the process @xmath115\}$ ] is uniquely determined by this duhamel s formula , which shows uniqueness in distribution . we finish this section with the proof of theorem [ t1 ] . in section [ s2 ] we showed tightness of the sequence of processes @xmath100\}_{n \in { { \mathbb n}}}$ ] . then in this section we first showed that any limit point @xmath115\}$ ] of this sequence satisfies the martingale problem stated in theorem [ t3.8 ] , and then we proved that this martingale problem has a unique solution in distribution . therefore , we conclude that the limit point of @xmath100\}$ ] is unique , thus proving actual sequential convergence to that unique limit point of the whole sequence . in this section we prove theorems [ t2 ] , [ t4.5 ] and [ t2.6 ] . the idea is to use the martingale characterization of the process @xmath115\}$ ] obtained in theorem [ t3.8 ] , and more precisely the martingale decomposition in order to rewrite the integral process @xmath400 as a martingale plus a function of @xmath15 and @xmath399 . for a given function @xmath401 , let @xmath402 denote the solution of the poisson equation @xmath403 if @xmath404 , then the solution @xmath405 of this equation belongs to @xmath373 , and therefore we can use it as a test function . then , by we have that @xmath406 now we can explain better what is the strategy of proof of theorem [ t2 ] . for each fixed time @xmath5 , @xmath15 is a white noise of variance @xmath407 . therefore , @xmath408 = \smash{\frac{1}{4 } } \|\psi_g\|^2 $ ] for any @xmath97 $ ] . the variance of the martingale @xmath17 is equal to @xmath409 . therefore , is stable under approximations in @xmath410 . we will use the following lemma . [ l4.1 ] let @xmath411 be the solution of the poisson equation @xmath412 satisfying @xmath413 . then , there exists a constant @xmath414 such that @xmath415 for any @xmath416 and any @xmath96 . we will obtain an explicit expression for the fourier series of the solution of . for @xmath417 , let @xmath418 be the trigonometric polynomial given by @xmath419 . then @xmath420 is an orthonormal basis of @xmath421 . for any function @xmath422 , we denote by @xmath423 the fourier series of @xmath424 : @xmath425 for any @xmath417 . then , @xmath426 . notice that the fourier series of @xmath427 is equal to @xmath428 . we conclude that @xmath429 the fourier series @xmath430 can be computed explicitly . in fact , @xmath431 where we use the convention @xmath432 for @xmath433 . by parseval s identity , for any function @xmath434 , @xmath435 we claim that there exists a universal constant @xmath436 such that @xmath437 for any @xmath438 . in fact , it is enough to notice that the gradient of the function @xmath439 is bounded in @xmath440 . since the function @xmath441 is bounded between @xmath442 and @xmath228 , we also have the trivial bound @xmath443 which is actually better than if @xmath444 . notice that @xmath445 we will split this sum into a sum over two sets , depending whether or is better . we start with the case on which the trivial bound is better . notice that the set @xmath446 is contained on the set @xmath447 let us define the new coordinates @xmath448 , @xmath449 . in these new coordinates , the set @xmath450 is given by for the image of the set @xmath450 under this change of coordinates . ] @xmath451 using the estimate @xmath452 , we have that @xmath453 we conclude that there is a constant @xmath454 such that for any @xmath455 small enough , @xmath456 now we estimate the sum in over the set @xmath457 . using , we see that @xmath458 putting the two pieces together , we end the proof of the lemma . at first glance , it seems that solving the poisson equation with @xmath459 instead of @xmath81 is not needed in order to prove this lemma , since the final bound depends only on @xmath459 and not on @xmath81 . actually , what we need is the integral condition @xmath460 , which is equivalent to @xmath459 being the actual derivative of a function @xmath81 . this condition means that the @xmath192-th order of the fourier s expansion of @xmath459 vanishes , allowing to divide by @xmath461 without worrying about the case @xmath462 , @xmath463 in various steps of the computations . what lemma [ l4.1 ] is saying , is that for any @xmath97 $ ] and any @xmath96 , the sequence @xmath464 is cauchy in @xmath465 , and therefore the limit @xmath466 exists in @xmath465 . this does not show the existence of the @xmath88-valued process @xmath116\}$ ] , nor the convergence of @xmath467\}$ ] to it . in order to obtain this convergence , we need a more refined argument . notice that @xmath468 therefore , for @xmath469 we have @xmath470 & \leq 2 e\big [ \big(q_t(\psi^\varepsilon_f-\psi^\delta_f)- q_0(\psi^\varepsilon_f-\psi^\delta_f)\big)^2\big]\\ & \quad + 2 e [ { { \mathscr w}}_t(\psi^\varepsilon_f-\psi^\delta_f)^2 ] . \end{split}\ ] ] the second expectation on the right - hand side is equal to @xmath471 , which is bounded by @xmath472 . the first expectation is computed in the following lemma : [ l4.2 ] for any @xmath473 , @xmath474 = \tfrac{1}{2 } { \langle}g , g - p_tg{\rangle}.\ ] ] let us assume that @xmath401 . by duhamel s formula , we have that @xmath475 & = e[q_0(p_t g - g)^2 ] + e\big [ \big ( \int_0^t d { { \mathscr w}}_s(p_{t - s}g)\big)^2\big]\\ & = \tfrac{1}{4 } \|p_tg - g\|^2 + \tfrac{1}{2 } \int_0^t \|\nabla(p_s g)\|^2 . \end{split}\ ] ] since @xmath476 , the lemma follows for @xmath401 . since @xmath15 and @xmath477 are continuous under approximations in @xmath421 , the lemma follows for @xmath473 by approximations . notice that @xmath478 and in particular @xmath479 . this shows that both expectations in are of the same order . therefore , we have the bound @xmath480 \leq ct \varepsilon \|f'\|^2,\ ] ] valid for any @xmath96 , any @xmath97 $ ] and any @xmath481 . using the representation @xmath482 we get the simple bound @xmath483 \leq \tfrac{1}{4 } t^2\|(f ' \otimes \delta ) \ast \iota_\varepsilon\|^2 \leq \frac{c t^2}{\varepsilon } \|f'\|^2.\ ] ] taking @xmath484 , choosing @xmath485 and combining with , we obtain the bound @xmath486 \leq c t^{3/2 } \|f'\|^2.\ ] ] these estimates are all we need in order to show that @xmath116\}$ ] is a well - defined , @xmath88-valued process and to show that @xmath128\}$ ] converges to @xmath116\}$ ] at the level of processes , see theorem 2.1 of @xcite for the details . let us denote by @xmath487 the solution of the poisson equation @xmath488 [ ec4.27 ] satisfying @xmath489 . by lemma [ l4.1 ] , @xmath490 . the various approximation results shown in the previous section allow to obtain the following representation formula for @xmath122 : @xmath491 for any @xmath97 $ ] . using this formula as a starting point , we can analyze the behavior of @xmath122 for small times @xmath5 . the variance of the first term on the right - hand side of is equal to @xmath492 , while the variance of the second term on the right - hand side of is equal to @xmath493 if the norm @xmath494 decays to @xmath192 like a power law , then the martingale term in turns out to be of lower order than the term @xmath495 . in fact , we have the following [ l4.4 ] for any smooth function @xmath20 we have that @xmath496 where @xmath138 notice that for any @xmath497 regular enough ( in fact , anything at least as regular as a measure on @xmath77 works ) , @xmath498 . therefore , by parseval identity we have that @xmath499 let us perform the change of variables @xmath500 , @xmath501 . notice that @xmath502 . we have that @xmath503 where the sum is over @xmath504 such that @xmath505 is even . let us define @xmath506 as @xmath507 rearranging terms in a convenient way , we can rewrite as @xmath508 therefore , it is enough to find the limit of the quantity @xmath509 notice that the sum between parentheses is a riemann sum for the integral @xmath510 the factor @xmath511 in front of it coming from the fact that the sum is over one half of the integers . we can check that there exists a finite constant @xmath512 such that @xmath513 for any @xmath514 . therefore , @xmath515 uniformly in @xmath40 . in particular , we have the diagonal limit @xmath516 for any @xmath517 . using the bound @xmath518 we can exchange the limit and the summation to obtain that @xmath519 which proves the lemma . with lemma [ l4.4 ] in hands , we can prove theorem [ t4.5 ] : recall the discussion after . since @xmath520 decays to @xmath192 as @xmath521 , the second term on the right - hand side of is of smaller order than the first term . in other words , @xmath522 whenever one of the limits exists . the process @xmath399 is gaussian and centered , and therefore it is easy to check its convergence . in fact , this is exactly what lemma [ l4.4 ] accomplishes . we conclude that @xmath523 converges in distribution as @xmath12 to a gaussian random variable of mean zero and variance @xmath524 . since the process @xmath525 has stationary increments , we have proved convergence in the sense of finite - dimensional distributions . in order to get convergence at the level of processes , we only need to show tightness of the sequence @xmath526 . but this follows from the comment after . notice that the process @xmath527 is a _ fractional brownian motion _ of hurst exponent @xmath528 . this is exactly the same process arising as the scaling limit of occupation times of the exclusion process @xcite . this fact is a quantitative version of the formal claim which states that fluctuations of additive functionals of local functions of degree 1 ( like the occupation time ) are of the same nature as fluctuations of additive functionals of extensive functions of degree 2 ( like the quadratic field ) . in the previous section we have seen how we can extract non - trivial information about the quadratic field @xmath133 for small times using the representation . now we will see what can we say about the variance of @xmath529 when @xmath5 is large . actually , it will be easier to work with the representation @xmath530 this relation can be obtained by passing to the limit in . the variance of @xmath531 ( and also of @xmath532 ) is equal to @xmath533 , while the variance of @xmath534 is equal to @xmath535 . therefore , we see that @xmath536}{t } = \tfrac{1}{2 } \|\nabla \psi_f \|^2.\ ] ] the following lemma tells us the behavior of @xmath537 : [ l4.6 ] there exists a function @xmath538 such that @xmath539 and such that @xmath540 notice that @xmath541 using the change of variables @xmath500 , @xmath542 and rearranging terms , we can rewrite this expression as @xmath543 the second sum is a riemann sum for @xmath544 . therefore , it is enough to choose @xmath545 the fact that @xmath546 allows us to identify the limiting variance of @xmath122 as the energy associated to a sort of _ fractional laplacian operator _ of order @xmath547 . in a more precise way , we can write @xmath548 where @xmath72 is a continuous operator in @xmath549 . the appearance of this compact operator is just a finite - volume effect . this work was written during a visit of the author to universit aix - marseille . the author would like to thank the warm hospitality at the institut de mathmati - ques de marseille and to etienne pardoux for valuable comments and for helping with the proof of theorem [ t3.8 ] . a. de masi , n. ianiro , a. pellegrinotti , and e. presutti . a survey of the hydrodynamical behavior of many - particle systems . in _ nonequilibrium phenomena , ii _ , stud . statist . mech . , xi , pages 123294 . north - holland , amsterdam , 1984 .

stellar mass loss is of critical importance to the later stages of stellar evolution . stars evolving up the red giant branch ( rgb ) lose mass via stellar winds . the remaining mass of a star leaving the rgb tip is the primary determinant of its position on the horizontal branch ( hb ; e.g. @xcite ) . subsequent mass loss on the asymptotic giant branch ( agb ) ejects the star s entire hydrogen envelope , creating a post - agb star and ( perhaps ) a planetary nebula ( pn ) . the timescale and end - point of agb evolution may therefore not be determined by the nuclear burning rate , but by mass loss @xcite . winds from these stars are the most significant producers of interstellar dust and therefore provided the heavy elements present in population i stars , including the sun and solar system @xcite . debate exists around many finer points of the mass loss process , particularly its variation with fundamental parameters such as luminosity , temperature and metallicity . firstly , the amount of mass loss that occurs on the rgb remains largely unmeasured . rgb mass loss appears to determine the ` second parameter ' ( after metallicity ) that defines the morphology of hbs in globular clusters ( gcs ) @xcite . secondly , it is not proven where dust formation stars , and whether rgb stars can form significant amounts of dust in their ejecta . if many rgb stars do produce dust , it would add a large number of dust factories ejecting dust into a galaxy s interstellar medium ( ism ) ( cf . the missing dust source in many galaxies ; e.g. @xcite ) . @xcite suggest that dust production occurs episodically over a large portion of the rgb , while @xcite have presented evidence that dust production is confined to the rgb tip , though gaseous mass loss is still present all along both the rgb and agb @xcite . dust can also help drive a wind through radiation pressure on grains ( e.g. @xcite ) , though it is not certain whether winds which lack high - opacity amorphous carbon can be accelerated in this manner @xcite . dustless stars typically lose mass via magneto - acoustic processes in their chromospheres @xcite . finally , the composition of circumstellar dust and the chemical processes by which it forms are only partly defined . of particular interest is the rle that metallic iron grains play in dust formation , and how metallicity affects dust production and composition @xcite . in this study , we aim to address these points . to do this , we explore dust production in the galactic globular cluster 47 tucanae ( ngc 104 ) . globular clusters are unique laboratories with which we can study late - stage evolution of low - mass stars : globular clusters contain stars that are among the oldest in the universe , and most clusters are comprised of a single ( or at least dominating ) stellar population with little internal variance in age or chemical composition . comparisons between clusters therefore probe differences in mass loss created by metallicity , while comparisons within clusters probe differences created by stellar temperature and luminosity . throughout this paper , we will refer to similar studies : @xcite , which covers the cluster @xmath0 centauri ( ngc 5139 ; hereafter paper i ) ; @xcite , which covers the cluster ngc 362 ( hereafter paper ii ) ; and a work which accompanies this paper ( submitted apjs ; hereafter paper iii ) . the cluster 47 tuc itself is among the galaxy s most massive and , co - incidentally , one of the closest to us . its distance has been estimated photometrically as 41904700 pc @xcite and kinematically as 4000 pc @xcite , with 4500 pc being the median distance determination in the literature . its mass is 69 @xmath1 m@xmath2 @xcite . its metallicity is determined to be roughly [ fe / h ] = 0.7 , though various studies have placed it anywhere from [ fe / h ] = 0.89 to 0.66 @xcite . the cluster suffers from little interstellar reddening ( @xmath3 mag @xcite ) and is estimated to be over 11 gyr in age @xcite . the cluster hosts an ionised intra - cluster medium ( icm ; @xcite ) . this ionised medium is undetected in any other cluster ( m15 has a neutral medium with a dusty component ; @xcite ) . integrating the observed intracluster electron density over the cluster core yields @xmath40.1 m@xmath2 of h ii , representing @xmath55% of the intracluster hydrogen @xcite . the total stellar mass loss from 47 tuc s stars therefore gives us insight into the fate of gas and dust expelled by stars in this cluster . paper iii more - precisely examines which stars have mid - infrared excess , indicative of dust , finding again that we can not corroborate the presence of dust around the fainter giants . in this work , we use the data and results from paper iii to derive mass - loss rates for individual stars and ( by summation ) the cluster as a whole , and investigate the composition of the circumstellar dust produced by analysing archival mid - infrared ( mid - ir ) spectra . the data for this study come from the accompanying paper iii , which identifies those stars in 47 tuc which have infrared excess , indicative of circumstellar dust . this list of 22 objects is an exhaustive list of those stars in the cluster which we can be relatively certain are producing dust , given currently available data . these objects are listed in table [ mdottable ] , which also includes the parameters we determine in later sections . @l@c@c@c@ c@ c@ c@ c@ c@ c@ c@c@ c@ c@ c@ c@ cl@ star & @xmath6 & @xmath7 & @xmath8 & @xmath9 & @xmath10 & & @xmath11 & @xmath12 & @xmath13 & @xmath14 & & notes + & ( mag ) & ( km / s ) & ( days ) & ( k ) & ( l@xmath15 & @xmath16 & [ 3.6 ] & [ 8 ] & ( % ) & 10@xmath17m@xmath2yr@xmath18 ) & ( kms@xmath18 & ( k ) & sil . & fe & al@xmath19o@xmath20 & feo & + v1 & 4.03 & 20 & 221 & 3623 & 4824 & 0.936 & 1.173 & 2.213 & 3.6 & 21.0 & 4.0 & 900 & 12 & 88 & 0 & 0 & + v8 & 1.6 & 16 & 155 & 3578 & 3583 & 0.904 & 1.210 & 2.135 & 5.9 & 14.7 & 4.0 & 900 & 7 & 85 & 7 & 1 & + v2 & 2.78 & 23 & 203 & 3738 & 3031 & 1.103 & & & 4.3 & 12.0 & 3.8 & 900 & 5 & 95 & 0 & 0 & @xmath21 + v3 & 4.15 & 22 & 192 & 3153 & 2975 & 0.954 & 1.198 & & 3.2 & 9.4 & 3.2 & 1000 & 0 & 100 & 0 & 0 & @xmath22 + v4 & 1.5 & 18 & 165 & 3521 & 2603 & 0.932 & 1.167 & 1.902 & 5.6 & 11.7 & 3.6 & 900 & 2 & 95 & 2 & 1 & @xmath4rgb - tip + v26 & 0.3 & & & 3500 & 2541 & 1.073 & 0.807 & 0.995 & 0.9 & 5.9 & 3.8 & 1000 & & 100 & & & @xmath23 + lw10 & 1.2 & & 221 : & 3543 & 2324 & 1.073 & 0.876 & 1.017 & 0.6 & 4.2 & 3.2 & 1000 & 0 & 100 & 0 & 0 & + v21 & 1.0 & 7 & 76 : & 3575 & 2301 & 0.977 & 1.083 & 1.355 & 1.1 & 4.9 & 3.6 & 1000 & 0 & 93 & 5 & 2 & + lw9 & 1.1 & & 74 & 3374 & 2204 & 0.973 & 1.003 & 1.228 & 1.1 & 6.0 & 3.3 & 950 & & 100 & & & + v27 & 0.3 & & 69 & 3374 & 2140 & 1.023 & 0.929 & 1.218 & 1.4 & 3.3 & 3.6 & 1050 & & 100 & & & + lee1424 & & & 65 & 3565 & 2122 & 1.006 & 1.182 & 1.390 & @xmath244 & 4.4 & 4.0 & 1050 & & 100 & & & + a19 & 0.7 & & 60 & 3526 & 2096 & 1.034 & 0.992 & 1.240 & 0.8 & 3.6 & 4.2 & 1100 & & 100 & & & + lw12&0.6&&116 & 3713 & 2079 & 1.082 & 0.967 & 1.093 & @xmath251 & low + x03 & & & & 3816 & 1640 & & 0.958 & 1.156 & 3.2 & 5.1 & 1.8 & 750 & & 100 & & & @xmath26 + lw19 & 0.1 & & 40 & 3738 & 1638 & 1.078 & 0.991 & 1.080 & @xmath251 & low + v20 & 0.8 & & 232 : & 3602 & 1575 & 1.053 & 0.928 & 1.039 & @xmath251 & low&&&&&&&@xmath27 + v22 & 0.5 & & 62 & 3684 & 1528 & & 2.504 & 2.834 & @xmath251 & low & & & & & & & @xmath23 + v6 & 0.6 & 7 & 48 & 3763 & 1374 & 1.001 & 1.000 & 1.167 & @xmath251 & low & & & & & & & + v5 & 0.7 & 8 & 50 & 3741 & 1363 & 1.003 & 1.100 & & @xmath251 & low&&&&&&&@xmath23 + v18 & 0.3 & 5 & 83 : & 3692 & 1297 & 0.989 & 1.004 & & 2.4 & 5.3 & 2.4 & 800 & 33 & 67 & 0 & 0 & @xmath26 + v23 & 0.5 & & 52 & 3775 & 1259 & 1.023 & 0.985 & & @xmath251 & low&&&&&&&@xmath23 + v13 & & 12 & 40 & 3657 & 1029 & 0.990 & 1.106 & & 1.2 & 4.1 & 2.2 & 800 & 0 & 85 & 15 & 5 & @xmath26 + + accurate or even approximate mass - loss rates are notoriously difficult to measure . one must assume that the wind has certain properties . perhaps the most crucial of these is the relationship between wind velocity and distance from the star . we here assume a constant , spherically - symmetric , purely radiatively - driven wind , starting from a near - stationary , dust - forming layer . as discussed previously ( paper i ; @xcite ) , this may not be the most appropriate description , but suffices due to lack of contrary evidence : we return to this point in the discussion . before calculating an estimated mass - loss rate , we can create a measure of the amount of circumstellar dust surrounding a star , irrespective of most modelling uncertainties . to do this , we define a quantity @xmath11 , the fraction of the star s luminosity that is reprocessed by circumstellar dust . for low optical depths , this should be directly proportional to the surface area of dust grains , therefore approximately proportional to the mass of circumstellar dust . the value of @xmath11 can be determined by taking the difference between the observed sed and model photospheric sed in the infrared ( yielding the infrared excess ) and dividing this by the total photospheric luminosity ( the value given in paper iii ) . we list this quantity in table [ mdottable ] . errors in the value of @xmath11 are introduced by difficulties in determining the integrated luminosity from photometry alone and from the uncertainty in the photospheric spectrum ( both temperature and luminosity ) . the former can be approximated as @xmath28 , or roughly 50% of @xmath11 , but is dominated by photometric uncertainty at short ( 2.210-@xmath29 m ) wavelengths . the latter gives a constant uncertainty of around @xmath30% . we reduce this latter error by fixing the photospheric luminosity to the @xmath31-band flux for all stars except v26 and x03 , which are fixed to the 3.6-@xmath29 m flux . the value @xmath11 could not reliably be measured for several stars with low mass - loss rates . the stars appear to fall into three main categories : those with large ( @xmath45% ) reprocessed flux ( v14 , v8 ) , those with low ( @xmath42% ) reprocessed flux concentrated toward long wavelengths ( v13 and v18 ) , and those stars with low ( @xmath322% ) reprocessed flux concentrated toward shorter wavelengths ( the remainder ) . given the limitations of our understanding and of the model presented above , we stress that the mass - loss rates and velocities we derive are based only on one model for wind acceleration , but should be comparative both internally and with other works which use these same assumptions . under these assumptions , we calculate mass - loss rates by modelling the sed using the radiative transfer package , dusty @xcite . as inputs , we use our interpolated marcs model of the stellar spectrum , and a standard mrn grain size distribution @xcite . within our dusty model , we vary the dust composition , temperature in the dust - forming region ( the inner edge of the dust envelope ) and optical depth of the wind . we change these parameters iteratively until a best fit is achieved . we relate the dusty mass - loss rate ( @xmath33 ) to the actual mass - loss rate ( @xmath34 ) as previously ( paper i ; paper ii ) , using the following formula @xcite : @xmath35 , \label{mdoteq2}\end{aligned}\ ] ] for luminosity @xmath10 , dust - to - gas ratio @xmath36 , grain density @xmath37 , and terminal wind velocity @xmath38 . we assume that @xmath39 g @xmath40 for silicates , 4 g @xmath40 for aluminium oxides , 5.7 g @xmath40 for feo , and 7 g @xmath40 for metallic iron . for these three species , we use the optical constants from @xcite ; @xcite ; and @xcite , respectively . the errors in the above assumptions are substantial , and are discussed in section [ mdoterrorsect ] . the dust - to - gas ratio , @xmath36 , can be estimated as : @xmath41 } , \label{psieqn}\ ] ] which is approximately 1/1000 for 47 tuc . this assumes that the fraction of condensed metals is identical for all metallicities . we discuss the accuracy of this assumption in section [ mdoterrorsect ] . table [ mdottable ] shows our calculated mass - loss rates and inferred wind parameters and dust compositions alongside literature pulsation properties . the striking fact about the dust composition is the fraction of iron grains present . the presence of metallic iron in these stars is debatable , as its featureless infrared emission can be mimicked by many other phenomena , including large silicate grains and emission from thick circumstellar molecular layers @xcite . we consider iron to be the most likely source of emission , though the fractions of metallic iron listed in table [ mdottable ] may not be particularly accurate , and we discuss this further in [ dustcompsect ] . many of the mass - loss rates are too low to be sensibly measured : we can merely say that some circumstellar dust probably exists . those rates that can be measured have their seds shown in figure [ sedsfig ] . the photometry is taken from the sources cited in paper iii . mid - ir spectra are included from @xcite ( hereafter vlmo ) , and @xcite ( hereafter lph ) . these objects are summarised in the following luminosity - ordered list : = 1em v1 : this star is clearly identified as the cluster s dominant dust producer by vlmo , lph and @xcite ( hereafter itm ) . v1 has clear excess at all wavelengths @xmath42 m . this excess , which we attribute to metallic iron dust , dominates even over its strong silicate feature . v8 : no silicate feature was seen in this star by vlmo , but a clear silicate and aluminium oxide feature was seen by lph . it has previously been suggested ( lph ; paper i ) that v8 may have variable mass loss . however , the _ akari _ photometry at 3 , 4 , 7 , 11 , 18 and 23 @xmath29 m shows no variation in the ( dominant ) iron contribution . this means that if the mass loss is variable , strong variation only occurs in the silicate production . v8 has a complex ir spectrum showing peaks from silicates , aluminium oxide , and from crystalline minerals at 9.7 , 11.3 , 13.1 , 20 and 32 @xmath29 m . we discuss the carrier(s ) of these bands in [ dustcompsect ] . note that the amplitude of the 10-@xmath29 m silicate peak is substantially over - estimated in our model compared to the 20-@xmath29 m silicate peak . this implies a lower temperature for the silicate grains than the metallic iron grains , which we also discuss in [ dustcondsect ] . v2 : a silicate dust feature was observed by lph , though it was too weak to have been seen by vlmo . iron dominates the ir flux , however . the 10- and 20-@xmath29 m silicate peaks are not well modelled , as in v8 ( see also [ dustcondsect ] ) . v3 : no silicate emission was seen by either lph or vlmo , but strong , featureless ( metallic iron ) emission is present , which is confirmed by _ akari _ photometry . v4 : the spectrum in lph shows complex dust mineralogy , akin to that in v8 . this star is roughly at the rgb - tip , as estimated from the luminosity function . v26 : blended with a star half its brightness , 2.2@xmath43 to the east , the near - ir , _ akari _ and 24-@xmath29 m photometry are affected . the amount of ir excess is difficult to determine here , but is likely to be quite small . no discernible features are present in the spectrum from vlmo . lw10 : also within the dense cluster core , it appears that the _ akari _ photometry ( at 9 & 11 @xmath29 m ) for this object is affected . no features are detectable in the spectrum from vlmo . v21 : observed by lph , the spectrum shows the same features as v8 . these features , and the silicate contribution , are considerably weaker in v21 . v11 : no obvious features were seen in the spectrum of vlmo , this star was only covered by _ spitzer _ at 3.6 and 5.8 @xmath29 m . 9-@xmath29 m photometry suggests it may have some excess there , but both the _ spitzer _ 8-@xmath29 m photometry and the vlmo spectrum ( which covers 9-@xmath29 m ) show no excess . the derived mass - loss rate can probably be treated as a rough upper limit . lw9 & v27 : an excess consistent with circumstellar dust is present in both the _ spitzer _ and _ akari _ photometry . the chemistry can not be well - constrained by this photometry , but it appears that the flux contribution from components other than iron is comparatively small . lee1424 : lacks any data redwards of 9 @xmath29 m , making constraining the mass - loss rate , dust temperature and dust chemistry very difficult . the values given should be treated as order - of - magnitude estimates only . a19 : as lw9 & v27 , however the _ akari _ 7-@xmath29 m data suggests little to no mid - ir excess in this star . v20 & v22 : not shown in figure [ sedsfig ] , they have no near - ir photometry attributed to them , meaning it is difficult to determine the amount of ir excess present . despite this , the dust temperature is relatively - well constrained by the presence of _ spitzer _ 24-@xmath29 m photometry . x03 : located in the cluster core , this star suffers from the same problems as v20 & v22 , but clearly displays a stronger excess , making an approximate determination of the mass - loss rate and wind characteristics possible . v18 : observed by both lph and vlmo , this star has a clear silicate feature . this feature was not claimed in vlmo due to poor signal - to - noise , but the detection appears to match that from lph well . it shows substantial mass loss compared to other stars of similar luminosity . the 10- and 20-@xmath29 m silicate peaks are not well modelled , as in v8 ( see also [ dustcondsect ] ) . v13 : highlighted in both lph and itm , the spectrum of this star shows an excess similar to that of v8 . the contribution of iron dust is comparatively small , though would appear to be non - zero . this star is isolated from the cluster core , but has poor - quality photometry due to its comparative faintness . the accuracy of the wind parameters we find is hard to determine . the major sources of error are listed below : 1 . a 30% uncertainty inherent in dusty due to an unspecified gravitational correction . this uncertainty is exceeded in stars with @xmath44 km s@xmath18 , where the effect of stellar gravity will change the variation of dust particle density with radius from the star ( the stellar mass is unspecified in dusty ) . 2 . uncertainty in the underlying photospheric spectrum . as the iron dust contribution is essentially a modified blackbody , one can fit a dust temperature to the 3.6- to 8-@xmath29 m data and extrapolate this fit backwards to @xmath16-band , where the dust contribution is essentially zero . the photospheric contribution can then be set from the @xmath45 photometry , but the accuracy of this fit depends on the stellar photometry . ( this would equally hold for other possible dust species , e.g. amorphous carbon . ) uncertainty in the amount of ir excess . apart from identifying how much flux is generated by the photosphere , missing photometry and photometric errors cause uncertainty in the combined flux of the ( photosphere + dust ) contributions . 4 . uncertainty in the dust composition . it is not clear for a number of these stars ( v26 , lw9 , v27 , lee1424 , a19 , x03 and any star marked `` low '' ) what the dust chemistry is . many of these stars have insufficient coverage in the 923 @xmath29 m region to determine whether they contain any dust components other than metallic iron . lw10 and v11 are not covered by lph : ground - based spectra by vlmo show no obvious spectral features , so we presume a pure metallic iron wind , but the spectra are of low signal - to - noise . the presence of silicates can significantly increase the implied mass - loss rate due to their low opacity . any carbonaceous dust present would also be hidden by the modified blackbody of metallic iron . carbonaceous dust typically has a higher opacity : if it is present , the implied mass - loss rate would be decreased . uncertainty in grain size distribution and grain shape ( assumed here to be spherical , with radius @xmath46 ) . the relationship between the grains absorption cross - section to radiation ( @xmath47 ) and their mass ( @xmath48 ) . a high grain porosity would also imply a decreased dust density and therefore a higher mass - loss rate . particularly , differences in grain properties between the iron and silicate grains could explain why we model such a high proportion of iron grains in the wind ( see [ dustcompsect ] ) . uncertainty in dust temperature . the dust temperature can only be determined to within @xmath4100 k. as @xmath49 , a 15% error in temperature leads to an error of up to @xmath460% in the flux emitted by the circumstellar dust , depending on the implied error in the radius of the dust shell . this will be most significant in stars that are losing the least mass , as their dust temperatures are more difficult to determine . uncertainty in the wind driving mechanism . we presume here that the winds are sustained by radiation pressure alone . the very low outflow velocities present in at least three cases ( x03 , v18 and v13 ) would suggest that there is insufficient radiation pressure for it to be the sole driver of mass loss . additional energy input , for example from pulsation , would increase the wind velocity and ( by virtue of eq . ( [ mdoteq2 ] ) ) decrease the mass - loss rate . there is a related uncertainty in the dust - to - gas ratio , which is assumed to scale with metallicity . these uncertainties can not be quantified statistically . lllll parameter & @xmath50 & @xmath51 & parameter variation used & other parameters affected + grain size@xmath52 & @xmath5310% & @xmath5330% & low : @xmath54 , @xmath55 nm , @xmath56 nm & @xmath57 k , 90% iron , 10% silicates + & & & assumed : @xmath54 , @xmath58 nm , @xmath59 nm & @xmath60 k , 88% iron , 12% silicates + & & & high : @xmath61 , @xmath58 nm , @xmath59 nm & @xmath60 k , 80% iron , 20% silicates + grain density & 50% & 50% & low : 1.63 g @xmath40 & + & & & assumed ( solid ) : 6.52 g @xmath40 & + dust formation & @xmath5310% & @xmath5320% & low : 750 k & @xmath62 , 92% iron , 8% silicates + temperature & & & assumed : 850 k & @xmath63 , 88% iron , 12% silicates + & & & high : 950 k & @xmath64 , 82% iron , 18% silicates + dust : gas ratio & @xmath65 & @xmath66 & low : 1 / 5000 & + & & & assumed : 1 / 1000 & + & & & high : 1 / 200 & + velocity distribution & @xmath4 + 500% & n / a & assumed : 0 @xmath67 v@xmath68 km s@xmath18 & @xmath60 k + & & & high : 10 km s@xmath18 ( constant ) & @xmath69 k + photosphere & @xmath5312% & @xmath532% & assumed @xmath53 5% in luminosity & @xmath70 decreased from 0.45 to 0.42 + & @xmath5315% & @xmath536% & assumed @xmath53 250 k in temperature & + calculation error@xmath71 & @xmath5330% & & & + total & @xmath72 & @xmath73 & & + + the easiest way to see how these uncertainties affect a star s determined mass - loss rate is to propagate them through the dusty code . the results for v1 ( table [ mdoterrortable ] suggest that , in absolute terms , our mass - loss rates are little more than an educated guess . however , since a large number of other studies use the same assumptions , one can use these rates to trace differences between stars , providing these properties are largely invariant between stars . the parameters listed in table [ mdoterrortable ] have different effects . the grain size , for example , has little effect on the mass - loss rate but significant effects on other wind properties . conversely , the dust - to - gas ratio have significant effects on the mass - loss rate and velocity but little effect on the other wind parameters ( temperature , optical depth and minerology ) . clearly the largest effect on the mass - loss rate is whether the wind has an initial velocity driven by , for example , pulsation . determining the radial velocity profile of the wind would reduce the error substantially . while this has not been done accurately for such low - metallicity , dust - producing giants , inferences from chromospheric line modelling ( e.g. @xcite ; @xcite ; @xcite ) appears to show that wind velocities do not decrease substantially from @xmath410 km s@xmath18 in the stages shortly prior to dust formation . we point out , however , that the highest mass - loss rates ( and by inference the highest wind velocities ) may be ruled out on evolutionary grounds ( [ dustcompsect ] ) . we emphasise that the errors derived above are not formal and are only estimates . these compare relatively well to the adopted error of @xmath45@xmath74 for silicate - dominated winds ( e.g. @xcite ) , where the grain properties are somewhat better determined due to the long history of their study and constraints from their spectral features . in relative terms , however , we can probably assume that the star - to - star variations are much smaller . in particular , the driving mechanism and hence the wind s initial velocity should not differ as markedly as the uncertainties we assume . likewise , the dust - to - gas ratio will only depend on the condensation fraction in the wind , which should not vary as much as a factor of 5@xmath74 between stars at the same metallicity . this reduces the overall error among stars in the same cluster to a factor of @xmath42@xmath74 and allows us to make direct comparisons between stars . the wind parameters listed in table [ mdottable ] pose several problems for stellar wind theory , which we discuss in this section . firstly , the projected terminal velocities we derive are very low , implying radiation pressure from the star is not effective in driving the wind : we must consider whether other processes may contribute to , or dominate , the wind acceleration . secondly , when one considers the evolutionary timescales of the stars in question , our derived mass - loss rates imply that the stars are losing much more material than is predicted from stellar theory . finally , we can use these problems to constrain some of the wind parameters listed in table [ mdoterrortable ] . mass - loss rate , pulsation amplitude and pulsation period are all known to correlate with stellar evolution . this is demonstrated in figures [ hrdvarsfig ] , [ plfig ] and [ mdotrelfig ] . these can be used to help solve the above problems . a major problem when calculating mass - loss rates from metal - poor stars is that the terminal velocities predicted from radiative transfer codes are very low . this is a particular worry with dusty , as no correction is input for stellar gravity . in this situation , it is impossible not to have mass lost from a star when dust is present , whereas the real situation may be that radiation pressure on dust is insufficient for it to overcome the stellar gravity field . assuming the dust layers around these stars do have some outward velocity , a different method of wind acceleration may be required . dusty provides a minimum mass for a radiative transfer model . if a star is less massive than this , the error due to the unspecified correction for stellar gravity will be greater than the 30% inherent in the model . as this error increases to 100% , it is unclear whether radiation pressure is capable of driving the wind . this is a particular concern for the less - luminous stars with high mass - loss rates : x03 , v18 and v13 . while expansion velocities of 34 km s@xmath18 are not unreasonable when compared to halo carbon stars @xcite , the lower velocities we derive for those stars losing less mass become increasingly unreliable due to this error . @xcite predict that stars with metallicities below [ fe / h ] = 1 can not drive winds via radiation pressure , and must rely on acoustic ( pulsation ) driving to eject mass from the star ( see also @xcite ) . at [ fe / h ] = 0.7 , 47 tuc is near this hypothetical limit , meaning we must consider acceleration by pulsation and other sources of energy . the velocity amplitude of the photosphere in these variables ranges from 423 km s@xmath18 , with v1 , v2 and v3 showing line doubling in near - ir and/or line emission in the optical due to the propagation of shocks in the atmosphere ( table [ mdottable ] ; @xcite ; @xcite ) . while this velocity greatly exceeds our calculated wind velocity of @xmath44 km s@xmath18 , the bulk material does not necessarily attain this velocity . nevertheless , the presence of shocks in the extended atmosphere implies that energy is being transferred to larger radii , and implies that this energy is available for driving mass loss . a comparison between the presence of circumstellar dust and pulsation reveals some interesting results . all the stars listed in table [ mdottable ] are variable : lee1424 = acvs002331 - 7222.6 , while x03 shows a variation of @xmath75 mag ( n. matsunaga , private communication ) . conversely , 22 of 43 variable stars show at least probable evidence of ir excess . dust - producing stars therefore appear to be ( in 47 tuc ) exclusively variable , and variable stars tend to be dust producing . in both figures [ hrdvarsfig ] & [ plfig ] there appears no easy way of discerning rgb - pulsators from agb - pulsators : the outliers in figure [ hrdvarsfig ] are such due to crowding in the cluster core ( paper iii ) , while the pulsation periods in figure [ plfig ] do not show the bimodality that one might expect from having rgb and agb stars at different masses . however , a bimodality may exist in mass - loss rate with a transition occurring around 25003000 l@xmath2 ( figure [ mdotrelfig ] ; cf . a similar possible bimodality in the gas - production rate in @xcite ) . the five stars with elevated mass - loss rates are also those with the highest pulsation velocity amplitudes and among those with the longest pulsation periods . as the presence of pulsation and its amplitude both appear to correlate with mass - loss rate , an obvious conclusion would be that pulsation energy is dissipated in the extended atmosphere and helps drive the wind , though we caution that the strongest variables are also the most - luminous stars with the highest capacity for driving a wind through radiation pressure . further evidence that pulsation sets the mass - loss rate can be drawn from the apparent invariance of mass - loss over the narrow range of temperatures we probe ( figure [ mdotrelfig ] ) . the photospheric temperature sets the radius at which dust can condense and therefore ( if other parameters are invariant ) the density of the wind . higher - density winds lead to more - efficient dust formation , as the timescale for collision between dust - forming molecules is shorter . an invariance of mass - loss rate with temperature thus means that the formation of dust has little impact on the mass - loss rates present , providing further evidence that the mass - loss rate is ` hard - coded ' into the wind before the dust - formation radius . if this is true , the rle of radiation pressure on dust may simply be to affect the outflow velocity of the wind . if the wind is acoustically driven by pulsations , then the wind velocity and thus mass - loss rate is liable to be several times greater than the values listed in table [ mdottable ] , since these sources of energy are not included in our model . this poses potential problems , as mass - loss rates this high would lead to the dissipation of the star s atmosphere on much shorter timescales than is inferred from star counts . this problem may be solved if one invokes different parameters for dust in the wind , which could lead to mass - loss rates much smaller than the listed values ( see [ dustcompsect ] ) . in paper iii , we discussed the integrated rgb mass loss for a typical cluster star . we noted it was likely @xmath320.22 m@xmath2 and possibly , to first order , invariant with metallicity . this invariance would further suggest that the primary driver of rgb mass loss is not related to dust production . we had also estimated in paper iii , based on the cluster s luminosity function , that one star reaches the agb tip every 80 kyr . these figures , however , do not correlate with our derived mass - loss rates . we find here that the cluster s @xmath422 dust - producing stars are losing a total of @xmath761.2 @xmath77 m@xmath2 yr@xmath18 of dust - containing wind . above the rgb tip ( @xmath762000 l@xmath2 ) , where dust production is ubiquitous , 13 stars are producing @xmath78 m@xmath2 yr@xmath18 . this implies that a staggering @xmath760.8 m@xmath2 is lost in the @xmath761 myr which a star spends on the agb above the rgb tip . clearly , this is impossible , as this is the entire mass of the star . this argues for a mass - loss rate of order 48 times lower than we measure . if additional wind driving mechanisms are important , this factor will increase further . mass loss via dust - containing winds between 1000 and 2000 l@xmath2 can be similarly determined . here , nine out of the 56 stars in this luminosity range are producing @xmath42 @xmath79 m@xmath2 yr@xmath18 , implying 0.16 m@xmath2 is lost over a 4.5 myr timescale . if one assumes that mass loss from these stars is also lower than measured , and that some or all of these stars are agb stars , this implies that mass loss via non - dusty winds may be of similar or even greater importance in removing the estimated @xmath320.22 m@xmath2 from the star between the middle of the rgb and middle of the agb . at the rates implied in table [ mdottable ] , the iron in the atmosphere of v1 would be entirely ejected within 1000040000 years , depending on the remaining envelope mass ( 0.080.3 m@xmath2 ) . v8 , v2 and v3 would similarly disperse the iron in their atmospheres within timescales of @xmath2590000 years . the padova and dartmouth isochrones used in paper iii predict that these stars ( especially v2 and v3 ) still have several hundred kyr left on the agb , implying that our iron mass - loss rates are ( again ) a factor of several too high . .maximum elemental abundances in dust [ cols= " < , > , > , > , > , > , > " , ] in [ windaccsect ] , we find radiation pressure may not be the dominant driver of winds from metal - poor stars , and is perhaps incapable of driving them in some cases . the simplest solution is to provide momentum from another source , such as stellar pulsations , in order to increase the wind velocity . unfortunately , this also has the effect of increasing the mass - loss rate , which is already higher than stellar evolutionary theory can accommodate ( [ dustevolsect ] ) . while we expect another momentum source to be present in the most - metal - poor stars ( simply because radiation pressure on dust is that inefficient ) , the limits implied by the mass - loss rates imply this is not the case in 47 tuc . a solution to this dilemma should address both these problems by decreasing the mass - loss rate without significantly decreasing the wind velocity . examining table [ mdoterrortable ] , the simplest change would be to bring the dust - to - gas ratio closer to unity , which would decrease the mass - loss rate and increase the wind velocity . alternatively , we may consider measures that would increase the wind velocity without appreciable increasing the mass - loss rate , such as changing the size distribution of grains . the dust - to - gas ratio , @xmath36 , given in eq . ( [ psieqn ] ) , assumes that the dust fraction scales with metallicity ( cf . this will hold true if every dust species condenses with the same efficiency . navely , one may imagine that the fraction of each element condensed would decrease toward low metallicity simply because there is less opportunity for metal - metal reactions to take place in a metal - poor environment . from our conventional understanding , it is difficult to see how the dust - to - gas ratio in the wind could be substantially closer to parity . we can , however , calculate a maximum amount of dust per unit gas volume , by taking the major dust - forming elements ( o , c , fe , si , mg , ca , al ) plus some other potential dust - forming metals ( ni , ti , cr , mn ) and determining how much dust they can form . for 47 tuc , we assume the abundances of most elements are scaled from the solar values of @xcite by [ z / h ] = @xmath80 dex , following @xcite . we take the following exceptions , however : [ fe / h ] = 0.7 dex @xcite ; [ he / h ] @xmath76 0.02 dex @xcite ; [ o / fe ] = + 0.2 dex @xcite ; [ mg / fe ] @xmath76 [ si / fe ] @xmath76 [ ti / fe ] = + 0.23 dex and [ ca / fe ] = 0.0 dex @xcite . table [ psitable ] shows the stages of this calculation , by which we conclude that @xmath81 . here we make the simplistic assumptions that all dust - forming elements start out in the gas phase , that all carbon forms co , that all silicon forms enstatite ( mgsio@xmath20 ) , that iron and nickel condense entirely as pure metals , and that the leftover metals all form their respective oxides . chemical reactions must be calculated on the basis of number density , but these must be converted to mass densities in order to calculate a dust - to - gas ( mass ) ratio . for reference , the same calculation using solar metallicities arrives at @xmath82 . this approach neglects several points . carbon may be enhanced in these stars due to dredge - up 2 m@xmath2 ) , carbon - enhanced stars of @xmath40.8 m@xmath2 exist in globular clusters ( e.g. @xcite ) . ] , but unless it is sufficiently abundant ( c / o @xmath5 0.75 ) to bind all the remaining free oxygen , this does not affect the end result . we also neglect water condensation , which could use up free oxygen , but this would serve to decrease @xmath36 . forsterite ( mg@xmath19sio@xmath83 ) may condense out in preference to enstatite , and calcium and aluminium may form into silicates before enstatite , but these would also decrease @xmath36 where they affect it at all . the formation of feo and iron - rich silicates would increase @xmath36 , but it seems that only a small percentage of iron condenses into non - metallic forms in these stars . our stated value of @xmath84 is therefore a relatively robust maximum value . table [ mdottable ] suggests that the number density of iron grains to other dust species in the winds is very high ( typically 10:1 , where it can be measured ) , but the analysis shown in table [ psitable ] would suggest it is of order 1:2 if all dust is condensed . this would suggest that the fraction of silicate dust that is condensed is quite low ( @xmath45% ) , however we have not yet accounted for two possibilities : that the temperatures of the grain populations may be limited , or that the size distributions ( specifically the surface - area - to - volume ratios ) of silicates and iron may be different . differing grain temperatures between dust species may exist when the wind is sparse enough that dust - gas collisions can not equalise the thermal energy . this would mean that more - opaque grains , like iron , are warmer ; more - transparent grains , like silicates are colder for a given radius , and could explain the difficulties we have in fitting both the metallic - iron continuum , and the 10- and 20-@xmath29 m silicate bands simultaneously ( [ mdotsect ] ) . more significantly , however , if thermal energy can not be equalised via collisions , this will lead to a significant velocity drift between the gas and dust components of the wind . this will substantially decrease the projected mass - loss rate , while the dust outflow velocity will also increase accordingly . if the gas outflow velocity is below the turbulent velocity in the wind , this may allow gas to preferentially fall back , making the star more metal poor . post - agb stars and planetary nebulae are known to be metal - poor for similar reasons ( e.g. @xcite ) , but it is not clear that this process is happening while the stars are still on the agb where the outflow velocities are lower and the wind densities at the dust - formation radius are greater . conversely , if dust grain temperatures are equalised and the formation temperatures of silicates and iron are roughly similar , we may examine the effect of different grain sizes by taking two limiting cases . if we take the limiting case that both silicates and iron grains follow the same continuous distribution of ellipsoids , the silicate condensation fraction is @xmath45% , and we arrive at a value of @xmath85 to 1/3000 if all the iron is condensed this would increase our derived @xmath34 values by @xmath7660% while decreasing @xmath38 by @xmath7630% . this is the opposite effect to that needed to reconcile our observations with evolutionary theory , therefore we consider this scenario unlikely . in the converse limit , silicate and iron dust has condensed efficiently , but the iron grains have average radii 20@xmath74 smaller . this increases @xmath34 by a moderate ( 3060% ) amount , but increases @xmath38 by much more ( 100200% ) , depending on the exact size distribution . if one is prepared to accept that iron grains grow preferentially in one direction , e.g. if magnetic effects are important , needle - like cylinders would be created ( cf . this would lead to a larger @xmath86 , but in this case without an increased @xmath34 . however , if identical fractions of silicates and iron are condensed , the length / radius ratio of the iron grains would need to be a staggering @xmath761200 , meaning that the grains must be smaller as well . for example , a population of grains 5@xmath74 smaller with a length / radius ratio of 10 allows a similar fraction of silicates and iron to be condensed with only a slight increase in @xmath34 . there is no obvious mechanism that would cause iron to be _ preferentially _ condensed at low metallicities , though magnetic effects and/or the basic problem of forming grains may be important ( an iron grain only needs two iron atoms to bond , whereas a silicate grain requires a series of chemical reactions to form mgsio@xmath20 ) . higher partial pressures of iron may also assist here , though it is not clear why this would happen in a metal - poor environment . note that in all cases above , we have assumed grains are homogeneous : our calculation of the wind parameters in [ mdotsect ] assumes all grains are a uniform mix of the included species , while our discussion above takes the opposite limit : where grains of each species exist separately in pure form . in reality , grains aggregate and form inclusions inside each other , which complicates the above arguments . in general , however , the evolutionary constraints provided in [ dustevolsect ] would suggest that dust condensation efficiency is high compared to solar - metallicity stars ( suggesting @xmath87 ) , and the iron grains are smaller and/or more - elongated than both the silicate grains and the assumed mrn distribution . a combination of these effects is likely required to reduce @xmath34 by the factor of 48 required without reducing the outflow velocities , @xmath86 , we derive ( table [ mdottable ] ) . where we are able to determine it , the composition of circumstellar dust and its variation with luminosity is similar to that we see in other clusters in @xcite ( see also figure [ fehfig ] ) . the evolution of dust composition in 47 tuc broadly agrees with that seen in other clusters at this metallicity . metallic iron appears to be ubiquitous around dust - producing stars , while amorphous silicates are produced by most stars without any obvious dependance on luminosity . the published spectra of several stars ( v4 , v8 , v13 , and weakly in v21 ; see @xcite ) show sharp features around 9.7 , 11.3 , 13.1 , 20 and 32 @xmath29 m . @xcite attribute the 20-@xmath29 m feature to iron - rich wstite ( mg@xmath88fe@xmath89o ) , where @xmath90 ) . recent work by niyogi , speck & onaka ( submitted ) suggests that all these features , with the exception of the 13.1 @xmath29 m feature ( which is probably due to an aluminium oxide @xcite ) , can be explained by iron - rich crystalline silicates . here , the 20-@xmath29 m feature is still produced by an fe - o bond , but it exists in a large crystalline lattice of fayalite ( fe@xmath19sio@xmath83 ) or ferrosilite ( fesio@xmath20 ) . iron - rich silicates are not expected to form in circumstellar environments , as their condensation temperatures are below that of both magnesium - rich silicates and metallic iron @xcite . if iron - bearing crystalline silicates are a significant component of some winds , they must exist alongside metallic iron and amorphous silicate species . one way this could occur is if cooling of the wind proceeds at a faster rate than dust condensation , which may occur in the tenuous winds of low - luminosity , metal - poor stars . figure [ fehfig ] shows that globular cluster stars which show a 20-@xmath29 m feature may be confined to luminosities below @xmath44000 l@xmath2 . it is difficult to say much about their metallicity dependance due to insufficient data , especially given silicate production appears to decline below [ fe / h ] @xmath4 1 . interestingly , the winds found in such stars ( x03 , v13 and v18 ) appear to be quite cool . in more - luminous stars , we find wind temperatures to be higher ( @xmath41000 k ) . the condensation of metallic iron before silicates remains a mystery : for pressures where iron condenses around @xmath41000 k , magnesium - rich silicates should have already formed closer to the star ( e.g. @xcite , their figure 2 ) . significant departures from lte , or unusually - high partial pressures of iron ( caused , perhaps , by magnetic confinement or clumping in the wind ) may offer ways to circumvent this problem , but to invoke them as solutions would be speculative without further data . by whichever process metallic iron is created , if radiation pressure is effective in wind driving , metallic iron condensation should have a dramatic effect on the wind . its condensation should cause a rapid acceleration of the wind , dropping the density and preventing other dust species from condensing . this may be what we see in moderately - luminous stars , such as v3 , lw10 and v11 . if radiation pressure is not an effective driver , the general absence of other dust species may be explained if silicates form using iron grains as seed nuclei . in this case , denser or more - metal - rich winds effectively condense silicates onto iron grains via collisions , while less - dense or metal - poor winds expand before much condensation occurs , leaving bare iron grains . this scenario may explain the iron grains found in the interplanetary dust particles known as gems ( glasses with embedded metal and sulphides ; @xcite ) . in the most - evolved stars ( v1 , v8 , v2 and v4 ) , however , we do see silicate emission making an increasingly - strong presence . meanwhile , there is a hint that the temperature in the dust formation zone ( as traced by metallic iron condensation ) may be decreasing . as noted in [ dustcondsect ] , different dust species may also have different temperatures . more detailed modelling of the winds of these stars , allowing for thermal decoupling of the grain species , is required if we are to speculate on factors which cause this temperature change . it could , however , be linked to stronger pulsation of the star and opacity shielding within the wind . in a cooler wind , however , silicates once again condense out in increasing preference to metallic iron , which contradicts our tentative finding that the silicate grains are cooler than the iron grains in these stars . in this work , we have used spectral energy distributions and infrared spectroscopy to establish mass - loss rates and investigate mineralogy around the dust - producing stars in 47 tuc . we summarise our conclusions as follows : the total mass - loss rate is estimated to be @xmath41.2 @xmath74 10@xmath91 m@xmath2 yr@xmath18 , mainly in the form of metallic iron , but this is probably over - estimated by a factor of 48 due to unconstrained properties of the dust grains . it remains unclear whether radiation pressure on dust is capable of driving a wind alone , due to the low projected terminal velocities of the stellar winds . the existence of circumstellar dust appears to correlate with the presence of stellar pulsation . we theorise that such pulsation may inject energy into the wind acoustically . variations in the wind chemistry broadly concur with those found in @xcite , with silicates found preferentially in high - luminosity , high - metallicity winds , while iron is found preferentially in low - luminosity , low - metallicity winds . iron may condense preferentially at low metallicity . alternatively , iron grains may be coated by silicates in dense , metal - rich winds , but this coating process may not be effective in winds around lower - luminosity and metallicity stars . * acknowledgments : * we are grateful to thomas lebzelter for sharing his original data with us . this paper uses observations made using the _ spitzer space telescope _ ( operated by jpl , california institute of technology under nasa contract 1407 and supported by nasa through jpl ( contract number 1257184 ) ) ; observations using _ akari _ , a jaxa project with the participation of esa ; and data products from the two microns all sky survey , which is a joint project of the university of massachusetts and ipac / cit , funded by nasa and the nsf . , s. 2007 , in astronomical society of the pacific conference series , vol . 378 , why galaxies care about agb stars : their importance as actors and probes , ed . f. kerschbaum , c. charbonnel , & r. f. wing , 145

actual causation has puzzled philosophers since at least the work by . one way of phrasing the problem is as follows : suppose we know the causal laws that govern some domain , and that we then observe a story that takes place in this domain ; when should we now say that , in this particular story , one thing actually caused another ? recent work by has also garnered interest in this topic in the ai community . their account ( which i will refer to as _ hp _ ) constructs a formal definition that tries to capture this intuition in the context of structural models @xcite . to be more concrete , it defines when some random variables @xmath0 of the structural model having values @xmath1 can be counted as an actual cause for @xmath2 . in previous work , i have tried to show that the knowledge representation properties of pearl s structural models can be improved by borrowing representations and techniques from logic programming . in particular , introduced the probabilistic logic programming language of logic programming with annotated disjunctions , for which riguzzi ( 2008 ; 2010 ) implemented sld and slg based resolution algorithms . further analysis of this language has lead to a reformulation of its semantics , called _ cp - logic _ , which attempts to clarify its causal aspects and examine its relation to pearl s work @xcite . a more recent paper @xcite showed that pearl s analysis of interventions and counterfactuals in the context of structural models can be elegantly redone in the context of cp - logic , yielding better results for a number of examples , most notably when cyclic causalities are involved . the goal of this paper is to examine the notion of actual causation in the context of cp - logic . section [ sec : mot ] will start with some motivation , by explaining a few of the differences between structural models and cp - logic , and offering some hand - waving arguments for why cp - logic might offer a more appropriate setting for the study of actual causation . the semantics of cp - logic is briefly recalled in section [ sec : prel ] . then , section [ sec : hp ] dives into the details of actual causation by discussing the hp definition , while section [ sec : ac ] gives my own account . the traditional way of testing such a definition is to run through a number of `` tricky '' examples and checking whether the obtained answers are intuitively plausible . in section [ sec : ex ] , i will follow suit . finally , section [ sec : imp ] briefly comments on a naive implementation in prolog that can be downloaded to play with my definitions a structural model is based on a set of random variables ( rvs ) . each rv has an associated domain of possible values . the simplest case are boolean rvs , which have @xmath3 as their domain , and can therefore be thought of as propositional symbols or ground atoms . boolean rvs suffice for typical examples of actual causation , so i will from now on restrict attention to just these . a structural model then consists of a set of equations @xmath4 , which define the value of the rv @xmath5 in terms of the rvs @xmath6 by a boolean function @xmath7 . the rvs that appear in the left - hand side of such an equation are called _ endogenous _ , and the other ones _ exogenous_. typically , these sets of equations are assumed to be acyclic . their meaning is formalized by the obvious possible world semantics : each assignment of values to the rvs that satisfies all of the equations is a possible world . an acyclic set of equations has the useful property that an assignment of values to the exogenous rvs uniquely determines a single possible world . pearl uses structural models to represent causal relations : each equation @xmath8 is taken to mean that the causes for @xmath5 taking value @xmath9 are all assignments @xmath10 for which @xmath11 . this representation is used by pearl to great effect , studying interventions , counterfactuals , and of course also actual causation . despite its successes , however , there is something peculiar about the structural model representation of causal relations : it does not take into account their dynamic nature . suppose , for instance , that you make the causal claim that dropping a glass causes it to break . if i do nt believe you , i might challenge you to prove your claim . how would you do this ? presumably , you would first hold out an unbroken glass . then , you would drop it , so that i could watch it fall , hit the floor , and break . in other words , you would show me a transition from a state of the world in which the glass is whole to a one in which it is broken . if you can convince me that it was indeed your dropping the glass that initiated this transition , then you have proven your causal claim . what this little thought experiment shows is that the idea of a transition from one state of the world to the next is inherently part of the way in which we interpret causal statements . however , structural models have nothing to do with such transitions . for instance , the causal claim about glasses breaking would just be represented by an equation @xmath12 , which determines two possible worlds : @xmath13 and @xmath14 . in this sense , structural models make complete abstraction of the dynamic aspects of causality , until all that remains is a static picture of how the values of different random variables can be defined in terms of each other . other approaches to causality do not share this static worldview . for instance , gives an explicitly dynamic account of causation . he represents causal systems by means of probability trees , in which edges represent transitions between states of the world . for example , in the following picture , the edge going from @xmath15 to @xmath16 represents a transition from a state in which joe hasnt yet taken a swing at the ball to one in which he has and missed : @xmath17\\ & n_1 \ar[ld]_{\text{joe misses ball}}^{0.75 } \ar[rd]^{\text{joe hits ball}}_{0.25}\\ n_2 \ar[d ] & & n_3\ar[d]\\ \ldots & & \ldots } \end{split}\ ] ] the edge @xmath18 represents a transition of the same state @xmath15 to a state where joe has hit the ball . together , these two edges represent a non - deterministic event , namely that of joe s taking a swing at the ball , which may result in one of these two outcomes . the edges are labeled with the probabilities of the outcomes : the probability of joe s swing missing is @xmath19 and that of it hitting is @xmath20 . this paper will use _ cp - logic _ as its formal language , which is essentially just a modular , syntactic representation for such shaferian probability trees . a theory in cp - logic represents the causal structure of a domain by means of a set of _ causal probabilistic laws _ ( cp - laws , for short ) . each such cp - law is a blue - print for a class of non - deterministic events . for instance , the following cp - law : @xmath21 states that , for every player @xmath22 and ball @xmath23 , player @xmath22 s taking at swing at ball @xmath23 causes a non - deterministic event , which has as one possible outcome that @xmath22 hits @xmath23 ( and this happens with probability @xmath20 ) , and as its other possible outcome that @xmath22 misses @xmath23 ( which happens with probability @xmath19 ) . if @xmath22 and @xmath23 are instantiated to , respectively , a particular player and a particular ball , say joe and the twelfth pitch , the we obtain a description of one particular event ( that may or may not happen , depending on whether joe decides to swing ) : @xmath24 this instantiated cp - law can be seen as a textual representation of picture , provided that , of course , node @xmath15 represents a state of the domain where joe has decided to take a swing at this particular pitch . in this way , each instantiation of a cp - law describes a piece of probability tree . as will be explained in more detail later , an entire cp - theory describes a class of probability trees , each of which can be constructed by putting these small pieces together . unlike structural models , the formal semantics of cp - logic therefore does provide mathematical objects that represent transitions between states of the world . as argued by , these transitions are important for a study of actual causation . indeed , when the goal is to figure out what caused what in a given _ story _ , it is obviously convenient to have a language whose formal semantics already offers objects that correspond in a natural way to stories . a branch of a shaferian probability tree is precisely such an object , because , like a story , it is a description of a sequence of events that change the state of the world . there is also a second argument in favour of cp - logic . the goal of actual causation is to explain why things happened . typically , though , not everything is in need of explanation . a detective solving a murder case , for instance , will be interested in why the victim is dead , but he wo nt care about why he was ever alive in the first place . the detective s causal model will therefore list causes for dying ( poison , gun shot , ) , but not for living ( sexual intercourse , ivf , ) . in more technical terms , the detective considers living to be the _ default _ state of a person , and he is only interested in _ deviations _ from this default . many authors , such as or , have argued that actual causation should be studied under the assumption that each rv has such a default state . structural models make no distinction between the different values of a rv . consequently , a rv @xmath25 with values @xmath26 and @xmath27 , and a rv @xmath28 with values @xmath27 and @xmath26 are completely interchangeable . in cp - logic , this is not the case . here , each rv ( i.e. , ground atom ) has * f * as its default value . this means that the mere existence of the atom @xmath29 implies that the default condition is for @xmath30 to be dead , and that he can only come to life when there is a sufficient cause for this . by contrast , the detective s theories will contain atoms such as @xmath31 , indicating that living is the default and death is in need of causal explanation . in the probability trees generated by a cp - theory , an atom always starts out at its default value , and only deviates from this when it has sufficient cause to do so . lacking space for a full review of cp - logic , i will only summarize the main ideas and refer to @xcite for details . the general form of a cp - law is : @xmath32 here , @xmath33 is a first - order formula and the @xmath34 are atoms , such that the tuple of variables @xmath1 contains all free variables in @xmath33 and the @xmath34 . the @xmath35 are non - zero probabilities with @xmath36 . such a cp - law expresses that @xmath33 causes some ( implicit ) non - deterministic event , of which each @xmath34 is a possible outcome with probability @xmath35 . if @xmath37 , then at least one of the possible effects @xmath34 must result if the event caused by @xmath33 happens ; otherwise , it is also possible that the event happens without any ( visible ) effect on the state of the world . for the purpose of this paper , the propositional fragment of cp - logic suffices , so i will from now on restrict attention to cp - laws in which the tuple of variables @xmath1 is empty . for a cp - law @xmath38 , we refer to @xmath33 as the _ body _ of @xmath38 , and to the sequence @xmath39 as the _ head _ of @xmath38 . we denote these objects as @xmath40 and @xmath41 , respectively , and also write @xmath42 for the set of all @xmath34 for which there exists an @xmath35 such that @xmath43 . for cp - laws that are vacuously caused , @xmath40 may be omitted . if a cp - law has a deterministic effect , i.e. , it is of the form @xmath44 , it is also written simply as @xmath45 . a _ cp - theory _ is a finite set of cp - laws . such a cp - theory describes the non - deterministic evolution of a domain , which is formally represented by a shaferian probability tree . initially , all rvs of this domain ( i.e. , all ground atoms ) are in their default state . this means that we can describe the initial state of the domain , which corresponds to the root of the probability tree , by the interpretation that assigns * f * to each of them . we then extend this root by picking a cp - law @xmath38 whose precondition @xmath40 is satisfied according to this interpretation and creating a child node for each pair @xmath46 in @xmath41 . the edge to child @xmath47 is labeled with the probability @xmath35 and the corresponding new state of the domain is constructed from the previous state by switching @xmath48 to its deviant state * t*. the cp - law @xmath38 has now happened , and will not happen again . we repeat this process of adding children to one of the leaf nodes of the current tree , until this is no longer possible , i.e. , until for all leaves @xmath49 of the current tree it is the case that all rules @xmath38 that have not yet happened in @xmath49 have a precondition @xmath40 that is false in @xmath49 . the resulting trees are called the _ execution models _ of the cp - theory . for a node @xmath50 of the tree , i denote by @xmath51 the interpretation that corresponds to the state of the world at that node , and , if @xmath50 is not a leaf , by @xmath52 the cp - law that was used to create the children of this node . the construction of execution models is quite non - deterministic , in the sense that in any particular node of the tree , there can be many cp - laws that may be used to extend it . the question is now whether each of these trees actually reflects a sensible way in which a domain described by the cp - theory might evolve . the answer is a qualified `` yes '' , and depends on precisely how we choose to interpret negation appearing in the body of a cp - law . consider the following example : @xmath53 we could take the body of this cp - law to mean that this transition may happen in any state where @xmath54 is still at its default state * f * , such as , by definition , the initial state in which suzy has no yet made up her mind about throwing . taking this view , every probability tree constructed according to the above principles can indeed be seen as a sensible description of how the domain might evolve . however , this is not very useful . as argued by , it is more interesting to read negation in a slightly different way , namely , as not just saying that @xmath54 is still at its default value in the current state , but that it can actually _ never _ deviate any more . in other words , according to this reading , the above cp - law will only be applicable _ after _ suzy has decided that she will not refuse the throw . this idea is formalized in the semantics of cp - logic by means of a construction similar to the gelfond - lifschitz reduct . we use this to compute , for each state @xmath50 , an overestimate @xmath55 of all atoms that can still be caused in this @xmath50 . only if an atom @xmath56 does not belong to @xmath57 , do we then say that @xmath58 holds in @xmath50 . if there are no loops containing double negation ( i.e. , some @xmath59 causing @xmath60 and @xmath61 causing @xmath62 ) , then it is the case that , in any branch of a probability tree , each cp - law must either happen at some point , or else become impossible . showed that there is a close connection between the resulting semantics and the well - founded model construction for a logic program each probability tree @xmath63 defines , in the obvious way , a probability distribution @xmath64 over its leaves . for an execution model @xmath63 of a cp - theory @xmath65 , this distribution @xmath64 induces a probabilistic possible world semantics : the probability @xmath66 of an interpretation @xmath67 is @xmath68 , where the sum is taken over leaves @xmath49 of @xmath63 . showed that each execution model @xmath63 of a cp - theory @xmath65 defines the same possible world semantics @xmath64 . for instance , the two trees shown on page are execution models of the same theory and , even though they are not isomorphic , they both define the same @xmath64 . in this way , each cp - theory @xmath65 defines a unique probability distribution , which is denoted as @xmath69 . the probability of a formula @xmath33 can then be defined as @xmath70 . the fact that @xmath69 does not depend on the choice of any particular execution model @xmath63 may help to explain why structural models choose to ignore the dynamic aspects of causality in the first place . indeed , this result shows precisely that , for applications which only care about properties of the final state that the domain will eventually reach , the details of how this final state came about can be safely ignored . as i attempt to show in this paper , though , actual causation is _ not _ such an application . like structural models , cp - logic also makes a distinction between exogenous and endogenous random variables . with @xmath5 the set of all exogenous atoms , the semantics of a cp - theory now becomes relative to an interpretation @xmath71 for these atoms . in particular , an execution model for @xmath65 _ given @xmath71 _ is defined as a execution model that starts not from a root in which _ all _ atoms are * f * , but instead starts with only the endogenous atoms being * f * and the exogenous atoms being interpreted by @xmath71 . have shown that for each interpretation @xmath71 for the exogenous predicates of a cp - theory @xmath65 , all execution models @xmath63 given @xmath71 define the same probability distribution @xmath64 , which is denoted as @xmath72 . this section briefly recalls the hp account . their paper starts with this example : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ suppose that two arsonists drop lit matches in different parts of a dry forest , and that both cause trees to start burning , until the entire forest burns down . both matches are necessary to burn down the forest ; with only one match , the fire would die down . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is clear that both arsonists are an actual cause of the forest burning down . hp reach this conclusion as follows . to represent the causal structure of the example , they use a structural model consisting of a single equation : @xmath73 the particular story under consideration is then represented by the following assignment of values to the exogenous rvs : @xmath74 . this of course also uniquely determines the values of the endogenous rvs : @xmath75 . the hp definition is now reproduced below . in it , @xmath76 is a structural model with endogenous rvs @xmath77 , @xmath78 an assignment of values to the exogenous rvs , @xmath0 a tuple of endogenous rvs , and @xmath33 a boolean formula in the rvs . the notation @xmath79 \phi$ ] means that @xmath33 holds in @xmath80 after the intervention of assigning @xmath1 to @xmath0 is performed , i.e. , each @xmath81 has its defining equation removed from @xmath76 and replaced by @xmath82 . [ hp ] @xmath83 is an actual cause of @xmath33 in @xmath84 if the following three conditions hold . * @xmath85 ( that is , both @xmath86 and @xmath33 are true in the actual world . ) * there exists a partition @xmath87 of @xmath88 with @xmath89 and some setting @xmath90 of the variables in @xmath91 such that if @xmath92 for all @xmath93 , then both of the following conditions hold : * * @xmath94\lnot\phi$ ] . in words , changing @xmath91 from @xmath95 to @xmath96 changes @xmath33 from true to false . * * @xmath97\phi$ ] for all subsets @xmath98 of @xmath99 and all subsets @xmath100 of @xmath101 . in words , setting any subset of variables in @xmath99 to their values in @xmath102 should have no effect on @xmath33 , as long as @xmath0 is kept at its current value @xmath1 , even if all the variables in an arbitrary subset of @xmath101 are set to their original values in the context @xmath78 . * @xmath0 is minimal ; no subset of @xmath0 satisfies conditions ac1 and ac2 . minimality ensures that only those elements of the conjunction @xmath103 that are essential for changing @xmath33 in ac2(a ) are considered part of a cause . with @xmath104 , @xmath105 and @xmath106 , this definition provides the result that @xmath107 actually caused @xmath108 , since if we change @xmath5 to @xmath109 , while leaving @xmath110 at its original value ( this trivially satisfies ac2(b ) ) , we obtain @xmath111 as required by ac2(a ) . in other words , in this example , we get actual causation from a simple counterfactual dependency : if it had nt been for @xmath107 , the forest would nt have burned down . hp also consider a disjunctive variant of this example , where a single match already suffices to burn down the forest ( @xmath112 ) . this causes the straightforward counterfactual criterion to fail , since stopping only one of the arsonists does not stop the forest burning down . this motivates the additional machinery of the above definition . by considering the context in which @xmath113 , that is @xmath110 and @xmath114 , we can re - establish the counterfactual dependency of @xmath115 on @xmath104 . as we have seen , a question of actual causation can only be asked in the presence of two pieces of information : a causal model of a domain ( the @xmath76 of definition [ hp ] ) and a story that takes place in this domain ( the @xmath78 ) . my definition will of course assume that the causal model is given in the form of a cp - theory @xmath65 . in the context of cp - logic , the most obvious formal counterpart of a `` story '' is a branch of an execution model of @xmath65 . already , this allows us some more room for nuance than hp , as the following example from shows . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ suzy and billy might each decide to throw a rock at a bottle . if suzy does so , her rock shatters the bottle with probability @xmath116 . billy s aim is slightly worse and he only hits with probability @xmath117 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this domain corresponds to the following set of cp - laws , where @xmath118 and @xmath119 are exogenous : @xmath120 assuming that suzy and billy both throw , there still exist two different execution models of the theory . representing the states in which the bottle is broken by an empty circle , and those in which it is still whole by a full one , they look like this : + [ trees ] @h=0.8 cm & & ^0.9|(.3 ) _ 0.1|(.3 ) + & _ 0.8|(.3 ) ^0.2|(.3 ) & & _ 0.8|(.3 ) ^0.2|(.3 ) + & & & & @h=0.8 cm & & ^0.8|(.3 ) _ 0.2|(.3 ) + & _ 0.9|(.3 ) ^0.1|(.3 ) & & _ 0.9|(.3 ) ^0.1|(.3 ) + & & & & + in the left execution model , suzy s rock reaches the bottle before billy s does , whereas in the right one , it is billy s rock that gets there first . as discussed at the end of section [ sec : prel ] , this difference is irrelevant if we are only interested in the final outcomes that might be reached : the probability of the bottle shattering is @xmath121 in both models . however , the difference becomes relevant when we want to judge actual causation . indeed , in the left execution model , it is possible for suzy s rock to _ actually _ break the bottle even though billy s also would have ( in particular , this happens in the leftmost branch of the tree ) . according to the execution model on the right , however , this is impossible : here , suzy s rock can only actually break the bottle if billy s rock fails to do so . goes on to consider the following story : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ suzy and billy both pick up rocks and throw them at a bottle . rock get there first , shattering the bottle . since both throws are perfectly accurate , billy s would have shattered the bottle had it not been preempted by suzy s throw . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this story tells us precisely that we are in the leftmost branch of the left execution model above . hence , suzy s rock should be the actual cause of the bottle breaking , and not billy s . before showing how i reach this conclusion in the context of cp - logic , let me first remark that things are more difficult for the hp account . their paper first tries the following straightforward structural model : @xmath122 here , there is no such thing as one execution in which suzy s rock reaches the bottle first and one in which billy s is first . hall s story therefore seems to say nothing more than that all five rvs are @xmath123 , and the phrase `` suzy s rock gets there first '' contributes nothing . of course , because it is precisely this phrase that determines which rock actually broke the bottle , this causal model does not work . hp fix the problem by introducing two new random variables : @xmath124 ( `` billy s rock hits the ( unbroken ) bottle '' ) and @xmath125 . the order in which the two rocks actually reach the bottle can then be encoded _ in the structure of the model _ : @xmath126 to me , this does not seem the right way to go . the order in which the rocks arrive is a purely contingent matter , which belongs to the details of the particular story that is being told , and _ not _ to the general causal structure of the domain . saying that suzy s rock arrives before billy s should not be placed on the same level of causal discourse as the statement that throwing rocks at bottles causes them to break . this is not just a matter of taste , but also has practical consequences . if we would want to know whether suzy s rock would still have been the actual cause of the bottle breaking if billy s rock had gotten there first , then in the hp account we would not just have to look at a different story in the same domain , but we would have to change the structure of our causal model . such hand - tailoring of the causal model to the question under consideration is undesirable , and , as i will now show , it is not needed in cp - logic . my definition too will be heavily based on the intuition of counterfactual dependency from a cause @xmath65 to an effect @xmath127 . therefore , i first formalize the following criterion : if all events happen in the way they actually happened _ with the exception that @xmath65 is somehow prevented from occurring _ , then @xmath127 will no longer occur . ( * ) this requires some mathematical machinery . first , we need to be able to fix the outcome of certain events . for a cp - law @xmath38 of the form @xmath128 , we write @xmath129 to denote the deterministic cp - law @xmath130 . if we now have a branch @xmath23 that tells us what actually happened , then we can define as follows a theory that fixes the outcome of all events that happened to their actual outcome . let @xmath131 be a branch of an execution model of a cp - theory @xmath132 . we define @xmath133 as the union of two disjoint sets @xmath134 and @xmath135 , where @xmath134 contains all cp - laws from @xmath132 that did not happen in branch @xmath23 , i.e. , @xmath136 , and @xmath135 consists of all @xmath137 for which @xmath38 caused @xmath138 in @xmath23 , i.e. , @xmath139 we also need an antonymical transformation , which prevents some @xmath34 from occurring . for an @xmath38 of the same form as above , we write @xmath140 for : @xmath141 to prevent an atom @xmath138 entirely , it now suffices to apply this transformation to all cp - laws that might cause it . given a theory @xmath132 , we therefore define @xmath142 as : @xmath143 by combining this transformation with the previous one , we can now construct a theory @xmath144 which corresponds precisely to the counterfactual eventuality that everything happens precisely as it did in branch @xmath23 , with the exception that @xmath65 is somehow prevented from occurring . i thus formalize the counterfactual criterion ( * ) , by expressing that , according to this new cp - theory @xmath144 , @xmath127 will not occur . let @xmath131 be a branch of an execution model of a theory @xmath132 . for two atoms @xmath65 and @xmath127 , such that both @xmath65 and @xmath127 hold in @xmath145 , we say there is a _ counterfactual dependency _ from @xmath65 to @xmath127 if @xmath146 where @xmath147 and , to cover the case where @xmath65 is exogenous , @xmath148 is @xmath149 . here , saying that @xmath150 is of course equivalent to @xmath127 being false in each leaf @xmath49 of each execution model of @xmath151 . this intuition of counterfactual dependency forms the core of the concept of actual causation , but as discussed above , it is in itself not enough . the additional aspect is the idea of _ relevance_. a causal model might make provisions for a large number of eventualities , many of which may not have been relevant in the actual course of events . it is typical for judgments of actual causation that truly irrelevant causal mechanisms are ignored , even when they might appear to become relevant in a counterfactual context . the typical case where this intuition manifests itself is when counterfactual dependencies are masked by _ redundant causation _ : there is some back - up mechanism waiting in the wings , which will ensures that the effect happens anyway , even if we preempt its actual cause . the example of suzy and billy is a good illustration of this . the reason why we nevertheless insist that suzy is the actual cause of the bottle shattering is precisely a criterion of relevance : because suzy s rock got to the bottle first , billy s was irrelevant , so we ignore it . tried to formalize this same intuition by means of the concept of a _ causal beam _ , which is meant to encompass precisely the relevant parts of the causal model . however , the formal details proved hard to get right , and the refinement that eventually became part of the hp definition seems to be a significant source of complexity , which considerably clouds the otherwise rather simple idea of counterfactual dependency . in the explicitly dynamic context of cp - logic , something much more simple is possible . let us ask again why intuitions feels that billy s rock is irrelevant if suzy s rock gets to the bottle first and shatters it . i suggest the blindingly obvious answer : it just got there _ too late_. by the time billy s rock reached the bottle , the damage was already done , the bottle lay in pieces , and there was nothing left to shatter . in other words , one simply can not cause what is already the case . my notion of relevance will comprise just this : whatever happened _ after _ the effect is irrelevant , and whatever happened _ on the way to _ the effect is counted as relevant . of course , this is not yet a complete dichotomy , since it does not rule on the status of those events that did not happen at all . recall that if some cp - law does not happen in a particular branch , this means that , somewhere along the way , its precondition must have become impossible . whether an event that did not happen is considered relevant will depend on when its precondition became impossible : if this was _ before _ the effect arose , then it is relevant , otherwise not . this leads to the following definition . let @xmath131 be a branch of an execution model of a theory @xmath132 . let @xmath65 and @xmath127 be two atoms that both hold in the final state @xmath152 of @xmath23 , i.e. , @xmath153 . @xmath65 is an _ actual cause _ of @xmath127 in branch @xmath23 if @xmath154 with @xmath155 and @xmath156 , where @xmath151 is constructed as follows . if @xmath157 is the smallest @xmath158 for which @xmath159 , then @xmath160 . in words , @xmath65 is an actual cause of @xmath127 if there is a counterfactual dependency from @xmath65 to @xmath127 , according to the theory @xmath161 that consists of both those events that happened before @xmath127 was caused , and those events that had already become impossible by then . [ def : actc ] it is quite easy to check whether this definition is satisfied : you look at the given branch , find the place where @xmath127 first appeared , discard all events that had not yet happened then but still were possible , and check whether the remaining theory exhibits a counterfactual dependency between @xmath65 and @xmath127 or not . to illustrate , consider again the leftmost branch @xmath162 of the left execution model for the billy and suzy example . the bottle breaks in node @xmath163 , i.e , @xmath164 . therefore , @xmath165 and @xmath166 . according to @xmath167 , there now is indeed a counterfactual dependency from @xmath118 to @xmath168 , so the first is an actual cause of the second . as this example illustrates , it is important that a branch @xmath169 of an execution model not only records the successive states @xmath170 of the domain , but also the events @xmath171 that caused each of the state transitions . recall that the hp setting offers no mathematical objects that correspond to a complete story about what happened , so their definition is always just given the final outcome in the form of an assignment of values to the rvs . in this case , we can not always say with certainty whether some potential cause actually caused an effect or not . indeed , if we get only the final interpretation @xmath145 instead of the full branch @xmath172 , then the best we can do is this : [ partial ] let @xmath132 be a cp - theory and @xmath71 an interpretation for its vocabulary . let @xmath173 be the set of all branches of all execution models of @xmath132 that end in a state @xmath50 for which @xmath174 . if @xmath65 actually causes @xmath127 in _ at least one _ branch @xmath175 , we say that @xmath65 is a _ possible _ actual cause for @xmath127 . if @xmath65 actually causes @xmath127 in _ all _ branches @xmath176 , we say that @xmath65 is a _ certain _ actual cause for @xmath127 . if , in the bottle breaking example , we are only told that eventually @xmath118 , @xmath119 and @xmath168 all hold , we find ourselves faced with precisely the same problem as hp s first structural model : all that we can say is that both are possible actual causes , but neither is a certain actual cause . this is typical for redundant causation patterns , and fits well with intuition here : without knowledge about the order in which events happened , we can not say which of the redundant causes actually `` got there first '' . so far , we have only considered actual causation as it applies to atoms causing atoms . often , it is also interesting to wonder which omissions contributed to an effect ( `` did the doctor s failure to treat the patient cause his death ? '' ) or why some effect was in fact not caused ( `` did the doctor s treatment prevent the patient s death ? '' ) . extending the framework to also address such questions is easy enough : * to extend our definition of actual causation to allow also literals @xmath177 to act as effects , we need to specify when such a @xmath177 `` happens '' for the first time , such that we may discard all later events when making counterfactual judgments to determine what caused @xmath177 . the obvious cut - off point is when @xmath127 no longer belongs to the overestimate @xmath178 . * to also allow literals @xmath179 to act as causes , we need to define precisely how we will check the counterfactual dependency in this case . to assume that @xmath179 was not the case , we need to assume that @xmath65 has somehow occurred , which we can do formally by just adding a new cp - law `` @xmath180 '' that always causes @xmath65 . due to space restrictions , formal details are left to the reader . there is a large literature about actual causation , with many examples , counterexamples , and counter - counterexamples . while e.g. have argued that the importance of such small examples should not be exaggerated , it nevertheless remains useful to check that my approach behaves sensibly for them . due to space restrictions , i will limit myself to those examples that most clearly illustrate the difference between my approach and the hp account . it is common practice in research on actual causation to formulate examples in terms of neuron diagrams . a neuron can be in one of two states , one is the default `` off '' state and the other is the deviant `` on '' state in which the neuron `` fires '' or `` is active '' . different kinds of links between two nodes define how the state of one affects the other . for instance , in the following figure , @xmath127 fires if and only if @xmath181 fires , and @xmath181 fires if at least one of @xmath138 or @xmath65 fires . [ fill](0,1)0.25 cm [ fill](1,1)0.25 cm [ fill](2,1)0.25 cm [ stroke](0,0)0.25 cm neuron diagrams typically record not only this causal structure , but also the state of the neurons . in the figure above , nodes that are `` on '' are represented by full circles and nodes that are `` off '' are shown as empty circles . so , @xmath138 , @xmath181 and @xmath127 all fire , whereas @xmath65 does not . in the language that we have developed so far , a neuron diagram therefore places us in the partial information setting of definition [ partial ] : we are given a causal model of a domain together with the final state that has been reached , but are not told precisely how this state has come about . shows a number of counterexamples to hp , and introduces an alternative account , which he formalizes for neuron diagrams only . one of his counterexamples concerns the following two diagrams : in both diagrams , the edges from @xmath181 to @xmath182 and from @xmath182 to @xmath127 are _ blocking _ edges : if @xmath181 fires , then @xmath182 will never fire , regardless of its other incoming edges . in the left diagram , both @xmath138 and @xmath65 cause @xmath127 : @xmath138 causes it directly and @xmath65 causes it by stopping @xmath183 from preventing @xmath127 . in the right diagram , however , @xmath65 also causes the very `` threat '' to @xmath127 that it prevents . therefore , hall argues , in this diagram it should not be counted as a cause for @xmath127 . the hp account correctly handles the left diagram , but fails for the right one , since taking @xmath184 and @xmath185 allows us to create the context @xmath186 in which there is a counterfactual dependency from @xmath65 to @xmath127 . to see how my definition fares , here are the obvious cp - logic versions . in the first , @xmath187 and @xmath183 are all exogenous , while in the second only @xmath138 and @xmath65 are . + @xmath188 @xmath189 first , consider the left theory . here , @xmath127 can only be caused after has already happened and both and have become impossible . therefore , all these cp - laws are relevant and we end up having to check whether there is a counterfactual dependency from @xmath65 to @xmath127 in the original theory . clearly , this is the case , since no tree that starts from a root in which the exogenous predicates @xmath183 and @xmath138 are * t * and @xmath65 is * f * can produce @xmath127 . in the second theory , the event may either happen before @xmath127 is caused or after . this means we either have to check for a counterfactual dependency in the theory @xmath190 or in @xmath191 . in neither theory we find a counterfactual dependency , so @xmath65 is correctly judged to certainly not be an actual cause of @xmath127 . hp s problems with this example are caused by the fact that , lacking an explicitly dynamic semantics , they have to resort to interventions to eliminate irrelevant events from consideration . as an undesired side - effect , they end up allowing the possibility that @xmath183 itself is relevant for judging the impact of @xmath65 on @xmath127 , but the link between @xmath65 and @xmath183 is not . the following is an example of _ bogus prevention _ @xcite , taken from . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ assassin is in possession of a lethal poison , but has a last minute change of heart and refrains from putting it in victim s coffee . bodyguard puts antidote in the coffee , which would have neutralized the poison had there been any . victim drinks the coffee and survives . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ here , hp , as well as others such as , erroneously designate the bodyguard s unnecessary antidote as an actual cause for victim s survival . as i will now show , my account handles this correctly . since the example states that assassin has his @xmath192 before the @xmath193 is administered , i will not make these exogenous atoms , but instead include them as endogenous atoms that are vacuously caused with some unknown ( and irrelevant ) probability . @xmath194 the example now tells the story that first event happens , which is then followed by . however , as soon as happens , both @xmath195 and @xmath196 become impossible , so is considered irrelevant in the actual course of events and will not be part of the theory in which we check for a counterfactual dependency . hence , preventing @xmath193 in this theory has no effect whatsoever upon the victim s survival , so it is not an actual cause of victim s survival ( but @xmath192 is ) . note that if the antidote were administered before the assassin s change of heart , then it _ would _ be considered relevant , but still not an actual cause of victim s survival because then would no longer be relevant . to make the antidote an actual cause of victim s survival , it would have to be administered after the assassin has _ failed _ to have a change of heart . a prototype implementation can be downloaded from the following url : http://people.cs.kuleuven.be/~joost.vennekens/actcaus/act.pl[http://people.cs.kuleuven.be/@xmath197joost.vennekens/actcaus/act.pl ] this small program computes whether an atom is a possible / certain actual cause for an effect in the partial information setting , or an actual cause in the complete information setting . it was written in swi - prolog , but should also run in sicstus or yap . currently , it only handles ground theories without disjunction in rule bodies . in the partial information setting , this prototype performs a simple backtracking search over all branches that might generate the given observations . obviously , this is not an approach that would scale well for larger examples . the goal of this prototype , however , is just to allow interested people to experiment with my definition , in order to see whether it corresponds to their intuition . as such , it is not meant to handle problems larger than the examples typically considered in the actual causation literature . future work may investigate better algorithms , e.g. , by means of an integration into riguzzi s @xcite query answering algorithm . this paper has tried to argue that the hp account of actual causation is flawed for two reasons , both of which stem from their choice of structural models as the formal language to express causal relations . first , structural models fail to make the distinction between default and deviant values , which has been argued by many authors to play a key role in a correct understanding of actual causation . second , the static world - view of structural models is ill - suited to handling dynamic concepts , such as the stories that are part of the input to an actual causation problem . since the hp paper first appeared , it has received a great deal of attention among researchers interested in actual causation , and many counterexample and alternative approaches have been presented . most of these , such as or , recognize the importance of the deviant / default distinction . the problems caused by the mismatch between the static formalism of structural equations and the dynamic problem of actual causation have achieved less attention , even though they are recently also pointed out by . nevertheless , also these more recent approaches still use static formalisms such as neuron diagrams or variants of structural models . the main point i hope to make in this paper is that for problems that , like actual causation , require reasoning about the way in which a domain evolves , it pays to have a language with a formal semantics that contains mathematical objects that correspond to such evolutions . i have tried to illustrate this by defining a notion of actual causation in the context of cp - logic , a probabilistic logic programming language which can be seen as a modular syntactic representation for shaferian probability trees , which offer precisely the kind of dynamic representation that is perfectly suited for a study of actual causation . my definition is based on a counterfactual criterion similar to hp s , but is able to leverage the dynamic nature of cp - logic s semantical objects to come up with a very straightforward notion of relevance , namely , it only considers as relevant those events that happened ( or became impossible ) _ before _ the effect first arose . this is much simpler than the relevance criterion of hp , since i do not have to rely on complex manipulations by means of interventions . while lacking space for an elaborate review of examples from the literature , i have shown that there are three examples where my definition beats hp : already for simple examples of redundant causation , it offers a more elegant account due to its ability to distinguish the complete and partial information settings ; it is also able to detect fake causes that simply prevent themselves from preventing the effect ; and it also handles bogus prevention . of course , that is not to say my approach is perfect . for instance , the railroad switch example from the hp paper can not be handled , because it contains a rv ( @xmath198 ) whose default and deviant values switch in the middle of the story . i am also offering a prototype implementation of my definition , in the hope that it may help to find further examples where it does not correspond to intuition . feedback will be appreciated . inference with logic programs with annotated disjunctions under the well founded semantics . in _ logic programming , 24th international conference , iclp 2008 , udine , italy , december 9 - 13 , 2008 , proceedings_. lecture notes in computer science , vol . springer , 667771 .

the scenario of @xmath0 users wanting to communicate to a single destination using cooperative strategies . for simplicity , we consider one of the @xmath0 users as the source and the remaining @xmath1 users as relays . predominantly there are two types of relaying strategies discussed in the literature , i.e. , 1 ) amplify and forward and 2 ) decode and forward . in this work , we focus only on amplify and forward based relaying strategies wherein the relays are allowed to perform linear processing of the received signal . amplify and forward based relaying is simpler and moreover does not require the relays to have knowledge of the codebook used by the source or the knowledge of the source to relay channel fading gains . recently in @xcite , the idea of space - time coding for collocated mimo ( multiple input multiple output ) systems has been applied in the setting of cooperative wireless relay networks in the name of distributing space - time coding , wherein coding is performed across users and time . this strategy provides each user a diversity order equal to the number of relays even though all the users are only equipped with a single antenna . the diversity thus achieved is called as cooperative diversity . however , such strategies require that the destination have complete knowledge of the fading coefficients from all the users to itself as well as that of the fading coefficients between users . obtaining the knowledge of the fading coefficients between the users at the destination requires additional resources . to solve this problem , in @xcite , kiran _ et al _ have proposed a differential encoding / decoding setup for cooperative wireless networks that does not require the knowledge of fading coefficients between the users at the destination . such codes were named as partially coherent distributed space - time codes in @xcite . in a recent work @xcite , it has been shown that the same strategy of @xcite offers full diversity with a suboptimal receiver that does not require the knowledge of any of the fading coefficients . in an independent work @xcite , jing and jafarkhani have proposed a differential encoding / decoding setup for cooperative wireless relay networks which is more general than the setup proposed in @xcite and they have also provided few code constructions . we call the class of differential space - time codes ( dstcs ) which can be used in a distributed manner for cooperative diversity as distributed dstcs ( and denote by ddstcs ) to differentiate them from dstcs for collocated mimo systems . the problem of designing ddstcs is more challenging than that of dstcs for collocated mimo systems , since in this scenario we have additional constraints to be satisfied which are due to the cooperative diversity protocol . in @xcite , the setup of @xcite has been generalized to the case when the source and destination nodes have multiple antennas . in @xcite , the authors propose the use of a random diagonal unitary matrix codebook in general and for certain specific number of relays ( @xmath2 , @xmath3 and @xmath4 ) propose the use of cyclic codes . however , a random code is not guaranteed to offer full diversity . moreover designing good cyclic codes becomes difficult for large number of relays . in @xcite , the authors propose the use of alamouti code and @xmath5 code for the specific case of @xmath6 and @xmath7 relays . for an arbitrary number of relays , the authors construct fully diverse ddstcs based on circulant matrices . but , except for the alamouti code all other ddstc constructions in the literature @xcite have large decoding complexity . thus , a general full diversity code construction targeting the requirements of low encoding complexity as well as low decoding complexity is not available . this issue gains significant importance especially if the number of cooperating terminals is large , which is quite expected in applications like wireless sensor networks . in this paper we address this issue and present the following contributions : * the notion of encoding complexity for space - time codes in terms of @xmath8-group encodability is introduced ( definition [ defn1 ] ) and its significance is highlighted . * the differential encoding / decoding setup introduced by kiran _ et al _ @xcite , oggier _ et al _ @xcite and jing _ et al _ @xcite for wireless relay networks that use codebooks consisting of unitary matrices is generalized to allow codebooks consisting of scaled unitary matrices . this generalization aids in reducing the encoding / decoding complexity of ddstcs . when the codebook of scaled unitary matrices is used in the jing - hassibi protocol @xcite for differential encoding at the source , the extra conditions involving the interrelationship between relay matrices and the codeword matrices that need to be satisfied are identified ( eqn . ) . * using the algebraic framework of extended clifford algebras , a new class of ddstcs satisfying these extra conditions ( constructions [ cons_design ] and [ cons_signalset ] ) is proposed which achieves full diversity with a low complexity sub - optimal receiver ( theorem [ thm_relay ] ) . explicit construction of multidimensional signal sets that lead to full diversity ( theorem [ thm_signalset ] ) are also provided . to the best of our knowledge , this is the first known low decoding complexity ddstc scheme for cooperative wireless networks which is available for all power of two number of relays . section [ sec2 ] introduces the system model for the distributed differential encoding / decoding setup employing a differential scaled unitary matrix codebook at the source . in section [ sec3 ] , the notion of @xmath8-group encodable and @xmath8-group decodable space - time codes is discussed for the system model of section [ sec2 ] and the problem statement for ddstcs with low encoding and decoding complexity is given . the extra conditions on the code structure imposed by the cooperative diversity protocol of @xcite are then described . in section [ sec4 ] , we explicitly construct a family of dstcs using extended clifford algebras and prove using algebraic techniques that they satisfy the extra conditions needed for ddstcs . simulation results and discussion on further work comprise sections [ sec5 ] and [ sec6 ] respectively . for two sets @xmath9 and @xmath10 , @xmath11 denotes the cartesian product of @xmath9 and @xmath10 . for a matrix @xmath12 , the matrices @xmath13 , @xmath14 and @xmath15 denote the conjugate , transpose and conjugate transpose of @xmath12 respectively . for a complex number @xmath16 , @xmath17 and @xmath18 denote its in - phase and quadrature - phase components respectively . for a vector @xmath19 , @xmath20 $ ] denotes the @xmath21 entry of @xmath19 . an all zero matrix of appropriate size will be denoted by @xmath22 . a complex gaussian vector with zero mean and covariance matrix @xmath23 will be denoted by @xmath24 . in this section , we briefly explain the distributed differential encoding / decoding setup proposed in @xcite with a slight modification . we consider a network consisting of a source node , a destination node and @xmath1 other relay nodes which aid the source in communicating information to the destination . all the nodes are assumed to be equipped only with a single antenna and are half duplex constrained , i.e. , a node can not transmit and receive simultaneously at the same frequency . the wireless channels between the terminals are assumed to quasi - static and flat fading . the channel fading gains from the source to the @xmath25-th relay , @xmath26 and from the @xmath27-th relay to the destination @xmath28 are all assumed to be independent and identically distributed complex gaussian random variables with zero mean and unit variance . moreover , we assume symbol synchronization among all the nodes . every transmission cycle from the source to the destination comprises of two stages . in the first stage , the source transmits a @xmath29 length vector @xmath30 which the relays receive . here , @xmath31 denotes the total average power spent by all the relays and the source . the fraction of total power @xmath31 spent by the source is denoted by @xmath32 . the vector @xmath33 satisfies @xmath34=1 $ ] and is a function of the information that the source intends to communicate . the exact relation of @xmath33 with the information sent by the source will be made precise in the sequel . the received vector at the @xmath27-th relay node is then given by @xmath35 . in the second half of the cycle , all the relay nodes are scheduled to transmit together . the @xmath27-th relay node transmits a @xmath36 length vector @xmath37 which is a function of @xmath38 . the relays are only allowed to linearly process the received vector @xmath38 or its conjugate @xmath39 . to be precise , the @xmath27-th relay node is equipped with a @xmath40 unitary matrix @xmath41 ( called relay matrix ) and it transmits either @xmath42 or @xmath43 . without loss of generality , we may assume that @xmath44 relays linearly process @xmath38 and the remaining @xmath45 relays linearly process @xmath39 . here , @xmath46 denotes the fraction of total power @xmath31 spent by a relay . if the quasi - static duration of the channel is much greater than @xmath47 time slots , then the received vector at the destination is given by @xmath48 where , @xmath49,\\ \mathbf{h}=&\left[\begin{array}{ccccccc}f_1g_1 & \dots f_mg_m & f_{m+1}^*g_{m+1 } & \dots & f_r^*g_r\end{array}\right]^t,\\ \mathbf{n}=&\sqrt{\frac{\pi_2p}{\pi_1p+1}}\left(\sum_{j=1}^{m}g_j\mathbf{a_jv_j}+\sum_{j = m+1}^{r}g_j\mathbf{a_jv_j}^*\right)+w \end{array}\ ] ] and @xmath50 represents the additive noise at the destination . the power allocation factors @xmath32 and @xmath46 are chosen in such a way to satisfy @xmath51 . throughout this paper , we chose @xmath52 and @xmath53 as suggested in @xcite . the differential encoding is performed at the source as follows . an initial vector @xmath54 known to the destination which satisfies @xmath55 is transmitted by the source in the first cycle . the initial vector @xmath56 should be such that the initial matrix @xmath57\ ] ] is unitary . the transmitted vector at the @xmath58-th cycle is then given as follows @xmath59 where , @xmath60 is the codeword containing the information at the @xmath58-th cycle which satisfies @xmath61 . the codebook @xmath62 at the source consisting of scaled unitary matrices should satisfy @xmath63=1 $ ] in order to meet the power constraint at the source and relays . the originally proposed coding strategies in @xcite force @xmath64 for all codewords . the received vector at the destination in the @xmath58-th cycle can then be written as @xmath65 where , . if @xmath66 then , substituting for @xmath67 from we get @xmath68 . if the channel remains approximately constant for @xmath69 channel uses , then we can assume that @xmath70 . then , @xmath71 where , @xmath72 . now we propose to decode the codeword @xmath73 as follows @xmath74 where , @xmath75 can be estimated from the previous decision . note that the above decoder does not require the knowledge of the channel gains and the relay matrices @xmath76 . also , note that this decoder is not a maximum - likelihood ( ml ) decoder since the equivalent noise vector @xmath77 is dependent on the previous codeword @xmath78 through @xmath75 . however , for the collocated mimo case , it has been shown in @xcite that there is no significant performance loss due to this . in this setup , we call @xmath62 a ddstc to distinguish it from dstcs for collocated mimo systems . in this section , we describe the problem statement of designing ddstcs with low encoding complexity and low decoding complexity . in order to benefit from low encoding complexity and low decoding complexity , we propose to choose the ddstc @xmath62 to be a @xmath79group encodable ( @xmath80 ) linear stbc . [ defn1]a linear design @xmath81 of size @xmath82 in @xmath83 real indeterminates or variables @xmath84 is a @xmath82 matrix with entries being a complex linear combination of the variables @xmath84 . to be precise , it can be written as follows : @xmath85 where , @xmath86 are square complex matrices called the weight matrices . a linear stbc @xmath62 is a finite set of @xmath82 complex matrices which can be obtained by taking a linear design @xmath81 and specifying a signal set @xmath87 from which the information vector @xmath88^t$ ] take values , with the additional condition that @xmath89 . a linear stbc @xmath90 is said to be @xmath8-group encodable ( or @xmath91 real symbol encodable or @xmath92 complex symbol encodable ) if @xmath8 divides @xmath83 and if @xmath93 where each @xmath94 . thus , given a @xmath8-group encodable stbc , it defines a natural partitioning of the real variables of its associated linear design into @xmath8-groups . this partitioning defines a partitioning of the set of weight matrices of @xmath95 into @xmath8-groups , the @xmath96-th group containing @xmath97 matrices . for simplicity we assume the simplest partitioning of the information symbol vector as @xmath98^t$ ] where , . now @xmath95 can be written as , @xmath99 minimizing the decoding metric corresponding to , @xmath100 is in general not the same as minimizing @xmath101 for each @xmath102 individually . however if it so happens , then the decoding complexity is reduced by a large extent . [ defn2 ] a linear stbc @xmath90 is said to be @xmath8-group decodable ( or @xmath91 real symbol decodable or @xmath92 complex symbol decodable ) if it is @xmath8-group encodable and if its decoding metric in can be simplified as in . @xcite the decoding metric in can be simplified as in if @xmath103 for all weight matrices @xmath104 and @xmath105 belonging to two different groups . we illustrate the significance of the above definitions concerning encoding and decoding complexity by giving few examples . consider the golden code for @xmath6 transmit antennas . it has @xmath106 real variables . for the coherent mimo channel , the signal set used is qam for each complex variable . hence , this code is a @xmath106-group encodable ( since qam is a cartesian product of two pam signal sets ) and @xmath107-group decodable linear stbc . however , if we now impose the requirement for unitary codewords , then we will have to solve for signal sets which will yield unitary codewords inside the division algebra . this may amount to entangling all the @xmath106-real variables which will make the code @xmath107-group encodable and @xmath107-group decodable . this approach was recently attempted in @xcite . [ eg_alamouti ] let us take the example of the alamouti code for @xmath6 transmit antennas . it has @xmath7 real variables . if the signal set is chosen to be psk for every complex variable , then all the codewords become unitary matrices . hence the resulting code is @xmath6-group encodable and @xmath6-group decodable . however , note that if we take square qam to be the signal set for each complex variable , then we get a @xmath7-group encodable ( since square qam is a cartesian product of two pam signal sets ) and @xmath7-group decodable code , but now the codewords are scaled unitary matrices as opposed to unitary matrices . thus relaxing the codewords to be scaled unitary matrices allows us to lower the encoding and decoding complexity . the above two examples show that _ the choice of signal sets is crucial in obtaining low encoding and decoding complexity_. example [ eg_alamouti ] explicitly shows how allowing scaled unitary codebooks aids in reducing the encoding / decoding complexity . the ddstc design problem is then to design a @xmath8-group decodable linear stbc @xmath108\right)|\mathbf{x}\in\mathscr{a}\right\}\ ] ] of size @xmath40 to be used at the source such that 1 . all codewords are scaled unitary matrices satisfying the transmit power constraint . 2 . the parameters @xmath83 and @xmath8 are maximized . increasing @xmath83 is motivated by the need for high rate transmission and maximizing @xmath8 is motivated by the requirement for low decoding complexity . [ cond_ddstc4 ] there exist @xmath1 unitary matrices @xmath109 of size @xmath40 such that the first @xmath44 of them satisfy and the remaining @xmath45 of them satisfy . 4 . [ cond_ddstc5 ] there exists an initial vector @xmath54 such that the initial matrix @xmath110 is unitary . @xmath111 is maximized , i.e. , the coding gain is maximized . observe that the requirements for designing ddstcs are much more restrictive than that for dstcs . note that condition and condition are not required for designing dstcs . as an additional requirement it would be nice to have a single design @xmath112 and a family of signal sets , one for each transmission rate such that all the required conditions are met . this means that we need to be able to find @xmath1 relay matrices satisfying the required conditions irrespective of the size of the code @xmath113 . in this section we give an explicit construction of full diversity @xmath7-group decodable ddstcs for all power of two number of relays . optimizing the coding gain for the proposed ddstcs is difficult in general and is beyond the scope of this paper . constructing @xmath79group decodable ddstcs for arbitrary @xmath8 appears to be rather difficult and a brief discussion on the issues involved is given is section [ sec6 ] . in subsection [ subsection_design ] , the construction of the proposed linear designs using tools from extended clifford algebras is given . in subsection [ subsection_signalset ] , an explicit construction of signal sets @xmath114 leading to full diversity for arbitrary transmission rate is provided for the linear designs given in subsection [ subsection_design ] , thus completely describing the construction of the linear stbc @xmath62 . in subsection [ subsection_relay ] , we use algebraic techniques to explicitly construct the @xmath1 relay matrices satisfying the required conditions for arbitrary size of the codebook @xmath62 . in this subsection , we briefly describe a construction of a class of rate one , linear designs satisfying the conditions for four - group decodability which were first obtained using extended clifford algebras in @xcite . extended clifford algebras and linear designs from them are addressed in detail in @xcite . however @xcite chooses signal sets as applicable for coherent communication in a distributed space - time coding setup . a brief explanation of extended clifford algebras and linear designs using them is provided in appendix [ appendix1 ] . this algebraic framework given in appendix [ appendix1 ] is needed to explicitly construct the relay matrices and the initial vector in subsection [ subsection_relay ] . given a @xmath82 linear design @xmath115 in @xmath116 complex variables @xmath117 , one can construct a @xmath118 linear design @xmath119 as follows : @xmath120 } \ ] ] where , the linear design @xmath121 is identical to the linear design @xmath115 except that it is a design in a different set of complex variables @xmath122 . we call this construction as the abba construction which was first introduced in @xcite . [ cons_doubling ] given a @xmath82 linear design @xmath115 , one can construct a @xmath118 linear design @xmath123 as follows . @xmath124 } \ ] ] where , the linear design @xmath121 is identical to the linear design @xmath115 . we call this construction as the doubling construction. we are now ready to describe the construction of the proposed linear designs . [ cons_design ] to obtain a linear design for @xmath125 relays , follow the steps given below . + step 1 : starting with the linear design @xmath126 $ ] in complex variable @xmath127 , keep applying abba construction iteratively on it till a @xmath128 linear design @xmath119 is obtained . + step 2 : apply doubling construction on @xmath119 and scale it by @xmath129 to obtain the linear design . following the steps given above for @xmath130 relays , we get @xmath131.\ ] ] in this subsection , we construct signal sets for the linear designs constructed in the previous subsection that lead to full diversity stbcs . in general , the signal sets should be designed such that the resulting linear stbc @xmath62 meets the following requirements . 1 . scaled unitary codeword matrices meeting power constraint . four - group encodable and four - group decodable . 3 . difference of any two different codeword matrices should be full rank . we call such a code to be fully diverse. 4 . the minimum determinant of the difference of any two codewords matrix should be maximized ( coding gain ) . we shall first illustrate the signal set construction procedure for @xmath7 relays and derive important insights form it . then , we generalize the ideas thus obtained for constructing signal sets for any @xmath125 relays . the design for @xmath7 relays according to the construction in the previous subsection is given by @xmath132.\ ] ] let us look at @xmath133 shown in at the top of the next page to find out the conditions on the signal sets under which the codewords are scaled unitary matrices . @xmath134\ ] ] ' '' '' from , we see that the signal set should be chosen such that the following condition is satisfied for all the signal points . @xmath135 firstly , we identify the grouping of the real variables into groups such that the corresponding weight matrices in each group satisfy . according to the construction , the four groups or real variables are as follows : first group : @xmath136 , second group : @xmath137 , third group : @xmath138 , fourth group : it is important to choose signal sets such that they do not enforce joint constraints on variables from different groups . the requirement for scaled unitary codewords as in can be satisfied without disturbing @xmath7-group encodability if all the signal points satisfy the following equation : @xmath140 where , @xmath141 and @xmath142 are positive real constants . then , the average power constraint requirement can be met by satisfying the conditions , @xmath143 a common set of solutions for and can be obtained by taking points on the intersection of circles and hyperbolas . to meet the third requirement of full diversity we make use of the structure of the constructed designs . note that @xmath144\right|\\ \\ & = \frac{1}{4}\left|\left[\begin{array}{cc}\delta \mathbf{a}^h\delta \mathbf{a}+\delta \mathbf{b}^h\delta \mathbf{b } & \mathbf{0}\\ \mathbf{0 } & \delta \mathbf{a}\delta \mathbf{a}^h+\delta\mathbf{b}\delta \mathbf{b}^h\end{array}\right]\right|\\ & \geq \frac{1}{4}\mathrm{max}(|\delta \mathbf{a}|^*|\delta \mathbf{a}|,|\delta \mathbf{b}|^*|\delta \mathbf{b}|)^2 \end{array } } \ ] ] where , the notation @xmath145 has been used to denote the difference matrix which will have the same form as the associated linear design . the second equality in follows because the linear designs @xmath146 and @xmath147 commute . this can be proved as follows . let @xmath148 $ ] and . then , we have @xmath149\left[\begin{array}{cc}\mathbf{w ' } & \mathbf{x'}\\\mathbf{x ' } & \mathbf{w ' } \end{array}\right]=\left[\begin{array}{cc}\mathbf{ww'}+\mathbf{xx ' } & \mathbf{wx'+xw'}\\\mathbf{xw'+wx ' } & \mathbf{xx'+ww ' } \end{array}\right],\\ \\ \left[\begin{array}{cc}\mathbf{w ' } & \mathbf{x'}\\\mathbf{x ' } & \mathbf{w ' } \end{array}\right]\left[\begin{array}{cc}\mathbf{w } & \mathbf{x}\\\mathbf{x } & \mathbf{w } \end{array}\right]=\left[\begin{array}{cc}\mathbf{w'w}+\mathbf{x'x } & \mathbf{w'x}+\mathbf{x'w}\\\mathbf{x'w}+\mathbf{w'x } & \mathbf{x'x}+\mathbf{w'w } \end{array}\right ] . \end{array } } \ ] ] thus we observe that linear designs @xmath146 and @xmath147 commute if their constituent sub - matrices commute . applying this argument recursively , the claim follows . thus @xmath123 is fully diverse , i.e. , @xmath150 if either of the submatrices @xmath146 or @xmath147 is fully diverse . for the example of @xmath7 relays , we have . hence we can guarantee full diversity by ensuring that @xmath151 and @xmath152 . just like before , we should be careful not to disturb @xmath7-group encodability in the process . we take care of that requirement also by satisfying the following conditions : @xmath153 putting together all the conditions that need to be met , we have to choose constellation points satisfying , and . the solution can be found simply by finding the intersection of points on the unit circle @xmath154 with a hyperbola @xmath155 , where @xmath156 on the two dimensional @xmath157 plane . this is illustrated in fig . [ fig_signal2dimstruct ] . observe from fig . [ fig_signal2dimstruct ] that the hyperbola intersects the circle at four different points . but the full diversity criterion demands that @xmath158 . after enforcing this condition , only two points survive out of the four points . they can be either the set of points marked a or the set of points marked b in fig . [ fig_signal2dimstruct ] . thus we have obtained a signal set containing @xmath6 points . supposing that we need more than two points , we can draw more circles ( centered at origin ) with radii such that the average power constraint is met and then find those points intersecting with the hyperbola . more precisely , to get @xmath159 points , we draw @xmath160 concentric circles with increasing radii @xmath161 such that @xmath162 . then we find those points intersecting with the hyperbola @xmath155 where , @xmath163 is a positive number less than @xmath164 . in this manner we can get the desired signal set for the variables @xmath165 and @xmath166 . the signal set for the variables @xmath167 and @xmath168 can be obtained by considering a different hyperbola . this is illustrated in fig . [ fig_signal2dim ] shown at the top of the next page . based on empirical studies , we propose to choose points in the set * a * and set * b * alternatively on each circle . also , choosing the constants @xmath169 seems to give the largest coding gain . similar observations have also been made in @xcite for another class of linear stbcs applicable in co - located mimo systems . implementing these two suggestions makes the signal identical for all the @xmath7-groups and the resulting signal set for the case of @xmath106 points is shown in fig.[fig_signal2dim2 ] . note that the proposed signal set construction is far from general and the most general solution would correspond to the following equations : @xmath170 in general , we need to choose the parameters @xmath171 and the solution points such that the coding gain is maximized . based on the detailed illustration of signal set construction for @xmath7 relays antennas , a natural generalization of it to higher dimensions is given in construction [ cons_signalset ] as follows . [ cons_signalset ] suppose that we want a @xmath172 points signal set@xmath173 for the constructed design for @xmath125 relays . then the resulting signal set @xmath173 should be a cartesian product of @xmath7 signal sets in @xmath174 since we insist on @xmath7-group encodability . in our case , we choose all the four sets to be identical and each will contain @xmath175{q}$ ] points . let the signal points in @xmath174 be labeled as @xmath176{q}$ ] . if @xmath177 , for some integers @xmath178 and @xmath179 where @xmath180 , then @xmath181 is given by @xmath182=0\ \forall j\neq(q\ \mathrm{mod}\ 2^{\lambda-1})+1\\ \mathbf{p}_i[(q\ \mathrm{mod}\ 2^{\lambda-1})+1]=+r_{q+1},\ \mathrm{if}\ r=1\\ \mathbf{p}_i[(q\ \mathrm{mod}\ 2^{\lambda-1})+1]=-r_{q+1},\ \mathrm{if}\ r=2 \end{array}\ ] ] where , @xmath183{q}}{2}$ ] are positive real numbers such that @xmath184{q}}{2}-1 $ ] and @xmath185{q}}{2}}r_i^2=\frac{\sqrt[4]{q}}{2}$ ] . let @xmath186 and @xmath187 . thus the rate of transmission of this code will be @xmath188 bits per channel use . the design is given in . the corresponding four dimensional signal set is described below . @xmath189^t,~ \mathbf{p}_2=\left[\begin{array}{cccc}-r_1 & 0 & 0 & 0\end{array}\right]^t,\\ \mathbf{p}_3=\left[\begin{array}{cccc}0 & r_2 & 0 & 0\end{array}\right]^t,~ \mathbf{p}_4=\left[\begin{array}{cccc}0 & -r_2 & 0 & 0\end{array}\right]^t,\\ \mathbf{p}_5=\left[\begin{array}{cccc}0 & 0 & r_3 & 0\end{array}\right]^t,~ \mathbf{p}_6=\left[\begin{array}{cccc}0 & 0 & -r_3 & 0\end{array}\right]^t,\\ \mathbf{p}_7=\left[\begin{array}{cccc}0 & 0 & 0 & r_4\end{array}\right]^t,~ \mathbf{p}_8=\left[\begin{array}{cccc}0 & 0 & 0 & -r_4\end{array}\right]^t,\\ \mathbf{p}_9=\left[\begin{array}{cccc}r_5 & 0 & 0 & 0\end{array}\right]^t,~ \mathbf{p}_{10}=\left[\begin{array}{cccc}-r_5 & 0 & 0 & 0\end{array}\right]^t,\\ \mathbf{p}_{11}=\left[\begin{array}{cccc}0 & r_6 & 0 & 0\end{array}\right]^t,~ \mathbf{p}_{12}=\left[\begin{array}{cccc}0 & -r_6 & 0 & 0\end{array}\right]^t,\\ \mathbf{p}_{13}=\left[\begin{array}{cccc}0 & 0 & r_7 & 0\end{array}\right]^t,~ \mathbf{p}_{14}=\left[\begin{array}{cccc}0 & 0 & -r_7 & 0\end{array}\right]^t,\\ \mathbf{p}_{15}=\left[\begin{array}{cccc}0 & 0 & 0 & r_8\end{array}\right]^t,~ \mathbf{p}_{16}=\left[\begin{array}{cccc}0 & 0 & 0 & -r_8\end{array}\right]^t \end{array}\ ] ] where , @xmath190 , @xmath191 , @xmath192 , @xmath193 , @xmath194 , @xmath195 , @xmath196 , @xmath197 . though the dimension of the signal set is @xmath7 , due to the special structure of the signal set , we can study the two dimensional projections of the signal points which is graphically shown in fig . [ fig_signal4dim ] at the top of the next page . the proposed multidimensional signal sets in construction [ cons_signalset ] are far from generality and obtaining a general solution targeting to maximize the coding gain appears to be difficult . [ thm_signalset ] construction [ cons_signalset ] provides fully diverse signals sets for the designs given in construction [ cons_design ] . the constructed designs for @xmath198 complex variables has the structure as shown in construction [ cons_doubling ] . the partitioning of the real variables into four groups is as follows . 1 . first group : @xmath199 2 . second group : @xmath200 3 . third group : @xmath201 4 . fourth group : @xmath202 the signal set is a cartesian product of four smaller dimensional signal sets @xmath203 . since and because the designs @xmath146 and @xmath147 are identical , it is sufficient to show that the design @xmath115 gives rise to fully diverse scaled unitary codewords for the chosen signal set . let the given signal set be denoted by @xmath204 . then the following properties are satisfied @xmath205 . @xmath206x[j]&=&0,~ 1\leq i\neq j\leq l\\ \label{eqn_fulldiv1 } 1\leq|\left\{i|~(x - y)[i]\neq 0 \right\}|&\leq&2\\ \label{eqn_fulldiv2 } ( x - y)[i]&\neq&\pm(x - y)[j],~ i\neq j\end{aligned}\ ] ] the design @xmath115 can be obtained by repeatedly applying abba construction on the linear design @xmath126 $ ] . the off - diagonal entries of the matrix @xmath207 are sum of terms like @xmath208 . but we have , which is equal to zero for the signal set @xmath209 by virtue of . thus we are guaranteed of scaled unitary codewords . due to the abba structure of the linear design @xmath115 , the determinant can be obtained in an iterative manner . let @xmath210 $ ] . then we have @xmath211\right|\\ & = & \left(|\mathbf{a_1}+\mathbf{a_2}|\right)\left(|\mathbf{a_1}-\mathbf{a_2}|\right ) . \end{array}\ ] ] the above equation suggests that @xmath212 can be computed recursively . for @xmath213 , . using this , we can easily show that @xmath212 is a product of @xmath116 terms . a typical term in the product looks like @xmath214 . thus to ensure @xmath215 , each term in the product should not equal zero . thus we need . let us look at the real part of this expression . we get @xmath216 . for the given signal set @xmath209 , from and , we have . thus the theorem is proved . [ thm_relay ] for the linear designs given by construction [ cons_design ] , there exist @xmath1 relay matrices satisfying condition for arbitrary signal sets . moreover , if the initial vector , then the initial matrix @xmath110 given in becomes unitary . see appendix [ appendix2 ] . [ eg_4relay ] let @xmath217 . then the ddstc @xmath62 is obtained using the design shown in and the signal set given in construction [ cons_signalset ] . the signal set is a cartesian product of four @xmath6-dimensional signal sets . the relay matrices are given as follows : @xmath218,\\ \mathbf{a_3}=\left[\begin{array}{ccrr}0 & 0 & -1 & 0\\ 0 & 0 & 0 & -1\\ 1 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\end{array}\right],~\mathbf{a_4}=\left[\begin{array}{ccrr}0 & 0 & 0 & -1\\ 0 & 0 & -1 & 0\\ 0 & 1 & 0 & 0\\ 1 & 0 & 0 & 0\end{array}\right ] . \end{array}\ ] ] the initial vector @xmath219^t$ ] and the initial matrix @xmath220 . this ddstc is single complex symbol decodable ( or 2 real symbol decodable ) . in this section , we compare the error performance of the proposed ddstc in example [ eg_4relay ] with that of the circulant codes in @xcite and the cyclic codes in @xcite for transmission rates of @xmath107 bit per channel use ( bpcu ) and @xmath221 bpcu . the signal set chosen for the proposed code is as given by construction @xmath222 with parameters @xmath223 , @xmath224 for a transmission rate of @xmath107 bpcu and for the case of rate=1.5 bpcu , the signal set parameters are @xmath225 , @xmath226 , @xmath227 and @xmath228 . the circulant code of @xcite is given by @xmath229 . the relay matrices @xmath230 are taken to be powers of the @xmath231 circulant matrix . the signal set @xmath232 is chosen to be @xmath233 radians rotated version of @xmath234-psk for a transmission rate of @xmath107 bpcu and @xmath235-psk for a rate of @xmath221 bpcu . the rotation angles are chosen as follows to guarantee full diversity : @xmath236 , @xmath237 , @xmath238 , @xmath239 . another reference for comparison is the cyclic codes proposed in @xcite whose codewords for rate=@xmath107 bpcu and rate=@xmath221 bpcu are given as follows : + @xmath240^i , i=0,\dots,255\right\}$ ] and @xmath241^i , i=0,\dots , 4095\right\}$ ] where , and @xmath242 . for simulations , we have assumed a block fading channel which is quasi - static for @xmath243 channel uses and varies independently from one block to another . shows the error performance curves of the proposed codes in comparison with those of @xcite . it can be observed from fig . [ fig_simulation ] that for a transmission rate of @xmath107 bpcu , the proposed code outperforms the cyclic code by about @xmath244 db and the circulant code by more than @xmath245 db . similarly for a transmission rate of @xmath221 bpcu , it can be observed that the proposed code outperforms the cyclic code by about @xmath244 db and the circulant code by more than @xmath246 db . it is worthwhile to note that for rate=@xmath107 bpcu , the decoding search space for the proposed code is only @xmath7 whereas it is @xmath247 for the other two codes . similarly for rate=@xmath221 bpcu , the decoding search space for the proposed code is only @xmath106 whereas it is @xmath248 for the other two codes . because of the exponential growth of decoding search space with increasing transmission rate for the codes in @xcite , it takes a prohibitively large time to complete a error performance simulation for higher rates . we have thus constructed a class of four group decodable ddstcs for any power of two number of relays using algebraic techniques . it is important to note that relaxing the unitary matrix codebook to scaled unitary matrix codebook has aided us in obtaining decoding complexity benefits . a limitation of this algebraic method is that it is available only for power of two number of relays . extending this method to obtain @xmath8-group decodable ddstcs for @xmath249 appears to be difficult for two reasons : 1 . the rate of the known linear designs in the literature for @xmath249 fall below @xmath107 and obtaining @xmath1 unitary relay matrices satisfying condition is difficult and it is not clear to the authors whether such matrices exist at all . for example , constructing the relay matrices satisfying condition for orthogonal designs is difficult . 2 . even if the relay matrices are obtained , an initial vector satisfying condition may not exist for the chosen relay matrices . for the specific case of @xmath250 , things were possible because the linear designs correspond to left regular matrix representation of an algebra over @xmath251 and hence using algebraic techniques , it was possible to construct the relay matrices and the initial vector satisfying all the required conditions . optimizing the signal sets for coding gain might be possible for small number of relays and for small number of points . however it becomes difficult for larger number of relays and/or larger number of points because it involves many parameters and a general closed form solution is difficult . hence optimizing the signal sets for coding gain is an important direction for further work . let @xmath252 . an extended clifford algebra denoted by @xmath253 is the associative algebra over @xmath254 generated by @xmath255 objects @xmath256 and @xmath257 which satisfy the following relations : * @xmath258 ; @xmath259 * @xmath260 ; @xmath261 * @xmath262 . the classical clifford algebra , denoted by @xmath263 , is obtained when only the first two relations are satisfied and there are no @xmath264 . thus @xmath263 is a sub - algebra of @xmath253 . let @xmath265 be the natural @xmath254 basis for this sub - algebra . @xmath266 then a natural @xmath254 basis for @xmath253 is : + @xmath267 we need a unitary matrix representation for the symbols @xmath107 , @xmath268 , @xmath269 , @xmath270 , , in the algebra @xmath271 . such matrices are naturally provided by the left regular representation . we first view @xmath271 as a vector space over @xmath251 by thinking of @xmath268 as the complex number @xmath272 . a natural @xmath251 basis for @xmath271 is given by : @xmath273 thus the dimension of @xmath271 seen as a vector space over @xmath251 is @xmath274 . we have a natural embedding of @xmath271 into @xmath275 ( the set of all @xmath251-linear maps from @xmath271 to itself ) given by left multiplication as shown below . @xmath276 since @xmath277 is @xmath251-linear , we can get a matrix representation of @xmath277 with respect to the natural @xmath251 basis @xmath278 . left regular representation yields unitary matrix representations for the required symbols in the algebra . the resulting linear designs are precisely those given by construction [ cons_design ] . this has been explicitly shown in more detail in @xcite . let us prove the theorem assuming that all the complex variables of the linear design take values from the entire complex field . we use the fact that the linear design for @xmath125 relays was obtained as a matrix representation of the extended clifford algebra @xmath279 . thus @xmath280 . we choose @xmath281 . the @xmath44 relay matrices are explicitly given by the union of the elements of the sets @xmath282 and @xmath283 . by virtue of the property that @xmath284 is a ring homomorphism , these matrices are guaranteed to commute with all the codewords because they are matrix representations of elements belonging to the center of the algebra @xmath279 . to obtain the remaining @xmath45 relay matrices , we need to find unitary matrices which satisfy @xmath285 where , @xmath12 is any codeword . but the codeword @xmath12 is simply a matrix representation of some element belonging to the extended clifford algebra . one method to get these relay matrices is to take them from within the extended clifford algebra itself . by doing so , we can translate the condition in into a condition on elements of the algebra which will then provide us a handle on the problem . towards that end , we first identify a map in the algebra which is the analogue of taking the conjugate of the matrix representation of an element . note from appendix [ appendix1 ] that we used the fact that @xmath268 can be thought of as the complex number @xmath272 . when we take the conjugate of a matrix , we simply replace @xmath25 by @xmath286 . hence the analogue of this action in the algebra is to replace @xmath268 by @xmath287 . thus , the analogous map @xmath288 in the algebra is defined as follows : @xmath289 where , the element @xmath290 is obtained from @xmath16 by simply replacing @xmath268 by @xmath287 in the expression of @xmath16 in terms of the natural @xmath291-basis of extended clifford algebra . now the problem is to find @xmath45 distinct elements denoted by @xmath292 of the algebra @xmath279 which satisfy @xmath293 . the elements of the union of the following two sets satisfy the above required condition . @xmath294 this can be proved by using the anti - commuting property , i.e. , @xmath295 . hence the matrix representation of these specific elements gives the unitary relay matrices @xmath296 . suppose we plug in these relay matrices to form a linear design where , @xmath297^t$ ] , it turns out that we get exactly the same linear design which is used at the source . thus the initial vector choice of @xmath219^t$ ] guarantees that the initial matrix @xmath110 is an identity matrix . however , we would like to point out that there are also other elements of the algebra which satisfy . for example , consider the union of the elements of the sets @xmath298 and @xmath299 . the authors thank prof . hamid jafarkhani , dr . yindi jing and dr . frdrique oggier for providing us with preprints of their recent works @xcite . the authors are grateful to the anonymous reviewers for providing constructive comments which helped in improving the presentation of this paper . f. oggier , b. hassibi , " cyclic distributed space - time codes for wireless relay networks with no channel information , submitted for publication . m. tao and r. s. cheng , differential space - time block codes , " proceedings of _ ieee globecom 2001 _ , vol . 2 , pp . 1098 - 1102 , san antonio , usa , nov 25 - 29 , 2001 . o. tirkkonen , a. boariu and a. hottinen , `` minimal non - orthogonality rate 1 space - time block code for @xmath300 tx antennas , '' in _ ieee int . symp . on spread - spectrum tech . & _ , new jersey , sep 6 - 8 , 2000 , pp . 429 - 432 . g. susinder rajan and b. sundar rajan , `` algebraic distributed space - time codes with low ml decoding complexity , '' proceedings of _ ieee intl . inform . theory _ , nice , france , june 24 - 29 , 2007 , pp . 1516 - 1520 . g. susinder rajan ( s2006 ) was born in chennai , india in 1983 . he completed his b.e . degree at the college of engineering guindy , anna university , india in 2005 . he is currently a ph.d . student in the department of electrical communication engineering , indian institute of science , bangalore , india . his primary research interests include space - time coding for mimo channels and wireless relay channels with an emphasis on algebraic code construction techniques . b. sundar rajan ( s84-m91-sm98 ) was born in tamil nadu , india . he received the b.sc . degree in mathematics from madras university , madras , india , the b.tech degree in electronics from madras institute of technology , madras , and the m.tech and ph.d . degrees in electrical engineering from the indian institute of technology , kanpur , india , in 1979 , 1982 , 1984 , and 1989 respectively . he was a faculty member with the department of electrical engineering at the indian institute of technology in delhi , india , from 1990 to 1997 . since 1998 , he has been a professor in the department of electrical communication engineering at the indian institute of science , bangalore , india . his primary research interests are in algebraic coding , coded modulation and space - time coding . rajan is an editor of ieee transactions on wireless communications from 2007 and also a editorial board member of international journal of information and coding theory . he is a fellow of indian national academy of engineering and is a member of the american mathematical society .

two dimensional integrable field theories have become an area of extensive study . their rich underlying mathematical structure allows for their exact solution . this in turn provides valuable information about the wide range of physical phenomena which integrable field theories can be used to model , and more generally about non - perturbative field theory . a number of models have been studied in the presence of a boundary . this has led to results for both the classical and quantum scattering of objects , like particles and solitons , off the boundary . in particular toda models @xcite , the sine - gordon model @xcite , and the sinh - gordon model@xcite amongst others have been studied on the half - line . in this paper we study the classical two dimensional complex sine - gordon ( csg ) model in the presence of a boundary . the model in the bulk is described by the following lagrangian @xmath0 where @xmath1 , @xmath2 , @xmath3 , @xmath4 . the field @xmath5 is complex , @xmath6 is a real coupling constant and the parameter @xmath7 can be changed by rescaling @xmath5 and will henceforth be set to one . the lagrangian is invariant under global phase rotations of the field @xmath5 and this leads to a conserved @xmath8 charge . as such , the complex sine - gordon model comprises an integrable generalisation of the sine - gordon theory , with an internal @xmath8 degree of freedom . it was first introduced independently by lund and regge as a model of relativistic vortices in a superfluid @xcite and by pohlmeyer in a dimensional reduction of a o(4 ) non - linear @xmath9-model @xcite . it belongs to a class of generalisations of the sine - gordon theory , which in the literature is referred to as homogeneous sine - gordon theories . the latter appear as @xmath10 gauged wess - zumino - witten models perturbed by a potential . these generalisations describe integrable perturbations of @xmath11 conformal field theories and have been studied in the bulk in @xcite . the csg model appears as the simplest case where @xmath12 and describes integrable perturbations of * z*@xmath13 parafermions by the first thermal operator @xcite . the quantum case was also studied by dorey and hollowood in @xcite , and maillet and de vega in @xcite . the csg model has found many applications in different fields of physics , from general relativity @xcite , to the description of propagating optical pulses in a non - linear medium @xcite . in the latter , the csg theory was used as a generalisation of the pioneering work of mccall and hahn @xcite , where the simple sine - gordon ( sg ) was used as a field theory description . the csg theory provides for a more realistic description of optical pulses incorporating effects like frequency detuning and modulation , while at the same time the theory can be extended beyond the description of two - level atom systems to multi - level atom systems . in the next section of this paper some aspects of the theory in the bulk will be reviewed . classical solutions , including particles , solitons and breathers will be presented for the model in the bulk clarifying previous treatments . unlike the sine - gordon model where the soliton solutions are topological in nature , the csg model solitons which carry a nether @xmath8 charge , have no obvious topological charge associated with them . this follows from the trivial vacuum structure of the model for @xmath14 , although as we shall explain chargeless solitons do carry the topological charge of the sine - gordon model in a subtle way . we also discuss the existence of breathers and find a class of charged breather solutions . in section 3 , we consider the effect of introducing a boundary and suitable boundary conditions are found which preserve the integrability of the model.in order to do this we explicitly construct low - spin conserved charges by abelianising the lax pair description of the model , and ensure that these are conserved on the half - line by imposing boundary conditions . we find a natural generalisation of the boundary conditions that ghoshal and zamolodchikov @xcite introduced for the sine - gordon theory . the final section in the main body of the paper deals with the scattering of particles and solitons . we find the corresponding solutions and use them to calculate the classical time - delay . we also find solutions corresponding to boundary bound states . the paper concludes with some general remarks about the results presented , as well as a few open questions that are related with some interesting properties of the model . in this section we shall review some important features of the csg theory and establish our notation . we present the equation of motion , the relation of the model with the sine - gordon theory and review the mathematical background in which the csg model can be regarded as a perturbed gauged wess - zumino - witten model . moreover vacuum , soliton and multi - soliton solutions are written down in a compact form . finally , we clarify the existing confusion concerning the existence of breathers within this model and present explicit breather solutions . the lagrangian of the model was presented in ( [ eq : csglagr2 ] ) . we assume without loss of generality @xmath15 since the parameter can be dropped by rescaling the field variables . the equation of motion that follows is @xmath16 the relation between this model and sine - gordon becomes obvious if we substitute [ eq : u2phi ] u = e^2 i , with @xmath17 and @xmath18 real fields . we use this notation here which gives a more standard connection with the sine - gordon theory but differs from previous treatments by @xmath19 . this takes ( [ eq : csglagr2 ] ) to [ eq : lagphi ] = + 4 ^2 -4 ^2 . this is essentially the form of the lagrangian derived by lund and regge , and pohlmeyer . by taking the field @xmath18 to be constant the sine - gordon lagrangian emerges . as was demonstrated by bakas @xcite the theory can be reformulated in terms of a gauged wzw action . the corresponding action principle can be written as [ eq : wzw1 ] s = s_gwzw+s_pot . the action term @xmath20 is the well known gauged wzw action @xmath21 this action is defined in a three - dimensional manifold @xmath22 whose boundary is our compactified normal two - dimensional space @xmath23 . the field @xmath24 is an @xmath25 group element and @xmath26 is the extension of @xmath24 to the three dimensional manifold . the last term introduces gauge fields @xmath27 and @xmath28 which act as lagrange multipliers . the @xmath29 term is s_pot = ( g _ 3 g^-1 _ 3 ) . this term breaks conformal invariance and thus gives rise to massive states . varying the action yields the csg equations of motion which can be expressed in a zero curvature form @xcite [ eq : lax ] [ + ( g + w g + i _ 3 ) , + ( |w- _ 3 g ) ] = 0 . from the variation of the gauge fields @xmath27 and @xmath28 , two constraint equations arise @xmath30 which are critical in order to make the identification with the csg theory . the connexion between the @xmath25 matrix @xmath24 and the complex field @xmath5 of ( [ eq : csgeq1u ] ) is given by [ eq : g ] g = ( rr u & -i + -i v & ) , where @xmath31 . the field variable @xmath32 should not be considered as an independent field but rather as an auxiliary field that is properly defined up to a constant through the constraint equations ( [ eq : const2 ] ) . in the gauge where @xmath33 , the constraint equations take the form = -i , = -i , whilst the equation of motion now becomes [ eq : lax2 ] [ - a , - ] = 0 , where [ eq : a ] a= - ( g + i _ 3 ) , = _ 3 g . this compact zero - curvature form of the equations of motion demonstrates the integrability of the model and will prove useful when we come to consider transformations and conserved quantities in later sections . from ( [ eq : csglagr2 ] ) it is easy to see that the energy of the csg model in the bulk is [ eq : bulkenergy ] h_bulk = dx ( + 4 u ) . the most suitable candidate for a vacuum , would be a constant value for the field @xmath5 that would force the kinetic term involving derivatives to vanish and at the same time minimize the potential term . it is clear to see from ( [ eq : u2phi ] ) and ( [ eq : lagphi ] ) that for @xmath14 the obvious choice is @xmath34 , while for @xmath35 the choice should be @xmath36 if we insist that @xmath37 . the sign of the parameter @xmath6 divides the theory into two sectors . in the matrix potential formalism , both sectors are treated simultaneously as the diagonal and off - diagonal parts of the field variable @xmath24 . in this context the fields @xmath5 and @xmath38 are both solutions to the csg equation each derived for a specific choice of @xmath6 , and each corresponding to a different vacuum . this is because the two sectors are connected by a duality transform which interchanges the sign of the coupling constant @xmath6 @xcite and simultaneously interchanges the role of @xmath5 and @xmath38 . that is to say the theory is invariant under the change [ eq : dual1 ] g g = i _ 1 g = ( cc v & i + i u & ) , - , the latter representing a transform akin to the krammers - wannier duality of the @xmath39 parafermion theory @xcite . taking into account the invariance of the theory under this duality transform , we shall concentrate in this paper on the @xmath40 sector which corresponds to the diagonal part of the matrix formalism . a suitable vacuum solution would be [ eq : vacuum ] g_vac = ( cc 0 & i e^- i + i e^i & 0 ) . this selection is consistent with the choices appearing in the beginning of this section with the diagonal @xmath14 sector , corresponding to the @xmath34 vacuum , while the off - diagonal @xmath35 to @xmath41 . it is noted that the apparent singular behaviour of the lagrangian at @xmath42 does not appear as a problem embedded in the theory but is a direct consequence of the fact that the gauge fields @xmath43 are ill defined at the specific point . the csg model , like the sine - gordon theory , possesses both particle and soliton solutions . when small perturbations around the vacuum are considered u = 0 + ( x , t ) , the theory becomes linear when higher order terms in @xmath44 are ignored @xmath45 where @xmath46 . the solution to the above equation is the familiar plane waves solution [ eq : planewave ] ( x , t ) = e^-i t ( a e^ikx + b e^-ikx ) , where @xmath47 and @xmath48 are related through @xmath49 different techniques have been used for the construction of soliton solutions like the inverse scattering method @xcite and the hirota method @xcite . however both methods yield results that are both cumbersome and difficult to manipulate . the transformation for the csg model provides a more elegant way to obtain soliton solutions and can be written in terms of two matrix variables @xmath24 and @xmath50 @xcite @xmath51 & = & 0 \ , \\ \db g \g \s - \s \db f f^{-1 } + \frac{\sqrt{|\b|}}{\delta}(g f^{-1 } \s - \s g f^{-1 } ) & = & 0 \ .\end{aligned}\ ] ] it is easy to show that both @xmath50 and @xmath24 satisfy the csg equation as well as the constraint equation in the specific gauge choice . taking @xmath50 to be an already known solution , one can generate a new solution through the equations presented above . one - soliton solutions can be derived by applying the transformation on the vacuum solutions @xmath52 of ( [ eq : vacuum ] ) . each sector of the theory provides us with two sets of two first order differential equations that can be integrated , in order to provide the one - soliton solutions . the diagonal elements of @xmath24 which correspond to the @xmath14 sector give @xmath53 the one - soliton solution that emerges is [ eq : usoliton ] u = , where @xmath54 and @xmath55 are real parameters associated with the velocity and charge of the soliton respectively . this solution was originally derived by getmanov @xcite for the @xmath56 case . in addition an expression for the phase @xmath32 which appears in its dual field @xmath38 , is also obtained [ eq : theta ] = - - ( ( a ) ( 2 ( a ) ) ) . respectively for the off - diagonal elements that correspond to the @xmath35 sector , the set of equations is @xmath57 that finally produce a different solution [ eq : vsoliton ] v = - e^i ( ( a ) ( 2 ( a ) ) + i ( a ) ) , with @xmath58 a real parameter associated with the vacuum of the theory . this is the solution that was derived by lund and regge @xcite when considering the @xmath59 case . a two - soliton solution can be obtained through a non - linear superposition technique . starting from the vacuum of the theory and by the application of the transformation twice , a set of parameters \{@xmath60 } is used respectively in each step . the same procedure is followed again where the two parameters are used in the opposite order . by demanding that the two results are equal , one ends up with an equation for the two - soliton solution in matrix form g_2s = _ 3 ( _ 1 g_2 - _ 2 g_1 ) g_vac _ 3 ( _ 1 g_1 ^ -1 - _ 2 g_2 ^ -1 ) ^-1 . the matrix field variables @xmath61 are of the general form of ( [ eq : g ] ) , with elements [ eq:1soliton ] u_k = , v_k = -e^i ( + i ) . the identification one must make is : @xmath62 where @xmath63 is a total phase . as expected @xmath64 has the same general form of equation ( [ eq : g ] ) @xmath65 the two - soliton solution and its complex conjugate are given by the diagonal elements of @xmath64 [ eq : u2s ] u_2s= , [ eq : ub2s ] _ 2s= , while the off diagonal elements represent the dual field and its conjugate v_2s= , _ 2s= . the expressions above represent two - soliton solutions to the equation of motion and are related through the duality transformation of ( [ eq : dual1 ] ) . multi - soliton solutions can also be obtained by following the same technique . instead of the vacuum solution , one can start from any given n - soliton solution @xmath66 and add solitons through the method described above , ending up with a @xmath67 solution . nevertheless , multi - soliton solutions for this model are quite large and their calculation is beyond the scope of this paper . in this section we argue that soliton - soliton and soliton - antisoliton solutions presented in the literature by previous treatments , do not represent distinct classes of solutions . charged solitons are non - topological solutions , therefore a distinction between a soliton and an antisoliton is impossible . on the other hand chargeless solutions may be realised as topological solitons and identified with the sine - gordon solitons . the sine - gordon theory appears as the limit of the csg model when the charge parameter @xmath55 is set to zero . we can substitute in the equation of motion of ( [ eq : csgeq1u ] ) [ eq : sg ] u= e^2i , where @xmath18 is now a constant to recover the sine - gordon model in the usual form ^2 - ^2 + 2 2 = 0 . the sine - gordon theory has topological solitons ( both kinks and antikinks ) interpolating between its degenerate vacuaa . in contrast the csg theory has a single vacuum for @xmath14 , and therefore its solitons are not topological in nature , but are stable because of integrability alone . the topological nature is hidden within the mapping of ( [ eq : sg ] ) and one has to be careful when trying to recover the sine - gordon soliton as a limit of the csg theory . nonetheless a subtle remnant of the topology survives the mapping to the complex sine - gordon theory . to see this consider how a sg soliton is mapped to csg soliton this is shown in fig.[fig : vacuum ] . consider now how the potential term behaves as @xmath68 increases for the single soliton solution . as a function of @xmath5 we can express the potential as 2 = 2 = 2 u , where @xmath18 has been ignored as a total phase . note that we should take the branch cut with opposite signs on each side of the point @xmath69 , @xmath70 . we shall see that the changing sign of the branch cut for a chargeless soliton will be important when we come to consider the theory with a boundary . in some sense the topology of the chargeless @xmath5-soliton is embedded in the the branch cut that appears at the singular point @xmath70 . the choice of branch corresponds to a different vacuum for @xmath17 and therefore to a different topological charge . however , when the charge parameter @xmath55 is not zero , then the @xmath5-soliton does not reach the sick point @xmath70 and remains non - topological . in this case no real distinction can be made between a soliton and an antisoliton . in the sine - gordon case , the antikink solution is derived from the kink by changing the sign of the parameter @xmath71 of the transformation . this effectively corresponds to a parity and time reversal transformation which finally produces an antikink solution . examining ( [ eq:1soliton ] ) we see that in the csg case this change actually leads to the complex conjugate solution , by changing the sign of the complex phase . the change of sign in both @xmath72 and @xmath68 , can be cancelled by taking the charge parameter @xmath73 . it is thus clear that instead of changing the sign of @xmath71 , one could effectively change the sign of @xmath55 to derive an antisoliton . since the soliton solution is a smooth function in @xmath55 , the antisoliton is not a distinct object but can be identified with the soliton itself . this also has an effect on the two - soliton solution . if we follow the same steps as in the sine - gordon two - soliton solution then the solution @xmath74 of ( [ eq : u2s ] ) corresponds to both a soliton - soliton and a soliton - antisoliton solution depending on the choice of sign for the parameter @xmath75 . the parameter @xmath75 can be chosen in such a way as to describe one of the following @xmath76 here we have taken the two solitons to have equal and opposite velocity ( in general this is not the centre of mass since differently charged solitons have different masses , but it will be convenient for our discussion when we introduce a boundary later on ) . also for reasons of simplicity we will refer to the soliton - soliton solution as @xmath77 ( @xmath78 ) and to the soliton - antisoliton as @xmath79 ( @xmath80 ) . however since no topological distinction exists between the soliton and antisoliton sector , it is possible to find a transformation of the parameters of the solution which effectively acts as a change of sign for the parameter @xmath75 . in fact , a set of transformations exists that maps @xmath77 to @xmath79 but we restrict ourselves to the simplest cases . before introducing the transformation , we need to introduce arbitrary shifts in @xmath68 , which are crucial not only for this mapping but also later when we consider breathers and soliton reflections . the shifts appear in exponentials , so it is more helpful to consider the shifts in the following forms @xmath81 the parameters @xmath82 are directly related with both @xmath54 and @xmath55 and correspond to the arbitrary initial positions in @xmath68 , in the real ( @xmath83 ) and imaginary phases ( @xmath84 ) respectively , that appear in the one - soliton solution . specifically the parameter @xmath85 represents a translation in @xmath68 , while the @xmath86 parameter represents a phase shift in the internal @xmath8 space . for reasons of simplicity , we include in the definition of @xmath86 the total phase @xmath87 which appears in ( [ eq:1soliton ] ) . henceforth these parameters will be referred as phase shifts , since they are directly related to the time - delay effect of the scattering process . now that we have defined the arbitrary phase shifts we start with the soliton - soliton solution @xmath88 which comes from the two - soliton solution @xmath74 when we choose @xmath89 . we consider the following transformation @xmath90 although this is enough to change a single soliton to an antisoliton , this is not the case for the two - soliton solution . the phase shifts have also to be fixed in a specific way to complete the mapping between @xmath77 and @xmath79 @xmath91 this effectively changes the sign of @xmath75 in the expression @xmath77 converting one of the solitons to an antisoliton . in contrast with the single soliton where the antisoliton can not be properly defined , in the two - soliton case there is a point of reference . a distinction between a soliton and an antisoliton can only be realised as a specific choice of the relative sign between the parameters @xmath92 and @xmath54 which does not in any case lead to topologically distinct solutions . the same mapping between @xmath77 and @xmath79 can also be achieved by making the following transformation a_2 a_2 + , which effectively changes the sign of all trigonometric functions involving the parameter @xmath93 sending the solution @xmath94 . this transformation will be used again on a later section when we come to consider soliton reflections , to demonstrate exactly the equivalence of the two sets of solutions . there are conflicting views in the literature concerning the existence of breathers @xcite . the problem arises because the transformation @xmath95 which is usually used to generate breathers from a two - soliton solution traveling with equal and opposite velocities , does not necessarily lead to a solution of the equations of motion . while the technique has been widely used before on other models , the fact that the csg equation involves both @xmath5 and @xmath96 , implies that naively analytically continued solutions do not necessarily satisfy the equation of motion . so it is not clear , for instance , that all the breather like `` solutions '' of @xcite do satisfy the csg equations of motion . however , since the sine - gordon is embedded in csg model by taking @xmath5 to be chargeless , the sine - gordon breather solutions do satisfy the csg equations of motion . in fact a family of charged , complex breather solutions does exist in csg model . although it is quite hard to actually check if a general breather solution satisfies the equation of motion , a trick can be used instead . we consider the two - soliton solution of ( [ eq : u2s ] ) and we demand that this solution is even in @xmath54 so that is effectively a function of @xmath97 . now the transformation @xmath98 , does nt change the reality properties of the solution but simply introduces an overall minus sign into the arbitrary parameter @xmath97 , which is irrelevant . making the solution even in @xmath54 , means that a few restrictions have to be imposed . firstly , the charge parameters have to be taken equal or opposite according whether @xmath99 is plus or minus one respectively . secondly , some of the arbitrary position parameters , have now to be fixed . however , up until now all the arbitrary phase shifts that appeared were either complex ( @xmath100 ) or real ( @xmath101 ) and there was no distinction between the shifts that originated from the space or time part of the phase . however , when constructing a breather solution , by analytical continuation of the @xmath54 parameter a separation between the space and time shifts is induced . all shifts that associated with space end up as real parameters , while time shifts become imaginary . we can restrict ourselves to shifts only in the @xmath68 direction . one could also consider more general phases which are complex and also depend on time and the parameter v. these however correspond to either @xmath8 rotations or time translations which make their use obsolete . the arbitrary shift parameters are now both real @xmath102 and should be compared with the general form of ( [ eq : shifts ] ) . in order to make a breather solution from the soliton - soliton case the following relations are required @xmath103 where the signs in these equation are correlated . it should be noted that more breather solutions may exist . it is possible that through certain restrictions a more general breather solution can be obtained , but a direct confirmation through the equations of motion is rather difficult . an analysis of the quantum csg model @xcite suggests that the soliton can be identified with the elementary particle since the vacuum of the theory and the one - soliton are not topologically distinct solutions . evidence of this conjecture exists even in the classical picture . from our experience with the sine - gordon model , we would expect to identify the particle with the lowest energy breather solution . it would seem to follow that the breather whose energy and charge correspond to that of a single particle should be equivalent to a single soliton . this remarkable fact can be shown as follows . we consider the static single - soliton solution [ eq : staticsoliton ] u_s = , where @xmath104 . the mass of the static soliton is given by ( [ eq : bulkenergy ] ) which after integration gives m_s = 4 m ( a_s ) , while the charge of the soliton is given by q_s = i = 4 ( [ a_s ] - a_s ) . the mass of a breather is twice the mass of a single soliton solution at velocity @xmath54 , which has been analytically continued @xmath105 m_b = . the breather solution is effectively constructed from two one - soliton solutions , each with charge q_b = 4 ( [ a_b ] - a_b ) . in order to have a chance of identifying the breather with the soliton , we demand that the mass of a breather is equal to the mass of a static single soliton and that their charges also coincide m_s = m_b , q_s = 2 q_b . from the above relations , one can solve for the parameter @xmath106 if this value is substituted into the breather , then the solution collapses to a static single - soliton carrying double the charge @xmath107 . in other words , the single - soliton can always be considered as a bound state of two single - solitons carrying half the charge . the argument can be used recursively so that a soliton can be regarded as an infinite collection of solitons carrying fractions of the original charge . at each level a soliton is identified with a breather emerging out of a soliton pair of half the original charge . in the classical picture this process can be carried out indefinitely , but in the quantum case the finite character of the mass states restricts this procedure . this is not surprising since the static single - soliton of ( [ eq : staticsoliton ] ) can be viewed as a bound state due to the oscillation effect which creates a breather - like behaviour . this is consistent with the fact that any breather can collapse to to this solution when the parameter @xmath54 is properly fixed . it can therefore be realised as a breather solution after the collapse , exhibiting all of its former properties . one point that has to be emphasized is that breathers constructed with the method described in the previous section are not chargeless . this is due to the fact that the choice of the charge parameters @xmath108 is such that both solitons that are combined to create a breather have the same charge . this is confirmed by the above demonstration in which a breather collapses to a single soliton solution which carries double the charge of the breather s solitons . neutral breathers do exist but only at the chargeless limit and can be identified with the breathers of the simple sine - gordon theory . in the following sections we consider the effect of introducing a boundary into the csg model . although we are free to add any boundary potential to the lagrangian , we choose to investigate those potentials and their corresponding boundary conditions which preserve integrability . such choices allow the non - perturbative solution of the model which is one of the main motivation for its study . once such suitable conditions are introduced we examine the scattering of particles and solitons , and determine the necessary conditions for the existence of boundary bound states . we shall consider a boundary condition to have preserved the integrability of the csg model , if we can still construct an infinite number of commuting conserved charges . in contrast with the theory in the bulk , the introduction of a boundary destroys the translation invariance of the model but preserves the time translation invariance . it is thus expected that the momentum will not be conserved , whilst the energy will . this situation also holds for the higher - spin conserved quantities . all energy - like , parity - even quantities can be conserved , unlike their momentum - like , parity - odd partners . nevertheless , since there is an infinite number of conservation laws , the main goal would be to concentrate on the conservation of the parity - even quantities . the presence of the spectral parameter @xmath109 in the lax pair of ( [ eq : lax2 ] ) implies the existence the infinite conserved currents in the bulk that can be determined through the method used by turok and olive @xcite . this is achieved by performing a gauge transformation @xmath110 [ eq : gauge2 ] = u a u^-1 + u u^-1 , in such a way that the commutator of the transformed gauge fields @xmath111 and @xmath112 of the lax pair to be zero . the equation of motion becomes ( | - ) = ( + ) , where the normal time and space derivatives are used . in the theory in the bulk we integrate over @xmath68 . if @xmath113 , then q( ) = ^_- ( | - ) dx = ^_- ( + ) dx = ^_- . since at infinity the fields are taken to vanish so that @xmath111 and @xmath114 approach a fixed value , it follows that the right - hand side vanishes . as @xmath111 and @xmath112 can be expanded as an infinite laurent series in @xmath115 , the coefficients of each power of @xmath115 , provides us with an infinite number of conserved charges . when a boundary is introduced the left - hand side involving the spatial derivative does not vanish since now the integration takes place over the semi - infinite interval . instead one is left with an equation of the form ^0_- ( |a - a ) dx = _ x=0 where the left - hand side is evaluated at the boundary . instead of demanding that the right - hand side vanishes , we instead ask that it can be expressed as a total time derivative with the help of suitable conditions , thus leading to a conserved quantity . we begin by finding explicit expressions for `` low - spin '' conserved charges of the csg model in the bulk by solving for the abelianizing gauge transformation @xmath110 order by order in the spectral parameter . let @xmath110 be a general real @xmath25 matrix , with @xmath116 . the diagonal elements of @xmath110 can be taken equal due to residual gauge freedom which leave @xmath111 and @xmath114 in an abelian form . thus @xmath110 takes the form u= ( cc 1 & + ) , where @xmath117 is a function of the fields and should not be associated with the space variable . we demand that @xmath110 diagonalises both @xmath118 and @xmath119 at the same time . the transformed fields lie both in the @xmath120 direction and the non - zero diagonal elements can be identified with the conserved currents . taking @xmath118 to be a= ( cc i & e + - & -i ) , with @xmath121 and @xmath122 , we demand that the non - diagonal part of @xmath123 vanishes @xmath124 the conserved quantities can also be written in terms of @xmath117 and @xmath125 j= -i - + . the same matrix @xmath110 , should also diagonalise @xmath126 , which is given by ( [ eq : a ] ) |a = ( cc d & p + & -d ) , where @xmath127 and @xmath128 . the choice of @xmath129 , @xmath130 and @xmath131 is not accidental . they actually represent the electric field , the polarization and the population inversion field variables respectively , when this theory is used to describe the propagation of optical pulses in a non - linear medium @xcite . when @xmath110 acts on @xmath119 , we again demand the off diagonal parts to vanish . examining the matrix explicitly yields @xmath132 it is easy to see that these equations are equivalent to equations ( [ eq : offdiag1 ] ) . the diagonal part yields the other component of the conserved current |j = ( 2 d + 2 ) + . in order to solve the two sets of equations ( [ eq : offdiag1 ] ) or equivalently ( [ eq : offdiag2 ] ) , we consider an expansion of @xmath117 and @xmath125 in powers of @xmath133 @xmath134 the coefficients @xmath135 and @xmath136 can be determined by direct substitution into ( [ eq : offdiag1 ] ) and ( [ eq : offdiag2 ] ) , and by demanding that the coefficients in all powers of @xmath133 vanish . up to order o(@xmath137 ) one finds @xmath138 now that @xmath117 and @xmath125 have been defined , we can also express the conserved quantities as a series in @xmath115 . each order of @xmath115 , provides a conserved quantity and since the series of @xmath115 in @xmath117 and @xmath125 does not terminate , we thus have an infinite number of conserved quantities as expected from the integrability of the csg model . the two components of the conserved current up to @xmath139 ) can be read off as coefficients in the following expansion of @xmath86 and @xmath140 @xmath141 and it can be checked that this current is conserved explicitly from the equation of motion . in the above we have constructed conserved currents that lead to conserved charges in the bulk . however , as we have previously argued , conserved charges on the half line are expected to take the form of an integral over a _ parity - even _ conserved current . the conserved currents above are neither parity even or odd . to rectify this we note that our system of equations and constraints possess a @xmath142 invariance involving parity transformations which can be used to construct a reflected " set of conserved currents . the reflected " set of conserved currents is easily obtained through the substitution @xmath143 in the expressions ( [ eq : currents1 ] ) including those derivatives involved in the definition of @xmath129 . the new set of currents @xmath144 , @xmath145 , can now be combined with the former set to produce pure parity odd and even currents . in the presence of a boundary only parity even quantities are conserved . the desired form of the equations to emerge is ( parity even ) = ( parity odd ) . by combining the two sets of currents one can separate the odd and even quantities for all powers of @xmath109 . = we examine the @xmath146 term in the expansion which gives ( e + + 2 ( d+ ) ) = ( - e + 2 ( d- ) ) , where @xmath147 , etc . after integration over the semi - infinite interval , the right hand side representing the parity odd part is ( u , ) = ( ) u + ( ) . this is a total derivative provided that = , = . the conserved quantity at hand , in terms of @xmath5 and @xmath96 , is then [ eq : conserv1 ] h = _ -^0 ( 2 + 4 ( 2 |u|^2 - 1 ) ) dx - [ w]_x=0 , since this quantity actually represents the energy of the system , @xmath148 can be identified with the energy contribution of the boundary term . when constructing the odd and even quantities of the @xmath149 term , one ends up with @xmath150 once more the parity - odd right hand side which after integration yields @xmath151 should be written as a total time derivative in order to force the currents to be conserved at the boundary . a set of boundary conditions have to be introduced to ensure that his is the case ( app . [ app1 ] ) . using the equations of motion , the parity - odd part of ( [ eq:2ndrhs ] ) can be written as a total derivative if the following restrictions are enforced @xmath152 the boundary constant @xmath153 is defined by the theory and is responsible for the way fields react to the boundary . consistency of the two equations in ( [ eq : bc ] ) implies that @xmath153 should be considered a real parameter . when one makes the transformation described in ( [ eq : u2phi ] ) , the new boundary conditions for the fields @xmath17 and @xmath18 are = - c ( ) , = 0 , which clearly shows that @xmath153 has to be real . it has to be pointed out that ( [ eq : bc ] ) is not the only set of boundary conditions that can be derived . a number of isolated dirichlet "- like conditions also exists . however , we restrict ourselves only to cases where the space derivatives of the fields appear . if we take the field @xmath5 to be real , the system is reduced to the sine - gordon equation with a boundary condition @xmath154 . this is the subset of integrable boundary conditions of the sine - gordon model presenting the @xmath142 symmetry @xmath155 . the corresponding conserved quantity , is rather large and is omitted . since the necessary conditions for the integrability of the model have been established , we study the scattering of particles and solitons off the boundary . we begin this section with the effects of introducing a boundary potential to the vacuum of the theory . we continue with the scattering of particles and solitons and derive the phase shifts induced by the process . finally , we investigate the necessary conditions for the existence of boundary bound states . when a boundary term is introduced , the vacuum of the theory that we discussed in section ( 2.2 ) , does not necessarily remain unchanged . it is exactly this contribution that needs to be carefully examined before any statements are made about the minimum energy configuration . although , the vacuum solution of the theory in the bulk is a strong candidate , soliton solutions could also be considered in the attempt to both minimize the energy functional and satisfy the boundary conditions of ( [ eq : bc ] ) . we begin by first determining the energy contribution of the boundary term . the full lagrangian of the model is now _ tot = l + _ b . the boundary term @xmath156 , can be determined by the variation principle of the total action . the variation of the @xmath157 term yields = u + + ( _ u ) + ( _ ) . when the euler - lagrange equations are used , two terms survive since the model is considered in the semi - infinite interval where the fields do not vanish at the boundary = _ ( u ) + _ ( ) . from the variation of the boundary term one has = u + . the variation of the action vanishes when the remaining terms evaluated at the boundary are forced to cancel . the two interrelated equations that emerge are @xmath158 by substituting the boundary conditions of ( [ eq : bc ] ) , these can easily be solved for the boundary term _ b = 2 c . we now consider the total energy of the system , now comprising of two parts [ eq : totenergy ] h_tot = h_bulk + h_b , where the term @xmath159 , represents the energy in the bulk and the second term @xmath160 represents the energy contribution from the boundary [ eq : hb ] h_b = - 2 c , which is evaluated at @xmath161 . this energy contribution makes the determination of the vacuum difficult . the sign of the boundary constant @xmath153 is not set , which could provide either a positive or negative contribution to the total energy of the system . this clearly shows that although the original choice for a vacuum should not be discarded , one should also consider other static solutions which in conjunction with the sign of @xmath153 could provide a lower energy vacuum than before . apart from the original choice for a vacuum , one can consider static multi - soliton solutions . we restrict ourselves to one - soliton solutions since experience with similar models usually makes multi - soliton solutions unsuitable candidates . when considering one - soliton solutions , one has the equations of the transformation ( [ eq : back1 ] ) which are always true to simplify expressions . in particular we first consider the @xmath162 term representing the energy in the bulk _ bulk = dx ( + m^2 u ) , with @xmath163 . by direct substitution of the equations a simplified expression of the bulk energy is acquired [ eq : bulkenergy2 ] h_bulk = dx ( 2 m^2 u ) . when the above expression is integrated throughout space , the result can be identified with the mass of the soliton solution @xmath5 . however , now the integration is over the half line and specifically over the @xmath164 $ ] region . the same equations can be used to express the @xmath165 term of ( [ eq : hb ] ) . specifically , the boundary constant @xmath153 is determined by direct comparison of the equations of ( [ eq : back1 ] ) and the boundary condition which appears in ( [ eq : bc ] ) c = ( e^i(+ ) + e^-i(+ ) ) . with @xmath32 given by ( [ eq : theta ] ) . at @xmath161 and assuming that @xmath166 , the above expression simplifies to [ eq : c ] c = . this implies that @xmath167 . since both @xmath168 and @xmath153 are defined by the theory , the above relation is true only for specific choices of the boundary parameter @xmath153 . alternatively , one can think of this restriction emerging from the fact that for @xmath169 , no choice of @xmath170 satisfies the boundary condition . in the case where we choose @xmath34 as a possible vacuum , the only remaining term in the total energy is _ tot = - 2 c . alternatively , one can consider a one - soliton solution where v is set to zero which appears in ( [ eq : staticsoliton ] ) . in this case , both terms of ( [ eq : totenergy ] ) depend on the initial position of the soliton . however , after some calculations , the @xmath170 dependence drops out and the total energy is given by a the following expression _ tot = 2 m ( a ) . it is far from obvious , which vacuum choice provides the minimum energy configuration . to determine this , one has to look at the expression of the boundary constant @xmath153 in ( [ eq : c ] ) . this can be rewritten in the following form [ eq : vfun ] y^2 + = ^2(a ) , where f^2 = ^2 ( m ( a ) x_0 ) , y = . in the above relation @xmath168 and @xmath153 should be treated as fixed parameters , while @xmath171 can be varied through @xmath170 . the left hand side of ( [ eq : vfun ] ) is monotonically decreasing as @xmath171 increases since @xmath172 . we observe the following @xmath173 this shows that moving the soliton away from the boundary decreases the energy of the system . on the extreme case where the soliton is placed at infinity , the model behaves as if no soliton exists , and the only contribution is the boundary term which coincides with the vacuum solution of @xmath34 . on the contrary as the soliton is placed closer to the boundary the energy increases . the maximum energy occurs when @xmath174 at which point @xmath175 so that @xmath176 which is greater than the energy @xmath177 of the @xmath34 vacuum . although the choice of vacuum in the bulk seems to be the most suitable choice in the boundary case too , one can not rule out multi - soliton solutions that might provide lower energy configurations . this demands tedious calculations and remains as one of the open questions for this model . in this section we investigate the reflection of solitons from the boundary . mathematically this can be represented by a two - soliton solution satisfying the boundary condition . one of the solitons represents the incoming soliton whilst the other represents the reflected one . the point where the two solitons actually meet along the whole line as well as the phase shift due to their collision create an overall time - delay effect which can be calculated directly through the parameters of the scattering . this time - delay can be attributed to the interaction of the soliton with the boundary . however , the most difficult step is to determine the restrictions that have to be imposed so that the two - soliton solution satisfies the boundary condition [ eq : bc_2s ] u_2s = - c u_2s . energy and charge conservation laws demand that both the mass and the charge of the soliton are conserved by the boundary . this restricts the choice of the charge parameters @xmath178 to be either equal or opposite . due to the large expressions involved in the calculation , one is forced to expand both sides of the equation ( [ eq : bc_2s ] ) to a taylor series in exponentials of @xmath72 , and match each term of the same order . each term provides us with an equation involving the boundary parameter @xmath153 . as mentioned in the previous section , the boundary constant has to be a real parameter . the real and imaginary parts of the equation yield two constrains on the parameters . let us consider this in more detail . we begin with a two - soliton solution , where the parameters are chosen in such a way so as to describe a soliton - soliton scattering . in this case , the charge parameters are taken to be opposite @xmath179 and the parameters to be @xmath180 . furthermore , we adopt the following parametrisation which is more natural = e^ , = e^i , v = ( ) . after both sides of the boundary equation are expanded as a taylor series in time , we can discard the imaginary parts from all terms by using the following relation [ eq : res1 ] ( ) = - ( ) ( a ) ( ) . when the above equation is used the infinite set of equations collapse to a single constraint [ eq : res2 ] c = m ( a ) ( ) ( ) . when the shift parameters are fixed according to the above relations , the two - soliton solution satisfies the boundary condition and this process describes a soliton being reflected by the boundary . the fact that only relative shifts in both normal and internal @xmath8 space are important should be expected from time translational and @xmath8 invariance of the model . the non - topological solitons in the csg theory are reflected as solitons carrying the same charge @xmath181 . this is because the boundary potential does not break the @xmath8 symmetry since it depends only on @xmath182 . we shall consider @xmath55 and @xmath183 as fixed parameters and use equations ( [ eq : res1 ] ) and ( [ eq : res2 ] ) to determine the parameter @xmath115 in terms of @xmath153 . by eliminating the parameter @xmath184 , one ends up with the following quadratic equation for @xmath153 ^2 - 2+(()^2+(a)^2 ) m^2 = 0 . the solutions of the above equation can be plotted to present the dependence on @xmath115 . the plot involves two branches ( fig . [ plot1 ] ) due to the sign ambiguity , which are mutually exclusive . the plot shows that a soliton can always be reflected by the boundary . the branches meet at the points c= m , ( ) = . in the limit @xmath185 the two branches of the plot can be identified with the soliton - soliton and soliton - antisoliton sector of the reflection process at the sine - gordon limit ( fig . [ plotsg ] ) . for fixed values of @xmath183 and @xmath186 , it is the value of the boundary constant @xmath153 which determines whether a soliton is reflected as a soliton or an antisoliton . for @xmath153 small , a soliton is reflected as an antisoliton ( neumann boundary conditions for @xmath187 ) , while for @xmath153 large a soliton is reflected as a soliton ( dirichlet boundary conditions for @xmath188 ) . for @xmath189 the branches do not meet as in the csg case . this specific value of @xmath153 corresponds to a logarithmic divergence that appears in the classic time delay for the sine - gordon case . these results coincide with the results derived by previous treatments of the boundary sine - gordon model @xcite . the time delay which appears at the scattering of a soliton off the boundary , can be calculated directly from the asymptotic values of the solution at @xmath190 . we begin with the two - soliton solution and change to a frame of reference which moves with the incoming soliton ( i.e. @xmath191 . in the limit @xmath192 the solution becomes [ eq : sminus ] s_- = _ t- u_2s = , where a_1= , ( b_1)= - and p = , + r = . the parameters @xmath193 and @xmath194 represent regular shifts that were introduced in ( [ eq : shifts ] ) . the solution , as expected , describes a single incoming soliton at early time far away from the boundary . we repeat the same calculation , but now we change to the frame of reference of the outgoing soliton ( i.e. @xmath195 ) and calculate the limit of the two - soliton solution at @xmath196 which yields [ eq : splus ] s_+ = _ t=+ u_2s = , where a_2 = , ( b_2)= , and q = . once again this is a single soliton solution representing the reflected soliton far away from the boundary wall . the asymptotic solutions @xmath197 , @xmath198 contain all the information needed to calculate the time - delay . the latter is a combination of two separate events . firstly , a phase shift is induced during the scattering of the two solitons . before the two solitons re - emerge as two separate entities , the reconfiguration of the solution creates a phase shift which is equivalent to a time delay . secondly , the centre of mass of the two - soliton solution does not necessarily lies at the boundary . this implies that the two solitons actually meet at a different point than @xmath161 . this creates again a time delay which may be either positive or negative corresponding to an attractive or repulsive boundary potential respectively . ignoring any interaction between the two solitons , one can project the trajectories of @xmath197 and @xmath198 on @xmath199 diagram and find the point where these cross ( fig . [ fig : scat1 ] ) . the distance of this point from the boundary is proportional to the time delay @xmath200 which in the diagram is given by the distance ( ac ) . the time delay corresponds to the time interval in which the soliton appears to be absorbed by the boundary before it reemerges as a well defined entity . the two solitons move across the following lines @xmath201 as dictated by ( [ eq : splus ] ) and ( [ eq : sminus ] ) . the lines cross at x_0 = ( x_1 - x_2 + ) = + , since @xmath202 . the time delay is finally _ csg = = + . in the expression above , the first term of the right - hand side represents the time delay caused by the non - symmetric character of the solution with respect to the boundary . in the special case where @xmath203 , the centre of mass lies on the boundary and the term vanishes . the second term is independent of the initial position of the two solitons or the boundary potential and is caused by the phase shift of the scattering process . the relative position of the two solitons are however fixed according to the constraint equations ( [ eq : res1 ] ) and ( [ eq : res2 ] ) which ensure that solution satisfies the boundary condition . specifically the parameter @xmath115 corresponds exactly to the @xmath204 difference up to the overall factor @xmath130 . it is thus possible to express the time delay in terms of the boundary constant , by solving the constraint equations and substituting the relative position of the solitons . we choose to express the velocity parameter @xmath54 in terms of the rapidity @xmath183 for simplicity reasons . after a few straightforward calculations we recover the following expression for the time delay _ csg = where q= in the special limit of @xmath186 , the time delay for the sine - gordon model is recovered _ sg = ( ^2 ( ) ) . this is exactly the time delay calculated for the sine - gordon theory in the presence of a boundary @xcite for the restricted class of boundary conditions which admit @xmath205 as a vacuum to which the chargeless limit of csg correspond . in this section we examine the spectrum of bound states . once again , for the the boundary condition to be satisfied we need to restrict some of the parameters in the solution . the simplest bound state that we can have is the static single soliton that was introduced in ( [ eq : staticsoliton ] ) . the solution is not really static , as the imaginary phase survives the setting of the speed parameter @xmath54 to zero . the solution is static only in the sense that the centre of mass does nt translate in the @xmath68 direction , although the wave oscillates with fixed angular velocity @xmath206 . when a boundary is introduced a static soliton can satisfy the boundary condition for @xmath207 when its position is fixed according to equation ( [ eq : c ] ) . at the chargeless limit any time dependence vanishes and the solution collapses to a static single soliton of the sine - gordon theory , fixed at the boundary . breathers that have been constructed by the method described in section ( [ sec : e1 ] ) can also be shown to satisfy the boundary condition . the condition that @xmath153 is real still holds . however , all the arbitrary phase shifts are now real numbers and constrained . we examine breather solutions that emerge from the soliton - soliton case . just as before , the solution does satisfy the boundary condition with some restrictions involving the arbitrary parameters . once more a taylor expansion of the boundary equation is needed . the parametrization used in this case is k_1=e^ , j_1 = e^ , v= ( ) , while the parameters @xmath208 and @xmath209 have been been properly fixed so that this is a breather solution . the first restriction needed for the solution to satisfy the boundary condition is ( 2 ) = - ( a ) ( ) ( 2 ) . the parameter @xmath183 plays the role of the rapidity , which has now been analytically continued . the second restriction which completes the necessary requirements for a boundary bound state is c= m ( a ) ( ) ( ) . both relations can be recovered by analytical continuation of the corresponding relations of ( [ eq : res1 ] ) and ( [ eq : res2 ] ) after the necessary restrictions for a breather solution have been already taken into account . it is instructive to examine the relation between the parameters @xmath153 and @xmath115 ( fig . [ fig : clbreather ] ) since it provides valuable insight to the structure of bound states . there are two distinct regions of values of @xmath153 that do not correspond to any bound state . this regions are defined by the limit values c = m ( ( a ) ( ) ( a ) ( ) ) . at the chargeless limit , the regions collapse to the single values c = m , which coincided with the logarithmic divergence appearing in the time delay for the soliton reflection . in this section we consider the spectrum of particles and their reflection factors in the presence of a boundary . for small fluctuations around the vacuum @xmath34 , the boundary condition ( [ eq : bc ] ) becomes _ 1 ( x , t ) = - c ( x , t ) . in order to calculate the reflection factor when particles bounce of the boundary wall , we substitute in the last relation the particle solutions presented in ( [ eq : planewave ] ) . the constant @xmath118 of the right propagating waves is taken to be one , since it has to do with the characteristics of the particle beam . the reflection factor is identified with the constant @xmath22 , which corresponds to a phase change as the particles encounter the boundary b = . the reflection factor , as expected , depends on @xmath153 which as stated before , appears as a free , real parameter in the boundary condition . for @xmath210 , the reflection factor is equal to @xmath211 and no phase appears between the two waves upon their scattering off the boundary . this is consistent with the fact that the boundary term is proportional to the boundary constant , so when @xmath153 is set to zero , the boundary term vanishes . particle solutions can be related to bound states through the pole appearing in @xmath22 . indeed one may choose @xmath212 and apply this to a solution of the form @xmath213 . the remaining terms depend explicitly on the boundary constant ( x , t ) = e^-i(t+-icx ) . this solution is square integrable only for a specific range of values for the boundary constant . specifically if @xmath153 is positive then the solution is not square integrable since it is exponentially increasing as @xmath214 . when @xmath215 , then @xmath216 represents a square integrable exponentially decreasing solution as @xmath217 . it oscillates with constant angular velocity @xmath218 and is therefore a stable bound state . it can also be viewed as the tail of a static one - soliton solution satisfying the boundary condition , with the parameters adjusted in such a way its centre of mass goes to positive infinity . examining the condition ( [ eq : c ] ) for the static soliton to obey the boundary condition , this limit can be achieved as @xmath219 , i.e. we must take the charge is such a way that @xmath220 . finally in the region @xmath221 , the solution can increase exponentially in time . this shows that the vacuum solution @xmath34 is no longer stable . in fact the particle behaviour which corresponds to a small perturbation around the vacuum seems to be ever increasing . this instability can be understood through a rather impressive mechanism in which a chargeless soliton is emitted from the boundary , effectively changing the value of @xmath153 so that @xmath34 is now stable . recall from section ( [ sec : sad ] ) that for a chargeless soliton we should take the opposite sign for @xmath222 on each side of the centre of the soliton where @xmath42 . the instability can be viewed as a left moving chargeless soliton which approaches the boundary from @xmath223 . in the beginning while the soliton is far away from the boundary @xmath34 so that the boundary potential of ( [ eq : hb ] ) is @xmath224 . as the centre of the soliton passes through @xmath161 , the sign of the square root in the boundary potential changes . as the soliton moves to @xmath225 , @xmath5 returns to @xmath226 near the boundary but now we take the boundary energy with the opposite sign @xmath227 . effectively the sign of @xmath153 has been flipped to a positive value . the energy released from the boundary is @xmath228 , which is greater than the rest mass of a single chargeless soliton . at @xmath229 , the soliton is emitted with infinitesimal velocity . as @xmath153 decreases , more energy is given up by the boundary and the soliton can be emitted with larger @xmath54 . this process agrees with the infinite time - delay effect which was encountered in the soliton reflections section . the soliton emission represents the time reversal picture of that effect in the chargeless soliton case ( fig . [ plotsg ] ) . it follows that we need never consider the situation where @xmath221 . the csg model is one of the simplest generalisations of the sine - gordon theory , but nonetheless has a rich and fascinating mathematical structure . in the first part of the paper we have examined the spectrum of the theory in the bulk and written down explicit two - soliton solutions within the framework of the matrix potential . we also demonstrated how to construct breather solutions in an elegant way avoiding the problems that arise by the analytical continuation of the parameter @xmath54 . there are two ways in which soliton solutions of the csg model differ qualitatively from those of the sine - gordon model . firstly , the solitons can be charged , so that whilst there centre of mass remains fixed , the solutions are not static . the second feature is that the csg does not possess degenerate vacua , and so the solitons are non - topological . there is therefore no distinction between solitons and antisolitons which can be interchanged by a continuous variation of the charge parameter @xmath55 . nevertheless , the topological nature of the sine - gordon theory can be recovered as the choice of branch cut of @xmath230 in the chargeless limit as was demonstrated in section [ sec : sad ] . a direct consequence of the non - topological nature of the csg soliton is that the breather solution can collapse to a single soliton when the parameter v is properly fixed , tying in with the picture presented in @xcite that particle and solitons can be identified in the quantum limit . in the second part of the paper we introduced a boundary term in the csg lagrangian and demanded that the system remains integrable . first we constructed low - spin conserved quantities of the theory using abelianisation of the lax pair , and then derived suitable boundary conditions in order to preserve these . in the presence of a boundary , we examined the vacuum structure and showed that the bulk vacuum @xmath34 remained the true vacuum in the boundary case . soliton reflections off the boundary were also studied and the necessary constraint equations were written down in terms of the phase shift parameters . the set of equations was derived by demanding that the two - soliton solution satisfies the boundary condition . moreover the time delay induced by the scattering process was calculated in terms of the boundary constant c and was found to coincide in the chargeless limit with the time delay of the sine - gordon theory . finally we looked for classical solutions corresponding to boundary bound states . we found that it was possible to construct both bound soliton and bound breather solutions . we also found a bound state in the particle spectrum . this was unstable when parameter @xmath153 associated with the boundary energy was in the range @xmath221 . in this case the boundary emits a chargeless soliton , effectively changing the sign of the boundary parameter @xmath153 to @xmath231 . we end our discussion by pointing out a few aspects of the model that appear quite interesting and deserve further study . as mentioned in the introduction the csg theory was used to generalise the existing field theory approach of optical pulse propagating in a non - linear medium . a physical interpretation of the results appearing in this paper would be extremely interesting , especially the physical meaning of breather solutions and their application to physical geometries . an obvious extension of our results is to consider the quantum case of the boundary csg model . the s - matrix for the model in the bulk , which corresponds to perturbed @xmath39 parafermions , is known . it would be interesting to see if one can find a quantum reflection matrix , compatible with this s - matrix and with the classical results presented in this paper . as the simplest case in the family of homogeneous sine - gordon theories , the results might shed light on the more complicated models in the family . we would like to thank patrick dorey and roberto tateo for illuminating discussion during the preparation of this work . in order to construct an infinite number of conserved quantities when a boundary is introduced , we need to express the parity odd terms of the equations of motion as total time derivatives . the corresponding term of order @xmath232 is in terms of the fields @xmath5 and @xmath96 @xmath233 since we need the above expression to be a total time derivative we can eliminate any second order spatial derivatives of the fields by using the equations of motion of ( [ eq : csgeq1u ] ) @xmath234 we take advantage of the fact that we are free to add total time derivatives on this expression , since this represents a conserved quantity . the expression simplifies significantly by adding the following term ( 4 ) , which yields @xmath235 we are looking for boundary conditions that are of the form u = f(u , ) , = g(u , ) , where @xmath171 and @xmath236 are functions of the fields not involving derivatives . by direct substitution of the above into ( [ ap : l21 ] ) we get @xmath237 the expression above does represent a total derivative when all terms are forced to vanish by selecting suitable functions @xmath171 and @xmath236 . the two separate differential equations that appear involving the undefined functions @xmath238 can easily be solved to yield [ ap : l23 ] f(u , ) = s_1(u ) , g(u , ) = s_2 ( ) . in addition , the last two terms in ( [ ap : l22 ] ) imply that f = g , using the above relation and solutions of ( [ ap : l23 ] ) into the remaining terms of ( [ ap : l22 ] ) , we can determine the remaining undefined functions @xmath239 and @xmath240 . the final form of the boundary conditions are @xmath241 where c is a real constant . h. j. de vega , j. ramirez mittelbrunn , m. ramon medrano , and n. sanchez . the general solution of the 2-d sigma model stringy black hole and the massless complex sine - gordon model . , b323:133138 , 1994 .

distribution of licensed digital content ( e.g. , software , movies , music etc . ) has become increasingly popular in recent times . with this comes the need to protect the copyright of the distributor against unauthorized redistribution of the content ( _ piracy _ ) . to introduce the problem , we begin with an informal description . suppose the distributor has some content which he would like to distribute among a set of licensed users . one can think of a simple scheme where each licensed copy is identified by a unique mark ( _ fingerprint _ ) which is embedded in the content and is imperceptible to the users of the system . note that the distributed copies are identical except for the fingerprints . if a naive user distributes a copy of his fingerprinted content , then the pirated copy can easily be traced back to the guilty user and hence he will be exposed . tracing the guilty user becomes more difficult when a collection of users ( _ pirates _ ) form a _ coalition _ to detect the fingerprints and modify / erase them before illegally distributing the data . digital fingerprinting is a technique that assigns to each user a mark in a way that enables the distributor to identify at least one of the members of the coalition as long as its size does not exceed a certain threshold @xmath5 , which is a parameter of the problem . there are two main setups considered for the fingerprinting problem in the literature . the _ distortion _ setting is commonly used in applications relating to multimedia fingerprinting @xcite . in this model , the fingerprint is usually a `` covert signal '' which is superimposed on the original `` host '' data in such a way that the difference , or distortion , between the original and the fingerprinted copies is smaller than some threshold . the coalitions are restricted to creating a forgery which has distortion less than some threshold from at least one of the colluders fingerprinted copies . on the other hand , we have the _ marking assumption _ setting introduced in @xcite which will be our main interest in this paper . in this case , the fingerprint is a set of redundant digits which are distributed in some random positions ( unknown to the users ) across the information digits of the original content . the fingerprint positions remain the same for all users . it is assumed that these redundant digits do not affect the functionality of the content , while tampering with an information digit damages the content permanently . the motivation for this assumption comes from applications to software fingerprinting , where modifying arbitrary digits can damage its functionality . the coalition attempts to discover some of the fingerprint positions by comparing their marked copies for differences . if they find a difference in some position , it is guaranteed to be a redundant fingerprint digit . in the other positions , it could be either an information digit or a fingerprint digit . the _ marking assumption _ states that the coalitions may modify only those positions where they find a difference in their fingerprinted copies . hence , in analyzing this model , it becomes sufficient to just look at the fingerprint positions and ignore the information digits . the collection of fingerprints distributed to all the users of the system together with the strategy of decoding ( pirate identification ) used is called a _ code _ below . a code is said to be _ @xmath5-fingerprinting _ or _ collusion - secure _ against coalitions of @xmath5 pirates if the error probability of decoding approaches 0 as the code length tends to @xmath6 collusion - secure fingerprinting codes were introduced by boneh and shaw @xcite . it was shown in @xcite that for any single deterministic code , the probability of decoding error in the `` wide - sense '' formulation ( see section [ sect : cap - def ] ) is bounded away from zero . hence , it becomes necessary for the distributor to use some form of randomization , where the random key is known only to the distributor , in order to construct such fingerprinting codes . this paper also gave the first example of codes with vanishing error probability . further general constructions were proposed by barg et al . @xcite and tardos @xcite . the case of zero error probability was considered independently by hollmann et al . @xcite who termed them as codes with the _ identifiable parent property _ , or ipp codes . they were further studied in @xcite among others . in this paper , we are interested in computing the fundamental limits of the fingerprinting problem , i.e. , in establishing bounds on the capacity ( or maximum attainable rate ) of fingerprinting codes . we denote by @xmath7 the capacity of fingerprinting with @xmath8-ary codes against coalitions of size @xmath5 ( this quantity is defined formally later in the paper ) . the problem of determining the fingerprinting capacity was raised in @xcite . to date , only some lower bounds are known through constructions and existence results : @xmath9 @xcite ; @xmath10 @xcite ; @xmath11 @xcite . new capacity bounds of our paper are based on an information - theoretic view of the fingerprinting problem . they are established as follows . attainability results ( lower bounds ) are shown by random coding techniques which take into account the typical coalitions , i.e. , the coalitions that occur with high probability . this represents a new idea in fingerprinting which enables us to improve random choice arguments of various kinds used earlier in @xcite . for upper bounds we model fingerprinting as a multi - user communications channel a converse theorem for a transmission scenario that models some aspects of the fingerprinting problem is proved to establish an upper bound on the capacity of fingerprinting . it should be noted that a similar information - theoretic approach to finding the capacity of fingerprinting was previously studied in @xcite and @xcite . in @xcite , the authors obtain upper and lower bounds on the capacity of fingerprinting with distortion constraints as opposed to the marking assumption setting of this paper . paper @xcite uses the marking assumption setting , but it addresses a simpler problem whose results do not directly apply to fingerprinting . the rest of the paper is organized as follows . in section [ sect : cap - def ] , we recall the statement of the fingerprinting problem and give an information - theoretic formulation . we also prove several results related to the problem statement that justify various techniques used to derive bounds on the fingerprinting capacity later in the paper . in particular , lower bounds on @xmath12 are proved in section [ sect : cap - lb2 ] . sections [ sect : cap - ub - w ] and [ sect : cap - ub - s ] are devoted to upper bounds on @xmath7 for arbitrary @xmath13 and their specializations for @xmath14 in the case of the binary alphabet . random variables ( r.v.s ) will be denoted by capital letters and their realizations by low - case letters . the probability distribution of a r.v . @xmath15 will be denoted by @xmath16 if @xmath15 and @xmath17 are independent r.v.s , then their joint distribution is written as @xmath18 for positive integers @xmath19 @xmath20 will denote the collection of r.v.s @xmath21 and the shorthand @xmath22 $ ] will be used to denote the set @xmath23 boldface will denote vectors of length @xmath24 for example , @xmath25 denotes a vector @xmath26 and @xmath27 denotes a random vector @xmath28 the hamming distance between vectors @xmath29 will be written as @xmath30 we will denote the binary entropy function by @xmath31 and @xmath32 will represent the indicator function . let @xmath33 denote an alphabet of ( finite ) size @xmath34 let @xmath35 be the number of users in the system and let @xmath36 denote the length of the fingerprints . assume that there is some ordering of the users and denote their set by @xmath37 . let @xmath38 be a finite set whose size may depend on @xmath24 elements of the set @xmath38 will be called keys . for every @xmath39 let @xmath40 be an @xmath36-length _ code _ , i.e. , a pair of encoding and decoding mappings : @xmath41 @xmath42 where the decoder output 0 will signify a decoding failure . by definition , the fingerprinting system is formed by a _ randomized code _ , i.e. , a random variable @xmath43 taking values in the family @xmath44 . note that the dependence on @xmath36 has been suppressed in this notation for simplicity . the _ rate _ of this code is @xmath45 . the system operates as follows . the distributor chooses a key @xmath46 according to a probability distribution @xmath47 on @xmath38 and assigns the fingerprint @xmath48 to user @xmath49 . on receiving a forged fingerprint , the distributor uses the tracing strategy @xmath50 ( corresponding to the selected key ) to determine one of the guilty users . we need randomization because : ( a ) deterministic fingerprinting codes do not exist in certain formulations @xcite , and ( b ) we allow the family of encoders and decoders and the distribution @xmath47 to be known to all users of the system . the only advantage the distributor has is the knowledge of the particular key being used . this assumption follows the accepted standards of cryptographic systems where it is usually assumed that the encryption / decryption algorithms are publicly available and that the only parameter kept secret by the system s constructor is the key . the fingerprints are assumed to be distributed inside the host message so that its location is unknown to the users . the location of the fingerprints , however , remains the same for all users . a _ coalition _ @xmath51 of @xmath5 users is an arbitrary @xmath5-subset of @xmath52 following accepted usage , we will refer to the members of the coalition as `` pirates '' . the coalition observes the collection of their fingerprints @xmath53 and attempts to create a fingerprint @xmath54 that does not enable the distributor to trace it back to any of the users in @xmath51 . note that although the fingerprint locations are not available to the pirates , they may attempt to detect some of these locations by comparing their copies for differences . thus , coordinate @xmath49 of the fingerprints is called _ undetectable _ for the coalition @xmath51 if @xmath55 and is called _ detectable _ otherwise . the _ marking assumption _ states that for any fingerprint @xmath56 created by the coalition @xmath51 , @xmath57 in every coordinate @xmath58 that is undetectable . in other words , in creating @xmath56 , the pirates can modify only detectable positions . for a given set of observed fingerprints @xmath59 the set of forgeries that can be created by the coalition is called the _ envelope_. its definition depends on the exact rule the coalition should follow to modify the detectable positions : * if the coalition is restricted to use only a symbol from their assigned fingerprints in the detectable positions , we obtain the _ narrow - sense envelope _ : @xmath60 * if the coalition can use any symbol from the alphabet in the detectable positions , we obtain the _ wide - sense envelope _ : @xmath61 we remark that there are further generalizations of the rules above where coalitions are also allowed to erase the symbols in detectable positions . this generalization is not considered below ; we refer the interested reader to @xcite . in the following , we will use @xmath62 to denote the envelope from any of the rules or their generalizations mentioned above . [ rem : intro - bin ] the definition for a fingerprinting code depends on the envelope considered . therefore , different problems can arise for each definition of the envelope . the binary alphabet is of special interest because of its wide use in practical digital applications . for this special case , it is easy to see that the narrow - sense and wide - sense envelopes are exactly the same . suppose that the coalition @xmath51 uses a randomized strategy @xmath63 to create the new fingerprint , where @xmath64 gives the probability that the coalition creates @xmath56 given that it observes the fingerprints @xmath65 . a strategy @xmath66 is called admissible if @xmath67 let @xmath68 denote the class of admissible strategies . such randomized strategies model any general attack the coalition is capable of and also facilitate mathematical analysis . the distributor , on observing the suspect fingerprint @xmath56 , uses the decoder @xmath50 while using the key @xmath46 . then the probability of error for a given coalition @xmath51 and strategy @xmath66 averaged over the family of codes is defined as follows : @xmath69 where @xmath70 is the expectation with respect to the distribution @xmath71 [ def : cap - fingcode ] a randomized code @xmath43 is said to be @xmath5-_fingerprinting with @xmath72-error _ if @xmath73 we now formulate the fingerprinting problem as a communications problem in which the set of messages is identified with the set of users of the fingerprinting system . each message is mapped to a codeword which corresponds to the fingerprint of the user . any set of @xmath5 messages ( a coalition ) may be chosen , and they are transmitted over an _ unknown _ @xmath5-input - single - output channel defined by the strategy of the coalition . the class of possible channels will be defined by the marking assumption . the output of the channel ( that represents the strategy ) gives the forged fingerprint . the task of the decoder is to recover at least _ one _ of the transmitted messages to have produced the channel output . observe that this information - theoretic model differs from the traditional @xmath5-user multiple - access channel ( mac ) because : ( a ) the decoder makes an error only when its output does not match _ any of the transmitted messages _ , and ( b ) all channel inputs are required to use the same codebook . for a given @xmath5-user strategy @xmath74 the _ maximum _ probability of error is given by @xmath75 note that here the users @xmath76 are not necessarily distinct . it is straightforward to see that the @xmath5-fingerprinting condition ( [ eqn : cap - fingdef ] ) can now be expressed as @xmath77 [ def : cap - capall ] for @xmath78 a number @xmath79 is an @xmath80-_achievable rate _ for @xmath8-ary @xmath5-fingerprinting if for every @xmath81 and every @xmath36 sufficiently large , there exists a randomized @xmath8-ary code @xmath43 of length @xmath36 with rate @xmath82 and maximum probability of error satisfying ( [ cap - eqn : maxpe ] ) . the @xmath80-_capacity _ of @xmath8-ary @xmath5-fingerprinting @xmath83 is the supremum of all such @xmath80-achievable rates . the _ capacity _ of @xmath8-ary @xmath5-fingerprinting is the infimum of the @xmath80-capacities for @xmath84 , i.e. , @xmath85 to proceed with the capacity @xmath86 we wish to consider coalitions of size exactly @xmath87 first , given any @xmath5-user strategy @xmath74 define the maximum probability of error corresponding to _ coalitions of size @xmath5 alone _ as @xmath88 the capacity value @xmath89 corresponding to the above criterion can be similarly defined . [ prop : cap - prop]@xmath90 clearly , @xmath91 the proof of the opposite inequality is also almost obvious because any coalition of @xmath5 pirates can simply ignore any subset of @xmath92 pirates when devising a forged fingerpint @xmath93 a formal version of this argument is provided by lemma a.1 in the appendix . similarly to the above , let us consider the _ average _ error probability @xmath94 ( this quantity will be used in the derivation of upper bounds on the capacity @xmath7 ) and the probability @xmath95 for coalitions of size exactly @xmath5 . define the capacity @xmath96 with respect to the latter error probability . we make a remark on the relation between the average and maximum error criteria . in general , it is true that the average error criterion yields a higher capacity value compared to the maximum one . however , when randomization is allowed , it is well - known for single - user channels that the capacity value is the same for both the maximum and average error probability criteria ( cf . e.g. , ( * ? ? ? we now extend this argument to the current context of multi - user channels and fingerprinting to show that both ( [ eqn : cap - maxpe2 ] ) and ( [ eqn : cap - avgpe2 ] ) lead to the same capacity value . [ prop : cap - rccap ] @xmath97 a formal proof is available in the appendix . it follows because here we simply use a randomized code @xmath43 , which also includes all @xmath98 permutations of any specific realization of @xmath43 . because of the symmetry introduced by this , the error probability @xmath99 is the same for all coalitions for a given @xmath66 , and hence the average and the maximum probability are the same . in this section , we construct fingerprinting codes for @xmath100 with error probability decaying exponentially in @xmath36 and with higher rate than previous constructions . the improvement is obtained by tailoring the decoder for the _ typical _ coalitions , i.e. , the coalitions that occur with high probability . we will say that an event occurs with high probability if the probability that it fails is at most @xmath101 where @xmath102 is a positive constant . our aim is to construct a sequence of randomized codes @xmath103 with error probability @xmath104 decaying to zero . by proposition [ prop : cap - prop ] , it suffices to consider only coalitions of size exactly @xmath5 . suppose , for every @xmath105 there exists a set @xmath106 such that for any coalition @xmath51 of size @xmath2 the observed fingerprints @xmath107 belong to @xmath108 with high probability . we will refer to a set with this property as a _ typical _ set . thus , in constructing the required code it suffices to study the conditions that allow us to obtain vanishing probability @xmath109 for any coalition @xmath51 of size @xmath2 any typical @xmath5-tuple @xmath110 of observed fingerprints , and any forgery @xmath111 as @xmath112 our first result is a lower bound on the fingerprinting capacity with 2 pirates over the binary alphabet . [ thm : cap - lb-2 ] @xmath113 fix @xmath114 suppose that the encoding mapping @xmath115 assigns @xmath116 fingerprints to the users choosing them uniformly and independently from all @xmath117 different vectors . for @xmath118 the fingerprints will be distinct with high probability . given a small @xmath119 we define the typical set as the set of vector pairs which agree in @xmath120 positions , where @xmath121 .\ ] ] consider any two pirates @xmath122 and @xmath123 notice that their observed fingerprints form a typical pair @xmath124 with high probability for an arbitrarily small @xmath125 hence , @xmath124 agree in @xmath126 positions . to create a forged fingerprint @xmath56 the pirates must fill the remaining @xmath127 positions . let @xmath128 and @xmath129 then @xmath130 and therefore @xmath131 given a forgery @xmath132 the decoder only considers typical pairs @xmath124 from the codebook . namely , the decoder takes any pair of distances @xmath133 that satisfy ( [ eqn : cap - lb2 ] ) and constructs the full lists @xmath134 and @xmath135 of the fingerprints located at distances @xmath136 and @xmath137 from @xmath93 each pair @xmath138 is then discarded if they simultaneously disagree with @xmath56 in any position @xmath139 i.e. , @xmath140 all remaining pairs contain @xmath56 in their envelope . for each such pair @xmath141 the decoding is completed by choosing the pirate @xmath142 whose fingerprint @xmath143 has a smaller distance @xmath144 either user is chosen if @xmath145 . obviously , the fingerprints @xmath146 and @xmath147 that belong to the factual pirates will not be discarded by the above decoding algorithm . the following probabilistic analysis shows that for two innocent users , the decoder discards their observed fingerprints @xmath148 with high probability if the code rate @xmath149 indeed , for @xmath148 to be typical , they should agree in @xmath150 positions . in all these positions , @xmath151 should also agree with @xmath56 to fulfill the marking assumption . in each of the remaining @xmath127 positions , the vectors @xmath151 are represented by only two combinations , @xmath152 or @xmath153 the probability of choosing such a pair @xmath148 in our random code equals @xmath154 and has exponential order of @xmath155 for any @xmath156 furthermore , by the union bound , the total probability of choosing such a pair in a random code of size @xmath116 is at most @xmath157 this probability tends to @xmath158 exponentially fast for any rate @xmath159 similarly , consider a coalition @xmath160 that includes the fingerprint @xmath146 of an actual pirate and the fingerprint @xmath161 of an innocent user . recall that @xmath146 disagrees with @xmath56 in @xmath136 positions . then to be output instead of @xmath162 the fingerprint @xmath161 must agree with @xmath56 in these positions and disagree with it in another set of @xmath163 positions . the total probability of choosing such a fingerprint @xmath161 is at most @xmath164 since @xmath165 and @xmath163 , we have restriction @xmath166 in this case , the above probability tends to @xmath158 exponentially fast for any rate @xmath159 thus , at least one pirate will be chosen from each coalition , and with high probability , no remaining ( innocent ) users will be chosen as pirates . we note that considering typical coalitions enables us to improve the lower bound @xmath9 obtained in @xcite . next we establish a lower bound on the fingerprinting capacity with 3 pirates over the binary alphabet . [ thm : cap - lb-3 ] @xmath167 suppose again that the encoding mapping @xmath115 assigns @xmath116 fingerprints to the users choosing them uniformly and independently from all @xmath117 different vectors . for a triple @xmath168 let @xmath169:\;x_{1s}=x_{2s}=x_{3s}\},\]]@xmath170:\;x_{is}=x_{js}\},\hspace{0.5em}i , j=1,2,3,\hspace { 0.5em}i\neq j,\ ] ] and let @xmath171 , @xmath172 given a small @xmath173 we say that @xmath174 form a typical triple if @xmath175,\label{eqn : cap - lb3}\]]@xmath176.\label{eqn : cap - lb4}\ ] ] for any three users @xmath177 note that the observed fingerprints form a typical triple with high probability . using the same idea as before , we now take the observed fingerprints @xmath174 to be a typical triple . a forged fingerprint @xmath56 agrees with all the three fingerprints on @xmath178 and takes arbitrarily values @xmath179 on the remaining subset @xmath180\backslash\cl $ ] positions . let @xmath181 for @xmath182 note that every position in @xmath180\backslash\cl$ ] contributes 1 or 2 to the sum @xmath183 implying @xmath184 given a forged fingerprint @xmath132 the decoder considers only typical triples @xmath174 from the codebook . each triple is then discarded if the fingerprints in it simultaneously disagree with @xmath56 in any position @xmath139 i.e. , @xmath185 if a triple @xmath174 is left , decoding is completed by choosing the pirate whose fingerprint has the smallest distance to @xmath56 among @xmath186 . obviously , the fingerprints @xmath174 corresponding to the factual pirates will not be discarded by the decoder . the following probabilistic analysis shows that a randomly chosen code of rate @xmath187 enables the decoder to discard with high probability all typical triples @xmath188 of fingerprints formed by three innocent users . indeed , a typical triple can be identified only if the fingerprints in it simultaneously agree with @xmath56 in some subset of @xmath189 positions . to simplify our analysis in this case , we can even ignore the extra conditions ( [ eqn : cap - lb4 ] ) in any of the remaining @xmath127 positions . thus , we allow the vectors @xmath188 to take on any combination of binary symbols @xmath179 different from all zeros or all ones . given 6 such combinations , any typical triple is chosen with probability at most @xmath190 we further observe that the total probability of choosing such a triple in a random code of size @xmath116 equals @xmath191 and tends to @xmath158 exponentially fast for any rate @xmath192 . now consider a slightly more involved case when the decoder locates the pirate coalition @xmath174 along with another coalition @xmath193 that includes the fingerprint @xmath146 of an actual pirate and the fingerprints @xmath194 of two innocent users . in what follows , we prove that a random code of rate ( [ eqn : cap - lb6 ] ) satisfies at least one of the following two conditions : * the decoder chooses @xmath146 in the coalition @xmath193 with high probability . * the coalition @xmath193 has vanishing probability . recall that @xmath195 then an innocent user , @xmath161 say , can be output by the decoder if @xmath196 . the probability that any such @xmath161 is chosen among @xmath35 random codewords is obviously at most @xmath197 given a code of rate ( [ eqn : cap - lb6 ] ) , this probability vanishes if @xmath198 therefore , condition ( i ) above fails if @xmath199 now let us consider condition ( ii ) given this restriction . consider a typical coalition @xmath200 we have @xmath201:\;x_{1s}=z_{2s}=z_{3s}=y_{s}\}|,\]]@xmath202:\;z_{2s}=z_{3s}\neq x_{1s}\}|\ ] ] thus , the vectors @xmath194 have fixed values on the one subset of size @xmath203 where these vectors are equal to @xmath146 , and on the other non - overlapping subset of size @xmath204 , where the vectors @xmath194 are equal to the binary complement of @xmath205 according to conditions ( [ eqn : cap - lb3 ] ) and ( [ eqn : cap - lb4 ] ) , @xmath189 and @xmath206 in the remaining @xmath207 positions we have @xmath208 summarizing the above arguments , we conclude that the total probability of choosing such vectors @xmath194 in the random code is bounded above as @xmath209 straightforward verification shows that this quantity vanishes given conditions ( [ eqn : cap - lb3 ] ) , ( [ eqn : cap - lb4 ] ) , and ( [ eqn : cap - lb7 ] ) for a code of rate @xmath210 thus a random code of smaller rate ( [ eqn : cap - lb6 ] ) discards all mixed coalitions of the form @xmath193 with high probability . the last remaining case , of a mixed coalition @xmath211 , is analyzed in a similar fashion ( the analysis is simpler than the one above and will be omitted ) . in finding upper bounds on fingerprinting capacity ( here and also in section [ sect : cap - ub - s ] ) , we restrict our attention to _ memoryless _ coalition strategies in order to make the problem tractable and to obtain single - letter expressions . any upper bound on the capacity thus obtained will be also valid in the original problem . let @xmath212 denote the family of discrete memoryless channels ( dmcs ) @xmath213 with @xmath5 inputs that satisfy the marking assumption for a single letter , i.e. , @xmath214 note that the above definition corresponds to the wide - sense envelope @xmath215 defined in ( [ eqn : intro - wenv ] ) . for the narrow - sense envelope @xmath216 ( [ eqn : intro - nenv ] ) and other variations of the problem it is possible to define similar communication channels and study their upper bounds on capacity . observe that @xmath212 is a convex and compact set . let @xmath217 , called the `` state '' , be an index which identifies the particular @xmath218 . hence , we will often write @xmath219 for channels in @xmath212 . we model a coalition s strategy by a ( discrete memoryless ) arbitrarily varying channel ( avc ) , i.e. , the state of the channel can vary from symbol to symbol . for a given state sequence @xmath220 , the channel is given by @xmath221 we denote the family of such channels @xmath222 by @xmath223 since the state sequence @xmath224 completely identifies the channel , we will use @xmath225 to denote the error probability in ( [ eqn : cap - avgpe ] ) . [ thm : cap - ub - w ] let @xmath43 be a @xmath8-ary @xmath5-fingerprinting code with @xmath80-error ( @xmath226 ) of length @xmath105 rate @xmath227 and @xmath228 keys , such that @xmath229 then @xmath230 where @xmath231 @xmath232 are @xmath8-ary r.v.s , @xmath233 @xmath234 is a r.v . taking values over a set of cardinality @xmath228 and satisfying the markov chain @xmath235 and the maximization is over joint distributions @xmath236{l}p_{k x_{1 } \dots x_{t } } = p_{k } \times p_{x_{1}|k } \times\dots\times p_{x_{t}|k}\\ \text{with } p_{x_{1}|k } = \dots= p_{x_{t}|k}. \end{array}\ ] ] let @xmath38 be a set of keys and let @xmath237 be a family of codes with probability distribution @xmath47 over @xmath238 since @xmath43 is @xmath5-fingerprinting with @xmath80-error , it satisfies @xmath239 let @xmath240 be independent r.v.s uniformly distributed over the message set @xmath241 and let @xmath234 be a r.v . independent of @xmath242 and with probability distribution @xmath47 over @xmath38 . also , let @xmath243 fix some @xmath220 and let @xmath244 be such that @xmath245 . then , we have @xmath246 which follows from ( [ eqn : cap - perr ] ) . we also have the following markov chain @xmath247 now , @xmath248 because @xmath240 are independent and uniformly distributed over @xmath249 the second term in ( [ eqn : cap - ub - w1 ] ) can be bounded above as follows . define @xmath250 @xmath251 let @xmath252 @xmath251 since @xmath253 are known given @xmath254 @xmath255 equation ( [ eq : h ] ) holds true because @xmath256 are binary r.v.s and the term @xmath257 in ( [ eqn : cap - ub - w2 ] ) follows from the fact that there are at most @xmath258 remaining terms and each can be bounded above by @xmath259 using this in ( [ eqn : cap - ub - w1 ] ) , we obtain @xmath260 we now use the premise that @xmath261 together with ( [ eqn : cap - markov ] ) and the memoryless property of the channel , which results in @xmath262 moreover , since the above bound applies for every @xmath263 i.e. , for every @xmath264 @xmath265 because the minimization is over channels whose state may vary over @xmath212 for every letter . note that @xmath266 are independent and identically distributed ( i.i.d . ) given @xmath234 ( by ( [ eqn : cap - x1x2 ] ) ) . therefore , given @xmath267 for every @xmath268 $ ] , @xmath269 are i.i.d . hence , @xmath270 where the maximization is over joint distributions satisfying ( [ eqn : cap - ub - w3 ] ) . [ cor : cap - ub - w ] @xmath271 where @xmath232 are @xmath8-ary r.v.s , @xmath272 and the maximization is over joint distributions such that @xmath273 are i.i.d . as we prove only a min - max type result , it becomes sufficient to consider only `` fixed '' memoryless coalition strategies , i.e. , strategies that remain fixed at every symbol instead of varying arbitrarily . in the subsequent text , @xmath274 will denote the @xmath36-letter extension of a dmc @xmath275 consider any sequence of @xmath5-fingerprinting codes @xmath276 of rate @xmath277 and error @xmath278 where @xmath279 approaches 0 as @xmath36 increases . then @xmath280 fix some @xmath281 we find that ( [ eqn : cap - ub - wmain ] ) holds for every @xmath24 therefore by the arguments in theorem [ thm : cap - ub - w ] @xmath282 where both @xmath283 and @xmath284 approach 0 as @xmath285 considering the inner term , we note that @xmath286 where @xmath287 and @xmath288 are the coordinate and key which maximize the mutual information . the term on the r.h.s . is a function of @xmath289 for every @xmath268,$ ] @xmath269 are i.i.d . when conditioned on @xmath290 therefore this term is at most @xmath291 where @xmath292 are @xmath8-ary r.v.s with @xmath293 and the maximization is over i.i.d . r.v.s . finally , since ( [ eqn : cap - ub - w7 ] ) is true for every @xmath294 we obtain the stated result by taking @xmath285 consider the case where @xmath295 . we would like to evaluate the upper bound on @xmath296 given by corollary [ cor : cap - ub - w ] . computing the exact optimum in this formula is a difficult problem . instead of attempting this , we will use a particular channel @xmath297 in ( [ eqn : cap - ub - w ] ) and compute a maximum on the prior distribution @xmath298 for this channel . the resulting value of the rate gives an upper bound on capacity @xmath299 let @xmath297 be the `` uniform channel '' defined by @xmath300 where @xmath301 is the number of 1s among @xmath302 fig . [ fig : cap - unifch ] shows the uniform channel for @xmath303 intuitively this choice is the worst strategy of the coalition from the distributor s perspective . if @xmath273 are independent binary - valued r.v.s with @xmath304 and @xmath17 is the output of the uniform channel with inputs @xmath273 , we have @xmath305 and @xmath306 evaluating the maximum mutual information in ( [ eqn : cap - ub - w ] ) for this channel gives a closed - form upper bound : [ thm : cap - ub - w - bin ] @xmath307}\big\{h(p)-\sum_{i=0}^{t}\binom{t}{i}p^{i}(1-p)^{t - i}h\left ( \frac{i}{t}\right ) \big\}\label{eqn : cap - ub - w - bin}\\ & \leq\frac{1}{t\ln2}.\label{eqn : cap - ub - w - bin-2}\ ] ] a proof of the estimate ( [ eqn : cap - ub - w - bin-2 ] ) is given in the appendix . [ thm : cap - ub - s ] for any @xmath78 @xmath308 where @xmath232 are @xmath8-ary r.v.s , @xmath272 and the maximization is over joint distributions such that @xmath273 are independent . we borrow techniques from @xcite in this proof . the result is proved for the case @xmath309 to ease understanding . it is a straightforward extension for arbitrary @xmath87 in the proof , all logarithms are to the base @xmath34 consider a family of @xmath36-length codes @xmath237 for @xmath35 users with probability distribution @xmath47 over @xmath38 which is @xmath310-fingerprinting with @xmath80-error ( @xmath311 ) . therefore @xmath312 let @xmath313 denote the fingerprints and @xmath314 denote the decoding regions for @xmath315 @xmath316 then the above error criterion can be written as follows : for every @xmath317 @xmath318 fix some @xmath319 there exists a @xmath320 such that @xmath321 hereafter , we drop the superscript @xmath322 for simplicity . consequently , either @xmath323 must be true . let us assume ( [ eqn : cap - ub - s1 ] ) is true . we first find a subset @xmath324 of `` good '' pairs of users ( messages ) for @xmath275 define @xmath325 where @xmath326 is such that @xmath327 then @xmath328 next , we derive a subset @xmath329 of the `` good '' pairs where approximate independence holds between the fingerprints ( codewords ) corresponding to a pair of users ( messages ) uniformly distributed over this subset . this is needed to restrict the maximization in the final result ( [ eqn : cap - ub - s ] ) to joint distributions where the r.v.s are independent . [ lem : cap - wring ] @xcite let @xmath330 @xmath331 with @xmath332 @xmath333 then for any @xmath334 @xmath335 there exist @xmath336,$ ] where @xmath337 and some @xmath338 such that for @xmath339 \right\ } $ ] * @xmath340 , and * for all @xmath341 @xmath268,$ ] @xmath342 where @xmath343 is a pair of r.v.s with uniform distribution on @xmath344 applying lemma [ lem : cap - wring ] to @xmath324 as in ( [ eqn : cap - ub - s - defa ] ) with parameters @xmath345 @xmath346 we obtain @xmath347 for @xmath348 define @xmath349 observe that the subcode corresponding to @xmath350 is a `` good '' code for the single - user channel obtained by fixing the second input to @xmath351 thus , the single - user strong converse given below holds for this subcode . [ lem : cap - aug ] @xcite if @xmath352 is a code with codewords @xmath353 and decoding regions @xmath354 for the ( non - stationary ) single - user dmc @xmath355 such that for every @xmath315 @xmath356 @xmath357 then @xmath358 where @xmath27 is distributed uniformly on the set of codewords . using lemma [ lem : cap - aug ] on the subcode @xmath359 @xmath360 where @xmath343 are distributed as in lemma [ lem : cap - wring ] and @xmath361 furthermore , using ( [ eqn : cap - ub - s4 ] ) , we obtain @xmath362 since @xmath363 for @xmath268.$ ] we next establish a lower bound on the left - side term in order to obtain an inequality for @xmath364 @xmath365 now , @xmath366 by using ( [ eqn : cap - ub - s3 ] ) and ( [ eqn : cap - ub - s8 ] ) . using this inequality in ( [ eqn : cap - ub - ss1 ] ) , we get @xmath367 combining ( [ eqn : cap - ub - s5 ] ) , ( [ eqn : cap - ub - s7 ] ) and ( [ eqn : cap - ub - s8 ] ) , @xmath368 although the above inequality resembles what is needed in the theorem , note that @xmath369 and @xmath370 are not necessarily independent . for @xmath371,$ ] let @xmath372 be r.v.s with distribution @xmath373 for all @xmath374 from lemma [ lem : cap - wring](b ) , for @xmath375 and every @xmath268 $ ] @xmath376 @xmath377 thus , by the uniform continuity of mutual information , for all @xmath268,$ ] @xmath378 where @xmath379 as @xmath285 together with ( [ eqn : cap - ub - s9 ] ) and dividing by @xmath105 @xmath380 where @xmath381 as @xmath285 similarly , assuming ( [ eqn : cap - ub - s2 ] ) is true , one can prove @xmath382 since either ( [ eqn : cap - ub - s11 ] ) or ( [ eqn : cap - ub - s12 ] ) holds for every @xmath317 taking @xmath383 concludes the proof . fix @xmath295 . for the case of @xmath309 and @xmath384 we again pick the uniform channel and obtain upper bounds on the expression in theorem [ thm : cap - ub - s ] , which turn out to be stronger than the bounds resulting from ( [ eqn : cap - ub - w - bin ] ) . the calculations become quite tedious for larger @xmath87 for @xmath385 let @xmath386 be independent binary - valued r.v.s with @xmath387 and let @xmath17 be the output of the uniform channel with inputs @xmath388 and @xmath389 we have @xmath390 computing the maximum conditional mutual information gives @xmath391 a similar computation for @xmath392 yields @xmath393 and the maximization gives @xmath394 combining these upper bounds with our lower bounds from theorem [ thm : cap - lb-2 ] and theorem [ thm : cap - lb-3 ] we obtain : [ thm : cap-2bin ] @xmath395 @xmath396 in this paper , we prove new lower bounds on the maximum rate of binary fingerprinting codes for 2 and 3 pirates by considering typical coalitions which improves the random coding results obtained previously in the literature . we also prove several new upper bounds on fingerprinting capacity relying upon converse theorems for a class of channels which are similar to the multiple - access channel . our results establish for the binary case , @xmath397 combined with the result of @xcite this implies that @xmath398 for the general case with arbitrary alphabets , we have established some upper bounds on the capacity involving single - letter mutual information quantities . _ lemma a.1 : _ let @xmath43 be a randomized code of size at least @xmath399 assume that @xmath400 then for any @xmath401 @xmath402 for simplicity of presentation we take @xmath403 the general case of @xmath404 can be established with only minor changes to the proof below . for any @xmath405 let us define a @xmath406 where @xmath407 then , for any coalition @xmath51 of size @xmath408 and any user @xmath409 @xmath410 , \label{eq : n}\end{aligned}\ ] ] where @xmath411 the first term in the last equation satisfies @xmath412 by the assumption of the lemma . we will next show that the second term in ( [ eq : n ] ) is also at most @xmath413 suppose for the sake of contradiction that @xmath414 let @xmath415 and @xmath416 ( we assume that the size of the code is at least @xmath417 or at least @xmath418 in the general case ) . then @xmath419 but this contradicts our initial assumption ( [ eq : assumption ] ) . it is clear that @xmath420 therefore , it is enough to show that for every randomized code @xmath43 , there exists another randomized code @xmath421 of the same rate such that @xmath422 for every channel @xmath423 we are given @xmath424 let @xmath425 identify a particular permutation from the set of all permutations on the message set @xmath249 choose @xmath426 uniformly at random from @xmath427 and construct a new key @xmath428 define @xmath429 @xmath430 let @xmath421 be the randomized code corresponding to the family @xmath431 then , for every channel @xmath74 @xmath432 . furthermore , for any @xmath433 @xmath434 @xmath435 which does not depend on the subset @xmath51 because of the averaging over all permutations . this implies @xmath436 our goal is to estimate @xmath437}u(p , t),$ ] where we use the following notation@xmath438@xmath439 first , note that @xmath440 is a concave function , and therefore @xmath441 is non - negative for all @xmath442.$ ] bernstein proved that the sequence of polynomials @xmath443 where @xmath444 is a function continuous on @xmath445 $ ] , provides a uniform approximation to @xmath444 on @xmath445.$ ] his proof , found for instance in feller @xcite 7.2 , relies on the weak law of large numbers . refining the proof in the case of the function @xmath440 , we show that for any @xmath442 $ ] and any @xmath2 @xmath446 it suffices to consider the case @xmath447.$ ] given some @xmath448 let us write a quadratic taylor approximation for @xmath449 @xmath450 where the coefficient @xmath451 depends on @xmath452 since @xmath453 for some @xmath454.$ ] we shall also consider the residual function @xmath455 the main part of our proof is to show that for any @xmath456,$ ] @xmath457 the right inequality is obvious since @xmath458 for all @xmath459 the left inequality will be proven in two steps . let us take any point @xmath460.$ ] then we compare @xmath461 with the quadratic function @xmath462 on the entire interval @xmath463.$ ] we first prove that functions @xmath464 and @xmath465 coincide at only two points , namely @xmath466 and @xmath467 indeed , let us assume that there exists a third such point @xmath468 without loss of generality , let @xmath469 the functions @xmath464 and @xmath465 coincide at the ends of both intervals @xmath470 $ ] and @xmath471 $ ] ; therefore there exist two points @xmath472 and @xmath473 where both functions have equal derivatives : @xmath474 the left sides of both equalities represent a linear function of @xmath475 given by @xmath476 whereas the right sides represent a convex function @xmath477 a linear function can intersect a convex function at no more than two points . this leads to a contradiction , which shows that @xmath478 and that the functions @xmath464 and @xmath465 intersect at two points @xmath466 and @xmath467 our next step is to find the minimum @xmath479 for all @xmath463 . $ ] compare the function @xmath464 with @xmath480 for any parameter @xmath481.$ ] now we use the fact that both functions intersect @xmath482 at only two points , one of which is @xmath483 however , @xmath480 has its second intersection @xmath484 to the left of @xmath467 thus , @xmath485 for @xmath486 and therefore , @xmath487 ( see fig . [ fig : parabolas ] ) . now we conclude that @xmath488}a(x).\ ] ] , height=192 ] finally , we find @xmath489 using the equality @xmath490 which gives @xmath491 the second interval @xmath492 $ ] can be considered similarly@xmath4 again , we use the same arguments and conclude that the end point @xmath493 gives the minimum @xmath494}a(x).$ ] direct calculation also shows that the global minimum is achieved at @xmath158 as @xmath495 for all @xmath496 and @xmath497 for @xmath498 this gives us the left inequality in ( [ ld2 ] ) and shows that for any @xmath499 and any @xmath500,$ ] @xmath501 let us take @xmath502 and substitute the above estimate into the expression for @xmath503 in this substitution , we also use the first two moments of the binomial distribution @xmath504 which gives @xmath505 @xmath506 then @xmath507 finally , it is easy to verify that the function @xmath508 monotonically decreases on the interval @xmath509 $ ] and achieves its maximum @xmath510 at @xmath511 this establishes ( [ ld ] ) and hence the bound ( [ eqn : cap - ub - w - bin-2 ] ) . the authors would like to thank prakash narayan and gbor tardos for numerous illuminating discussions . r. ahlswede and n. cai , `` codes with the identifiable parent property and the multiple - access channel , '' _ general theory of information transfer _ , r. ahlswede at al . , lecture notes comput . , vol . 4123 , berlin : springer verlag , 2006 , pp . 249 - 257 . a. barg , g. r. blakley and g. kabatiansky , `` digital fingerprinting codes : problem statements , constructions , identification of traitors , '' _ ieee trans . inform . theory _ 852 - 865 , apr . a. barg , g. cohen , s. encheva , g. kabatiansky , and g. zmor , `` a hypergraph approach to the identifying parent property : the case of multiple parents , '' _ siam journal discrete math . 14 , no.3 , pp . 423 - 431 , 2001 . g. tardos , `` optimal probabilistic fingerprint codes , '' _ journal of the acm _ , to appear . preliminary version in _ proc . 35th annual acm symposium on theory of computing _ ( stoc 2003 ) , pp . 116 - 125 , 2003 .

over the past few years rapid progress has been made in the field of symbolic summation . the beginning was made by gosper s telescoping algorithm @xcite for hypergeometric terms and zeilberger s extension of it to creative telescoping @xcite . an algebraic clarification of gosper s setting has been carried out by paule @xcite . meanwhile various important variations or generalizations have been developed , like for @xmath0hypergeometric terms @xcite , the mixed case @xcite , or the @xmath1-finite case @xcite . in particular , karr s telescoping algorithm @xcite based on his theory of difference fields provides a fundamental general framework for symbolic summation . his algorithm is , in a sense , the summation counterpart to risch s algorithm @xcite for indefinite integration . karr introduced the so - called @xmath2-extensions , in which parameterized first order linear difference equations can be solved in full generality ; see below . as a consequence , karr s algorithm can not only deal with telescoping and creative telescoping over hypergeometric terms , but also over rational terms consisting of arbitrarily nested sums and products . more generally , it turned out that also parameterized linear difference equations can be solved in such difference fields @xcite . this enables to solve recurrence relations with coefficients in terms of indefinite nested sums and products ; it also gives rise to algorithms for rather general classes , for instance , holonomic sequences @xcite . an important general aspect of using difference field methods is the following : in order to exploit the full power of the algorithmic machinery , it is necessary to find for a summand , given in terms of indefinite nested sums and products , a `` good '' representation in a suitable @xmath3-field extension @xmath4 of @xmath5 ; note that some similar considerations for indefinite integration appeared in @xcite . based on the results of @xcite karr comes to the following somehow misleading conclusion @xcite : _ loosely speaking , if @xmath6 is summable in @xmath4 , then part of it is summable in @xmath5 , and the rest consists of pieces whose formal sums have been adjoined to @xmath5 in the construction of @xmath4 . this makes the construction of extension fields in which @xmath6 is summable somewhat uninteresting and justifies the tendency to look for sums of @xmath7 only in @xmath5 . _ in other words , following karr s point of view , one either succeeds to express a given sum of @xmath6 in @xmath5 , or , if one fails , one adjoins the sum formally to @xmath5 which leads to a bigger field @xmath4 . but , it turns out that karr s theory of difference field extensions can be refined . namely , as shown below , his strategy in general produces sum representations that are not optimal with respect to simplification ; see , e.g. , examples [ exp : truncatednaive ] and [ exp : depthincreaseexp ] . as a measure of simplification we introduce the notion of nested depth . and the main part of this article deals with the problem of finding sum representations which are optimal with respect to this property . based on results of @xcite we develop a refined version of karr s summation theory , which leads to the definition of the so called depth - optimal @xmath3-extensions . various important properties hold in such extensions which are relevant in symbolic summation . moreover , an efficient telescoping algorithm which computes sum representations with optimal nested depth is presented . throughout this article all these ideas will be illustrated by one guiding example , namely the identity @xmath8 which was needed in @xcite to generalize identities from statistics . we stress that our algorithms are of particular importance to simplify dalembertian solutions @xcite , a subclass of liouvillian solutions @xcite , of a given recurrence ; for applications see , e.g. , @xcite . furthermore , we obtain a refined version of creative telescoping which can find recurrences with smaller order ; for applications see , e.g. , @xcite . in addition , we show how our algorithms can be used to compute efficiently algebraic relations of nested sums , like harmonic sums @xcite @xmath9 @xmath10 . we illustrate by concrete examples @xcite from perturbative quantum field theory how our algorithms can evaluate efficiently feynman diagrams . the general structure of this article is as follows . in section [ sec : symbsum ] we introduce the basic summation problems in difference fields . in section [ sec : mainresults ] we present in summarized form our refined summation theory of depth - optimal @xmath3-extensions in which the central results are supplemented by concrete examples . some first properties of depth - optimal @xmath3-extensions are proven then in section [ sec : mainrespart ] . after considering a variation of karr s reduction technique in section [ sec : reduction ] we are ready to design algorithms to construct depth - optimal @xmath3-extensions in section [ sec : constructdeltaext ] . as a consequence we can prove the main results , stated in section [ sec : mainresults ] , in section [ sec : mainproofs ] . finally , we present applications from particle physics in section [ sec : application ] . let @xmath5 be a _ difference field _ with field automorphism @xmath11 . note that @xmath12 forms a subfield of @xmath5 ; we call @xmath13 the _ constant field _ of the difference field @xmath14 . subsequently , we consider the following two problems : 1 . ( _ sequence representation _ ) : * given * sequences @xmath15 ; try to * construct * an appropriate difference field @xmath14 with elements @xmath16 where the shift - behavior @xmath17 for @xmath18 is reflected by @xmath19 . ( _ parameterized telescoping _ ) : * given * @xmath14 with @xmath20 and @xmath21 ; * find * all @xmath22 and @xmath23 such that @xmath24 then reinterpreting such a solution @xmath23 with @xmath22 in terms of a sequence @xmath25 gives @xmath26 which then holds in a certain range @xmath27 . hence , summing this equation over @xmath28 gives @xmath29 if we restrict to @xmath30 in and search for a solution with @xmath31 , we solve the telescoping problem : * given * @xmath7 ; * find * @xmath23 such that @xmath32 moreover , zeilberger s creative telescoping @xcite can be formulated by translating @xmath33 into @xmath34 for a parameter @xmath35 which occurs in the constant field @xmath36 . karr s summation theory @xcite treats these problems in the so - called @xmath2-difference fields . in our work we restrict to @xmath3-extensions @xcite being slightly less general but covering all sums and products treated explicitly in karr s work . those fields are introduced by difference field extensions . a difference field @xmath37 is a _ difference field extension _ of a difference field @xmath38 if @xmath5 is a subfield of @xmath5 and @xmath39 for all @xmath7 ; since @xmath40 and @xmath41 agree on @xmath5 , we usually do not distinguish between them anymore . a difference field extension @xmath42 of @xmath14 is a _ @xmath43-extension _ ( resp . _ @xmath44-extension _ ) , if @xmath45 is transcendental over @xmath5 , @xmath46 ( resp . @xmath47 ) for some @xmath48 and @xmath49 ; if it is clear from the context , we say that @xmath45 is a @xmath43-extension ( resp . a @xmath44-extension ) . @xmath3-extension _ is either a @xmath44-extension or a @xmath43-extension . a _ @xmath44-extension ( resp . @xmath43-extension/@xmath3-extension ) _ @xmath50 of @xmath14 is a tower of such @xmath44-extensions ( resp.@xmath43-extensions/@xmath3-extensions ) . such an extension if defined over @xmath51 if @xmath51 is a subfield of @xmath5 and for all @xmath52 , @xmath53 or @xmath54 is in @xmath55 ; note : @xmath56 , if @xmath57 . a @xmath3-extension @xmath50 of @xmath14 is called _ generalized dalembertian _ , or in short _ polynomial _ , if @xmath58 $ ] or @xmath59 for all @xmath52 . @xmath3-field @xmath14 over @xmath36 _ is a @xmath3-extension @xmath14 of @xmath60 with @xmath61 . karr derived an algorithm @xcite that solves the following more general problem which under the specialization @xmath62 and @xmath63 gives . [ rem : sigmasit ] karr s algorithm or our simplified version @xcite can be applied if @xmath14 is a @xmath3-extension of @xmath64 where @xmath64 satisfies certain properties ; see @xcite . as a consequence , we obtain algorithms for problem pfde if @xmath64 is given as follows : ( 1 ) : : @xmath65 : as worked out in ( * ? ? ? 3.2,thm . 3.5 ) , we obtain a complete algorithm , if @xmath36 is as a rational function field over an algebraic number field . ( 2 ) : : the free difference field @xmath64 with @xmath66 , @xmath67 for @xmath68 , and @xmath36 is as in * ( 1)*. in this setting generic sequences can be treated ; see @xcite . ( 3 ) : : the radical difference field @xmath69 with @xmath70 and @xmath67 where @xmath71 ; @xmath36 is given as in * ( 1)*. this allows to handle @xmath72 ; see @xcite . ( 4 ) : : @xmath64 is a @xmath3-extension of one of the difference fields described in * ( 1)**(3)*. sum - product expressions can be represented in @xmath3-fields with the following result @xcite . [ thm : pisigma ] let @xmath42 be a difference field extension of @xmath14 with @xmath73 . ( 1 ) : : @xmath45 a @xmath43-extension iff @xmath74 and there is no @xmath23 s.t . . ( 2 ) : : @xmath45 is a @xmath44-extension iff @xmath75 , @xmath76 and there are no @xmath77 s.t . @xmath78 . consequently , we are allowed to adjoin a sum formally by a @xmath43-extension if and only if there does not exist a solution of the telescoping problem . the product case works similarly ; for further information and problematic cases we refer to @xcite . [ exp : truncatednaive ] we try to simplify the left hand side of by telescoping , or equivalently , by representing in a @xmath3-field . for simplicity of representation , it will be convenient to rewrite this expression as @xmath79 ( 1 ) : : we start with the difference field @xmath80 with @xmath81 for all @xmath82 , i.e. , @xmath83 is the constant field . since there is no @xmath84 with @xmath85 , we can define the @xmath43-extension @xmath86 of @xmath60 with @xmath70 . ( 2 ) : : since there are no @xmath87 and @xmath88 with @xmath89 ( for algorithms see @xcite ) , we can define the @xmath44-extension @xmath90 of @xmath86 with @xmath91 . similarly , we introduce the @xmath44-extension @xmath92 of @xmath90 with @xmath93 . by construction , @xmath40 reflects the shift in @xmath28 with @xmath94 and @xmath95 . ( 3 ) : : next , we try to simplify @xmath96 by telescoping . since we fail , i.e. , there is no @xmath97 with @xmath98 , we add the @xmath43-extension @xmath99 of @xmath92 with @xmath100 ; note that @xmath101 . ( 4 ) : : finally , we look for a @xmath102 such that + @xmath103 + since there is none , see example [ exp : truncatednaiverat ] , we adjoin the sum in form of the @xmath43-extension @xmath104 of @xmath99 with @xmath105 . summarizing , using this straight - forward approach the sum could not be simplified : the two nested sum - quantifier is reflected by the nested definition of @xmath104 . let @xmath14 be a @xmath3-extension of @xmath64 with @xmath106 where @xmath107 or @xmath108 for @xmath52 . the _ depth function for elements of @xmath5 over @xmath109 _ , @xmath110 , is defined as follows . 1 . for any @xmath111 , @xmath112 . if @xmath113 is defined for @xmath114 with @xmath115 , we define @xmath116 ; for @xmath117 , with @xmath118 $ ] coprime , we define @xmath119 for @xmath120 , @xmath121 . the _ depth of @xmath14 _ , in short @xmath122 , is given by @xmath123 . similarly , the _ extension depth _ of a @xmath3-extension @xmath124 of @xmath14 is @xmath125 . this extension is _ ordered _ if @xmath126 ; if @xmath127 , we call @xmath124 an _ ordered _ @xmath3-field . in the @xmath3-extension @xmath104 of @xmath60 from example [ exp : truncatednaive ] the depth ( function ) is given by @xmath128 , and @xmath129 . throughout this article the depth is defined over the ground field @xmath64 ; we set @xmath130 . we might use the depth function without mentioning @xmath109 . then we assume that the corresponding difference fields are @xmath3-extensions of @xmath64 . moreover , note that the definition of @xmath131 depends on the particular way the extension field @xmath5 is build from @xmath109 . we consider the sum and take the @xmath3-field @xmath99 with @xmath132 which we introduced in example [ exp : truncatednaive ] . now we proceed differently : we compute the @xmath3-extension @xmath133 of @xmath134 with @xmath135 in which we find the solution @xmath136 of ; for details see example [ exp : constructdepthoptext ] . note that @xmath137 and @xmath138 , in particular , @xmath139 . reinterpreting @xmath140 as a sequence and checking initial values produces . we emphasize that this way we have reduced the depth in since @xmath140 and the summand @xmath141 in have the same depth @xmath142 . this example motivates us to consider the following refined telescoping problem . [ exp : s12s21 ] our goal is to encode the harmonic sums @xmath143 and @xmath144 in a @xmath3-field . ( 1 ) : : first we express @xmath145 with the inner sum @xmath146 in the @xmath3-field @xmath147 with @xmath70 , @xmath148 and @xmath149 , i.e. , @xmath150 and @xmath151 represent @xmath152 and @xmath143 , respectively ; note that we failed to express @xmath143 in an extension with depth @xmath153 . ( 2 ) : : next , we consider @xmath154 . we start with @xmath155 and construct the @xmath43-extension @xmath156 of @xmath157 with @xmath158 . finally , we treat the sum @xmath144 and look for a @xmath140 such that @xmath159 . : : since there is no @xmath160 , we take the @xmath43-extension @xmath161 of @xmath156 with @xmath162 . : : we can compute the @xmath43-extension @xmath163 of @xmath164 with @xmath165 in which we find the solution @xmath166 . note that this alternative solution has the same depth , namely @xmath167 , but the underlying @xmath3-field is simpler . as result , we obtain @xmath168 to sum up , the following version of telescoping is relevant . we shall refine karr s theory such that we can find a common solution to dot and dot@xmath169 . a difference field extension @xmath170 of @xmath14 with @xmath171 is called _ depth - optimal _ @xmath43-extension , in short @xmath172-extension , if there is no @xmath43-extension @xmath37 of @xmath14 with extension depth @xmath173 and @xmath174 such that . a @xmath3-extension @xmath50 of @xmath14 is depth - optimal , in short a @xmath175-extension , if all @xmath43-extensions - extensions are @xmath43-extensions by theorem [ thm : pisigma].1 . note also that @xmath44-extensions are not refined here ; this gives room for further investigations ; see , e.g. , @xcite . ] are depth - optimal . a @xmath175-field consists of @xmath44- and @xmath172-extensions . our main result is that problems sr , dot and dot@xmath169 can be solved algorithmically in @xmath175-extensions . moreover , we will derive various properties that are of general relevance to the field of symbolic summation and that do not hold for @xmath3-extensions in general . _ in all our results 19 , stated below and proved in section [ sec : mainproofs ] , we suppose that @xmath14 is a @xmath175-extension of @xmath64 and @xmath176 . from an algorithmic point of view we assume that @xmath64 is @xmath40-computable : _ a difference field @xmath64 is _ @xmath40-computable _ , if one can execute the usual polynomial arithmetic of multivariate polynomials over @xmath109 ( including factorization ) , and if one can solve problem pfde algorithmically in any @xmath3-extension @xmath14 of @xmath64 . for instance , @xmath64 can be any of the fields given in remark [ rem : sigmasit ] . in our examples we restrict to the case @xmath177 , i.e. , @xmath14 is a @xmath3-field over @xmath109 . * 1 . construction . * problem sr can be handled algorithmically in @xmath175-extensions . [ exp : constructdepthoptext ] the @xmath3-field @xmath99 with is depth - optimal since dot with @xmath178 and @xmath179 has no solution . moreover , @xmath180 is a @xmath172-extension of @xmath99 with . with the solution @xmath136 of we represent the sum in a @xmath175-field ; for algorithmic details see example [ exp : truncatedfullrat ] . * 2 . reordering . * let @xmath181 be a @xmath3-extension of @xmath64 with @xmath107 or @xmath108 for @xmath52 . if there is a permutation @xmath182 with @xmath183 for all @xmath184 , @xmath185 is again a @xmath3-extension of @xmath64 and @xmath186 is isomorphic to @xmath187 as fields . in short , we say that @xmath181 can be _ reordered to _ @xmath185 ; on the field level we identify such fields . clearly , by definition of nested depth there is always a reordering that brings a given field to its ordered form , i.e. , @xmath188 for @xmath52 . note that reordering of @xmath175-extensions without destroying depth - optimality is not so obvious : putting @xmath43-extensions in front or removing them , might change the situation of problem dot@xmath169 . but one of our main results says that reordering indeed does not matter . [ exp : s12reordered ] let @xmath163 be the @xmath175-field from example [ exp : s12s21 ] . then , e.g. , the ordered @xmath3-field @xmath189 is depth - optimal . depth stability . * the following example illustrates the importance of result [ res : depthstable ] . [ exp : depthincreaseexp ] let @xmath161 be the @xmath3-field from example [ exp : s12s21 ] which is not depth - optimal . we find the solution @xmath190 of with @xmath191 . hence , @xmath192 . in other words , we obtained the identity @xmath193 where the depth is increased by telescoping ; compare . we remark that result [ res : depthstable ] can be exploited algorithmically : in order to find all solutions of , one only has to take into account those extensions with depth @xmath194 . * extension stability . * the most crucial property is the following : suppose we are given a @xmath172-extension @xmath195 of @xmath14 . then we can embed any @xmath43-extension @xmath196 of @xmath14 in a @xmath172-extension @xmath37 of @xmath195 without increasing the depth . the @xmath3-field @xmath14 from example [ exp : s12reordered ] with @xmath197 is depth - optimal , and @xmath195 with @xmath198 is a @xmath172-extension of @xmath14 . now consider in addition the @xmath3-extension @xmath196 of @xmath14 with @xmath199 ; see example [ exp : s12s21 ] . then we can take the @xmath172-extension @xmath37 of @xmath195 with @xmath200 , see example [ exp : s12reordered ] , and we can define the field homomorphism @xmath201 with @xmath202 for all @xmath7 , @xmath203 and @xmath204 . by construction , @xmath205 is injective and @xmath206 for all @xmath207 . in other words , we have embedded @xmath196 in @xmath37 with for all @xmath208 . more precisely , @xmath209 is called a _ @xmath40-monomorphism/@xmath40-isomorphism _ for @xmath14 and @xmath210 if @xmath205 is a field monomorphism / isomorphism with @xmath211 for all @xmath212 . + let @xmath14 and @xmath213 be difference field extensions of @xmath196 . an _ @xmath51-monomorphism_/_@xmath51-isomorphism _ @xmath209 is a @xmath40-monomorphism/@xmath40-isomorphism with @xmath214 for all @xmath208 . * depth optimal transformation . * by result [ res : extensiondepthstable ] any @xmath3-extension of @xmath14 can be transformed to a @xmath175-extension with the same or an improved depth - behavior . hence the refinement to @xmath175-extensions does not restrict the range of applications ; on the contrary , the refinement to @xmath175-extensions can lead only to better depth behavior . * product freeness . * @xmath44-extensions are irrelevant for problem dot . + * 7 . alternative definition . * thus we obtain the following equivalent definition . + * 8 . a common solution to dot and dot@xmath169 * can be found by result [ res : depthconstruct ] and + * 9 . refined parameterized telescoping . * + [ equ : creasol ] for @xmath215 we take the @xmath175-field @xmath216 over @xmath217 with @xmath70 , @xmath218 and @xmath219 ; then @xmath220 and @xmath221 can be represented by @xmath222 , respectively . we set @xmath223 and @xmath130 , and get @xmath224 . then we find the @xmath172-extension @xmath37 of @xmath225 with @xmath226 , @xmath227 and @xmath228 which fulfills the properties from result [ res : paratele ] . in particular , we get @xmath229 and @xmath230 s.t . . hence , for @xmath231 , we get @xmath232 . summation over @xmath28 from @xmath233 to @xmath234 gives @xmath235 for @xmath236 together with @xmath237 , we arrive at the recurrence relation @xmath238 and can read off the closed form @xmath239 note : only a recurrence of order two is produced for @xmath240 by using standard creative telescoping ; see ( * ? ? ? * how to proceed . * we will prove results 19 as follows . in section [ sec : mainrespart ] we first show weaker versions of results [ res : reordering][res : constructsigmaextontop ] ; there we impose that all @xmath175-extensions are ordered . after general preparation in section [ sec : reduction ] , these results allow us to produce theorem [ thm : extendforteleordered ] ( cf . result [ res : depthconstruct ] ) in section [ sec : constructdeltaext ] . given all these properties , we will show our results 19 in full generality in section [ sec : mainproofs ] . let @xmath14 be a difference field with @xmath20 , @xmath120 and @xmath241 . we write @xmath242 and @xmath243 . let @xmath244 be a subspace of @xmath5 over @xmath36 . the _ solution space _ for @xmath245 and @xmath246 in @xmath244 is defined by @xmath247 it forms a subspace of the @xmath36-vector space @xmath248 . in particular , note that the dimension is at most @xmath249 ; see @xcite . if @xmath250 , we write in short @xmath251 thus finding bases of @xmath252 or @xmath253 solves problem pt or pfde , respectively . let @xmath254 be a rational function field . for a polynomial @xmath255 $ ] the degree is denoted by @xmath256 ; we set @xmath257 . we define @xmath258 , \deg(p)<\deg(q)\ } } } ; & { \mathbb{f}}[t]_k&:={{\{p\in{\mathbb{f}}[t]\,|\,\deg(p)\leq k\}}}\end{aligned}\ ] ] for @xmath259 . let @xmath260^n$ ] . then @xmath261 gives the @xmath262-th coefficient of @xmath263 $ ] . moreover , we define @xmath264 * extensions and reordering.*[page : ordering ] we will exploit the following fact frequently : if @xmath265 is a @xmath3-extension of @xmath64 and @xmath266 is a @xmath3-extension of @xmath265 over @xmath109 , we can reorder @xmath266 to the @xmath3-extension @xmath267 of @xmath64 . + we call a difference field @xmath268 a @xmath3-extension ( resp . @xmath43-extension/@xmath44-extension ) of @xmath181 if there is a @xmath3-extension ( resp . @xmath43-extension/@xmath44-extension ) @xmath269 of @xmath181 over @xmath109 such that we get the difference field @xmath268 by reordering the difference field @xmath269 . note that here @xmath270 could be @xmath271 , but additional reordering might be possible . in a nutshell , we enlarge the ground field @xmath109 by additional @xmath3-extensions to the field @xmath270 where @xmath268 is still a @xmath3-extension of @xmath181 . * basic properties . * the next properties follow from karr s theory @xcite . [ lemma : karr ] let @xmath272 with @xmath273 be a @xmath3-extension of @xmath196 . let @xmath274 and suppose there is a solution @xmath275 with @xmath276 , but no solution in @xmath51 . if @xmath277 is a @xmath44-extension , then @xmath76 and @xmath278 for some @xmath279 and @xmath280 ; if @xmath277 is a @xmath43-extension , then @xmath281 and @xmath74 . [ cor : extendpioversigma ] let @xmath195 be a @xmath43-extension of @xmath14 . if @xmath42 is a @xmath44-extension of @xmath14 with @xmath47 , then @xmath282 is a @xmath44-extension of @xmath195 with @xmath47 . write @xmath283 with the @xmath43-extensions @xmath284 . if @xmath285 , nothing has to be shown . suppose that @xmath282 is not a @xmath44-extension of @xmath195 . then we find a @xmath286 with @xmath287 for some @xmath288 by theorem [ thm : pisigma].2 . by lemma [ lemma : karr ] it follows that @xmath23 . hence @xmath42 is not a @xmath44-extension of @xmath14 by theorem [ thm : pisigma].2 . [ prop : propertiesinpisi ] let @xmath37 be a @xmath3-extension of @xmath14 with @xmath13 and @xmath7 . ( 1 ) : : if there is a @xmath289 such that , then there is no @xmath23 such that . ( 2 ) : : let @xmath290 with @xmath291 or @xmath292 for @xmath52 . if @xmath174 s.t . , then @xmath293 where @xmath294 and @xmath295 ; moreover , @xmath296 , if @xmath59 . * ( 1 ) * assume there are such @xmath297 and @xmath289 . then @xmath298 , and thus @xmath299 , a contradiction that @xmath37 is a @xmath3-extension of @xmath14 . * ( 2 ) * is a special case of ( * ? ? ? * result , page 314 ) ; see also ( * ? ? ? * thm . 4.2.1 ) . [ prop : homproperties ] let @xmath14 , @xmath210 be difference fields with a @xmath40-isomorphism @xmath209 . ( 1 ) : : let @xmath42 and @xmath300 be @xmath43-extensions of @xmath14 and @xmath210 , respectively , with @xmath301 such that @xmath302 . then there is a @xmath40-isomorphism @xmath303 with @xmath304 and @xmath305 for all @xmath7 . ( 2 ) : : let @xmath42 be a @xmath3-extension of @xmath14 with @xmath306 . then there is a @xmath3-extension @xmath307 of @xmath213 with @xmath308 . moreover , there is the @xmath40-isomorphism @xmath309 where @xmath310 and @xmath311 for all @xmath212 . ( 3 ) : : let @xmath37 be a @xmath3-extension of @xmath14 then there is a @xmath3-extension @xmath312 of @xmath213 with a @xmath40-isomorphism @xmath309 where @xmath311 for all @xmath212 . * ( 1 ) * let @xmath313 , and let @xmath301 such that @xmath314 . by proposition [ prop : propertiesinpisi].2 there are a @xmath315 and a @xmath316 such that @xmath317 . since @xmath318 is transcendental over @xmath319 , also @xmath140 is transcendental over @xmath319 . therefore we can define the field isomorphism @xmath320 with @xmath304 and @xmath305 for all @xmath7 . we have @xmath321 and thus @xmath322 is a @xmath40-isomorphism . since @xmath323 , the first part is proven . * ( 2 ) * let @xmath42 be a @xmath3-extension of @xmath14 with @xmath306 . since @xmath205 is a @xmath40-isomorphism , there is a @xmath3-extension @xmath307 of @xmath210 with @xmath308 by theorem [ thm : pisigma].1 . we can construct the field isomorphism @xmath303 with @xmath310 and @xmath311 for all @xmath212 . since @xmath324 , @xmath322 is a @xmath40-isomorphism . iterative application of * ( 2 ) * shows * ( 3)*. we will show the first properties of depth - optimal @xmath3-extensions ; some of the following results and proofs are simplified and streamlined versions of @xcite [ prop : twonestedext ] a @xmath3-extension @xmath181 of @xmath64 with @xmath325 , @xmath326 and @xmath177 is depth - optimal . @xmath327 is depth - optimal . suppose that @xmath328 is not depth - optimal with @xmath329 . set @xmath330 . then there is a @xmath3-extension @xmath124 of @xmath14 with @xmath331 and @xmath332 such that . by prop . [ prop : propertiesinpisi].2 , @xmath333 for some @xmath334 . then @xmath335 ; a contradiction to theorem [ thm : pisigma].1 . [ lemma : depthstablesummation ] let @xmath37 with @xmath290 be an ordered @xmath175-extension of @xmath14 with @xmath336 and @xmath337 . let @xmath7 with @xmath338 . then for any @xmath174 with , @xmath339 . suppose we have with @xmath174 and @xmath340 . hence @xmath140 depends on one of the @xmath328 , i.e. , let @xmath341 and @xmath342 such that @xmath343 . by proposition [ prop : propertiesinpisi].2 , @xmath344 where @xmath207 , @xmath345 and @xmath346 . since the extension is ordered , @xmath347 . by proposition [ prop : propertiesinpisi].1 there is no @xmath348 with @xmath349 . therefore by theorem [ thm : pisigma].1 one can construct a @xmath43-extension @xmath350 of @xmath196 with @xmath171 where @xmath351 . note that @xmath352 with @xmath353 . hence @xmath328 is not depth - optimal , a contradiction . therefore @xmath339 . [ thm : depthstableordered ] let @xmath14 be an ordered @xmath175-extension of @xmath64 and @xmath354 . if @xmath23 as in , then . since @xmath355 , @xmath356 . if @xmath357 , then @xmath142 otherwise , since @xmath14 is ordered , we can split @xmath5 into the @xmath175-extension @xmath14 of @xmath196 with @xmath358 ( @xmath359 ) and the @xmath175-extension @xmath196 of @xmath64 where @xmath360 for each @xmath52 and @xmath361 . by lemma [ lemma : depthstablesummation ] , @xmath362 . [ lemma : shiftsumtoleft ] let @xmath50 be a @xmath175-extension of @xmath14 and @xmath363 be a @xmath43-extension of @xmath50 with @xmath364 for all @xmath52 . by reordering , @xmath365 is a @xmath175-extension of @xmath366 . we show the lemma by induction . if @xmath285 , nothing has to be shown . consider @xmath367 as claimed above with @xmath368 . then by the induction assumption @xmath369 is a @xmath175-extension of @xmath370 . note that @xmath371 is a @xmath3-extension of @xmath14 . if @xmath327 is a @xmath44-extension , we are done . otherwise , suppose that @xmath327 is a @xmath43-extension with @xmath372 which is not depth - optimal . then there is a @xmath43-extension @xmath196 of @xmath366 with extension depth @xmath373 and @xmath374 such that . since @xmath375 , @xmath196 is a @xmath43-extension of @xmath14 with extension depth @xmath373 . a contradiction that @xmath376 is a @xmath172-extension of @xmath14 . the following two propositions will be heavily used in section [ sec : constructdeltaext ] . [ prop : extensionontop ] let @xmath37 with @xmath290 be an ordered @xmath175-extension of @xmath14 where @xmath336 and @xmath377 . suppose that @xmath124 is a @xmath43-extension of @xmath14 with @xmath378 and @xmath379 for @xmath380 . then the following holds . ( 1 ) : : there is the @xmath43-extension @xmath381 of @xmath37 with @xmath382 for all @xmath383 with @xmath380 . ( 2 ) : : in particular , by reordering , we get the @xmath3-extension @xmath384 of @xmath14 ; @xmath384 is a @xmath175-extension of @xmath124 . * ( 1 ) * let @xmath385 be minimal s.t.@xmath386 is not a @xmath43-extension of @xmath37 . take @xmath387 s.t . hence @xmath339 , i.e. , @xmath389 by lemma [ lemma : depthstablesummation ] ; this contradicts thm . [ thm : pisigma].1 . iterative application of lemma [ lemma : shiftsumtoleft ] proves * ( 2)*. [ prop : reorderequaldepth ] let @xmath50 be a @xmath175-extension of @xmath14 with @xmath390 for @xmath52 and @xmath391 . for @xmath182 , @xmath392 is a @xmath175-extension of @xmath14 . let @xmath393 ( @xmath285 is trivial ) ; take @xmath394 with @xmath395 such that @xmath396 is not a @xmath172-extension of @xmath397 with @xmath398 . choose a @xmath43-extension @xmath399 of @xmath397 with @xmath400 and @xmath401 such that . note that @xmath402 is a @xmath43-extension of @xmath14 by reordering ; hence by prop . [ prop : extensionontop].2 , @xmath403 is a @xmath43-extension of @xmath50 . let @xmath404 . by prop . [ prop : propertiesinpisi].2 , @xmath405 where @xmath406 and @xmath407 . let @xmath408 be maximal such @xmath409 ; if @xmath296 for all @xmath410 , set @xmath411 . if @xmath412 , @xmath413 for @xmath414 . otherwise , @xmath415 with @xmath416 for @xmath417 . summarizing , @xmath418 is not a @xmath172-extension of @xmath419 , a contradiction . [ thm : constructsigmaextontopordered ] let @xmath14 be a @xmath3-extension of @xmath64 and let @xmath195 be a @xmath43-extension of @xmath14 which can be brought to an ordered @xmath175-extension of @xmath64 . then for any @xmath43-extension @xmath196 of @xmath14 with extension depth @xmath420 there is a @xmath43-extension @xmath37 of @xmath195 with extension depth @xmath421 and an @xmath5-monomorphism @xmath201 s.t . for @xmath208 . let @xmath422 be an ordered @xmath175-extension of @xmath14 that we get by reordering the @xmath43-extension @xmath195 of @xmath14 . moreover , let @xmath196 be a @xmath43-extension of @xmath14 with extension depth @xmath420 , i.e. , @xmath423 . suppose that @xmath424 , otherwise we can reorder it without loosing any generality . we show that there is a @xmath43-extension @xmath37 of @xmath422 with extension depth @xmath421 and an @xmath5-monomorphism @xmath201 with @xmath425 for @xmath208 . then reordering of @xmath422 proves the statement for the extension @xmath195 of @xmath14 . + if @xmath285 , i.e. , @xmath426 , the statement is proven by taking @xmath427 with the @xmath5-monomorphism @xmath428 where @xmath214 for all @xmath212 . otherwise , suppose that @xmath429 with @xmath430 and @xmath431 where @xmath432 and @xmath433 . moreover , assume that there is a @xmath43-extension @xmath312 of @xmath422 with extension depth @xmath434 together with an @xmath5-monomorphism @xmath435 such that for all @xmath436 . now define @xmath437 . by assumption @xmath438 * case 1 : * suppose that there is no @xmath439 as in . then we can construct the @xmath43-extension @xmath440 of @xmath312 with @xmath441 by theorem [ thm : pisigma].1 and can define the @xmath5-monomorphism @xmath442 such that @xmath311 for all @xmath436 and @xmath443 . with we get @xmath444 . by the induction assumption , @xmath445 for all @xmath446 . clearly , the @xmath43-extension @xmath440 of @xmath422 has extension depth @xmath421 . + * case 2 : * suppose there is a @xmath447 with @xmath448 . since @xmath312 is a @xmath43-extension of @xmath422 with extension depth @xmath434 ( @xmath449 ) we can apply lemma [ lemma : shiftsumtoleft ] and obtain by reordering of @xmath312 an ordered @xmath175-extension @xmath450 of @xmath451 where @xmath452 and @xmath453 for all @xmath454 . hence with we can apply lemma [ lemma : depthstablesummation ] and it follows that @xmath455 , i.e. , @xmath456 since @xmath205 is a monomorphism , there is no @xmath140 in the image @xmath457 such that . since @xmath458 is a difference field ( it is a sub - difference field of @xmath312 ) , @xmath459 is transcendental over @xmath457 by theorem [ thm : pisigma].1 . in particular , we get the @xmath5-monomorphism @xmath460 with @xmath311 for all @xmath436 and @xmath443 . with and our induction assumption it follows that @xmath445 for all @xmath446 . this completes the induction step . we modify karr s reduction for problem pt : given a @xmath3-extension @xmath461 of @xmath196 and @xmath462 ; find a basis @xmath463 of @xmath464 , as follows : first split @xmath465 by polynomial division in the form @xmath466 such that @xmath467 and @xmath468^n$ ] ; in short we write @xmath469^n.\ ] ] then the following lemma , a direct consequence of ( * ? ? ? * lemma 3.1 ) , is crucial . [ lemma : splitproblem ] let @xmath461 be a @xmath3-extension of @xmath196 with @xmath306 and . then @xmath470 , @xmath471 $ ] solve iff @xmath472 and @xmath473 . note that we get a first strategy : find bases for @xmath474 and @xmath475})}$ ] , and afterwards combine the solutions accordingly to get a basis of @xmath476 . as it will turn out , the following version , presented in figure [ fig : ratred ] , is more appropriate : first solve the rational problem ( rp ) ; if there is no solution , there is no solution for the original problem . otherwise plug in the rational solutions into our ansatz and continue to find the polynomial solutions ( problem pp ) ; for details see remark [ rem : ratdetails ] . @xmath477\txt{task : find a basis of\\ $ { \mathbb{v}}={{\rm v}({\boldsymbol{f}},{{\mathbb{h}}(t)})}$}\ar[d]\\ * + [ f-:<3pt>]{\txt{set up~\eqref{equ : splitf } by polynomial division}}\ar[d]&*+[f-:<3pt>]{\txt{set up $ { \boldsymbol{p'}}\in{\mathbb{h}}[t]^m$ by~\eqref{equ : polyf}.}}\ar[d]\\ * + [ f--]{\txt{{{\sf rp } } : find a basis $ r$ of $ { { \rm v}({\boldsymbol{h}},{{{{\mathbb{h}}(t)}_{\text{\tiny(r)}}}})}$}}\ar@{.>}|{r}@/_1.8pc/[dr]\ar[dd]_{r=\{\}}\ar[ru]|<(.3){r\neq\{\}}&*+[f--]{\txt{{{\sf pp } } : find a basis $ p$ of $ { { \rm v}({\boldsymbol{p'}},{{\mathbb{h}}[t]})}$. } } \ar@{.>}_{p}[d]\\ & * + [ f.]{\txt{combine $ r$ and $ p$ to a basis $ b$ of $ { \mathbb{v}}$}}\ar[d]\\ * + [ f-,]\txt{return : $ \{\}$}&*+[f-,]\txt{return $ b$ } } \ ] ] [ rem : ratdetails ] find a basis @xmath478 of @xmath474 ; note that @xmath479 . if @xmath480 , then @xmath481 by lemma [ lemma : splitproblem ] . otherwise , define @xmath482 and @xmath483 . then we look for all @xmath484 and @xmath485 $ ] such that @xmath486 since @xmath487 and @xmath488 by construction , problem is equivalent to looking for all @xmath484 and @xmath485 $ ] s.t.@xmath489 hence , set up @xmath490 where @xmath491^m$ ] and find a basis @xmath492 of @xmath493})}$ ] . note that @xmath494 , since there is the trivial solution @xmath495 ; define @xmath496 and @xmath497 . then with it follows that @xmath498 thus , if we define @xmath499 and @xmath500 , we get a set of generators @xmath501 that spans a subspace of @xmath464 . by simple linear algebra arguments it follows that @xmath463 is a basis of @xmath244 . [ exp : truncatednaiverat ] consider the ordered @xmath175-field @xmath99 over @xmath502 with and @xmath503 . by we have @xmath504 and @xmath505 . following the reduction of figure [ fig : ratred ] , we need a basis @xmath506 of @xmath507 ; obviously @xmath508 . thus @xmath509 by . in example [ exp : truncatednaivepoly ] we will show that @xmath510 is a basis of @xmath511)})}$ ] . by remark [ rem : ratdetails ] a basis of @xmath512 is @xmath513 . _ in @xcite the reversed strategy was proposed : first consider the polynomial and afterwards the rational problem . related to problems dot and dot@xmath169 , the following remark is in place . in lemma [ lemma : ratext ] we will show that the solutions of rp are independent of the type of extension that are needed to solve problems dot , dot@xmath169 . thus , we will consider problem rp first . if there is no solution , we can stop ; see corollary [ cor : polyratext ] . otherwise , we will attack problems dot , dot on @xmath514 which is usually simpler ( @xmath515 ) than @xmath516 . the next lemma will be needed in section [ sec : depthspeedup ] to solve problems dot , dot@xmath169 efficiently . [ lemma : entriesinpolcase ] let @xmath461 be a @xmath3-extension of @xmath196 and @xmath465 . if @xmath208 occurs in @xmath246 , then the reduction in figure [ fig : ratred ] can be applied so that @xmath517 occurs in @xmath514 . let @xmath208 occur in the @xmath383th position of @xmath246 . write @xmath518^n$ ] . then the @xmath383th entry in @xmath519 is zero . hence , we can take a basis @xmath506 for @xmath474 where the @xmath383th unit - vector is in @xmath506 . applying it follows that @xmath517 occurs in @xmath514 . subproblem rp has been solved in ( * ? ? ? * sections 3.4 , 3.5 ) . alternatively , this task can be accomplished by computing a basis @xmath520 of @xmath521 by using , e.g. , algorithm @xcite which is based on results from @xcite . namely , by ( * ? ? ? 1,2 ) it follows that @xmath522 hence a basis @xmath506 for @xmath474 can be derived by simple manipulation of the basis @xmath520 . + we remark that both approaches can be solved algorithmically if @xmath196 is @xmath40-computable . as in karr s reduction @xcite we bound the degree of the polynomial solutions . [ lemma : degbound ] let @xmath461 be a @xmath3-extension of @xmath196 and let @xmath523^m$ ] . then @xmath524_b})}={{\rm v}({\boldsymbol{p'}},{{\mathbb{h}}[t]})}$ ] for @xmath525 thus we set up @xmath526 and @xmath527^m_{r}$ ] and look for a basis @xmath528 of @xmath529_r})}$ ] . we will accomplish this task by solving instances of problem pfdein @xmath196 ; see also ( * ? ? ? 12 ) or ( * ? ? ? note that this is possible if @xmath196 is @xmath40-computable . if @xmath530 , we are already in the base case . otherwise , let @xmath531 . then we try to get all @xmath532_{r}$ ] and @xmath533 such that @xmath534 as follows . first , we derive the possible leading coefficients @xmath535 in @xmath196 , then we plug in the resulting solutions into @xmath534 and look for the remaining @xmath536 by recursion . the technical details are given in remark [ rem : polytechnical ] , and a graphical illustration is presented in figure [ fig : polyred ] : here the task of finding a basis of @xmath537 is reduced to finding a basis of the `` leading coefficients '' ( problem cp ) and to finding a basis of the `` remaining coefficients '' @xmath538 . @xmath539\txt{return $ b_r$}&*+[f=]{\txt{task : find a basis $ b_r$ of\\ $ { \mathbb{v}}_r:={{\rm v}({\boldsymbol{f_r}},{{\mathbb{h}}[t]_r})}$.}}\ar[d]&*+[f-,]\txt{return\\ $ \{{({0,\dots,0,1})}\}$}&*+[f-,]+<-1.3cm,0cm>\txt{return $ \tilde{b}_r$}\\ * + [ f.]\txt{combine $ b_{r-1}$ and $ \tilde{b}_r$\\ to get a basis $ b_r$ of $ { \mathbb{v}}_{r}$.}\ar@{.>}[u]&*+[f-:<3pt>]\txt{set up $ { \boldsymbol{\tilde{f}_r}}$ by~\eqref{equ : defineftilde}}\ar[r]&*+[f--]\txt{{{\sf cp } } : find a basis $ \tilde{b}_r$\\of $ { { \rm v}({{({\alpha^r,-1})}},{\boldsymbol{\tilde{f}_r}},{{\mathbb{h}}})}$ } \ar[ur]_<(0.5){r=0}\ar@{.>}@/^2pc/[ll]_{\tilde{b}_r}\ar[u]^{r>0\wedge\tilde{b}_r=\{\}}\ar[d]^{r>0\wedge\tilde{b}_r\neq\{\}}\\ & * + [ f=]{\txt{task : find a basis $ b_{r-1}$ of\\ $ { \mathbb{v}}_{r-1}:={{\rm v}({\boldsymbol{f_{r-1}}},{{\mathbb{h}}[t]_{r-1}})}$.}}\ar@{.>}@/^2pc/[lu]|<(.55){b_{r-1 } } & * + [ f-:<3pt>]\txt{set up $ { \boldsymbol{f_{r-1}}}$ by~\eqref{equ : definefdeltam}}\ar[l]\\ } \ ] ] [ rem : polytechnical ] the main task is to find a basis @xmath528 of @xmath540_r})}$ ] . first , define @xmath541 where @xmath542 . then find a basis @xmath543 of @xmath544 . if @xmath545 , then @xmath546 and @xmath547_{r-1}$ ] are the only choices , i.e. , @xmath548 . otherwise , if @xmath549 , define @xmath550 @xmath551 by construction , @xmath552_{r-1}^{\lambda}$ ] . now we proceed as follows . find all @xmath553_{r-1}$ ] and @xmath554 such that @xmath555 which is equivalent to @xmath556 therefore the subtask is to find a basis @xmath557 of @xmath558_{r-1}})}$ ] ; note that @xmath559 , since @xmath495 . given @xmath560 and @xmath561 , a basis for @xmath540_{r}})}$ ] can be constructed as follows . define @xmath562 , @xmath563^{l}_{r-1}$ ] , @xmath564 , and @xmath565_r^l$ ] . then by the set @xmath566 spans a subspace of @xmath567})}$ ] . by simple linear algebra arguments it follows that @xmath528 is a basis of @xmath537 . [ exp : truncatednaivepoly ] given the @xmath3-field @xmath350 over @xmath36 with @xmath568 and @xmath569 from example [ exp : truncatednaiverat ] , we compute a basis of @xmath570})}$ ] as follows . we start the reduction of figure [ fig : polyred ] with @xmath571 , see , and @xmath572 . @xmath573 : : : by we get @xmath574 ; a basis of @xmath575 is @xmath576 . @xmath577 : : : we get @xmath578 by and @xmath579 by . by another reduction in @xmath51 we compute the basis @xmath580 of @xmath581 . @xmath530 : : : @xmath582 by . clearly , @xmath583 is a basis of @xmath584 . finally , we combine @xmath585 , @xmath586 and @xmath587 and get the basis @xmath588 , @xmath589 of @xmath590_1})}$ ] , @xmath591_2})}$ ] , respectively . thus @xmath592 is a basis of @xmath593})}$ ] . + we that the summand @xmath594 of @xmath595 given in occurs in @xmath596 . this observation is crucial for our refined summation algorithm ; see example [ exp : truncatedfullpoly ] . [ rem : ratpolyred ] if @xmath530 , or if @xmath531 and @xmath45 is a @xmath43-extension ( @xmath597 ) , problem cp is nothing else than problem pt in the ground field @xmath196 . hence , we can apply again the reductions presented in the figures [ fig : ratred ] and [ fig : polyred ] to the subfield @xmath51 . more precisely , if @xmath181 is a @xmath3-extension of a @xmath40-computable @xmath64 , we reduce problem pt to pt in the fields below whenever possible , and change to the more general situation pfde only when it is necessary . this strategy will be the basis to construct @xmath172-extensions in section [ sec : constructdeltaext ] . the following lemmata are needed for our refined algorithms . lemma [ lemma : singeextplde ] is immediate by construction ; it is used to prove corollary [ cor : incremredref ] . lemma [ lemma : entriesingroundfield ] is crucial in section [ sec : depthspeedup ] . [ lemma : singeextplde ] let @xmath598 be a @xmath3-extension of @xmath14 with @xmath306 where @xmath599 . set @xmath600 , let @xmath531 , @xmath601^m_r$ ] and @xmath602 with . if the coefficients with the monomial @xmath603 in @xmath604_r})}$ ] are free of the @xmath605 , it suffices to take a basis of @xmath606 to get a basis of @xmath244 following the reduction in figure [ fig : polyred ] . [ lemma : entriesingroundfield ] let @xmath461 be a @xmath3-extension of @xmath196 with @xmath607 and @xmath491^m$ ] . suppose we succeed in reducing the problem to the base case @xmath530 with @xmath608 . ( 1 ) : : the reduction can be applied s.t . all entries of @xmath514 which are in @xmath5 also occur in @xmath609 . ( 2 ) : : moreover , if @xmath597 , we can guarantee that @xmath610 occurs in @xmath609 . * ( 1 ) * suppose that @xmath212 occurs in the @xmath383th position of @xmath514 . define @xmath611 by and set @xmath612 . if @xmath530 , nothing has to be shown . otherwise , suppose that @xmath531 . since @xmath208 , the @xmath383th entry in @xmath613 , defined by , is zero . hence we can choose a basis @xmath561 of @xmath614 where the @xmath383th unit - vector is in @xmath561 and all other elements of @xmath561 have zero in the @xmath383th position . with the corresponding @xmath615 and we get @xmath616 where @xmath517 pops up . this construction can be done for all such entries of @xmath514 which are in @xmath5 . if we continue with this refined reduction to the case @xmath609 , all @xmath5-entries of @xmath514 occur in @xmath609 . + * ( 2 ) * assume that @xmath45 is a @xmath43-extension . hence @xmath617 by . suppose we are in the reduction for @xmath577 . since @xmath581 contains @xmath618 , we can choose a basis @xmath586 with @xmath619 . by construction of @xmath609 it follows that @xmath620 occurs in @xmath609 . we focus on the problem when extensions do not contribute to solutions of pfde . [ lemma : ratext ] let @xmath461 be a @xmath3-extension of @xmath196 and @xmath621 be a @xmath3-ext . of @xmath461 over @xmath51 . for @xmath467 , @xmath622 suppose we find an additional solution in a @xmath3-extension @xmath621 of @xmath461 over @xmath51 , i.e. , there is a @xmath623 such that @xmath140 depends on @xmath624 and @xmath625 for some @xmath470 . take such a solution and define @xmath626 . now reorder the extension to the @xmath3-extension @xmath627 of @xmath196 . with proposition [ prop : propertiesinpisi].2 it follows that @xmath628 for some @xmath629 and @xmath630 . write @xmath631 with @xmath632 $ ] ; @xmath633 . then @xmath634 . since @xmath635 and @xmath636 are free of @xmath624 , @xmath637 , i.e. , @xmath638 . [ cor : polyratext ] let @xmath461 be a @xmath3-extension of @xmath196 and let @xmath621 be a @xmath3-extension of @xmath461 over @xmath51 . let @xmath639 and take @xmath464 and @xmath640 . write @xmath466 as in . ( 1 ) : : if @xmath641 , then @xmath642 . ( 2 ) : : otherwise , define @xmath491^m$ ] and @xmath643 by and . then @xmath644 iff @xmath524_b})}={{\rm v}({\boldsymbol{p'}},{{\mathbb{h}}(x_1)\dots(x_r)[t]_b})}.$ ] if @xmath506 , @xmath645 are bases of @xmath646 and @xmath647_b})}$ ] , respectively , we get a basis of @xmath648 as given in remark [ rem : ratdetails ] . by lemma [ lemma : ratext ] we have @xmath649 , i.e. , @xmath506 is also a basis of @xmath650 . thus , if @xmath480 , then @xmath642 ; see fig . [ fig : ratred ] . this proves * ( 1)*. otherwise , let @xmath514 and @xmath643 be as assumed . note that @xmath643 bounds the polynomial solutions in @xmath651 $ ] _ and _ @xmath652 $ ] by lemma [ lemma : degbound ] . hence , if @xmath524_b})}={{\rm v}({\boldsymbol{p'}},{{\mathbb{h}}(x_1)\dots(x_e)[t]_b})}$ ] , then by remark [ rem : ratdetails ] we get @xmath644 . conversely , if @xmath524_b})}\subsetneq{{\rm v}({\boldsymbol{p'}},{{\mathbb{h}}(x_1)\dots(x_e)[t]_b})}$ ] , then @xmath653 by construction . this proves * ( 2)*. consequently , if one wants to find an extension with additional solutions , one has to focus on problem cp ; see fig . [ fig : polyred ] . with lemma [ lemma : noextforpi ] we can refine this observation in corollary [ cor : incremredref ] . [ lemma : noextforpi ] let @xmath461 be a @xmath44-extension of @xmath196 and let @xmath621 be a @xmath3-extension of @xmath461 over @xmath51 . set @xmath654 . then for @xmath655 and @xmath531 , @xmath656 suppose that @xmath657 . then there are @xmath470 and @xmath658 for some @xmath659 such that @xmath660 . note : there is no @xmath661 with @xmath662 . ( otherwise , @xmath663 . since @xmath664 , we get @xmath665 for @xmath666 . but , by reordering we get the @xmath44-extension @xmath667 of @xmath668 , a contradiction by theorem [ thm : pisigma].2 . ) thus we can apply lemma [ lemma : karr ] : if @xmath334 is a @xmath44-extension with @xmath669 for some @xmath670 , @xmath671 for some @xmath280 and @xmath672 . hence , @xmath673 with @xmath674 . otherwise , if @xmath334 is a @xmath43-extension , @xmath675 . summarizing , @xmath676 for some @xmath677 . since @xmath678 is a @xmath44-extension of @xmath679 , this contradicts theorem [ thm : pisigma].2 . [ cor : incremredref ] let @xmath461 be a @xmath3-extension of @xmath196 and take a @xmath3-extension @xmath680 of @xmath461 over @xmath51 where for @xmath681 we have @xmath682 or @xmath683 . let @xmath531 , @xmath684^m_r$ ] , and set @xmath685_r})}$ ] and @xmath686_r})}$ ] . define @xmath613 and @xmath616 as in and . ( 1 ) @xmath687 : : : if @xmath688 and + @xmath689_{r-1}})}={{\rm v}({\boldsymbol{f_{r-1}}},{{\mathbb{h}}[t]_{r-1}})}$ ] , then @xmath690 . + given the basis @xmath557 of @xmath689_{r-1}})}$ ] and the basis @xmath691 of @xmath692 , one gets a basis of @xmath693 as stated in remark [ rem : polytechnical ] . ( 2 ) @xmath694 : : : if @xmath689_{r-1}})}={{\rm v}({\boldsymbol{f_{r-1}}},{{\mathbb{h}}[t]_{r-1}})}$ ] , then @xmath690 . given the basis @xmath557 of @xmath689_{r-1}})}$ ] and the basis @xmath691 of @xmath614 , one gets a basis of @xmath693 as stated in remark [ rem : polytechnical ] . * ( 1 ) * @xmath45 is a @xmath43-extension : by prop . [ prop : propertiesinpisi].2 the leading coefficients with degree @xmath262 in the solutions of @xmath693 are free of @xmath695 . hence , by lemma [ lemma : singeextplde ] it suffices to take a basis of @xmath692 to get a basis of @xmath693 following remark [ rem : polytechnical ] . thus , if @xmath696_{r-1}})}={{\rm v}({\boldsymbol{f_{r-1}}},{{\mathbb{h}}(x_1)\dots(x_l)(y_1)\dots(y_k)[t]_{r-1}})}$ ] and @xmath688 , then @xmath644 by remark [ rem : polytechnical ] . given @xmath557 , @xmath691 from above one gets a basis of @xmath693 . + * ( 2 ) * if @xmath697 , then @xmath698 by lemma [ lemma : noextforpi ] . hence , if @xmath689_{r-1}})}={{\rm v}({\boldsymbol{f_{r-1}}},{{\mathbb{h}}[t]_{r-1}})}$ ] , @xmath644 by remark [ rem : polytechnical ] . given the bases @xmath557 , @xmath691 as stated above , one gets a basis of @xmath693 . subsequently , we prove the following theorem which will establish result [ res : depthconstruct ] in section [ sec : mainproofs ] . [ thm : extendforteleordered ] let @xmath14 be an ordered @xmath175-extension of @xmath64 and @xmath7 . then there is a @xmath43-extension @xmath37 of @xmath14 such that @xmath37 can be brought to an ordered @xmath175-extension of @xmath64 and such that there is a @xmath174 as in . if @xmath64 is @xmath40-computable , such an @xmath37 and @xmath140 can be given explicitly . in order to accomplish this task , we consider the following more general situation . [ def : completea ] let @xmath14 be a @xmath3-extension of @xmath64 and @xmath699 . then @xmath14 is _ @xmath700complete _ , if for any @xmath3-extension are @xmath43-extensions . ] @xmath701 of @xmath14 with extension depth @xmath702 we have @xmath703 . that is to say , we show the following theorem . [ thm : algtheorema ] let @xmath14 be an ordered @xmath175-extension of @xmath64 , @xmath704 and @xmath699 . then there is a @xmath43-extension @xmath195 of @xmath14 which is @xmath700complete and which can be brought to an ordered @xmath175-extension of @xmath64 . if @xmath64 is @xmath40-computable , then such an @xmath195 and a basis of @xmath705 can be given explicitly . then theorem [ thm : extendforteleordered ] is implied by the following lemma . [ lemma : basiccomplete ] let @xmath14 be a @xmath3-extension of @xmath64 and @xmath699 . if @xmath14 is @xmath706complete , then @xmath707 . suppose @xmath708 , i.e. , there is a @xmath470 such that there is no @xmath23 with @xmath709 . thus there is the @xmath43-extension @xmath170 of @xmath14 with @xmath171 and @xmath710 . hence @xmath14 is not @xmath706complete . namely , we conclude by theorem [ thm : algtheorema ] that there is a @xmath43-extension @xmath195 of @xmath14 which is @xmath711complete and which can be brought to an ordered @xmath175-extension of @xmath64 . hence by lemma [ lemma : basiccomplete ] there is a @xmath712 such that . in most applications one works with a @xmath3-field over @xmath109 , with @xmath70 for some @xmath713 . in this case , the following shortcut can be applied ; the proof is similar to prop . [ prop : twonestedext ] . [ lemma : shortcut2 ] let @xmath14 be a @xmath3-field over @xmath109 and @xmath699 . if @xmath714 for some @xmath23 , then @xmath14 is @xmath715complete . the proof of theorem [ thm : algtheorema ] will be obtained by refining the reduction of section [ sec : reduction ] . namely , let @xmath716 , let @xmath37 with @xmath290 be an ordered @xmath175-extension of @xmath64 where @xmath285 or @xmath717 , and let @xmath718 . then loosely speaking , we will obtain an @xmath700complete extension @xmath195 of @xmath37 by constructing step by step a tower of extensions , say @xmath719 , where @xmath720 is a @xmath172-extension of @xmath721 for @xmath722 . within this construction problem cp in figure [ fig : polyred ] will be refined to the following subproblem : we are given a vector @xmath723 with entries from @xmath724 , and we have to enrich @xmath724 by @xmath172-extensions to @xmath725 such that @xmath725 becomes @xmath726-compete . note that during this extension process it is crucial that @xmath727 forms a @xmath175-extension of @xmath720 for each @xmath722 . in order to get a grip on this situation , we introduce the following definition , which reduces to definition [ def : completeb ] when @xmath285 . [ def : completeb ] let @xmath37 be a @xmath3-extension of @xmath14 with @xmath290 and @xmath699 . then @xmath37 is _ @xmath728complete _ , if for any @xmath3-extension @xmath729 of @xmath37 over @xmath5 with extension depth @xmath702 we have @xmath703 . subsequently , we prove the following theorem which implies theorems [ thm : algtheorema ] and [ thm : extendforteleordered ] . [ thm : algtheoremb ] let @xmath704 and let @xmath50 be an ordered @xmath175-extension of @xmath64 ; if @xmath368 , @xmath730 . let @xmath699 . then there is a @xmath43-extension @xmath731 of @xmath50 with extension depth @xmath702 such that @xmath731 is @xmath732complete and such that @xmath731 is an ordered @xmath175-extension of @xmath64 . if @xmath64 is @xmath40-computable , such an @xmath731 and a basis of @xmath733 can be given explicitly . we will show theorem [ thm : algtheoremb ] by induction on the depth @xmath734 . the base case @xmath735 is covered by lemma [ lemma : shortcut].1 ; the proof of lemma [ lemma : shortcut ] is immediate with lemma [ lemma : basiccomplete ] . [ lemma : shortcut ] let @xmath37 be a @xmath3-extension of @xmath64 , let @xmath699 and set @xmath736 . then the following holds . ( 1 ) : : @xmath37 is @xmath737complete . ( 2 ) : : if @xmath738 ( or @xmath739 , @xmath37 is @xmath740complete for all @xmath741 . ( 3 ) : : @xmath14 is @xmath742complete iff @xmath707 . in the following let @xmath716 and let @xmath50 be an ordered @xmath175-extension of @xmath64 ; if @xmath368 , then @xmath730 . _ simplification i. _ note that it suffices to restrict to the case that @xmath390 for @xmath52 . otherwise , let @xmath743 be maximal such that @xmath744 . then we show that there exists such a @xmath43-extension @xmath745 of @xmath746 as required . finally , by proposition [ prop : extensionontop ] we get the desired @xmath175-extension @xmath747 of @xmath64 . the induction step uses another induction on the number of extensions in @xmath5 with depth @xmath734 . the base case and the induction step of this `` internal induction '' are considered in sections [ sec : completephase ] and [ sec : reductionphase ] , respectively . the case @xmath748 ( including @xmath749 ) is covered by the following consideration . _ simplification ii . _ we can assume that @xmath750 by lemma [ lemma : shortcut ] : if we find such a @xmath43-extension @xmath751 of @xmath50 which is @xmath752complete , then @xmath753 , and thus @xmath751 is @xmath754complete for any @xmath741 . with this preparation the following lemma gives the key idea . [ lemma : depthoptbysubprob ] let @xmath716 and let @xmath37 with @xmath755 be an ordered @xmath175-extension of @xmath14 with @xmath756 and @xmath390 for @xmath52 . let @xmath120 and suppose that @xmath37 is @xmath757complete . then any @xmath43-extension @xmath758 of @xmath37 with @xmath759 for @xmath760 is depth - optimal ; in particular , @xmath761 . let @xmath758 be such a @xmath43-extension of @xmath37 with constant field @xmath36 . first note that @xmath761 for all @xmath380 : if there is an @xmath762 with @xmath763 and @xmath764 , then by reordering we get the @xmath43-extension @xmath765 of @xmath37 with extension depth @xmath766 ; a contradiction that @xmath37 is @xmath757complete . now suppose that @xmath767 , @xmath768 , is not depth - optimal with @xmath769 ; set @xmath770 . then there is a @xmath43-extension @xmath771 of @xmath196 with @xmath772 for @xmath773 such that there is a @xmath774 with @xmath775 . the @xmath43-extension @xmath776 of @xmath37 is obtained by reordering ( note that @xmath777 ) . by prop . [ prop : propertiesinpisi].2 , @xmath778 where @xmath294 , @xmath779 . hence @xmath780 for some @xmath470 , i.e. , @xmath37 is not @xmath757complete , a contradiction . namely , by our induction assumption we apply theorem [ thm : algtheoremb ] and take a @xmath43-extension @xmath781 of @xmath50 with extension depth @xmath766 which is an ordered @xmath175-extension @xmath64 and which is @xmath757complete . then we adjoin step by step @xmath43-extensions such that for all @xmath782 there is a @xmath140 with @xmath775 as follows : 1 @xmath783 . for @xmath782 do + 2 if @xmath784 and @xmath785 s.t . @xmath786 then adjoin the @xmath43-extension + @xmath787 of @xmath788 with @xmath789 ; i:=i+1 . fi + 3 od finally , we get a @xmath43-extension , say @xmath790 of @xmath50 ; note that this extension process can be constructed explicitly if @xmath64 is @xmath40-computable . we complete the base case ( of the internal induction ) by the following arguments . @xmath791 : : by lemma [ lemma : depthoptbysubprob ] @xmath790 is an ordered @xmath175-extension of @xmath64 . by prop . [ prop : reorderequaldepth ] we get the @xmath172-extension @xmath792 of @xmath196 with @xmath793 . hence @xmath792 is an ordered @xmath175-extension of @xmath64 . @xmath791 : : since @xmath794 , @xmath792 is @xmath795complete by lemma [ lemma : shortcut ] ; if @xmath64 is @xmath40-computable , a basis of @xmath796 can be given explicitly . [ exp : truncatedh ] take the ordered @xmath175-field @xmath797 over @xmath36 with @xmath798 and ; let @xmath799 . by lemma [ lemma : shortcut2 ] @xmath797 is @xmath715complete . since there is no @xmath800 with @xmath801 , we get the @xmath172-extension @xmath802 of @xmath797 with @xmath803 by lemma [ lemma : depthoptbysubprob ] . by proposition [ prop : reorderequaldepth ] we obtain the ordered @xmath175-field @xmath804 . by construction , @xmath805 , i.e. , @xmath806 . therefore @xmath804 is @xmath807complete . note : since @xmath808 with @xmath809 is an ordered @xmath175-field with @xmath810 , we get the ordered @xmath175-field @xmath811 by proposition [ prop : extensionontop ] . [ exp : truncatedh ] take the ordered @xmath175-field @xmath170 with @xmath812 from example [ exp : truncatedh ] ; let @xmath813 . first note that @xmath170 is @xmath814complete and that there is no @xmath815 with @xmath816 ; see example [ exp : truncatedhsubcom ] . hence we can construct the @xmath172-extension @xmath817 of @xmath170 with @xmath818 ; by reordering we get the ordered @xmath175-field @xmath819 . a basis of @xmath820 is @xmath821 . clearly , @xmath819 is @xmath822complete . we suppose that @xmath823 , i.e. , @xmath824 where @xmath461 is a @xmath175-extension of @xmath196 with @xmath825 and @xmath826 . as above , @xmath827 is a @xmath3-extension of @xmath461 ; in particular with the simplification i : if @xmath368 , then @xmath828 with the following definition and corollary [ cor : polyratext ] we obtain corollary [ cor : completeratcrit ] . let @xmath37 be a @xmath3-extension of @xmath196 with @xmath829 and @xmath830^n_r$ ] . then @xmath37 is _ @xmath831_r)$]complete _ , if for any @xmath3-extension @xmath729 of @xmath37 over @xmath5 with extension depth @xmath702 we have @xmath832_r})}={{\rm v}({\boldsymbol{f}},{{\mathbb{h}}[t]_r})}$ ] . [ cor : completeratcrit ] let @xmath716 and let @xmath827 be a @xmath3-extension of @xmath196 with @xmath833 ; if @xmath368 , then . let @xmath465 , and define @xmath519 and @xmath516 by . let @xmath506 be a basis of @xmath474 and set @xmath464 . ( 1 ) : : if @xmath480 , @xmath827 is @xmath834complete ; @xmath835 . otherwise : ( 2 ) : : take @xmath491^m , b$ ] by , . if @xmath836 is a @xmath3-extension of @xmath781 that is @xmath837_b,{\mathfrak{d}})$]-complete , @xmath836 is @xmath838-complete . if @xmath645 is a basis of @xmath839_b})}$ ] , a basis of @xmath244 can be constructed by remark [ rem : ratdetails ] . [ exp : truncatedfullrat ] consider the ordered @xmath175-field @xmath350 with @xmath840 and ; let @xmath503 . as in example [ exp : truncatednaiverat ] we get @xmath504 , @xmath505 , @xmath841 , @xmath842 and @xmath573 . in examples [ exp : truncatedfullpoly ] and [ exp : truncatedfullpoly2 ] we will construct the @xmath43-extension @xmath843 of @xmath350 with @xmath844 and where @xmath843 is an ordered @xmath175-field which is @xmath845_2)$]complete ; we obtain the basis @xmath846 of @xmath847_2)})}$ ] . hence @xmath843 is @xmath848-complete by corollary [ cor : completeratcrit ] ; a basis of @xmath849 is @xmath645 . let @xmath617 and set @xmath850 and @xmath527^m_r$ ] . define @xmath613 as in . if @xmath45 is a @xmath44-extension , we can apply corollary [ cor : makecomplete ] ( see below ) and obtain the reduction @xmath851 . otherwise , if @xmath45 is a @xmath43-extension , the following preprocessing step is necessary . by the induction assumption we can apply theorem [ thm : algtheoremb ] and get a @xmath43-extension of @xmath852 with extension depth @xmath766 which can be brought to an ordered @xmath175-extension @xmath853 of @xmath64 and which is @xmath854complete . by proposition [ prop : extensionontop ] we can adjoin the extensions @xmath855 on top and get the ordered @xmath175-extension @xmath856 of @xmath64 . now we are ready apply corollary [ cor : makecomplete ] ( @xmath857 is replaced by @xmath51 ) and proceed with the reduction @xmath851 . [ cor : makecomplete ] let @xmath716 and let @xmath827 be a @xmath3-extension of @xmath64 with @xmath833 ; if @xmath368 , then . let @xmath836 be a @xmath3-extension of @xmath827 with extension depth @xmath702 . let @xmath531 and @xmath684^m_r$ ] , and define @xmath613 and @xmath616 as in and . ( 1 ) @xmath687 : : : if @xmath196 is @xmath858complete and in addition @xmath836 is @xmath859_{r-1},{\mathfrak{d}})$]complete , then @xmath836 is @xmath860_r,{\mathfrak{d}})$]complete . + if @xmath561 and @xmath557 are bases of @xmath861 and @xmath862_{r-1}})}$ ] , respectively , we get a basis of @xmath863_r})}$ ] following remark [ rem : polytechnical ] . ( 2 ) @xmath864 : : : if @xmath836 is @xmath859_{r-1},{\mathfrak{d}})$]-complete , then it follows that @xmath836 is @xmath865_r,{\mathfrak{d}})$]complete . if @xmath561 and @xmath557 are bases of the solution spaces @xmath614 and @xmath862_{r-1}})}$ ] , respectively , we get a basis of @xmath863_r})}$ ] following remark [ rem : polytechnical ] . let @xmath37 be a @xmath3-extension of @xmath866 with extension depth @xmath734 . reorder it to @xmath867 with @xmath868 for @xmath722 and @xmath869 for @xmath681 . then by corollary [ cor : incremredref ] @xmath870 . summarizing , we obtain a reduction for @xmath871 , which can be illustrated in figure [ fig : polydepthred ] . @xmath872\txt{return\\ $ ( { \mathbb{h}}'_{r}(t)(t_1,\dots , t_e),b_r)$}&*+[f=]{\txt{task : find an $ ( { \boldsymbol{f_r}},{\mathfrak{d}},{\mathbb{h}}'_{r})$--complete { $ \sigma^*$}-extension\\ $ { ( { { \mathbb{h}}'_r(t)(t_1)\dots(t_e)},{\sigma})}$ of $ { ( { { \mathbb{h}}_{r}(t)(t_1)\dots(t_e)},{\sigma})}$ with\\ extension depth $ \leq{\mathfrak{d}}$ which is an ordered { $ \pi\sigma^{\delta}$}-ex-\\ tension of $ { ( { { \mathbb{g}}},{\sigma})}$ ; find a basis $ b_r$ of $ { { \rm v}({\boldsymbol{f_r}},{{\mathbb{h}}'_r[t]_r})}$.}}\ar[d]&*+[f-,]\txt{return\\ $ ( { \mathbb{h}}_{r}(t)(t_1,\dots , t_e)$,\\ $ \{{(\begin{smallmatrix}{0,\dots,0,1}\end{smallmatrix})}\})$}\\ * + [ f.]\txt{combine $ b_{r-1}$ , $ \tilde{b}_r$\\ to a basis $ b_r$ of $ { \mathbb{v}}_{r}$.}\ar@{.>}[u]&*+[f-:<3pt>]\txt{set $ { \boldsymbol{\tilde{f}_r}}$ by~\eqref{equ : defineftilde}}\ar[r]&*+[f--]\txt{{\bf solve problem~{{\sf ce } } } } \ar@{.>}@/^1.pc/[ll]|{\tilde{b}_r}\ar[u]^<(0.26){\tilde{b}_r=\{\}}\ar[d]^{\tilde{b}_r\neq\{\}}\\ & * + [ f=]{\txt{task : find an $ ( { \boldsymbol{f_{r-1}}},{\mathfrak{d}},{\mathbb{h}}'_{r-1})$--complete { $ \sigma^*$}-extension\\ $ { ( { { \mathbb{h}}'_{r-1}(t)(t_1)\dots(t_e)},{\sigma})}$ of $ { ( { { \mathbb{h}}_{r-1}(t)(t_1)\dots(t_e)},{\sigma})}$ with\\ extension depth $ \leq{\mathfrak{d}}$ which is an ordered { $ \pi\sigma^{\delta}$}-extension\\ of $ { ( { { \mathbb{g}}},{\sigma})}$ ; find a basis $ b_{r-1}$ of $ { { \rm v}({\boldsymbol{f_{r-1}}},{{\mathbb{h}}'_{r-1}[t]_{r-1}})}$.}}\ar@{.>}@/^2pc/[lu]|<(.33){b_{r-1 } } & * + [ f-:<3pt>]\txt{set $ { \boldsymbol{f_{r-1}}}$ by~\eqref{equ : definefdeltam}}\ar[l ] } \ ] ] [ exp : truncatedfullpoly ] we continue the reduction from example [ exp : truncatedfullrat ] with @xmath573 . @xmath573 : : : set @xmath873 . by we get @xmath574 . clearly , @xmath196 is @xmath874complete with the basis @xmath576 of @xmath575 . @xmath577 : : : we get @xmath875 by and @xmath579 by . we can construct the @xmath43-extension @xmath595 with @xmath137 which gives the ordered @xmath175-field @xmath876 and which is @xmath877complete ; a basis of @xmath878 is @xmath879 ; see example [ exp : truncatedsub ] . this gives @xmath880 . if @xmath530 , we need a @xmath43-extension @xmath881 of @xmath882 with extension depth @xmath702 which is an ordered @xmath175-extension of @xmath64 and which is @xmath883complete . [ exp : truncatedfullpoly2 ] we continue example [ exp : truncatedfullpoly ] for the case @xmath530 . by ex . [ exp : truncatedh ] we get the ordered @xmath175-field @xmath843 with @xmath884 which is @xmath885complete ; a basis of @xmath886 is @xmath887 . hence we obtain the bases @xmath888 of @xmath889 and @xmath890 of @xmath891 . note that @xmath609 is a vector in @xmath892 which is a smaller field in the following sense : @xmath893 is a @xmath43-extension of @xmath196 with extension depth @xmath766 , but the extension @xmath45 with @xmath894 is eliminated ( it pops up in the tower @xmath895 above ) . eventually , all extensions with depth @xmath734 are eliminated , and we get a difference field with depth @xmath896 ; see section [ sec : completephase ] . [ exp : truncatedsub ] we are given the ordered @xmath175-field @xmath897 over @xmath36 with @xmath898 and @xmath899 . following figure [ fig : ratred ] we get @xmath842 . the degree bound is @xmath900 by . we start the reduction of figure [ fig : polydepthred ] with @xmath577 , set @xmath901 , and get @xmath902 by . a basis of @xmath903 is @xmath904 . hence @xmath905 by . now we need an ordered @xmath175-field @xmath906 which is a @xmath43-extension of @xmath897 with extension depth @xmath907 and which is @xmath908complete ; note that @xmath643 is eliminated . by example [ exp : truncatedh ] we get the @xmath175-field @xmath906 with @xmath909 and the basis @xmath910 of @xmath911 . completing the reduction , we get the basis @xmath912 of @xmath913 . by construction , @xmath906 is @xmath814complete . we sum up the construction from above : the derived @xmath172-extensions are defined by entries of some vectors @xmath723 which occur within the reduction process ; see section [ sec : completephase ] . internally , those vectors @xmath723 are determined by the reduction presented in figure [ fig : ratred ] ( the rational reduction ) and figure [ fig : polydepthred ] ( the polynomial reduction ) . exploiting additional properties in difference fields , we can predict how the @xmath723 and therefore the derived @xmath172-extensions look like . the first result is needed in lemma [ lemma : swapxy ] . [ cor : nopiext ] let @xmath914 be a @xmath3-extension of @xmath64 with @xmath915 such that @xmath914 can be brought to an ordered @xmath175-extension of @xmath64 ; let @xmath699 and @xmath704 . then there is a @xmath43-extension of @xmath914 over @xmath5 with extension depth @xmath702 which can be brought to an ordered @xmath175-extension of @xmath64 and which is @xmath700complete . if @xmath916 , the corollary follows by theorem [ thm : algtheoremb ] and proposition [ prop : propertiesinpisi].2 . let @xmath917 . we refine the inductive proof in section [ sec : constructdeltaext ] . suppose that the reduction holds for @xmath896 . as in section [ sec : constructdeltaext ] we assume that @xmath50 is a @xmath3-extension of @xmath64 ; if @xmath368 , then . now reorder @xmath914 to an ordered @xmath175-extension @xmath213 of @xmath64 . if @xmath918 , we construct a @xmath43-extension over @xmath5 as required ; see section [ sec : completephase ] . if @xmath919 , set @xmath824 with @xmath825 as in section [ sec : reductionphase ] with @xmath306 ; note that @xmath920 , since @xmath917 . then by corollary [ cor : completeratcrit ] @xmath491^m$ ] . define @xmath643 by and set @xmath612 , @xmath850 . now we apply the reduction as given in figure [ fig : polydepthred ] . since @xmath613 is free of @xmath459 , we can apply the induction assumption : we can take as required a @xmath43-extension @xmath853 of @xmath921 where the new @xmath43-extensions do not depend on @xmath459 . note that @xmath922 is free of @xmath459 by proposition [ prop : propertiesinpisi].2 ( if @xmath45 is a @xmath43-extension ) or lemma [ lemma : noextforpi ] ( if @xmath45 is a @xmath44-extension ) . hence @xmath616 is free of @xmath459 . suppose we reach the base case ( after at most @xmath262 steps ) with @xmath923 , free of @xmath459 , where @xmath924 is a @xmath43-extension of @xmath827 which is free of @xmath459 . by the reduction of the extensions with depth @xmath925 , the corollary follows . corollary [ cor : polyext ] can be shown completely analogously by using the following lemma [ lemma : polyclosure ] . [ lemma : polyclosure ] let @xmath781 be a polynomial @xmath3-extension of @xmath196 . for all @xmath926 $ ] , @xmath927 $ ] iff @xmath926 $ ] . [ cor : polyext ] let @xmath781 be an ordered @xmath175-extension of @xmath64 with @xmath928 and @xmath929 such that the @xmath172-extension @xmath781 of @xmath196 is polynomial ; + let @xmath930^n$ ] and @xmath704 . ( 1 ) : : then there is @xmath43-extension @xmath931 of @xmath781 with @xmath932 for @xmath380 and @xmath933 which can be brought to an ordered @xmath175-extension of @xmath64 and which is @xmath700complete . ( 2 ) : : in particular , the extension @xmath931 of @xmath866 is polynomial . ( 3 ) : : we have @xmath934[t_1,\dots , t_e]$ ] . ( 4 ) : : it can be constructed explicitly , if @xmath64 is @xmath40-computable . consider the polynomial @xmath175-extension @xmath99 of @xmath86 with and let @xmath935 $ ] . by corollary [ cor : polyext ] our construction will always yield a @xmath43-extension @xmath37 of @xmath99 with @xmath936 where @xmath37 is a polynomial @xmath3-extension of @xmath86 ; in particular , any solution @xmath140 of is in @xmath937[s_1,\dots , s_r]$ ] . the building blocks from above can be summarized to algorithm [ alg : multiplesolvesolutionspace ] . [ alg : multiplesolvesolutionspace ] solving problem . ` finddepthcompleteext`(@xmath938 ) an ordered @xmath175-extension @xmath50 of a @xmath40-computable @xmath64 , @xmath699 , @xmath704 . a @xmath43-extension @xmath731 of @xmath50 s.t . @xmath731 is an ordered + @xmath175-extension of @xmath64 and is @xmath732complete ; a basis of @xmath939 . [ alg : basecase ] if @xmath735 or . ] ( @xmath940 and @xmath177 and @xmath70 for some @xmath713 ) then compute a basis @xmath463 of @xmath941 ; return @xmath942 . fi if @xmath393 and @xmath943 then let @xmath743 be minimal such that @xmath744 . execute @xmath944=`finddepthcompleteext`(@xmath945 ) . return @xmath946 fi if @xmath756 then @xmath947 [ alg : subext ] let . ] @xmath948=`finddepthcompleteext`(@xmath949 ) [ alg : extcommand]construct a @xmath43-extension @xmath792 of @xmath781 and a basis @xmath950 of + @xmath796 as in sec . [ sec : completephase ] . return @xmath951 fi let @xmath952 with @xmath953 ; if @xmath368 , then . [ alg : redphasestart ] follow the rational reduction as in figure [ fig : ratred ] . let @xmath506 be a basis of @xmath954 . if @xmath480 , then return @xmath955 fi [ alg : subext2 ] take @xmath491^m$ ] , @xmath643 by and . apply the polynomial reduction and @xmath531 , ce is solved by @xmath956:=`finddepthcompleteext`(@xmath957 . ] from figure [ fig : polydepthred ] for @xmath958 . if the reduction stops earlier , return the corresponding result . otherwise , take @xmath609 with the computed ordered @xmath175-extension @xmath924 of @xmath64 . [ alg : keepdepth ] execute @xmath959:=`finddepthcompleteext`(@xmath960 ) . [ alg : redphaseend ] finish the reductions from figures [ fig : polydepthred ] and [ fig : ratred ] ; let @xmath950 be the basis of @xmath961 . return @xmath962 . now suppose that we remove lines [ alg : subext ] and [ alg : extcommand ] and return instead @xmath963 where @xmath463 is a basis of @xmath941 . then this modified version of algorithm [ alg : multiplesolvesolutionspace ] boils down to the recursive reduction presented in section [ sec : reduction ] ; see remark [ rem : ratpolyred ] . in other words , the execution of lines [ alg : subext ] and [ alg : extcommand ] is the heart of our new algorithm . in the sequel , we will optimize this part further . for this task we will refine the reduction phase ( lines [ alg : redphasestart][alg : redphaseend ] ) as follows . let @xmath50 be a @xmath3-extension of @xmath14 over @xmath5 with @xmath964 for @xmath52 and @xmath756 ; moreover let @xmath965 and let @xmath966 . note : by executing ` finddepthcompleteext`(@xmath967 ) it calls itself with depth @xmath734 in line [ alg : keepdepth ] ; for all other recursive calls ( in line [ alg : subext2 ] , see footnote ) we use depth @xmath766 . finally , if we enter the completion phase ( lines [ alg : subext ] and [ alg : extcommand ] ) with depth @xmath734 we can assume that @xmath246 contains all the elements @xmath54 for @xmath968 . this follows by lemmata [ lemma : entriesingroundfield ] and [ lemma : entriesinpolcase ] . given such a refined reduction , we can simplify the completion phase as follows . [ lemma : basismani ] let @xmath37 be a @xmath3-extension of @xmath14 over @xmath5 with @xmath290 such that @xmath964 for @xmath52 and @xmath756 ; let @xmath20 and @xmath965 . let @xmath699 where the the last @xmath969 entries are @xmath291 for @xmath968 . then : ( 1 ) : : if @xmath14 is @xmath757complete , then @xmath37 is @xmath757complete . ( 2 ) : : if a basis of @xmath941 is given , by row - operations in @xmath970 over @xmath36 one can construct a @xmath43-extension @xmath758 of @xmath37 s.t . @xmath758 is @xmath971complete ; a basis of @xmath972 can be extracted with no extra cost . * ( 1 ) * suppose that @xmath37 is not @xmath757complete . then there is a @xmath43-extension @xmath381 of @xmath14 with @xmath973 and @xmath470 such that @xmath974 . by proposition [ prop : propertiesinpisi ] , @xmath975 for some @xmath976 and @xmath977 . thus there is @xmath978 such that @xmath979 and therefore @xmath14 is not @xmath757complete . + * ( 2 ) * suppose that the entries @xmath980 with @xmath968 occur in the last @xmath981 entries of @xmath246 ; in particular suppose that they are sorted in the order as the corresponding extensions @xmath855 occur in @xmath4 . take a basis of @xmath941 and apply row operations such that one gets a basis @xmath982 where @xmath983 is in reduced form . if @xmath615 is the identity matrix , @xmath37 is @xmath984complete by lemma [ lemma : shortcut].2 and we are done . + otherwise , let @xmath985 be the set of all integers @xmath986 such that the @xmath28th column does not have a corner element in @xmath615 . suppose that @xmath987 with @xmath988 . now consider the difference field @xmath758 extension of @xmath37 where @xmath989 is a rational function field and for all @xmath990 we have @xmath991 . we prove that this is a @xmath43-extension which is @xmath984complete . + let @xmath992 be minimal such that @xmath993 is not a @xmath43-extension . then there is a @xmath994 with @xmath995 . hence , by proposition [ prop : propertiesinpisi ] , @xmath996 with @xmath295,@xmath997 . hence @xmath998 for some @xmath999 ; thus @xmath1000 where 1 is at the @xmath1001th position . since @xmath615 is in row reduced form and the @xmath1001th position has no corner entry , we can not generate @xmath1002 ; a contradiction that @xmath463 is a basis . hence @xmath758 is a @xmath43-extension of @xmath37 . + we get besides @xmath463 the solutions @xmath1003 where the @xmath900 is at the @xmath1001th position and @xmath1004 where the @xmath900 is at the @xmath1005th position . since the @xmath249 elements in @xmath1006 are lin.independent , @xmath758 is @xmath984complete by lemma [ lemma : shortcut ] . clearly , @xmath1007 is a basis of @xmath972 . * crucial improvements . * the first consequence is that we can replace line [ alg : subext ] by executing the function call @xmath1008 then by prop . [ prop : extensionontop ] we get an ordered @xmath175-extension @xmath781 of @xmath64 . moreover , by lemma [ lemma : basismani].1 @xmath781 is complete , as needed in lemma [ lemma : depthoptbysubprob ] . finally , by lemma [ lemma : basismani].2 we simplify the construction of step [ alg : extcommand ] as follows : by analyzing the basis @xmath463 of @xmath1009 , that has been computed already in step [ alg : subext ] , we get a @xmath43-extension of @xmath1010 of @xmath64 which can be brought to an ordered @xmath175-extension @xmath1011 and which is @xmath1012complete ; we extract a basis of @xmath796 . [ exp : truncatedhsubcom ] in example [ exp : truncatedh ] we claim that @xmath170 with @xmath812 is @xmath814complete where @xmath813 . note that @xmath1013 . hence by lemma [ lemma : basismani].1 it suffices to show that @xmath14 is @xmath814complete . with our algorithm this can be easily checked ; during this check we get the basis @xmath1014 of @xmath941 . following the proof of lemma [ lemma : basismani].2 , @xmath817 is a @xmath43-extension of @xmath170 with @xmath1015 and we get the basis @xmath1016 of @xmath1017 . by lemma [ lemma : depthoptbysubprob ] @xmath1018 is depth - optimal . we emphasize that the modified algorithm differs from the reduction presented in section [ sec : reduction ] ( which is similar to karr s algorithm ) by just analyzing the sub - results and by inserting extensions if necessary . _ in a nutshell , running our new algorithm which computes an appropriate @xmath175-extension and which outputs the corresponding solution to problem pt is not more expensive than choosing such a @xmath175-extension manually and solving problem pt with the recursive algorithm from section [ sec : reduction ] ( or karr s algorithm ) . on the contrary , adjoining the extensions only when it is required during the reduction keeps the computations as simple and therefore as cheap as possible . _ we need the following preparation to prove result [ res : reordering ] . [ lemma : pinotneeded ] let @xmath914 be a @xmath44-extension of @xmath14 which can be brought to an ordered @xmath175-extension of @xmath64 . let @xmath7 and let @xmath37 be a @xmath43-extension of @xmath14 with extension depth @xmath734 and @xmath174 such that . then there is a @xmath43-extension @xmath195 of @xmath14 with extension depth @xmath702 and @xmath1019 such that @xmath349 and @xmath1020 . write @xmath1021 with @xmath1022 . since we can bring @xmath914 to an ordered @xmath175-extension of @xmath64 , we can apply corollary [ cor : nopiext ] : there is a @xmath43-extension @xmath195 of @xmath914 over @xmath5 with @xmath1023 which can be brought to an ordered @xmath175-extension of @xmath64 and in which we have @xmath1019 s.t . @xmath349 ; by proposition [ prop : propertiesinpisi].2 , @xmath1024 . by theorem [ thm : constructsigmaextontopordered ] we can take a @xmath43-extension @xmath1025 of @xmath195 and an @xmath1026-monomorphism @xmath1027 s.t . for all @xmath1028 . note that @xmath1029 . since @xmath1030 , @xmath1031 for some @xmath1032 therefore , @xmath1033 . since @xmath1034 for @xmath380 , @xmath1035 . by construction , @xmath1036 is a @xmath43-extension of @xmath14 with extension depth @xmath702 . [ lemma : swapxy ] let @xmath1037 be a @xmath175-extension of @xmath14 and suppose that @xmath366 and @xmath914 can be brought to ordered @xmath175-extensions of @xmath64 . then the @xmath3-extension @xmath1038 of @xmath14 is depth - optimal . first we show that @xmath914 is a @xmath175-extension of @xmath14 . if @xmath459 is a @xmath44-extension , we are done . otherwise , let @xmath459 be a @xmath43-extension with @xmath441 which is not depth - optimal . hence , we can take a @xmath43-extension @xmath1039 of @xmath14 with extension depth @xmath173 and @xmath286 such that . there are two cases . + * case 1a : * @xmath277 is a @xmath44-extension . by corollary [ cor : extendpioversigma ] @xmath1040 is a @xmath44-extension of @xmath1039 . by reordering , @xmath1041 is a @xmath43-extension of @xmath366 . consequently , @xmath1037 is not a @xmath172-extension of @xmath366 , a contradiction . + * case 1b : * @xmath277 is a @xmath43-extension . bring @xmath366 to an ordered @xmath175-extension @xmath195 of @xmath64 . hence , by thm . [ thm : constructsigmaextontopordered ] there is a @xmath43-extension @xmath37 of @xmath195 with extension depth @xmath173 and an @xmath5-monomorphism @xmath1042 with @xmath1043 . since @xmath1044 ( as fields ) , @xmath1037 is not a @xmath172-extension of @xmath366 ; a contradiction . second , we show that @xmath1038 is a @xmath175-extension of @xmath914 . if @xmath277 is a @xmath44-extension , we are done . otherwise , let @xmath277 be a @xmath43-extension . if @xmath459 is a @xmath43-extension and @xmath1045 , the statement follows by lemma [ lemma : shiftsumtoleft ] and by proposition [ prop : reorderequaldepth ] . what remains to consider are the cases that @xmath459 is a @xmath44-extension or that @xmath459 is a @xmath43-extension with @xmath1046 . now suppose that @xmath1038 is a @xmath43-extension of @xmath914 with @xmath1047 which is not depth - optimal . hence , we can take a @xmath43-extension @xmath1048 of @xmath914 with extension depth @xmath173 and @xmath1049 such that . + * case 2a : * @xmath459 is a @xmath44-extension . by lemma [ lemma : pinotneeded ] , there is a @xmath43-extension @xmath195 of @xmath14 with extension depth @xmath173 and @xmath1019 such that @xmath349 ; hence @xmath366 is not a @xmath172-extension of @xmath14 , a contradiction . + * case 2b : * @xmath459 is a @xmath43-extension with @xmath1050 . reorder @xmath914 to a @xmath175-extension of @xmath64 . by theorem [ thm : depthstableordered ] @xmath1051 . since @xmath1052 , @xmath140 is free of @xmath459 . hence , @xmath1039 is a @xmath43-extension of @xmath14 with @xmath286 such that , and therefore @xmath366 is not a @xmath172-extension of @xmath14 ; a contradiction . @xmath791 * result [ res : reordering ] . * if @xmath1053 nothing has to be shown . let @xmath1054 be a @xmath175-extension of @xmath64 with @xmath393 and suppose the theorem holds for @xmath393 extension . choose any possible reordering . if @xmath277 stays on top , by the induction assumption all extensions below are depth - optimal . @xmath277 remains depth - optimal , since the field below has not changed . this shows this case . otherwise , suppose that @xmath855 for some @xmath52 is on top . then we can reorder our field to the @xmath3-extension @xmath1055 of @xmath64 with @xmath1056 . by the induction assumption we can bring @xmath1057 and @xmath272 to ordered @xmath175-extensions of @xmath64 . thus , we can apply lemma [ lemma : swapxy ] and get the @xmath175-extension @xmath1058 of @xmath64 . by the induction assumption we can bring the extensions in @xmath51 to the desired order without changing the @xmath175-property . @xmath791 * result [ res : depthconstruct ] . * this follows by theorem [ thm : extendforteleordered ] , corollary [ cor : polyext ] and result [ res : reordering ] . in particular , @xmath37 and @xmath174 can be computed as follows . 1reorder @xmath14 to an ordered @xmath3-extension of @xmath64 . + 2execute @xmath1059:=`finddepthcompleteext`(@xmath1060 ) and extract @xmath140 from @xmath463 s.t . . @xmath791 * result [ res : depthstable ] . * this is a direct consequence of theorem [ thm : depthstableordered ] and result [ res : reordering ] . @xmath791 * result [ res : constructsigmaextontop ] . * this follows by theorem [ thm : constructsigmaextontopordered ] and result [ res : reordering ] . @xmath791 * result [ res : extensiondepthstable ] . * this is implied by the following more general statement : there is a @xmath43-extension @xmath195 of @xmath37 and a @xmath175-extension @xmath422 of @xmath14 with an @xmath5-isomorphism @xmath1061 as in for all @xmath1028 ; we can assume that @xmath4 is ordered . + we prove this result by induction on the number of extensions in @xmath4 . for @xmath1062 , take @xmath1063 and @xmath1064 with @xmath1065 . now suppose we have shown the result for @xmath1066 with @xmath393 . i.e. , we are given a @xmath175-extension @xmath422 of @xmath14 , a @xmath43-extension @xmath195 of @xmath37 with @xmath1067 and an @xmath5-isomorphism @xmath1061 as in for all @xmath1028 . let @xmath1068 be a @xmath3-extension of @xmath37 with @xmath1069 . case 1 : * suppose that @xmath45 is a @xmath44-extension with @xmath1070 . then by corollary [ cor : extendpioversigma ] we can construct the @xmath44-extension @xmath282 of @xmath195 . moreover , by proposition [ prop : homproperties].2 we can construct the @xmath44-extension @xmath1071 of @xmath422 with @xmath1072 and can extend the @xmath5-isomorphism @xmath205 to @xmath1073 with @xmath1074 . by reordering , we get the @xmath43-extension @xmath1075 of @xmath1068 with the @xmath5-isomorphism @xmath1076 . as @xmath1077 , it follows that @xmath1078 . hence for all @xmath1079 . + * case 2 : * suppose that @xmath45 is a @xmath43-extension with @xmath1080 . we consider two subcases + * case 2a : * if there is a @xmath712 with @xmath1081 , let @xmath659 be minimal such that @xmath1082 . then by theorem [ thm : pisigma].1 there is the @xmath43-extension @xmath1083 of @xmath1084 with @xmath1080 . furthermore , there is an @xmath1085-isomorphism @xmath1086 with @xmath1087 by prop . [ prop : homproperties].1 . by reordering we get the @xmath43-extension @xmath1088 of @xmath1068 . now we can construct a @xmath43-extension @xmath1025 of @xmath1088 with an @xmath1089-isomorphism @xmath1090 by prop . [ prop : homproperties].3 . hence we arrive at an @xmath5-isomorphism @xmath1091 with @xmath1092 . finally , observe that for all @xmath1028 we have @xmath1093 and @xmath1094 . since @xmath1095 , @xmath1096 by result [ res : depthstable ] . with @xmath1097 it follows that @xmath1098 . since @xmath1099 for all @xmath1028 , we get @xmath1100 for all @xmath1079 . + * case 2b : * suppose that there is no @xmath712 with @xmath1081 . then there is no @xmath1101 with @xmath1102 . by result [ res : depthconstruct ] there is a @xmath172-extension @xmath1103 of @xmath422 such that @xmath1102 for some @xmath1104 . moreover , by proposition [ prop : homproperties].3 it follows that there is a @xmath43-extension @xmath1105 of @xmath195 and an @xmath5-isomorphism @xmath1106 where @xmath311 for all @xmath1028 . furthermore , we can construct the @xmath43-extension @xmath1107 of @xmath1105 with @xmath1080 by proposition [ prop : propertiesinpisi].1 . finally , we can construct the @xmath5-isomorphism @xmath1108 with @xmath1109 for all @xmath1110 and @xmath1111 by proposition [ prop : homproperties].1 . by reordering of @xmath1112 we obtain the @xmath43-extension @xmath1025 of @xmath1068 with @xmath1113 . as above , @xmath1114 . since @xmath1115 for all @xmath1028 , @xmath1116 for all @xmath1028 . thus @xmath1116 for all @xmath1079 . + note that this construction can be given explicitly , if @xmath64 is @xmath40-computable . @xmath791 * result [ res : prodfree ] . * the induction base @xmath285 is obvious . suppose result [ res : prodfree ] holds for @xmath359 extensions , and consider a @xmath3-extension @xmath1117 of @xmath14 with @xmath1118 s.t . . then by assumption there is a @xmath172-extension @xmath1119 of @xmath376 with @xmath1120 such that @xmath1121 and @xmath349 . now we apply result [ res : extensiondepthstable ] ( if @xmath327 is a @xmath43-extension ) and lemma [ lemma : pinotneeded ] together with result [ res : extensiondepthstable ] ( if @xmath327 is a @xmath44-extension ) : it follows that there is a @xmath172-extension @xmath195 of @xmath14 with @xmath1122 s.t . @xmath1123 and @xmath1124 . since @xmath1125 , we are done . @xmath791 * result [ res : prodsumdef ] . * this is a direct consequence of results [ res : depthconstruct ] and [ res : prodfree ] . @xmath791 * result [ res : dot ] . * let @xmath37 be such a @xmath172-extension of @xmath14 with @xmath20 ; take @xmath174 as in . * ( 1 ) * let @xmath196 be a @xmath3-extension of @xmath14 with @xmath348 s.t . @xmath349 . by result [ res : prodfree ] there is a @xmath172-extension @xmath195 of @xmath14 with @xmath1126 and @xmath1127 such that @xmath1128 and @xmath1129 . by result [ res : constructsigmaextontop ] we get a @xmath172-extension @xmath312 of @xmath37 and an @xmath5-monomorphism @xmath1130 as in for all @xmath1131 . hence @xmath1132 . since @xmath1133 and @xmath1134 , @xmath1135 for some @xmath1136 . hence @xmath1137 . + * ( 2 ) * suppose in addition that @xmath1138 and @xmath1139 . by the above considerations , @xmath1140 for @xmath380 and @xmath1135 for some @xmath1136 . hence there is an @xmath383 with @xmath380 s.t . @xmath1141 occurs in @xmath1142 . hence @xmath1143 . @xmath791 * result [ res : paratele ] . * by theorem [ thm : algtheorema ] we can take a @xmath172-extension @xmath37 of @xmath14 which is @xmath700complete . now let @xmath196 be any @xmath3-extension of @xmath14 with extension depth @xmath702 and @xmath374 , @xmath470 s.t . . then by results [ res : depthconstruct ] and [ res : dot].1 we take a @xmath172-extension @xmath195 of @xmath14 with @xmath1019 such that @xmath1144 and @xmath1121 . moreover , by result [ res : constructsigmaextontop ] we take a @xmath43-extension @xmath312 of @xmath37 with extension depth @xmath702 and an @xmath5-monomorphism @xmath1130 s.t . @xmath1145 for @xmath1146 . since @xmath1128 and @xmath37 is @xmath700complete , @xmath1147 ; in particular , @xmath1148 . @xmath37 can be constructed explicitly : 1reorder @xmath14 to an ordered @xmath3-extension of @xmath64 . + 2@xmath1059:=`finddepthcompleteext`(@xmath1149 ) . we conclude our article by non - trivial applications from particle physics @xcite . for the computations we used the summation package sigma @xcite which contains in its inner core our new difference field theory . in massive higher order calculations of feynman diagrams @xcite the sum @xmath1150 where @xmath1151 denotes the beta function @xcite , arose . it turns out that our refined creative telescoping method produces analogous to example [ equ : creasol] a recurrence with minimal order : @xmath1152 note that standard creative telescoping produces a recurrence of order @xmath1153 only ; see @xcite . given this optimal recurrence of order @xmath900 , the closed form denotes the riemann zeta function at @xmath28 ; e.g. , @xmath1154 . ] @xmath1155\\ & -2\tfrac{s_3(n)-\zeta_3}{n+2 } -\tfrac{s_2(n)-\zeta_2}{n+2}s_1(n ) + \tfrac{2 + 7n+7n^2 + 5n^3+n^4 } { n^3(n+1)^3(n+2)}s_1(n ) + 2\tfrac{2 + 7n+9n^2 + 4n^3+n^4 } { n^4(n+1)^3(n+2 ) } \end{split}\ ] ] can be read off immediately . + we remark that in this example the algebraic object @xmath1156 occurs which can not be handled in a direct fashion in @xmath3-fields . as it turns out , our algorithmic framework can be slightly extended such that it works also in this case ; the technical details are omitted here . as worked out in @xcite sigma could reproduce the evaluation of a feynman diagram that occurred in @xcite during the computation of the third - order qcd corrections to deep - inelastic scattering by photon exchange . more precisely , in mellin space the related feynman diagram could be expressed in terms of the recurrence @xmath1160 \underbrace{-\frac{1}{n+1}\sum_{k=4}^n(-1)^k\underbrace{\sum_{j=4}^k\frac{(-1)^j ( 3 j-2)}{(j-2 ) ( j-1 ) j}\overbrace{\sum_{i=4}^j\frac{f(i-3)}{(3 i-5 ) ( 3 i-2)}}^{=a(j)}}_{=b(k)}}_{=c(n)}$$\end{gathered}\ ] ] for constants @xmath1161 . checking initial values shows that @xmath1162 , @xmath1163 and @xmath1164 . now the main task is to simplify further . with , e.g. , karr s algorithm @xcite the inner sum @xmath1165 can be eliminated and one gets a rather big expression for @xmath1165 in terms of single nested harmonic sums @xmath1166 . in other words , we obtain an expression for where the depth is reduced by one . to get a representation with optimal nested depth , we execute our refined algorithm ; the result is an expression for @xmath1167 in terms of two nested sum expressions only : @xmath1168 note that the depth optimality of the sum representation is justified by results from @xcite . finally , splitting these sums by partial fraction decomposition , we get the solution @xcite : @xmath1169 & + s_2(n ) \big(\frac{4 s_3(n)}{n+1}-\frac{4 \zeta_3 } { n+1}\big)+s_{-3}(n ) \big(\frac{8(1+(-1)^n)}{n^2 ( n+1)}-\frac{4(2+(-1)^n ) s_{-2}(n)}{n+1}\\ & -\frac{4 s_2(n)}{n+1}\big)+s_{-2}(n ) \big(\frac{-12 \zeta_3 n^3+(-1)^n ( 8 - 8 n^3 \zeta_3)+8}{n^3 ( n+1)}-\frac{(4 + 8 ( -1)^n ) s_3(n)}{n+1}\big)\\[-0.1 cm ] & + \frac{8 s_{-3,-2}(n)}{n+1}+ \frac{(4 - 4(-1)^n)s_{-3,2}(n)}{n+1}+\frac{(4 + 12 ( -1)^n ) s_{-2,3}(n)}{n+1}-\frac{8s_{2,3}(n)}{n+1}.\end{aligned}\ ] ] for further examples how one can simplify dalembertian solutions @xcite with our algorithms see , e.g. , @xcite . we note that in the derived result no algebraic relations between the harmonic sums occur . in the next section we show how sigma eliminates , or equivalently , finds such algebraic relations explicitly and efficiently . during the calculation of feynman integrals harmonic sums arise frequently ; see , for instance , @xcite for further literature . in order to derive compact representations of such computations , one can use , e.g. , results from @xcite where all relations of harmonic sums are classified in general and tabulated up to nested depth @xmath1170 . alternatively , we illustrate how this task can be handled efficiently in the general @xmath175-field setting . consider , e.g. , the sums @xmath1171 , @xmath1172 , @xmath1173 , @xmath1174 , @xmath1175 , @xmath1176 , @xmath1177 which are algebraically independent except the last one : here the relation @xmath1178 pops up . with the naive reduction from section [ sec : reduction ] , sigma finds by representing the sums in a @xmath3-field @xmath1179 where the depths of the @xmath43-extensions @xmath1180 are : @xmath1181 respectively . with a standard notebook ( 2.16 ghz ) we needed @xmath1182 seconds to construct this @xmath3-field in order to get the relation . applying our new algorithms , we can represent the harmonic sums in a @xmath175-field with again @xmath1183 extensions , but this time the depths are @xmath1184 respectively . e.g. , the extensions correspond to the sum representations and note that the representation of @xmath1172 and @xmath1171 in the corresponding @xmath175-field has been carried out in details in example [ exp : s12s21 ] . in total we needed 37 seconds ( instead of 772 seconds ) to construct the underlying @xmath175-field . based on this optimal @xmath175-field representation , by backwards transformation the relation can be found automatically . p. paule and a. riese . a mathematica q - analogue of zeilberger s algorithm based on an algebraically motivated aproach to @xmath0-hypergeometric telescoping . in m. ismail and m. rahman , editors , _ special functions , q - series and related topics _ , volume 14 , pages 179210 . fields institute toronto , ams , 1997 .

the large - order behavior of a perturbative expansion in gauge theories is inevitably dominated by the factorial growth of renormalon diagrams [ 1 - 4 ] . in the case of quantum chromodynamics ( qcd ) , the coefficients of perturbative expansions in the qcd coupling @xmath1 can increase dramatically even at low orders . this fact , together with the apparent freedom in the choice of renormalization scheme and renormalization scales , limits the predictive power of perturbative calculations , even in applications involving large momentum transfers , where @xmath1 is effectively small . a number of theoretical approaches have been developed to reorganize the perturbative expansions in an effort to improve the predictability of the perturbative qcd ( pqcd ) . for example , optimized scale and scheme choices have been proposed , such as the method of effective charges ( ech ) [ 5 ] , the principle of minimal sensitivity ( pms ) [ 6 ] , and the brodsky - lepage - mackenize ( blm ) scale - setting prescription [ 7 ] and its generalizations [ 8 - 20 ] . in ref.[4 ] , the resummation of the formally divergent renormalon series and the parametrization of related higher - twist power - suppressed contributions has been given . in general , a factorially divergent renormalon series arises when one integrates over the logarithmically running coupling @xmath2 in a loop diagram . such contributions do not occur in conformally invariant theories which have a constant coupling . of course , in the physical theory , the qcd coupling does run . among the fundamental predictions of qcd are asymptotic scaling laws for large - angle exclusive processes [ 21 - 28 ] . qcd counting rules were formalized in refs.[22,23 ] . these reactions probe hadronic constituents at large relative momenta , or equivalently , the hadronic wave function at short distances . in particular , the meson wave function determines the leading higher - twist contribution to meson production at high @xmath0 . the hadronic wave functions in terms of quark and gluon degrees of freedom play an important role in the quantum chromodynamics predictions for hadronic processes . in the perturbative qcd theory , the hadronic distribution amplitudes and structure functions which enter exclusive and inclusive processes via the factorization theorems at high momentum transfers can be determined by the hadronic wave functions , and therefore they are the underlying links between hadronic phenomena in qcd at large ( nonperturbative ) and small distances ( perturbative ) . if the hadronic wave functions were accurately known , then we could calculate the hadronic distribution amplitude and structure functions for exclusive and inclusive processes in qcd . the study of hadron production at large transverse momenta in hadronic interactions is a valuable testing ground of the perturbative regime of qcd , providing information on both the parton distribution functions ( pdfs ) in the proton , and the parton - to - hadron fragmentation functions ( ffs)[29 ] . the frozen coupling constant approach can be applied for investigation , not only for exclusive processes , but also for the calculation of higher - twist contributions to some inclusive processes , for example as large -@xmath3 meson photoproduction [ 30 ] , two - jet+meson production in the electron - positron annihilation [ 31 ] . in the frozen coupling constant approach was used refs.[30,31 ] for calculation of integrals , such as @xmath4 according to ref.[7 ] , in pqcd calculations , the argument @xmath5 of the running coupling constant should be taken equal to the square of the momentum transfer of a hard gluon in a corresponding feynman diagram , in both the renormalization and factorization scale . but defined in this way , @xmath6 suffers from infrared singularities . for example in ref.[32 ] , @xmath5 equals to @xmath7 and @xmath8 , where @xmath9 , @xmath10 are the subprocess s mandelstam invariants . therefore , in the soft regions @xmath11 , @xmath12 , the integrals in ( 1.1 ) diverge and we need some regularization methods of @xmath13 in these regions for their calculation . in ref.[33 ] , the authors investigated the phenomenology of infrared renormalons in inclusive processes . the dispersive approach has been devised to extend properly modified perturbation theory calculations towards the low - energy region [ 34 ] . connections between power corrections for the three deep inelastic scattering sum rules have also been explored in ref.[35 ] . investigation of the infrared renormalon effects in various inclusive and exclusive processes is one of the most important and interesting problems in the perturbative qcd . it is known that infrared renormalons are responsible for factorial growth of coefficients in perturbative series for the physical quantities . but , these divergent series can be resummed by means of the borel transformation [ 1 ] and the principal value prescription [ 36 ] , and effects of infrared renormalons can be taken into account by a scale - setting procedure @xmath14 at the one - loop order results . technically , all - order resummation of infrared renormalons corresponds to the calculation of the one - loop feynman diagrams with the running coupling constant @xmath15 at the vertices or , alternatively , to calculation of the same diagrams with nonzero gluon mass . studies of infrared renormalon problems have also opened new prospects for evaluation of power - suppressed corrections to processes characteristics [ 37 ] . power corrections can also be obtained by means of the landau - pole free expression for the qcd coupling constant . the most simple and elaborated variant of the dispersive approach , the shirkov and solovtsov analytic perturbation theory , was formulated in ref.[38 ] . by taking these points into account , it may be argued that the analysis of the next - to - leading order higher - twist effects on the dependence of the pion wave function in pion production at proton - proton collisions by the running coupling approach , are significant from both theoretical and experimental[39 ] points of view . in this work we apply the running coupling approach[40 ] in order to compute the effects of the infrared renormalons on the meson production in proton - proton collisions in next - to - leading order . this approach was also employed previously[41,42 ] to calculate the inclusive meson production in proton - proton and photon - photon collisions . the running coupling method in next - to - leading order pion and kaon electromagnetic form factor was computed in [ 43 ] . as we know , power - suppressed contributions to exclusive processes in qcd , which are commonly referred to as higher - twist corrections . the higher - twist approximation describes the multiple scattering of a parton as power corrections to the leading - twist cross section . we will show that higher - twist terms contribute substantially to the inclusive meson cross section at moderate transverse momenta . in addition , we shall demonstrate that higher - twist reactions necessarily dominate in the kinematic limit where the transverse momentum approaches the phase - spase boundary . a precise measurement of the inclusive charged pion production cross section at @xmath16 and @xmath17 is important for the proton - proton collisions program at the relativistic heavy ion collider ( rhic ) at the brookhaven national laboratory . another important aspect of this study is the choice of the meson model wave functions . in this respect , the contribution of the higher - twist feynman diagrams to a pion production cross section in proton - proton collisions is been computed by using various pion wave functions . also , higher - twist contributions which are calculated by the running coupling constant and frozen coupling constant approaches are been estimated and compared to each other . within this context , this paper is organized as follows : in sec.[ht ] , we provide formulas for the calculation of the contribution of the high twist diagrams . in sec . [ ir ] we present formulas and an analysis of the next - to - leading order higher - twist effects on the dependence of the pion wave function by the running coupling constant approach . in sec . [ lt ] , we provide formulas for the calculation of the contribution of the leading - twist diagrams . in sec . [ results ] , we give the numerical results for the cross section and discuss the dependence of the cross section on the pion wave functions . we present our conclusions in sec . the higher - twist feynman diagrams , which describe the subprocess @xmath18 for the pion production in the proton - proton collision are shown in fig.1 . in the higher - twist diagrams , the pion of a proton quark is directly observed . their @xmath19 power suppression is caused by a hard gluon exchange between pion constituents . the amplitude for this subprocess can be found by means of the brodsky - lepage formula [ 27 ] : @xmath20 in eq.(2.1 ) , @xmath21 is the sum of the graphs contributing to the hard - scattering part of the subprocess . the hard - scattering part for the subprocess under consideration is @xmath22 , where a quark and antiquark form a pseudoscalar , color - singlet state @xmath23 . here @xmath24 is the pion wave function , i.e. , the probability amplitude for finding the valence @xmath25 fock state in the meson carry fractions @xmath26 and @xmath27 , @xmath28 . remarkably , this factorization is gauge invariant and only requires that the momentum transfers in @xmath21 be large compared to the intrinsic mass scales of qcd . since the distribution amplitude and the hard - scattering amplitude are defined without reference to the perturbation theory , the factorization is valid to leading order in @xmath29 , independent of the convergence of perturbative expansions . the hard - scattering amplitude @xmath21 can be calculated in perturbation theory and represented as a series in the qcd running coupling constant @xmath30 . the function @xmath31 is intrinsically nonperturbative , but its evolution can be calculated perturbatively . in our calculation , we have neglected the pion and the proton masses . turning to extracting the contributions of the higher - twist subprocesses , there are many kinds of leading - twist subprocesses in @xmath32 collisions as the background of the higher - twist subprocess @xmath33 , such as @xmath34 , @xmath35 , @xmath36 etc . the contributions from these leading - twist subprocesses strongly depend on some phenomenological factors , for example , quark and gluon distribution functions in the proton and fragmentation functions of various constituents , etc . most of these factors have not been well determined , neither theoretically nor experimentally . thus they cause very large uncertainty in the computation of the cross section of process @xmath37 . in general , the magnitude of this uncertainty is much larger than the sum of all the higher - twist contributions , so it is very difficult to extract the higher - twist contributions . the production of a hadron at large transverse momentum , @xmath0 , in a hadronic collisions is conventionally analyzed within the framework of perturbative qcd by convoluting the leading - twist @xmath38 hard subprocess cross sections with evolved structure and fragmentation functions . the most important discriminant of the twist of a perturbative qcd subprocess in a hard hadronic collision is the scaling of the inclusive invariant cross section [ 22 - 25,44 ] , @xmath39 at fixed @xmath40 and center - of - mass angle @xmath41 . as we know in the original parton model [ 45 ] the power fall - off is simply @xmath42 since the @xmath43 subprocess amplitude for point - like partons is scale invariant , and there is no dimensional parameter as in a conformal theory . the detected hadron @xmath44 can be produced directly in the hard subprocess reaction as in an exclusive reaction , then such direct higher - twist processes can give a significant contribution since there is no suppression from jet fragmentation at large momentum fraction carried by the hadron , @xmath45 , and the trigger hadron is produced without any waste of energy[44 ] . the mandelstam invariant variables for subprocesses @xmath18 are defined as @xmath46 in our calculation , we have also neglected the quark masses . we have aimed to calculate the pion production cross section and to fix the differences due to the use of various pion model functions . we have used five different wave functions : the asymptotic wave function ( asy ) , the chernyak - zhitnitsky wave function [ 28,46 ] , the cleo wave function [ 47 ] , the braun - filyanov pion wave functions [ 48 ] and the bakulev - mikhailov - stefanis pion wave function [ 49 ] . it should be noted that the wave functions of pions also are developed in refs.[50 - 52 ] by the dubna group : @xmath47,\ ] ] @xmath48,\ ] ] @xmath49,\ ] ] @xmath50,\ ] ] @xmath51 @xmath52 @xmath53 where @xmath54 is the pion decay constant . here , we have denoted by @xmath55 , the longitudinal fractional momentum carried by the quark within the meson . then , @xmath56 and @xmath57 . the pion wave function is symmetric under the replacement @xmath58 . several important nonperturbative tools have been developed which allow specific predictions for the hadronic wave functions directly from theory and experiments . the qcd sum - rule technique and lattice gauge theory provide constraints on the moments of the hadronic distribution amplitude . however , the correct pion wave function is still an open problem in qcd . it is known that the pion wave function can be expanded over the eigenfunctions of the one - loop brodsky - lepage equation , _ i.e. _ , in terms of the gegenbauer polynomials @xmath59 @xmath60.\ ] ] the evolution of the wave function on the factorization scale @xmath61 is governed by the functions @xmath62 , @xmath63^{\gamma_n/\beta_0},\ ] ] @xmath64 in eq.(2.6 ) , @xmath65 are anomalous dimensions defined by the expression , @xmath66.\ ] ] the constants @xmath67 are input parameters that form the shape of the wave functions and which can be extracted from experimental data or obtained from the nonperturbative qcd computations at the normalization point @xmath68 . the qcd coupling constant @xmath13 at the one - loop approximation is given by the expression , @xmath69 here , @xmath70 is the fundamental qcd scale parameter , @xmath71 is the qcd beta function one - loop coefficient , respectively , @xmath72 the cross section for the higher - twist subprocess @xmath73 is given by the expression @xmath74 ^ 2}{{\hat s}^3}\left[\frac{1}{{\hat u}^2}+\frac{1}{{\hat t}^2}\right],\ ] ] where @xmath75+e_2\hat u\int_{0}^{1}dx\left[\frac{\alpha_{s}(\hat q_2 ^ 2)\phi_{\pi}(x,\hat q_2 ^ 2)}{1-x}\right].\ ] ] here @xmath76and @xmath77 , represent the momentum squared carried by the hard gluon in fig.1 , @xmath78 is the charge of @xmath79 and @xmath80 . the higher - twist contribution to the large-@xmath3 pion production cross section in the process @xmath81 is [ 53 ] : @xmath82 @xmath83 @xmath84 @xmath85 @xmath86 @xmath87 @xmath88 @xmath89 @xmath90 where @xmath91 is the transverse mass of pion , which is given by @xmath92 let us first consider the frozen coupling approach . in this approach , we take the four - momentum square @xmath93 of the hard gluon to be equal the pion s transverse momentum square @xmath94 . in this case , the qcd coupling constant @xmath95 in the integral ( 2.10 ) does not depend on the integration variable . after this substitution calculation of integral ( 2.10 ) becomes easy . hence , the effective cross section obtained after substitution of the integral ( 2.10 ) into the expression ( 2.9 ) is referred as the frozen coupling effective cross section . we will denote the higher - twist cross section obtained using the frozen coupling constant approximation by @xmath96 . we have extracted the following higher - twist subprocesses contributing to the two covariant cross sections in eq.(2.11 ) @xmath97 by charge conjugation invariance , we also have @xmath98 in this section , we shall calculate the integral ( 2.10 ) using the running coupling constant method and also discuss the problem of normalization of the higher - twist process cross section in the context of the same approach . as is seen from ( 2.10 ) , in general , one has to take into account not only the dependence of @xmath99 on the scale @xmath100 , but also an evolution of @xmath101 with @xmath100 . the meson wave function evolves in accordance with a bethe - salpeter - type equation . therefore , it is worth noting that , the renormalization scale ( argument of @xmath95 ) should be equal to @xmath102 , @xmath103 , whereas the factorization scale [ @xmath61 in @xmath104 is taken independent from @xmath105 , we assume @xmath106 . such a approximation does not considerably change the numerical results , but the phenomenon considered in this article ( effect of infrared renormalons ) becomes transparent . the main problem in our investigation is the calculation of the integral in ( 2.10 ) by the running coupling constant approach . this integral in the framework of the running coupling approach takes the form @xmath107 the @xmath108 has the infrared singularity at @xmath109 , if @xmath110 or @xmath111 , if @xmath112 and as a result integral @xmath113 diverges ( the pole associated with the denominator of the integrand is fictitious , because @xmath114 , and therefore , the singularity of the integrand at @xmath115 is caused only by @xmath116 ) . for the regularization of the integral , we express the running coupling at scaling variable @xmath116 with the aid of the renormalization group equation in terms of the fixed one @xmath13 . the renormalization group equation for the running coupling @xmath117 has the form [ 36 ] @xmath118 ^ 2-\frac{\beta_{1}}{16}[\alpha(\lambda q^2)]^3\ ] ] where @xmath119 the solution of eq.(3.2 ) , with the initial condition @xmath120 is [ 35 ] @xmath121^{-1}.\ ] ] this transcendental equation can be solved iteratively by keeping the leading @xmath122 and next - to - leading @xmath123 powers . for @xmath124 these terms are given by @xmath125}{[1+\ln\lambda / t]^2}\ ] ] the first term in eq.(3.4 ) is the solution of the renormalization group eq.(3.2 ) with leading power accuracy , whereas the whole expression ( 3.4 ) is the solution of eq.(3.2 ) with next - to - leading power accuracy . after substituting eq.(3.4 ) into eq.(2.10 ) we get : @xmath126 @xmath127 @xmath128 @xmath129^{\gamma_{n}/\beta_{0}}c_{n}^{3/2}(2x-1)\right]}{(1-x ) ( 1+\ln\lambda / t_{1})}-\ ] ] @xmath130^{\gamma_{n}/\beta_{0}}c_{n}^{3/2}(2x-1)\right ] \ln(1+\ln\lambda / t_{1})}{(1-x ) ( 1+\ln\lambda / t_{1})^2}+\ ] ] @xmath131^{\gamma_{n}/\beta_{0}}c_{n}^{3/2}(2x-1)\right]}{(1-x ) ( 1+\ln\lambda / t_{2})}-\ ] ] @xmath132^{\gamma_{n}/\beta_{0}}c_{n}^{3/2}(2x-1)\right]\ln(1+\ln\lambda / t_{2})}{(1-x ) ( 1+\ln\lambda / t_{2})^2}\ ] ] where @xmath133 and @xmath134 . it should be note that in [ 41 ] , we used @xmath135 . the integral ( 3.5 ) is common and , of course , still divergent , but now it is recast into a form , which is suitable for calculation . using the running coupling constant approach , this integral may be found as a perturbative series in @xmath13 @xmath136 the expression coefficients @xmath137 can be written as power series in the number of light quark flavors or , equivalently , as a series in power of @xmath71 : @xmath138 the coefficients @xmath139 of this series demonstrate factorial growth @xmath140 , which might indicate an infrared renormalon nature of divergences in the integral ( 3.5 ) and corresponding series ( 3.6 ) . the procedure for dealing with such ill - defined series is well known : one has to perform the borel transform of the series [ 15 ] @xmath141(u)=\sum_{n=0}^{\infty}\frac{d_n}{n ! } u^n,\ ] ] then invert @xmath142(u)$ ] to obtain the resummed expression ( the borel sum ) @xmath143 . after this we can find directly the resummed expression for @xmath144 . the change of the variable @xmath105 to @xmath145 , as @xmath146 . then @xmath147 @xmath148 the basic theoretical problem is how to define the borel sum @xmath142(u)$ ] of the integral in eq.(3.7 ) for the quantities we are interested in . in qcd this problem is usually solved using perturbative methods and calculating the corresponding multiloop feynman diagrams with a one - gluon line , dressed by the chains of fermion bubbles . for the calculation the expression ( 3.7 ) we will apply the integral representations of @xmath149 and @xmath150 [ 54,55 ] . after this operation , formula ( 3.7 ) is simplified and we can extract the borel sum of the perturbative series ( 3.6 ) and the corresponding borel transform in dependence from the wave functions of the meson , respectively . also after such manipulations the obtained expression can be used for numerical computations . it is convenient to use the following integral representation for @xmath149 and @xmath151 @xmath152 and @xmath153 where @xmath154 is the euler - mascheroni constant . after inserting eq.(3.8 ) and ( 3.9 ) into ( 3.7 ) . then , we obtain @xmath155 @xmath156 @xmath157 @xmath158 in the case of @xmath159 for @xmath143 , we get @xmath160-\frac{4\sqrt{3 } \pi f_{\pi } e_{1 } \beta_{1 } \hat t}{\beta_{0}^3 } \cdot\ ] ] @xmath161(1-c-\ln u-\ln { t_1})u+\ ] ] @xmath162-\frac{4\sqrt{3 } \pi f_{\pi } e_{2}\beta_{1 } \hat u}{\beta_{0}^3}\cdot\ ] ] @xmath163(1-c-\ln u-\ln { t_2})u.\ ] ] in the case of the @xmath164 wave function , we find @xmath165 @xmath166^{50/81 } \left[\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}- \left.\frac{20}{4-u}\right]\right]-\ ] ] @xmath167 @xmath166^{50/81 } \left[\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}- \left.\frac{20}{4-u}\right]\right](1-c-\ln u-\ln { t_1})u+\ ] ] @xmath168 @xmath166^{50/81 } \left[\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}- \left.\frac{20}{4-u}\right]\right]-\ ] ] @xmath169 @xmath166^{50/81 } \left[\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}- \left.\frac{20}{4-u}\right]\right](1-c-\ln u-\ln { t_2})u\ ] ] in the case of the @xmath170 wave function , we get @xmath171^{50/81}\cdot\ ] ] @xmath172 - 0.4125\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right]-\ ] ] @xmath174^{50/81}\cdot\ ] ] @xmath172 - 0.4125\left[\frac{\alpha_{s}(q^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right ] ( 1-c-\ln u-\ln { t_1})u+\ ] ] @xmath175^{50/81}\cdot\ ] ] @xmath172 - 0.4125\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right]-\ ] ] @xmath176^{50/81}\cdot\ ] ] @xmath172 - 0.4125\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right ] ( 1-c-\ln u-\ln { t_2})u\ ] ] also , in the case of the @xmath177 wave function , we get @xmath178^{50/81}\cdot\ ] ] @xmath172 - 0.244\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right]-\ ] ] @xmath179^{50/81}\cdot\ ] ] @xmath172 - 0.244\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right](1-c-\ln u-\ln { t_1})u+\ ] ] @xmath180^{50/81}\cdot\ ] ] @xmath172 - 0.244\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right]-\ ] ] @xmath181^{50/81}\cdot\ ] ] @xmath172 - 0.244\left[\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right]^{364/405}\cdot\ ] ] @xmath173\right ] ( 1-c-\ln u-\ln { t_2})u\ ] ] equations(3.11)-(3.14 ) is nothing more than the borel sum of the perturbative series ( 3.6 ) , and the corresponding borel transform in the case @xmath159 is @xmath141(u)=\frac{1}{1-u}-\frac{1}{2-u},\ ] ] in the case @xmath164 is @xmath141(u)=\frac{1}{1-u}-\frac{1}{2-u}+0.84\left(\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right)^{50/81 } \left(\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}-\frac{20}{4-u}\right),\ ] ] in the case @xmath170 is @xmath141(u)=\frac{1}{1-u}-\frac{1}{2-u}+0.405\left(\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right)^{50/81 } \left(\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}-\frac{20}{4-u}\right)-\ ] ] @xmath182 and in the case @xmath177 is @xmath141(u)=\frac{1}{1-u}-\frac{1}{2-u}+0.282\left(\frac{\alpha_{s}(p_{t}^2)}{\alpha_{s}(\mu_{0}^2)}\right)^{50/81 } \left(\frac{4}{1-u}-\frac{24}{2-u}+\frac{40}{3-u}-\frac{20}{4-u}\right)-\ ] ] @xmath183 the series ( 3.6 ) can be recovered by means of the following formula @xmath184(u)\mid_{u=0}\ ] ] the borel transform @xmath142(u)$ ] has poles on the real @xmath185 axis at @xmath186 which confirms our conclusion concerning the infrared renormalon nature of divergences in ( 3.6 ) . to remove them from eqs.(3.11)-(3.14 ) we applied the principal value prescription . hence , the effective cross section obtained after substitution of the expressions ( 3.10 - 3.14 ) into the expression ( 2.10 ) is referred as the running coupling effective cross section . we will denote the higher - twist cross section obtained using the running coupling constant approach by @xmath187 regarding the higher - twist corrections to the pion production cross section , a comparison of our results with leading - twist contributions is crucial . we take two leading - twist subprocesses for the pion production:(1 ) quark - antiquark annihilation @xmath188 , in which the @xmath189 meson is indirectly emitted from the gluon , @xmath190 and ( 2 ) quark - gluon fusion , @xmath191 , with subsequent fragmentation of the final quark into a meson , @xmath192 . the corresponding cross sections are obtained in @xmath193 @xmath194 for the leading - twist contribution , we find @xmath195 @xmath196 where @xmath197 @xmath198 and @xmath199 represents gluon and quark fragmentation functions into a meson containing gluon and quark of the same flavor . in the leading - twist subprocess , the @xmath189 meson is indirectly emitted from the gluon and quark with the fractional momentum @xmath45 . the @xmath200 function can be expressed in terms of the parton kinematic variables , and the @xmath45 integration can then be done . the final form for the cross section is @xmath201 @xmath202 @xmath203 @xmath204 in this section , we discuss the numerical results for next - to - leading order higher - twist effects with higher - twist contributions calculated in the context of the running coupling and frozen coupling approaches on the dependence of the chosen meson wave functions in the process @xmath205 . in the calculations , we use the asymptotic wave function @xmath206 , the chernyak - zhitnitsky @xmath207 , the cleo pion wave function [ 41 ] , the braun - filyanov pion wave functions [ 42 ] , and the bakulev - mikhailov - stefanis pion wave function [ 43 ] . for the higher - twist subprocess , we take @xmath208 and we have extracted the following four higher - twist subprocesses contributing to @xmath209 cross sections : @xmath210 @xmath211 , @xmath212 , @xmath213 contributing to cross sections . for the dominant leading - twist subprocess for the pion production , we take the quark - antiquark annihilation @xmath188 , in which the @xmath189 meson is indirectly emitted from the gluon and quark - gluon fusion , @xmath214 , with subsequent fragmentation of the final quark into a meson , @xmath215 . as an example for the quark distribution function inside the proton , the mrst2003c package [ 56 ] has been used . the higher twist subprocesses probe the meson wave functions over a large range of @xmath61 squared momentum transfer , carried by the gluon . therefore , in the diagram given in fig.1 we take @xmath216 , @xmath103 , which we have obtained directly from the higher - twist subprocesses diagrams . the same @xmath61 has been used as an argument of @xmath30 in the calculation of each diagram . the results of our numerical calculations are plotted in figs.2 - 15 . first of all , it is very interesting to compare the resummed higher - twist cross sections with the ones obtained in the framework of the frozen coupling approach . in figs.2 - 4 we show the dependence of next - to - leading order higher - twist cross sections @xmath217 calculated in the context of the running coupling constant approach and the ratios @xmath218/@xmath219 , @xmath217/@xmath220 as a function of the pion transverse momentum @xmath3 for different pion wave functions at @xmath221 . it is seen that the values of cross sections @xmath217 , and ratios for fixed @xmath222 and @xmath223 depend on the choice of the pion wave function . as seen from fig.2 the next - to - leading order higher - twist differential cross section is monotically decreasing with an increase in the transverse momentum of the pion . in figs.3 - 5 , we show the dependence of the ratios @xmath218/@xmath219 , @xmath217/@xmath220 , @xmath217/@xmath224 as a function of the pion transverse momentum @xmath3 for different pion wave functions . here @xmath217 , @xmath225 , @xmath224 are the higher - twist cross sections calculated in the context of the running coupling method , in the framework of the frozen coupling approach and is the leading - twist cross section , respectively . as seen from fig.5 , in the region @xmath226 next - to - leading order higher - twist cross section calculated in the context of the running coupling method is suppressed by about 1 - 2 orders of magnitude relative to the leading - twist cross section , but in the region @xmath227 is comparable with the cross section of leading - twist . in figs.6 - 8 , we have depicted next - to - leading order higher - twist cross sections@xmath217 , ratios @xmath228 , @xmath229 as a function of the rapidity @xmath222 of the pion at @xmath16 and @xmath230 . at @xmath16 and @xmath230 , the pion rapidity lies in the region @xmath231 . as seen from figs.6 - 8 , next - to - leading order higher - twist cross section and ratios have a different distinctive . in the region ( @xmath232 ) the next - to - leading order higher - twist cross section @xmath217 , is suppressed by about 1 - 2 order of magnitude relative to the leading - twist cross section . in the region ( @xmath233 ) , the ratio for all wave functions increase with an increase of the @xmath222 rapidity of the pion and has a maximum approximately at the point @xmath234 . besides that , the ratio decreases with an increase in the @xmath222 rapidity of the pion . as is seen from figs.7 - 8 , the ratio @xmath235 very sensitive to the choice of the meson wave functions . but , as seen from fig.8 , the ratio @xmath217/@xmath224 for all wave functions has a minimum approximately at the point @xmath234 . also , the distinction between @xmath236 with @xmath237 , @xmath238 , @xmath239 and @xmath240 have been calculated . for example , in the case of @xmath16 , @xmath221 , the distinction between @xmath236 with @xmath241(i = cleo , cz , bf , bms ) as a function of the pion transverse momentum @xmath3 is shown in table [ table1 ] . thus , the distinction between @xmath236 and @xmath242 is maximum at @xmath243 , with @xmath244 at @xmath245 but the distinction between @xmath236 with @xmath246 is minimum at @xmath245 , with @xmath244 at @xmath243 and increase with an increase in @xmath0 . such a behavior of @xmath235 may be explained by reducing all moments of the pion model wave functions to those of @xmath159 for high @xmath61 . also , we have calculated the distinction between @xmath236 with @xmath237 , @xmath238 , @xmath239 and @xmath240 as a function of the rapidity @xmath222 of the pion . for example , in the case of @xmath247 , @xmath248 the distinction between @xmath236 with @xmath241(i = cleo , cz , bf , bms ) as a function of the rapidity @xmath222 of the pion is presented in table [ table2 ] . we have also carried out comparative calculations in the center - of - mass energy @xmath17 . the results of our numerical calculations in the center - of - mass energies @xmath249 are plotted in figs.9 - 15 . analysis of our calculations at the center - of - mass energies @xmath16 and @xmath17 , show that with the increase in beam energy values of the cross sections , ratio @xmath250 , and contributions of next - to - leading order higher - twist to the cross section decrease by about 1 - 2 order . therefore the experimental investigation of higher - twist effects include renormalon effects conveniently in low energy . on the other hand , the higher - twist corrections and ratios @xmath235 a very sensitive to the choice of the pion wave function . analysis of our calculations at the center - of - mass energies @xmath16 and @xmath17 , show that values of the next - to - leading order cross sections decrease by about @xmath251 percent of magnitude relative to the leading - order cross sections . also , the distinction between @xmath236 with @xmath237 , @xmath238 , @xmath239 and @xmath240 have been calculated . for example , in the case of @xmath17 , @xmath221 , the distinction between @xmath236 with @xmath241(i = cleo , cz , bf , bms ) as a function of the pion transverse momentum @xmath3 is shown in table [ table3 ] . thus , the distinction between @xmath236 with @xmath241,(i= cleo , bf , ) is maximum at @xmath252 , with @xmath244 at @xmath253 , but the distinction between @xmath236 with @xmath238 , @xmath237 , @xmath239 is minimum at @xmath253 , with @xmath244 at @xmath254 and increase with an increase in @xmath0 . also , we have calculated the distinction between @xmath236 with @xmath237 , @xmath238 , @xmath239 and @xmath240as a function of the rapidity @xmath222 of the pion . for example , in the case of @xmath255 , @xmath256 the distinction between @xmath236 with @xmath241 as a function of the rapidity @xmath222 of the pion is presented in table [ table4 ] . the calculations show that the ratio @xmath241/@xmath236 , ( i = cleo , cz , bf , bms ) for all values of the transverse momentum @xmath0 of the pion identically equivalent to ratio @xmath257/@xmath258 . in our calculations of the next - to - leading order higher - twist cross section of the process the dependence of the transverse momentum of meson appears in the range of @xmath259 . therefore , higher - twist cross section obtained in our paper should be observable at rhic in this work we have calculated the inclusive meson production via higher - twist mechanism and obtained the expressions for the subprocess @xmath260 cross section for mesons with symmetric wave functions . for calculation of the next - to - leading order cross section we have applied the running coupling constant method and revealed infrared renormalon poles in the cross section expression . infrared renormalon induced divergences have been regularized by the means of the principal value prescripton and the resummed expression ( the borel sum ) for the next - to - leading order higher - twist cross section has been found . in the resummed higher - twist cross section differs considerably from that found using the frozen coupling approach , some regions . also we have demonstrated that next - to - leading order higher - twist contributions to meson production cross section in the proton - proton collisions have important phenomenological consequences . our investigation enables us to conclude that the higher - twist pion production cross section in the proton - proton collisions depends on the form of the pion model wave functions and may be used for their study . analysis of our calculations shows that the magnitude of next - to - leading order cross sections calculated in the running coupling approach in some regions is larger than the leading - twist cross sections in 1 - 2 order . further investigations are needed in order to clarify the role of high twist effects in this process . we have demonstrated that the resummed result depends on the pion model wave functions used in calculations . the proton - proton collisions provide us with a new opportunity to probe a proton s internal structure . in particular , meson production in proton - proton collisions takes into account infrared renormalon effects : this opens a window toward new types of parton distributions which can not be measured by the deep inelastic lepton - proton scatterings . finally , we discuss the phenomenological consequences of possible higher - twist contributions to meson production in proton - proton collisions in next - to - leading order at rhic . future rhic measurements will provide further tests of the dynamics of large @xmath0 hadron production beyond leading twist . the work presented in this paper was completed while one of the authors , a.i.ahmadov , was visiting the hecap section of the abdus salam ictp , trieste , italy . he would like to express his gratitude to the members of the section especially to the head prof . s. randjbar - daemi for their hospitality . financial support by ictp is also gratefully acknowledged . hooft , in the why s of subnuclear physics , erice , 1977 , edited by a. zichichi ( plenum , new york , 1979 ) , p.94 a. h. mueller , nucl . phys . * b250 * , 327 ( 1985 ) ; phys . lett . * b308 * , 355 ( 1993 ) . v. i. zakharov , nucl . b385 * , 452 ( 1992 ) . m. beneke , phys . rep . * 317 * , 1 ( 1999 ) . g. grunberg , phys . lett . * b95 * , 70 ( 1980 ) ; * b110 * , 501(e ) , ( 1982 ) ; phys . rev . * d29 * , 2315 ( 1984 ) . p. m. stevenson , phys . rev . * d23 * , 2916 ( 1981 ) . s. j. brodsky , g. p. lepage and p. b. mackenzie , phys . * d28 * , 228 ( 1983 ) ; g. p. lepage and p. b. mackenzie , ibid . * 48 * , 2250 ( 1993 ) . g. grunberg and a. l. kataev , phys . lett * b279 * , 352 ( 1992 ) . s. j. brodsky and j. rathsman , hep - ph/9906339 . h. j. lu , and c. a. r. sa de melo , phys . lett * b273 * , 260 ( 1991 ) ; * b285 * , 399(e ) ( 1992 ) ; h. j. lu , phys . rev . * d45 * , 1217 ( 1992 ) . c. p. lepage and p. b. mackenzie phys . * d48 * , 2250 ( 1993 ) . s. j. brodsky and h. j. lu , phys . rev . * d51 * , 3652 ( 1995 ) . j. rathsman , phys . rev . * d54 * , 3420 ( 1996 ) . m. neubert , phys . rev . * d51 * , 5924 ( 1995 ) . m. beneke and v. m. braun , phys . lett . * b348 * , 513 ( 1995 ) ; p. ball , m. beneke and v. m. braun , nucl . phys . * b452 * , 424 ( 1995 ) ; m. beneke , nucl . phys . * b405 * , 563 ( 1993 ) . c. n. lovett - turner and c. j. maxwell , nucl . phys . b432 * , 147 ( 1994 ) . s. j. brodsky , j. ellis , e. gardi , m. karliner , and m. a. samuel , phys . rev . * d56 * , 6980 ( 1997 ) . s. j. brodsky , g. t. gabadadze , a. l. kataev , and h. j. lu , phys . b372 * , 133 ( 1996 ) . s. j. brodsky , m. melles , and j. rathsman , phys . rev . * d60 * , 096006 ( 1999 ) . s. j. brodsky , m. s. gill , m. melles , and j. rathsman , phys . rev . * d58 * , 116006 ( 1998 ) . s. v. mikhailov , jhep 0706:009 ( 2007 ) . v. a. matveev , r. m. muradyan , and a. n. tavkhelidze , lett . nuovo cimento 7 , 719 ( 1973 ) . s. j. brodsky and g. r. farrar , phys . lett . * 31 * , 1153 ( 1973 ) . j. f. gunion , s. j. brodsky , r. blankenbecler , phys.rev . * d6 * , 2652 ( 1972 ) d. w. sivers , s. j. brodsky , r. blankenbecler , phys . rept . * 23 * , 1 ( 1976 ) . v. a. matveev , l. a. slepchenko , and a. n. tavkhelidze , phys . lett . * b100 * , 75 ( 1981 ) . g. l. lepage and s. j. brodsky , phys . rev . * d22 * , 2157 ( 1980 ) . v. l. chernyak and a. r. zhitnitsky , phys . rep . * 112 * , 173 ( 1984 ) . franois arleo and david denterria , phys * d78 * : 094004 ( 2008 ) . j. a. bagger and j. f. gunion , phys * d25 * , 2287 ( 1982 ) . v. n. baier and a. grozin , phys . lett . * b96 * , 181 ( 1980 ) ; s. gupta , phys . rev . * d24 * , 1169 ( 1981 ) . a. i. ahmadov , i. boztosun , r. kh . muradov , a. soylu and e. a. dadashov , int . j. mod e15 * , 1209 ( 2006 ) . m. maul , e. stein , a. schfer , l. mankiewich , phys . lett.*b401*,100 ( 1997 ) . y. l. dokshitzer , v. a. khoze and s. i. troyan , phys . rev . * d53 * , 89 ( 1996 ) . a. l. kataev , mod . * a20 * , 2007 ( 2005 ) . h. contopanagos and g. sterman , nucl . phys . * b419 * , 77 ( 1994 ) . w. greiner , s. schramm and e. stein , quantum chromodynamics , 2nd edn.(berlin , springer , 2002 ) , pp.551 . d. v. shirkov and i. l. solovtsov , phys . lett . * 79 * , 1209 ( 1997 ) . star collaboration ( j.adams et.al ) , phys b637 * , 161 ( 2006 ) . agaev , phys . lett . * b360 * , 117 ( 1995 ) ; * b369 * , 379(e ) ( 1996 ) . a. i. ahmadov , coskun aydin , sh . m. nagiyev , yilmaz a. hakan , and e. a. dadashov , phys * d80 * , 016003 ( 2009 ) . a. i. ahmadov , coskun aydin , e. a. dadashov and sh . m. nagiyev , phys . rev . * d81 * , 054016 ( 2010 ) . agaev , modern physics letters * a13 * , 2637 ( 1998 ) . franois arleo , stanley j. brodsky , dae suang hwang , and anne m.sickles , phys * 105 * , 062002 ( 2010 ) . s. m. berman , j. d. bjorken and j. b. kogut , phys . rev . * d18 * , 900 ( 1978 ) . v. l. chernyak and a. r. zhitnitsky , nucl . phys . * b201 * , 492 ( 1982 ) ; v. l. chernyak , a. r. zhitnitsky and i. r. zhitnitsky , nucl . phys . * b204 * , 477 ( 1982 ) . a. schmedding and o. yakovlev , phys d62 * , 116002 ( 2000 ) . v. m. braun and i. f. filyanov , z. phys , * c44 * , 157 ( 1989 ) . a. p. bakulev , s. v. mikhailov and n. g. stefanis , phys . lett , * b587 * , 91 ( 2004 ) . s. v. mikhailov and a. v. radyushkin , phys . rev . * d45 * , 1754 ( 1992 ) . bakulev , s. v. mikhailov , z. phys . * c68 * , 451 ( 1995 ) . a. e. dorokhov , jetp lett . * 77 * , 63 ( 2003 ) . j. f. owens , rev . phys . * 59 * , 465 ( 1987 ) . j. zinn - justin , phys . rept . * 70 * , 109 ( 1981 ) . a. erdelyi , higher transcendental functions ( mcgrow - hill book company , new york , 1953 ) , vol.2 . mrst2003c.f can be obtained from http://durpdg.dur.ac.uk/hepdata/pdf.html . see also , a. d. martin , r. g. roberts , w. j. stirling and r. s.thorne , hep - ph/0307262 ; r. s.thorne , hep - ph/0309343 .

post - main - sequence mass loss affects both present and future generations of stars . it determines the remnant state of stars of intermediate mass and below , as well as affecting the state and contents of the interstellar medium out of which new stars are formed . however , the rate and amount of mass lost by red giant branch ( rgb ) and asymptotic giant branch ( agb ) stars is a major uncertainty in stellar evolutionary theory . as @xcite and @xcite point out , existing formulations rely on empirical laws based on observations of population i stars . there is little theoretical guidance on how mass loss should be included in stellar evolutionary models , or on how it affects population synthesis models based on the stellar models . globular clusters ( gcs ) have long been the prime testing grounds for stellar evolution theory . with large populations of low - mass stars of the same age , metallicity , and distance , they should be ideal places to test ideas about mass loss . there are several observational confirmations of the theoretical prediction that stars more massive than @xmath2 lose a significant fraction of their mass after leaving the main sequence . these include initial - final mass estimates of white dwarfs ( e.g. * ? ? ? * ) and evidence from line profiles of red giants ( e.g. , * ? ? ? to explain the horizontal branch it was argued ( e.g. , * ? ? ? * ) that much of this mass is shed on the rgb , but as agb stars , post - agb stars , and pne are also observed in gcs @xcite some mass must be lost on the agb as well . there is also ample observational evidence that many gcs have stars which are currently losing mass , at rates of @xmath3 yr@xmath4 @xcite . from these observations it is difficult to obtain an overall picture of time - averaged mass loss rates and duty cycles and their relation to cluster properties such as metallicity , since only a few stars at a time lose mass . a more global picture of mass loss within gcs might be provided by the intra - cluster medium ( icm ) . mass lost from gc stars is expected to accumulate in the icm until the cluster passes through the galactic disk , when the icm will be stripped by ram pressure . the icm is expected to be mostly gaseous ( although the ionization state is unknown ) , with a small amount of dust formed in the extended atmospheres of red giants . the dust should be optically thin and heated by starlight to 5080 k @xcite ; its thermal emission should peak in the mid - to - far - infrared ( ir ) and could be detected as an ir excess at these wavelengths . although gc stars are clearly losing mass , detections of the icm have been surprisingly difficult . searches for intra - cluster gas have been conducted for several decades @xcite ; searches for dust @xcite are more recent but have yielded similar results , mostly upper limits . only one galactic gc , ( m15 ) , shows clear evidence for an infrared ( ir ) excess @xcite . diffuse infrared emission near may be associated with the cluster or with galactic cirrus , while has an infrared point source which may be intracluster dust @xcite . an additional cluster , , has a tentative detection of dust at submillimeter wavelengths @xcite . all of the detections imply dust masses well below the expected levels . this paper presents results of a _ @xcite survey of eight milky way gcs using 24 and 70 @xmath0 m observations with the multiband imaging photometer for _ spitzer _ ( mips ; * ? ? ? we describe our observational strategy and data analysis techniques in 2 . section 3 introduces our results , including the detection of excess ir emission in several of the target clusters and upper limits on the remaining clusters . section 4 explores the implication of these new mid - ir results , including a comparison with predicted dust masses . we conclude in 5 with a summary and suggested follow - up observations . the target list ( table [ tab : targets ] ) comprises the milky way globular clusters that we believed to be the best candidates for icm dust detection . these clusters are reasonably nearby , at high galactic latitude ( to reduce confusing fore- and back - grounds ) , and have high escape velocities . they are among the most massive galactic gcs . more massive clusters are more likely to harbor a larger number of stars in the short - lived superwind phase : if the icm is being removed from the cluster on short time scales , then we are more likely to see it in a more massive cluster where the icm material is being replenished more quickly . clusters with escape velocities that are much larger than agb wind speeds ( @xmath5 km s@xmath4 ) will also be more likely to retain icm material . at the time this program was planned , all had existing or scheduled _ spitzer _ mid - ir observations with the infrared array camera ( irac ; * ? ? ? observing clusters with a range of times since last disk passage is desirable , but we did not use this as a selection criterion because orbital information is only available for a fraction of milky way globular clusters ( e.g. , from * ? ? ? the clusters span a factor of thirty in metallicity ( 1.5 dex ) , important for testing the effects of metallicity on dust production and retention . all of the clusters are detected with iras @xcite , and both ngc 104 ( 47 tuc ) and ngc 362 contain stars with known circumstellar dust @xcite . irac photometry traces the rayleigh - jeans tail of the spectral energy distribution of globular cluster stars . although cold intracluster dust is not expected to radiate at irac wavelengths , understanding the near - ir sed is very useful for quantifying any ir excess from the dust . irac observations of the clusters were made over the period 2005 july 2006 september as part of _ spitzer _ program identification ( pid ) 20298 ( pi r. rood ) , except for ngc 5904 which was part of pid 3256 ( pi p. goudfrooij ) . dataset identifiers are given in table [ tab : iracphot ] . all but one of the clusters were observed in a single irac field of view ( @xmath6 ) in all four channels , with 90252 frames ( 12 s each ) per position ( 30 frames , 30 s each for ngc 5904 ) . ngc 104 was observed in a @xmath7 map . ngc 104 and ngc 6752 were observed in the high - dynamic - range ( hdr ) mode which includes short ( 0.6 s ) frames taken before the 12 s frames . the half - light radius of ngc 104 ( 190 ) is slightly too large to be fully covered by the pid 20298 observations . fortunately , observations of this cluster covering a wider field of view were available , taken as part of instrument commissioning in 2003 november ( pid 623 ) . these observations were taken in 12-second hdr mode with 3 frames per position , and cover a @xmath8 area . for all clusters , the _ spitzer _ science center ( ssc ) pipeline version s14 post - bcd mosaics were used for integrated photometry . apphot _ was used to measure integrated magnitudes within the half - light radii given in table [ tab : targets ] . for each cluster the background was measured in an annulus with inner and outer radii 108 and 117 , respectively , and subtracted from the integrated magnitudes . for ngc 104 and ngc 6752 , the mosaics made from the short - exposure frames were used to avoid bright star saturation ; the other clusters had no saturation problems . the magnitudes measured in the half - light radii were converted to total flux densities with multiplication by 2 , and corrected with application of the ` extended source calibration ' given by the ssc . the results are given in table [ tab : iracphot ] , with electronic links to the observational data . the uncertainties in the flux densities due to poisson noise from the cluster and sky background noise are very small ( @xmath9% ) ; uncertainties in the absolute ( a few percent ; * ? ? ? * ) and extended source calibrations ( 10% ) dominate the total flux measurements . the mips photometry astronomical observing template ( aot ) was used to produce @xmath6 images of the clusters at 24 and 70 @xmath0 m . this field size is comparable to the cluster sizes ( half - light radii from * ? ? ? * @xmath10 ) . although a somewhat larger field of view would have been desirable , this would have substantially increased the observing time . to set exposure times , we predicted the 24 and 70 @xmath0 m stellar fluxes from the clusters @xmath11-band fluxes ( measured from 2mass images ; * ? ? ? * ) assuming a rayleigh - jeans @xmath12 spectrum . the 24 @xmath0 m observations were required to characterize the cluster sed , so high - significance detections of individual stars were not required . the estimated 24 @xmath0 m flux was used to compute the expected average 24 @xmath0 m surface brightness inside the clusters half - light radii . the resulting values , 0.84.2 mjy sr@xmath4 , were expected to be detectable in 1 cycle of the photometry aot with 10 second frame times . the intra - cluster dust is expected to emit most strongly at 70 @xmath0 m ; however , even at this long wavelength , the integrated flux of the cluster stars is much brighter than the mips detection limits . we therefore targeted the exposure times to have strong detections of the integrated stellar photospheric flux ; any dust signal would appear as an ir excess at 70 @xmath0 m . the resulting predicted 70 @xmath0 m fluxes ranged from 6150 mjy , requiring 110 cycles of the mips photometry aot with 10 second frame times . the observations were made over the period 2006 june 2006 november under _ pid 30031 , a guaranteed time program of g. fazio . the data were processed with the s14 version of the mips pipeline . post - pipeline processing for the 24 @xmath0 m bcd images included ` self - calibration ' ( dividing by the normalized median image of ` off - source ' frames ) and mosaicking ; the 70 @xmath0 m bcd images were time- and column - filtered and then mosaicked . all mosaicking used the mopex program @xcite with the standard parameters recommended in ssc documentation . the 24 @xmath0 m mosaics have a pixel scale of 2.5 , and the 70 @xmath0 m mosaics have a scale of 5.0 . figures [ fig : images24 ] and [ fig : images70 ] show the cluster mosaic images . all of the clusters were detected at high signal - to - noise in the 24 @xmath0 m images . the same is not true at 70 @xmath0 m , and the major result of this paper is the lack of any strong dust signal from any of the eight clusters . as for irac , flux densities in the mips bands were measured with aperture photometry . inspection of figure [ fig : images24 ] shows that the mid - infrared light for several clusters ( ngc 5272 and ngc 6205 ) appears to have slightly different centers from the optical centers given by @xcite . we computed new centers by smoothing the 24 @xmath0 m images and computing the flux centroid of the smoothed images in a box of width 4 times the smoothed fwhm , centered on the optical center . the mid - ir centers are given in table [ tab : mipsphot ] . compared to the irac data , the choice of aperture for the mips images is less obvious : the intracluster medium seems more likely to sink to the center of the cluster potential well than to follow the same spatial distribution as the cluster stars . [ however , we note that both the extended emission detected in ngc 7078 by @xcite and the possible intracluster dust in ngc 6024 reported by @xcite are offset from the cluster centers . ] we decided to use two apertures for each cluster : the optical half - light radius as for the irac data , and a smaller ` core ' radius , to maximize signal - to - noise for detection of dust at the cluster center . the half - light radii are given in table [ tab : targets ] ; for these apertures , background annuli with inner and outer radii 150 and 175 were used . following @xcite , we chose a core aperture of 16 radius ( diameter twice the fwhm of the psf ) , with background annuli 120140 in radius . to allow for subtraction of stellar contribution to the 70 @xmath0 m signal ( see [ sec : nsflux ] ) , the 24 @xmath0 m photometry in this aperture was measured on versions of the images convolved to the 70 @xmath0 m resolution using a kernel described in @xcite . the results of the aperture photometry are listed in table [ tab : mipsphot ] , with electronic links to the observational data ; no color corrections have been made to these values . the only aperture correction made is a multiplicative factor of 1.13 applied to the large - aperture 70 @xmath0 m photometry . again following @xcite , uncertainties in the mips flux density were calculated as @xmath13 where @xmath14 accounts for the noise correlation between pixels due to re - sampling during mosaicing ( @xmath15 for 24 @xmath0 m and 2 for 70 @xmath0 m ) , @xmath16 is discussed below , @xmath17 is the standard deviation of the pixels in the background annulus , and @xmath18 and @xmath19 are the number of pixels in the photometry and background apertures , respectively . ( poisson noise from the clusters is negligable compared to the background noise . ) after examining noise in their 70 @xmath0 m background and photometry apertures , @xcite concluded that ` excess noise ' corresponding to @xmath20 was present in their photometry apertures , a conclusion validated by their distribution of signal - to - noise ratios . although this precise value may not apply to our observations , which were made in a slightly different manner , we adopt it here to be conservative . the uncertainties in table [ tab : mipsphot ] do not include uncertainties in the absolute calibration ( 2% at 24 @xmath0 m and 5% at 70 @xmath0 m ; * ? ? ? * ; * ? ? ? confusion noise due to extragalactic sources is not expected to affect these observations : @xcite derived the 70 @xmath0 m confusion limit due to extragalactic sources as 0.3 mjy , well below the flux regime of the observations presented here . confusion due to galactic cirrus , however , is an issue : @xcite computed the mips 70 @xmath0 m point source detection limits due to galactic cirrus confusion at the latitudes of our clusters ( @xmath21 ) as @xmath22 mjy , again below our detection limits . for larger apertures , however , the confusion noise and detection limits scale as @xmath23 where @xmath24 is the aperture size and @xmath25 is the cirrus power spectrum index . the scaling factors range from about 6 to several hundred for the half - light radius apertures used , and hence cirrus confusion will affect our ability to detect extended emission for clusters with large angular sizes . figure [ fig : seds ] shows the cluster spectral energy distributions ( seds ) from 0.3 to 100 @xmath0 m . the new irac measurements from table [ tab : iracphot ] and half - light radius mips measurements from table [ tab : mipsphot ] ( multiplied by 2 to convert to total fluxes , even though this may not be appropriate for non - stellar emission ) are shown as filled symbols . also plotted are extinction - corrected total optical and near - ir magnitudes given by @xcite and @xcite . iras measurements from the faint source catalog @xcite are also shown ; the 12 and 25 @xmath0 m measurements have been corrected for the cluster extent following the procedure used by @xcite and flux densities in all iras bands have been color - corrected using the values for a 5000 k blackbody given in the _ iras explanatory supplement_. upper limits to the 90 @xmath0 m flux densities given by @xcite are shown for the five clusters in common between the two samples . overplotted are black - body seds derived from fits to the flux densities ( equally - weighted ) at @xmath26 @xmath0 m . all are consistent with the expected average temperatures for globular cluster stars ( 45005500 k ) . a first check of our measurements is to compare them with previous work . spitzer _ and iras results are mutually consistent except for ngc 6341 , where the 70 @xmath0 m data are slightly above the iras 60 @xmath0 m upper limits . we suggest that the iras limit might be slightly too low if there is extended emission from the cluster . for ngc 104 , both the 60 and 70 @xmath0 m observations are below the predictions based on the black - body fit , possibly because of the large spatial extent of the cluster . two of our target clusters were observed with the phot instrument on iso by @xcite : using those authors method to convert surface brightness to flux density , their derived 70 @xmath0 m flux densities are @xmath27 mjy for ngc 104 and @xmath28 mjy for ngc 7252 . our measurements for these clusters are consistent with these values . the 70 @xmath0 m measurements are generally consistent with the 90 @xmath0 m upper limits given by @xcite , except for ngc 6341 , where the 70 @xmath0 m value appears to be too large . the detection for this cluster is at @xmath29 significance and may be spurious . figure [ fig : mips_compare ] compares the predicted ( from the black - body fit ) and measured flux densities ( the same values as in the previous figure ) at 24 and 70 @xmath0 m . we expected that any icm dust would likely be too cool to be detectable at 24 @xmath0 m ; however , circumstellar dust should be warmer and possibly detected at this wavelength . all of the 24 @xmath0 m flux densities are within 20% of the prediction ; six clusters are within 10% . this is reasonable accuracy considering the difficulties in deriving integrated measurements of these resolved objects . the non - detection of 24 @xmath0 m excesses in integrated photometry may simply reflect the small number of mass - losing agb stars compared to the bulk of the population . point - spread - fitting photometry of the 24 @xmath0 m images , while beyond the scope of this work , may reveal which stars are surrounded by circumstellar dust . seven of the eight clusters in the sample have measurements or upper limits on 70 @xmath0 m flux that are more than a factor of 2 above the black - body prediction . the exception is ngc 104 , for which the 70 @xmath0 m upper limit is _ lower _ than the predicted photospheric value . the problem could be background subtraction : the half - light radius is almost as large as the 70 @xmath0 m image and a `` background '' region is difficult to define . figures [ fig : seds ] and [ fig : mips_compare ] clearly demonstrate that only one of the clusters in our sample ngc 6341has a significant infrared excess . to quantify this , we computed ( limits on ) the 70 @xmath0 m non - stellar emission in both the half - light and core apertures . to compute the non - stellar emission in the half - light aperture , we assumed that any photospheric emission at 70 @xmath0 m follows the optical emission , i.e. , half of the total emission is contained within @xmath30 . as an estimate of the stellar emission at 70 @xmath0 m we use the predicted value from the black - body fit described above . thus the non - stellar emission at 70 @xmath0 m in the half - light radius is @xmath31 the fraction of stellar emission contained within the core aperture is not as straightforward to compute ; to estimate it , we scale the flux measured at 24 @xmath0 m in the core aperture ( recall that the image used to measure this flux was convolved to the 70 @xmath0 m resolution ) by the ratios of the predicted total fluxes at 70 and 24 @xmath0 m . therefore the non - stellar emission at 70 @xmath0 m in the core aperture is : @xmath32\ ] ] where the factor of 1.76 is the aperture correction for a point source measured in a 16 aperture . table [ tab : ns_dust ] gives the derived non - stellar fluxes for each cluster , together with the dust mass predictions and observational limits derived in [ sec : dust_limit ] . only one cluster shows even tentative evidence for the detection of excess ir emission at 70 @xmath0 m : ngc 6341 . figure [ fig : images70 ] shows that the apparent 70 @xmath0 m emission in this cluster comes from an extended region offset from the cluster core by roughly 43 arcsec , or about @xmath33 . within the emission region there are three separate peaks which contain about 60% of the total flux ; it is unclear whether these peaks are separate point sources ( none has a co - located 24 @xmath0 m source ) or merely represent structure in the emission . in comparison , the dust emission detected in ngc 7078 by @xcite is much rounder and is located closer to the center of that cluster ( @xmath34 ) ; however , the possible dust cloud in ngc 5024 detected by @xcite is also far from the cluster center , at @xmath35 . the iris 60 @xmath0 m sky maps do show some nebulosity near ngc 6341 , so it is possible that the detected flux is in fact not associated with the cluster ; however ngc 1851 , at a similar galactic latitude , also has some nearby nebulosity and does not have a 70 @xmath0 m detection . future data from the akari all - sky survey should have sufficient spatial resolution to address the issue of whether there is diffuse emission near ngc 6341 . the lack of any excess infrared emission in the 90 @xmath0 m akari observations of ngc 6341 by @xcite ( see fig . [ fig : seds ] ) casts doubt on the reality of our low - significance detection ; however , we retain it for analysis in the remainder of this paper . while only some of the clusters have spatially diffuse ir emission within @xmath30 , all of the 70 @xmath0 m images except those of ngc 104 and ngc 5904 do contain some faint point sources which are not concentrated toward the cluster centers . [ most of the point sources identified by @xcite are off the fields of view of our images . ] what is the nature of these sources ? the 70 @xmath0 m data provide only marginal detections and poor spatial localization , but the 24 @xmath0 m data provide additional clues to the nature of the sources . initially , visual examination was used to generate lists of possible point sources in each cluster image . using an estimated 70 @xmath0 m position as a starting point , we performed aperture photometry on the 24 and 70 @xmath0 m images using iraf / phot , with the photometry routine allowed to re - center the sources . several 70 @xmath0 m sources were off the edges of the 24 @xmath0 m images , or not visible at the shorter wavelength : the latter group could either be very red sources , or noise fluctuations in the 70 @xmath0 m images . standard photometric apertures as given in the mips data handbook were used ( at 70 @xmath0 m , aperture radius 18 and background annulus @xmath36 ; at 24 @xmath0 m , aperture radius 6 and background annulus @xmath37 ) and appropriate aperture corrections were applied . we did not attempt irac photometry for the 70 @xmath0 m sources : the irac images of the gcs are significantly more crowded , and we were doubtful about the reliability of measurements in these regions . the 70 @xmath0 m sources in the globular cluster fields could be stars within the cluster , unrelated galactic stars , or background galaxies . to discriminate between these possibilities , a color - magnitude diagram was made from the mips photometry , shown in figure [ fig : mips_cmd ] . for comparison , the agb and post - agb models ( restricted to oxygen - rich models with @xmath38 , which are most appropriate for comparison with gcs ) of @xcite are also shown . the model fluxes have been scaled down by a factor of 5 to reflect a dust - to - gas ratio appropriate for gc metallicities . the model fluxes are scaled to a common fiducial distance ( 8.5 kpc ) so the same practice was followed for the cluster sources . in addition to the evolved star models , we show the ( unscaled ) flux densities and ratios for 70 @xmath0 m sources in the elais - n2 field of the swire legacy survey @xcite ; these sources are predominantly extragalactic . all but one of the cluster field 70 @xmath0 m sources are very red , with typical values of @xmath39 , compared to the rayleigh - jeans value of @xmath40 . the cluster source colors are more similar to those of the galaxies than to most of the evolved - star models , and the stellar models with similar colors are significantly brighter . from the colors , we tentatively conclude that the 70 @xmath0 m sources in the cluster fields are likely background galaxies . the high galactic latitude of the clusters ( @xmath41 ) supports this conclusion , as do the results of @xcite from their study of 70 @xmath0 m sources near @xmath42 cen and those of @xcite . the number density of sources is also consistent with this conclusion : there are 23 sources in a total of 346 arcmin@xmath43 for a density of 0.066 arcmin@xmath44 . the number counts given by @xcite for the goods - north field show that there are 0.07 galaxies arcmin@xmath44 to flux limits of 11 mjy ( appropriate for our average field ) , comparable to the number in the cluster fields . @xcite observed ngc 104 and 362 at mid - infrared wavelengths from 2.424 @xmath0 m . those authors found eight red sources ( @xmath45 ) in ngc 362 ; no such sources were found in ngc 104 . comparing our images of ngc 362 to those of @xcite , we find that their red sources a , d and f are within our field of view . only source d is detected at 70 @xmath0 m ; it has @xmath46 , consistent with the colors of a galaxy . the non - stellar 70 @xmath0 m fluxes computed in [ sec : nsflux ] allow us to estimate the amount of dust present in these clusters . for the clusters with negative non - stellar fluxes , we computed the dust mass upper limits as @xmath47 . we follow @xcite in assuming any dust to be optically thin , so the dust mass is given by : @xmath48 } \frac{d^2_{\rm kpc}}{{\kappa}_{\nu } b_\nu(t_d)}\ ] ] where @xmath49 is the dust absorption coefficient and @xmath50 the planck function , both in cgs units . our data do not constrain the dust temperature , so we use the value found by @xcite for ngc 7078 ( @xmath51 k ) , and the same @xmath52 cm@xmath43 g@xmath4 . different assumptions about these two quantities can change the derived dust masses by several orders of magnitude ; for example , @xcite estimated the dust mass in ngc 7078 by using @xmath54 for amorphous fayalite at 50 k ; at 70 @xmath0 m this substance has @xmath55 cm@xmath43 g@xmath4 , or five times the value used by @xcite . on the opposite extreme , @xcite assume silicate grains with @xmath56 k ; with @xmath57 , the absorption coefficient at 70 @xmath0 m is 36 cm@xmath43 g@xmath4 . table 5 of @xcite shows the effects of different assumptions about the dust temperature on derived dust masses : changing from @xmath51 k to @xmath58 k , for the same composition , increases the inferred @xmath59 by about a factor of ten , while increasing from @xmath51 k to @xmath60 k decreases @xmath59 by about a factor of 2.5 . uncertainties in the dust composition and temperature therefore result in uncertainties of @xmath61 orders of magnitude in derived dust masses . table [ tab : ns_dust ] gives the dust mass upper limits derived from the 70 @xmath0 m measurements in both the half - light and core apertures . all values reported are @xmath62 upper limits , except for the possible ( @xmath63 ) detection of total dust mass in ngc 6341 . this is our only observation resulting in even a tentative detection ; we show it as a value rather than an upper limit in the table and following plots in order to give a visual estimate of the uncertainties . because the cluster emission is generally very faint , we have not attempted to subtract any point source emission from the total fluxes ; if point sources within the apertures are unrelated to the clusters then the true upper limits would be even lower . for all of the clusters in our sample , the limits on the total dust masses are @xmath64 ; the limits on core masses are an order of magnitude lower , @xmath65 . @xcite gave upper limits to the dust masses for ngc 104 and ngc 5272 ; our total value for ngc 104 is slightly above their limits while our core value is slightly below . our total value for ngc 5272 is comparable to that of @xcite for 70 k dust . our value for ngc 5272 is also below that given by @xcite for 80 k dust based on their observations of this cluster at millimeter wavelengths . our upper limits are above those of @xcite for the five clusters in common , due to the greater sensitivity of their observations . the dust mass limits we infer are consistent with measurements and upper limits on gas masses . converting dust masses to gas masses by multiplying by @xmath66}}$ ] , we derive gas masses or upper limits for our sample clusters in the ranges @xmath67 ( core ) and @xmath68 ( total ) . the only confirmed h i detection , @xmath69 in ngc 7078 @xcite , is nicely within this range . our inferred limits on the gas mass are consistent with the results of previous searches for intra - cluster neutral and ionized hydrogen , as summarized in @xcite . two clusters have reported detections of ionized gas : ngc 104 @xcite and ngc 1851 @xcite . @xcite derived a total gas mass of @xmath70 and stated that they expected essentially all the gas in the cluster to be ionized ; @xcite estimated the total mass of ionized hydrogen in ngc 1851 as @xmath71 . our gas mass upper limit for ngc 104 is about the same as the @xcite detection while our upper limit for ngc 1851 is above the @xcite level . to understand the significance of the dust masses and upper limits derived from the _ spitzer _ observations , we need theoretical estimates of the expected dust content of the clusters in the sample . various estimates of the dust mass within a globular cluster are possible , but little detailed modeling has been done with the exception of the work of @xcite . those authors combined dynamical models of star clusters with radiative transfer models to predict dust grain temperatures and luminosities for a few individual clusters , given the presence of @xmath72 of dust . while it would be valuable to repeat such modeling with modern radiative transfer codes and understanding of dust properties , for present purposes a simpler estimate of globular cluster dust content will suffice . following @xcite , the dust mass may be estimated by : @xmath73}}{100 } \label{eq : dustpred}\ ] ] where @xmath74 yr is the horizontal branch lifetime , @xmath75 is the time since last crossing of the galactic plane , @xmath76 is the number of stars on the horizontal branch , @xmath77 is the gas mass lost from each star , and @xmath78}$ ] is the cluster metallicity . the last factor scales for the change in gas - to - dust ratio with metallicity . values for @xmath75 are listed in table [ tab : targets ] ; for the two clusters not listed by @xcite we assume a lower limit of @xmath79 yr . to estimate @xmath76 we use the ` specific evolutionary flux ' method of @xcite ( see also * ? ? ? * ) : @xmath80 assuming @xmath81 stars yr@xmath4 l@xmath82 as appropriate for globular clusters , and @xmath83 @xcite using @xmath84 and @xmath85 given by @xcite . for one cluster , we can compare this dust mass estimate to the current dust production rate : @xcite have recently used _ data to identify the mass - losing stars in ngc 104 . summing their mass loss rates for the brightest stars ( @xmath86 ) and converting to a dust production rate with a gas - to - dust ratio of 200 , we estimate @xmath87 yr@xmath4 . when multiplied by @xmath88 yr , this yields a total dust content of @xmath89 , about twice that predicted from eq . [ eq : dustpred ] . the observations of @xcite are likely to be affected by blending ; correcting for this would lower the inferred dust mass to a value closer to the prediction . even without such a correction , the comparison gives us confidence that the dust mass predictions are of the correct order of magnitude . figure [ fig : dustpred ] compares the predicted and observed dust masses and upper limits . to augment our sample we analyse additional clusters observed by other groups : @xcite , @xcite , and ngc 7078 @xcite , and the six other clusters observed by @xcite : , , , , and . where several values for dust mass or upper limits are given in published works , only the one with assumed dust temperature closest to 70 k is used . all of our measurements are well below the predicted values ; for most clusters the disagreement is more than an order of magnitude . the same is true for clusters observed by other groups , in that only two clusters have the inferred dust mass close to the predicted value : ngc 6356 @xcite ( only for specific assumptions about the dust distribution ) , and ngc 5634 @xcite , one of the lowest - mass , lowest - metallicity clusters shown . while assuming a lower dust temperature or higher absorption coefficient would increase the dust masses inferred from the observations , it seems unlikely that this can explain the three order of magnitude discrepancy for clusters such as ngc 5272 . the other explanation is that one or more of the assumptions involved in eq . [ eq : dustpred ] is incorrect : what if globular cluster stars lose much less than @xmath90 on average , the number of horizontal branch stars does not scale with @xmath91 , the dust - to - gas ratio does not scale with metallicity , or the dust is removed from clusters on a timescale much shorter than the disk passage time @xmath75 ? direct observations of red giant mass loss and color - magnitude diagram modeling are consistent with mass losses of approximately @xmath90 @xcite , and the scaling of @xmath76 with @xmath91 is well - grounded in stellar evolutionary theory . we address the remaining factors below . with a small sample of clusters , and an even smaller number of low - significance detections , deriving conclusions about the effects of gc properties on mass loss is difficult . figure [ fig : masslim ] shows the dust masses and limits plotted as a function of @xmath78}$ ] , @xmath91 , @xmath75 , and @xmath92 . ( in this and the following figure , we plot the more stringent dust mass limits of @xcite for the five clusters in common with our sample . ) there are no clear correlations : while the clusters with large inferred dust masses tend to have large @xmath75 , consistent with the expectation that intracluster dust builds up between cluster passage through the disk , other clusters with similar values of @xmath75 have much smaller inferred masses . this agrees with the finding of @xcite that the icm dust in ngc 7078 must have been collecting for a time @xmath93 , and with the computation by @xcite that cluster dust lifetimes @xmath94 . the clusters with dust detections do not have the highest central escape velocities , suggesting that possession of a deep potential well is not in itself enough to guarantee the presence of an icm . the clusters cover a relatively small range in bolometric luminosity ; this parameter does not seem to have a strong effect on inferred dust mass . how does metallicity relate to intracluster dust ? the two clusters with the largest inferred dust masses ( ngc 7078 and ngc 6356 ) have the lowest and highest metallicities in the sample . ngc 5024 , with a tentative dust detection , also has a low metallicity . the globular cluster data do not help to clarify the controversy over the connection between metallicity and dust production : while some studies show that metallicity and mass loss are not correlated in carbon - rich agb stars in the magellanic clouds @xcite , others find that metal - poor carbon stars , like oxygen - rich giants , form less dust @xcite . the clusters considered here are almost all more metal - poor than the magellanic clouds , so their agb stars are expected to be oxygen , rather than carbon - rich ; however the effects ( if any ) of metallicity on dust mass loss are not evident from the present data . observations of mass - losing stars within gcs show that mass loss is occurring at roughly the expected rate , so the non - detection of the icm means that it must somehow be removed . one possible mechanism is ram - pressure stripping of the cluster icm by the galactic halo . support for this mechanism is provided by detections of diffuse x ray emission near several galactic globular clusters @xcite ; @xcite show that the temperature and luminosity of the diffuse emission is consistent with it arising from stripped intracluster gas , and that the energetics require nearly all of the cluster gas to be removed . those authors also state , to show diffuse x ray emission , a cluster should have a velocity relative to the galactic halo @xmath95 km s@xmath4 . we computed @xmath96 for our sample clusters from the space velocities given in @xcite , @xcite , or @xcite assuming a non - rotating halo ; these are plotted against dust masses and limits in the left panel of figure [ fig : dust_remove ] . the high velocities of both ngc 7078 and ngc 5024 appear to be problematic for halo stripping ; a measurement of the space velocity of ngc 6356 might help to clarify the situation . there is certainly no evidence that clusters with low @xmath96 have more dust ; however , the correlation of cluster kinematics with metallicity combined with any dependence on metallicity of the dust mass limits may also confuse the issue . another possible icm removal mechanism is kinetic energy injected by stellar collisions , as proposed by @xcite for m15 . those authors use an n - body model to estimate the time between collisions in the cluster core ; their result is of the same order of magnitude as a less - sophisticated estimate based on eq . 8 - 125 of @xcite : @xmath97 \frac{\sigma_c}{\rho_0 ^ 2 r_c^3}\ ] ] where @xmath98 is the number of stars within the cluster core ( of radius @xmath99 pc ) , @xmath100 is the central velocity dispersion in km s@xmath4 , and @xmath101 is the central mass density in @xmath102 pc@xmath103 . using the values for @xmath99 , @xmath101 , and @xmath104 , from from king - model fits given in @xcite , we computed @xmath105 for each cluster in the sample ; it is plotted against @xmath59 in the right panel of fig . [ fig : dust_remove ] . again , there is no clear correlation of the time between collisions and dust masses : while both ngc 6356 and ngc 5024 have long @xmath106 , so do many other clusters . ngc 7078 is core - collapsed and should have a short central collision time . we conclude that neither ram pressure stripping and stellar collisions as the cause of icm removal can be ruled out by the current data . the intracluster medium of globular clusters remains elusive . our sample comprised eight galactic globular clusters thought to have the best chance for retaining an intracluster medium . no significant extended emission is detected from seven of the eight target clusters : any such emission may well be hidden underneath confusion noise from galactic cirrus . one cluster , ngc 6341 , has tentative evidence for excess ir emission from cool dust , although the lack of a 90 @xmath0 m excess casts doubt on the detection . all of our inferred dust masses and upper limits are @xmath107 , compared to predicted dust masses ranging from @xmath108 for the same clusters . searches for intracluster gas have been similarly unsuccessful ( e.g. , * ? ? ? the lack of intracluster dust and gas implies that either ( 1 ) evolved stars in globular clusters make less dust than predicted , or ( 2 ) the icm escapes from the clusters more quickly than expected . the first possibility seems unlikely because of the evidence to the contrary in both ngc 7078 and ngc 5139 , where two of the three dustiest , most mass - losing stars have @xmath78 } < -2.0 $ ] @xcite . it is important to point out that icm escape is not just a trivial occurrence in the life of a globular cluster : substantial mass loss over a gc s history could affect its dynamical evolution and perhaps even its survival as a bound system @xcite . several mechanisms for icm escape have been discussed above , with energy injected by stellar collisions perhaps being favored over ram pressure from galactic halo gas . stellar winds provide another possible mechanism ; giant stars with wind velocities above the escape velocity of a typical cluster have been observed @xcite . while there is observational evidence that metallicity does not strongly affect the mass loss rate of individual stars @xcite , stellar wind velocities do appear to scale with metallicity @xcite . faster winds in agb stars in metal - rich clusters could therefore increase the rate at which the icm is driven out of the cluster . the existing small sample of clusters does not permit us to distinguish between the possibilities for icm removal . decades of searching have yielded few detections of the elusive globular cluster icm . a possible strategy for future observations hoping to detect the icm is to study the removal mechanisms in detail and identify those clusters most likely to retain some icm . observations of distant clusters which have had few interactions with the milky way could place limits on the importance of halo stripping , although the larger distance of such clusters reduces the effective sensitivity . while infrared facilities available in the near future will not greatly improve on the sensitivity of _ spitzer _ , observing a larger sample of clusters would allow a more thorough exploration of the dust content as a function of cluster properties . observations with future radio facilities such as the square kilometer array will place more stringent limits on the gas content . together with more sophisticated modeling efforts and detailed examination of mass loss in individual cluster stars , a multi - wavelength observational approach provides the best chance for finding and explaining the elusive intracluster medium in globular clusters . we thank the anonymous referee for a careful reading of the manuscript and helpful comments and r.t . rood for early access to the irac data on ngc 6205 and ngc 6341 . this work is based on observations made with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology under nasa contract 1407 . p. b. was partially supported by a discovery grant from the natural sciences and engineering research council of canada . m. l. b. , c. e. w. , e. p. , and r. d. g. are supported in part by nasa through _ spitzer _ contracts 1276760 , 1256406 , and 1215746 issued by jpl / caltech to the university of minnesota . lrlrrrr ngc 104 ( 47 tuc ) & 4.5 & @xmath109 & 65.9 & 190.1 & @xmath110&@xmath111 + ngc 362 & 12.1 & @xmath112 & 46.7 & 49.1 & @xmath113&@xmath114 + ngc 1851 & 8.5 & @xmath115 & 48.7 & 30.4 & @xmath116 & + ngc 5272 ( m3 ) & 10.4 & @xmath117 & 27.6 & 138.4 & @xmath118&@xmath119 + ngc 5904 ( m5 ) & 7.5 & @xmath120 & 37.2 & 105.9&@xmath121&@xmath119 + ngc 6205 ( m13 ) & 7.7 & @xmath122 & 32.1 & 101.4 & @xmath123&@xmath124 + ngc 6341 ( m92 ) & 8.2 & @xmath125 & 35.9 & 61.4 & @xmath126&@xmath111 + ngc 6752 & 4.0 & @xmath127 & & 114.8 & @xmath128 & + lrrrrl ngc 104 & 107 & 67.9 & 46.4 & 24.7 & , + ngc 362 & 7.10 & 4.82 & 3.44 & 1.73 & + ngc 1851 & 3.57 & 2.31 & 1.64 & 0.82 & + ngc 5272 & 8.67 & 5.68 & 3.63 & 1.85 & + ngc 5904 & 9.25 & 6.56 & 4.81 & 2.42 & + ngc 6205 & 9.66 & 6.43 & 4.35 & 2.17 & + ngc 6341 & 5.05 & 3.38 & 2.31 & 1.16 & + ngc 6752 & 13.25 & 8.41 & 6.32 & 3.42 & + llrrrrl ngc 104 & 00:24:07.4 @xmath12972:04:48&@xmath130 & @xmath131&@xmath132 & @xmath133 & , + ngc 362 & 01:03:14.8 @xmath12970:50:51&@xmath134 & @xmath135 & @xmath136 & @xmath137 & , + ngc 1851 & 05:14:06.8 @xmath12940:02:49&@xmath138 & @xmath139 & @xmath140 & @xmath141 & , + ngc 5272 & 13:42:11.6 + 28:22:47&@xmath142 & @xmath143 & @xmath144 & @xmath145 & , + ngc 5904 & 15:18:33.0 + 02:04:50&@xmath146 & @xmath147 & @xmath148 & @xmath149 & , + ngc 6205 & 16:41:39.9 + 36:27:28&@xmath150 & @xmath151 & @xmath152 & @xmath153 & , + ngc 6341 & 17:17:07.8 + 43:08:11&@xmath154 & @xmath155 & @xmath156 & @xmath157 & , + ngc 6752 & 19:10:52.0 @xmath12959:59:02&@xmath158 & @xmath159 & @xmath156 & @xmath160 & , + lrrrrr ngc 104 & @xmath161 & @xmath162 & @xmath163&@xmath164&@xmath165 + ngc 362 & @xmath166 & @xmath167 & @xmath168&@xmath169 & @xmath170 + ngc 1851 & @xmath171 & @xmath172 & @xmath173&@xmath174 & @xmath175 + ngc 5272 & @xmath176 & @xmath177 & @xmath178&@xmath179 & @xmath180 + ngc 5904 & @xmath181 & @xmath182&@xmath183&@xmath184 & @xmath185 + ngc 6205 & @xmath186 & @xmath187 & @xmath188&@xmath189 & @xmath190 + ngc 6341 & @xmath191 & @xmath192 & @xmath193&@xmath194 & @xmath195 + ngc 6752 & @xmath196 & @xmath197&@xmath198&@xmath199 & @xmath200 +

data from the large area telescope ( lat ) on the _ fermi _ gamma - ray space telescope have shown definitively that millisecond pulsars ( msps ) are gamma - ray emitters similar to young pulsars @xcite , and more than a quarter of all known msps outside of globular clusters were discovered in radio searches targeting _ fermi _ unidentified sources only within the past six years @xcite . _ fermi _ has been especially successful in uncovering msps in black widow and redback binary systems , where the pulsar s intense particle wind gradually ablates the companion . the main difference between the two classes is the mass and type of the companion . while in black widow systems the companion is degenerate , with a mass of 0.010.05 _ @xmath0 , redback pulsars have non - degenerate companions whose mass is an order of magnitude larger @xcite . both types are believed to be transitional between accreting low - mass x - ray binaries ( lmxbs ) , rotation - powered binary msps , and isolated msps . nineteen of 25 known black widows and nine of 13 known redbacks in the galactic field were discovered in radio searches of _ fermi _ unidentified sources . black widows and redbacks are relatively rare objects because they are selected against in undirected ( `` all sky '' ) surveys due to their intrinsic properties ( radio eclipses , high line - of - sight acceleration from orbital motion ) and because this stage in msp evolution is short compared to an msp s lifetime . consequently , there is much still to be learned about the details of the transition between accreting and rotation - powered msps ( @xcite , @xcite ) . this is especially true of redback pulsars , as they are the rarer and more recently discovered type . in this paper we present a detailed multiwavelength study of psr j1048 + 2339 , a redback radio pulsar found in a _ fermi _ source by @xcite . the pulsar s prolonged radio eclipses around superior conjunction as well as `` mini - eclipses '' at other orbital phases are reminiscent of the redback class archetype , psr j1023 + 0038 , whose transition from lmxb to radio msp was observed almost in real time ( @xcite , @xcite ) . in addition , psr j1048 + 2339 exhibits significant orbital period variations similar to those observed in the redbacks psr j1023 + 0038 , psr j2339@xmath30533 @xcite , and the original black widow pulsar b1957 + 20 @xcite . in section [ sec_radio ] we present two years of radio timing observations and polarimetry . in section [ sec_optical ] we focus on the optical properties of the system , and in section [ sec_gamma ] we discuss the gamma - ray counterpart . psr j1048 + 2339 was discovered in a search at 327 mhz with the arecibo telescope of high - latitude _ fermi _ unidentified sources from a preliminary version of the 3fgl _ fermi _ lat source catalog @xcite . the source selection criteria were ( 1 ) non - variability in gamma - rays , ( 2 ) pulsar - like gamma - ray spectrum , and ( 3 ) an error box 15 across or smaller , corresponding to the diameter of the arecibo beam at 327 mhz . the pulsar was discovered in a 15-minute observation targeting the _ fermi _ source 3fgl j1048.6 + 2338 on 2013 august 4 . psr j1048 + 2339 has a pulse period of 4.66 ms and a dispersion measure ( dm ) of 16.65 pc @xmath4 . according to the ne2001 model of electron density in the galaxy @xcite , this dm corresponds to a distance of 0.7 kpc . we performed timing observations of psr j1048 + 2339 at 26 epochs between 2013 august 4 and 2015 september 20 using the arecibo radio telescope at 327 mhz . the observations used the puppi backend in `` fast4k '' mode , with 4096 channels and a 81.92 @xmath5s sampling time . since the minimum bandwidth provided by puppi , 100 mhz , is larger than the receiver bandwidth of 68 mhz , only 2680 channels were recorded to disk , corresponding to the part of the puppi band covering the receiver bandwidth . after deriving a timing solution in 2015 july , we observed psr j1048 + 2339 at 1430 mhz at two epochs in coherent folding , full - stokes mode using puppi with a bandwidth of 700 mhz , 448 channels , and a 64 @xmath5s sampling time . many observations spanned eclipse ingress , egress , or showed significant short - timescale emission variations , especially near ingress ( fig . [ fig_ingress ] ) . we did not detect any radio emission in the orbital phase range @xmath6 0.020.49 , corresponding to the main eclipse when the pulsar is near superior conjunction with respect to the companion . sporadic `` mini - eclipses '' occur frequently at other orbital phases and their duration ranges from tens to hundreds of seconds . at two epochs we did not detect the pulsar even at orbital phases where it is normally visible , @xmath7 0.60.8 ( mjd 57343 and 57345 ; sessions 36 and 37 in fig . [ fig_det ] ) , indicating that the pulsed radio emission may be obscured by dense ionized gas for extended periods of time . in order to derive a phase - connected timing solution that accounts for every rotation of the pulsar between observations , we extracted times - of - arrival ( toas ) from folded 327 mhz data using a pulse profile template consisting of four gaussian components , two each for the main pulse and interpulse . we used the tempo software package and the btx binary model ( @xcite ; d. j. nice , unpublished ) to perform least - squares fitting of various pulsar parameters by minimizing the square of the differences between expected and measured toas , after excluding three points with large , apparently dispersion - induced delays . we scale the toa uncertainties by a factor of 1.47 in order to bring the reduced @xmath8 of the fit to unity . the resulting best - fit parameters and other derived pulsar properties are listed in table [ tab_pulsars ] . figure [ fig_res ] shows the timing residuals for all 327 mhz observations . there are no significant trends in the remaining timing residuals , indicating that at the current timing precision and for the current data set , the ephemeris presented in table [ tab_pulsars ] describes the toas . the most unusual timing property of j1048 + 2339 is its orbital period variation necessitating the fitting of five orbital period derivatives in order to obtain a timing model that fits the data well over the full time span of radio observations ( @xmath9 years ) . the relatively short span of the solution and the large number of orbital period derivatives required to model the data mean that there are significant correlations between some of the timing parameters ( e.g. the astrometric and orbital parameters ) . as a result , the tempo uncertainties of the celestial coordinates are significantly underestimated ( see section [ sec : energetics ] ) . two toas near eclipse egress ( orbital phase @xmath10 ) show an extra delay indicating that pulses are passing through the outer layers of the companion . one toa near orbital phase @xmath11 also shows an extra delay . such features are common in systems where gas turbulence and outflows may introduce short - lived dispersive delays in the radio emission at various orbital phases . we note that this feature is present in only one of several observations spanning that part of the orbit . we calculate that the extra delays observed for these three toas correspond to a dm excess of 0.008 , 0.003 , and 0.005 pc @xmath4 . assuming that the dm excess is incurred over a linear distance equivalent to the projected semimajor axis of the orbit ( or the radius of the companion , which is of the same order of magnitude ) , we obtain an intra - binary electron density on the order of @xmath12 @xmath4 . in comparison , @xcite find an intra - binary electron density of @xmath13 @xmath4 for the eclipsing msp b1744@xmath324a , and @xcite find a similar value for psr j2140@xmath32310a . we performed three observations of psr j1048 + 2339 using the arecibo l - wide receiver and the puppi backend with a bandwidth of 700 mhz centered on 1430 mhz . one of the observations was in search mode , for a duration of 15 minutes . the remaining two observations were in coherent dedispersion full - stokes mode , for 40 minutes each . the coherently dedispersed l - wide data were analyzed with psrchive @xcite and figure [ fig_pol ] shows the calibrated full - stokes pulse profile averaged from the two observations . the brightest component of the main pulse exhibits significant circular polarization , while the interpulse is linearly polarized . the pulse profile and linear polarization vector position angle are consistent between the two observations . the calibrated period - averaged flux densities are @xmath14 mjy and @xmath15 mjy , assuming a 20% uncertainty . the estimated rotation measure is @xmath16 rad m@xmath17 , consistent with zero . we computed the flux density from our search - mode observations at other frequencies ( where calibrated flux density measurements were not available ) using the radiometer equation @xcite . at 327 mhz , out of 80 data sets we excluded those spanning ingress , egress , or containing `` mini - eclipses '' and calculated a period - averaged flux density from each of the remaining 47 data sets . the 327 mhz arecibo receiver has gain @xmath18 k / jy . nominal receiver temperature is 95 k , including spillover . scaling the 408 mhz sky temperature @xcite to 327 mhz yields @xmath19 k. overall , @xmath20 k. the range of flux density values we obtain is @xmath21 0.55.3 mjy , with a mean of 1.6 mjy and a standard deviation of 1.1 mjy . we also analyzed two robert c. byrd green bank telescope ( gbt ) observations at 820 mhz from an earlier search of the _ fermi _ source 3fgl j1048.6 + 2338 where the pulsar was not detected in a blind periodicity search , but a pulsed signal became evident after the data were folded with our radio ephemeris . the observations used the guppi backend with a bandwidth of 200 mhz . the 820 mhz receiver has @xmath22 k / jy and @xmath23 k. the observation times are 10 and 99 minutes , and we obtain period - averaged flux densities of @xmath24 and 0.11 mjy , respectively . the difference in flux densities between observations at each of the three observing frequencies indicates significant scintillation and/or variability . because of this , and the fact that we have only two observations at 820 mhz and three at 1430 mhz , we refrain from calculating a spectral index , but it appears compatible with that of a typical pulsar . figure [ fig_multifreq ] shows averaged pulse profiles at 327 , 820 , and 1430 mhz aligned in phase using the radio ephemeris . there is significant profile evolution with frequency : the pulse at phase @xmath25 is brighter at 327 mhz but becomes weaker than the interpulse as the frequency increases . within the interpulse at phase @xmath26 in figure [ fig_multifreq ] , the trailing component is brighter at 327 mhz but the leading component dominates at the higher frequencies . using the radio timing position , we identified a 20th magnitude candidate optical counterpart to in the catalina real - time transient survey ( crts , @xcite ) second data release . the catalina i d number of the star is 1123054021805 . we extracted 8 years of crts data on this star , consisting of 403 observations on 119 nights from 2005 april 10 to 2013 june 3 ( figure [ fig : periodogram]a ) . on a typical night , four unfiltered 30 s exposures of a field are taken within 30 minutes , and the photometry is expressed as @xmath27 magnitudes ( discussed in section [ sec : interpret ] ) . the @xmath28 limiting magnitude is @xmath29 as we infer from those points whose error bars are @xmath30 mag . we first transformed the times of the data to barycentric dynamical time for consistency with the radio pulsar ephemeris . magnitudes were converted to fluxes , which were normalized by the variance of the data @xcite , and a lomb - scargle periodogram @xcite was calculated around the radio pulsar s orbital period . in this analysis we omitted high points that are clear outliers from the mean light curve , because they add noise at all frequencies . this was done crudely by excluding the 25 points with @xmath31 . although some may just be statistical fluctuations , 19 of them can be considered confirmed outliers because there are at least two on the same night . examples of excluded points are the four brightest in figure [ fig : periodogram]a , with magnitudes 18.1217.59 on mjd 54536 ( 2008 march 11 ) . they clearly represent some sort of flare , a non - stationary process . this simple procedure may cause us to underestimate the normal maximum flux , and it does not account for smaller flares that may occur during fainter phases of the orbit ; i.e. , we do not iterate the selection of outliers based on their phase in the folded light curve . a strong signal ( figure [ fig : periodogram]b ) is detected at @xmath32 d , consistent with the more precise orbital period of the radio pulsar , @xmath33 d. the noise power in the periodogram is measured to be exponentially distributed with a mean of 2.06 ( not 1 as expected for white noise , due to the closely correlated times of the nightly points ) , so the peak power of 42.0 has a single - trial chance probability of @xmath34 . since the optical search found a period consistent with the radio pulsar s orbital period , we folded the data on the precise radio ephemeris . more precisely , we used the orbital period and time of ascending node , but not the period derivative(s ) , as the latter are known to wander on long timescales in such systems and can become either positive or negative , as seen in @xcite and @xcite . for the purpose of displaying the folded data ( figure [ fig : fold]a ) , we restored the high points previously omitted as flares in the periodogram analysis . finally , we calculated a mean light curve by averaging the points ( by flux , but not weighted by error bars ) in 30 bins in orbital phase while excluding the flaring points . the mean light curve is displayed in the inset of figure [ fig : fold]a . we also extracted observations of from the palomar transient factory ( ptf , @xcite ) . only nine @xmath35-filter measurements were made over 2012 march 2327 , but they fortuitously cover a large fraction of the orbit , with smaller errors than crts . as shown in figure [ fig : fold]b , the shape of the modulation is very much like the mean crts light curve , and has the same total amplitude of @xmath36 mag . the sloan digital sky survey ( sdss , @xcite ) photometric catalog lists the optical counterpart of psr j1048 + 2339 at position ( j2000.0 ) r.a.@xmath37 , decl.@xmath38 . the observation occurred on 2005 february 4 ( mjd 53405.39 ) at phase 0.96 of the binary orbit . the sdss psf magnitudes are @xmath39 . we used the equations of lupton ( 2005 ) to transform these to the johnson system , giving @xmath40 . this is useful , as the crts unfiltered magnitude @xmath27 is not a good approximation to johnson @xmath41 for red stars . knowing the color , a more precise calibration of crts is given as @xmath42 @xcite . at @xmath43 , @xmath44 from the binned light curve in figure [ fig : fold]a , therefore @xmath45 , in excellent agreement with @xmath46 from the sdss . the total range of the light curve , neglecting any color changes , is then @xmath47 . in principle , one could get some information about the @xmath48 color variation around the orbit by comparing the crts and ptf light curves . but the sparseness of the ptf data , and the poor sensitivity of the crts data around the minimum of the light curve , do not enable such a variation to be measured robustly . psr j1048 + 2339 was also detected by the pan - starrs ps1 survey @xcite . the pan - starrs survey has been photometrically calibrated using a two - stage approach . in the first stage , photometrically stable nights are identified and the system zero points and a modest number of other parameters are determined based on the relative photometry overlaps , resulting in systematic uncertainties of ( @xmath49 , @xmath50 , @xmath51 , @xmath52 , @xmath53 ) = ( 8.0 , 7.0 , 9.0 , 10.7 , 12.4 ) mmag @xcite . in the second stage , the resulting zero points are used to tie the remainder of the data to this internal system @xcite , with the overall photometric system tied to the ab system via spectrophotometric standards @xcite . there have been to date three image processing pipeline versions , designated pv1 , pv2 , and pv3 . pv2 is currently the most reliable astrometrically , but yields fewer total detections than pv3 because it covers fewer epochs . the shorter wavelengths give smaller uncertainties in position . using 16 detections in the @xmath49 and @xmath50 filters ( mean epoch mjd 55674 ) and assuming a 50 mas error radius to allow for unknown systematics , we find ( j2000.0 ) r.a.@xmath54 , decl.@xmath55 . pv3 is preferred for photometry , being better calibrated and more complete . for psr j1048 + 2339 , pv3 yields 75 useful detections ( by a point - spread function goodness - of - fit criterion ) , typically obtained on 68 nights per filter , and spanning mjd @xmath56 ( 2010 february 15 2015 january 21 ) . figure [ fig_grizy ] shows the light curves of psr j1048 + 2339 in all five bands , with orbital phase calculated using the same method as for the crts and ptf data ( see section [ sec : crts ] ) . phase coverage is non - uniform in all filters . nevertheless , it is evident that the amplitude of modulation is a decreasing function of wavelength ; @xmath49 and @xmath50 span more than 1.0 mag , while the other bands are less variable . this is expected for a heating effect , where brighter regions are hotter . in all filters , the minimum occurs at @xmath57 and the maximum at @xmath58 . on the @xmath50 filter data in figure [ fig_grizy ] we superpose the mean unfiltered crts light curve from the inset of figure [ fig : fold ] , without any renormalization . these two light curves bear some resemblance , although both @xmath49 and @xmath50 have considerably higher amplitude of modulation than crts . the small number but high accuracy of the pan - starrs points leaves an impression that intrinsic variation may be frequent , while it is smoothed out in the crts mean light curve that comprises a much larger number of less precise measurements . we performed a spectral analysis of the gamma - ray source 3fgl j1048.6 + 2338 to characterize its spectrum and perform an optimized pulsation search using the weighted h - test @xcite . we used a subset of the recently released pass 8 lat data set starting from 2013 august 4 ( the beginning of our phase - connected radio timing solution ) and extending through 2015 july 15 ( met 397267203 458611204 ) . we selected source class events ( ` evclass = 128 and evtype = 3 ` ) only during the intervals of good science data ( ` data_qual=1 and lat_config=1 ` ) and restricted our events to those with a zenith angle less than 90@xmath59 and energies from 100 mev to 100 gev . we selected events from a 15@xmath59 radius around the source and performed a binned likelihood analysis over a @xmath60 region with 0.1@xmath59 pixels . the initial model we used was based on the 3fgl catalog @xcite . we modified the spectral model for the target source to be an exponentially cutoff power law of the form = n_0 ^- ( - ) . to perform the maximum likelihood fit , we used the ` p8r2_source_v6 ` instrument response functions with the _ fermi _ science tools version 10 - 00 - 05 and the ` newminuit ` fitting function . in the initial fit , we held all parameters fixed at the 3fgl catalog values except for the spectral parameters for the target source , and the normalization for sources within 6@xmath59 of the target or flagged in the 3fgl catalog as being variable . with the exponentially cutoff power - law ( ecpl ) model , the source is detected with a test statistic ( ts ) of 228 , corresponding to 15 @xmath61 @xcite . integrating the spectral model over the range 100 mev to 100 gev yields a photon flux of @xmath62 ph cm@xmath17 s@xmath64 and an energy flux of @xmath65 erg cm@xmath17 s@xmath64 . the photon spectral index is @xmath66 and the cutoff energy is @xmath67 gev . figure [ fig_gspec2 ] shows the spectral fit . when we fit with a simple power law and integrate over the same energy range , we find a photon flux of @xmath68 ph cm@xmath17 s@xmath64 , an energy flux of @xmath69 erg cm@xmath17 s@xmath64 , and a spectral index of @xmath70 . the ecpl model is weakly preferred with the cutoff detected at a significance of only 1.3 @xmath61 . for the gamma - ray timing analysis , we selected photons from a region of radius 2@xmath59 around the pulsar and assigned weights based on the best - fit spectral model . we compute a pulse phase for each selected lat photon using the ` fermi ` plugin for tempo2 @xcite and the best - fit radio timing model . the resulting weighted h - test value was 18.8 , corresponding to a significance of 3.45@xmath61 , not enough for a high - confidence claim of a pulsed detection . it is not feasible to extend the span of gamma - ray timing analysis earlier than the start of our radio timing solution due to the significant orbital period variations that are the hallmark of psr j1048 + 2339 . however , the h - test value improves as the radio timing solution and correspondingly the gamma - ray timing span are extended forward with the accumulation of more radio observations . this suggests that a confident detection of gamma - ray pulsations is likely with @xmath71 more years of radio timing data and an improved timing model that includes proper motion and possibly higher - order orbital period derivatives . a single _ swift _ xrt observation exists of the field in figure [ fig_sources ] , taken on 2011 january 19 in photon counting mode for an exposure of 1.2 ks ( obsid 00041540001 ) . the mid - time of the exposure falls at @xmath72 , which is near the maximum in the optical light curves . only three counts were detected inside a circular radius 47^@xmath73 around the position of psr j1048 + 2339 which is consistent with a non - detection of any x - ray source . we derive an absorbed 0.53.0 kev 3-@xmath61 upper limit of @xmath74 erg s@xmath64 cm@xmath17 for a 0.2 kev blackbody neutron star surface with the estimated galactic hydrogen column density for the pulsar position , @xmath75 cm@xmath17 @xcite . for the agn nvss j104900 + 233821 we derive an upper limit of @xmath76 erg s@xmath64 cm@xmath17 for an assumed x - ray power - law index of 1.8 and the same galactic hydrogen column density . the x - ray source 1rxs j104838.7 + 234351 , also shown in figure [ fig_sources ] , is within the error ellipse of the _ fermi _ source 3fgl j1048.6 + 2338 . since the position uncertainties for both the x - ray source and psr j1048 + 2339 are much smaller than the distance between the two objects , an association can be ruled out . the sources for our radio search were selected from a preliminary version of the _ fermi _ 3fgl catalog @xcite . in the final version of this catalog @xcite , the source 3fgl j1048.6 + 2338 is listed as associated with the bl lac active galactic nucleus nvss j104900 + 233821 . however , the association was made by a pipeline based on positional proximity and the radio flux density distribution of nvss sources . while most pulsars exhibit gamma - ray spectra that can be fitted by a power law with an exponential cutoff , for weaker gamma - ray sources the spectrum shape may be ambiguous and be equally well - fitted with a power law , which is typical of agns . the discovery of psr j1048 + 2339 well within the _ fermi _ source error ellipse ( figure [ fig_sources ] ) means that the agn association is likely spurious . a secure detection of gamma - ray pulsations from this pulsar will confirm the identification with 3fgl j1048.6 + 2338 and allow a search for emission from the nvss source by gating the data to include only the off - pulse phase of the pulsar . the sdss and pan - starrs optical positions are ( j2000.0 ) r.a.@xmath37 , decl.@xmath38 at epoch 2005 and ( j2000.0 ) r.a.@xmath54 , decl.@xmath55 at epoch 2011 , respectively . while they are consistent within their respective uncertainties , the radio timing position in table [ tab_pulsars ] , with end figures @xmath77(1 ) and @xmath78 , is offset from them by @xmath79 mas . the ppmxl catalog @xcite includes the optical counterpart of psr j1048 + 2339 , with components of proper motion @xmath80 mas yr@xmath64 , @xmath81 mas yr@xmath64 . the apop catalog of proper motions outside of the galactic plane @xcite also has an entry for this object , with @xmath82 mas yr@xmath64 , @xmath83 mas yr@xmath64 . the discrepancies between these results give some idea of the systematic errors involved , since they both derive from the same plate material , the palomar observatory sky survey . the pulsar position , proper motion , and orbital period derivatives are highly covariant in the timing analysis . despite having two years of radio timing data on psr j1048 + 2339 , we are still not able to confidently fit for proper motion along with the five orbital period derivatives listed in table [ tab_pulsars ] . this means that any actual proper motion is subsumed in the fits for @xmath84 and position . the timing solution in table [ tab_pulsars ] uses the apop proper motion as a fixed parameter and results in the smallest rms residual . however , we are also able to obtain phase - connected timing solutions using several alternative sets of free parameters : ( 1 ) position , proper motion , and @xmath85 , ( 2 ) position , @xmath85 , and @xmath86 , and ( 3 ) position , @xmath85 , @xmath86 , and @xmath87 . based on the spread of position fits between these timing solutions , we conclude that the position in the timing solution listed in table [ tab_pulsars ] is consistent with the sdss and pan - starrs optical positions within a remaining systematic error of @xmath88 mas . the present - epoch error ellipse for the ppmxl position , taking into account the proper motion uncertainty , encompasses best - fit positions from all of our attempted timing solutions described above . the present - epoch error ellipse for the apop position is within our systematic error of @xmath89 mas from the best - fit positions of all timing solutions . when it becomes possible to fit confidently for proper motion in our timing solution , the resulting position will very likely converge to that of the optical counterpart . the observed rotation period derivative @xmath90 includes a kinematic component @xmath91 @xcite , where @xmath5 is the total proper motion . at the ne2001-estimated distance of 0.7 kpc , we calculate @xmath92 s s@xmath64 using ppmxl and @xmath93 s s@xmath64 using apop total proper motions , which contribute , respectively , 4% or 12% of the observed @xmath94 s s@xmath64 . given the evident systematic errors , we assume only an upper limit on proper motion of @xmath95 mas yr@xmath64 , corresponding to a tangential velocity @xmath96 km s@xmath64 at @xmath97 kpc . this would imply a @xmath98% reduction to the apparent pulsar spin - down luminosity of @xmath99 erg s@xmath64 ( @xmath100 ) . the gamma - ray luminosity can be computed following @xcite , equation 15 . assuming @xmath101 , and using the dm distance of 0.7 kpc , the gamma - ray luminosity is @xmath102 erg s@xmath64 . this yields an efficiency @xmath103 of 2.4% . this is consistent with , but on the low end of the distribution of , efficiencies observed for lat - detected msps @xcite . any kinematic corrections to the spin - down will serve to increase the efficiency . the most consistent feature of the optical light curves is the minimum at @xmath57 . this accords with the expectation for pulsar wind heating of the companion star , as phase 0.25 corresponds to inferior conjunction of the companion , when we are viewing its cold side . the consistent alignment also implies that a constant radio period serves well to extrapolate the orbital phase over the previous 9 yr . on the other hand , the light curves do not have the symmetric heating maximum around @xmath104 expected in this scenario . rather , the flux is higher at @xmath105 and decreases toward @xmath106 . this resembles a similar sloping feature in the optical and x - ray light curves of the transitioning millisecond pulsar psr j1023 + 0038 when in its radio - pulsar state @xcite . it could indicate that the heating is asymmetric because it does not come directly from the pulsar , but is mediated by an intrabinary shock that has a skewed shape near the companion . alternatively , the shape of the light curve may be affected by the magnetic field of the companion directly channeling the pulsar wind to localized regions on its surface @xcite . spectral hardening around @xmath105 seen in the pan - starrs data is generally consistent with heating due to irradiation from the pulsar . the amplitude of modulation is about 0.8 magnitudes in @xmath51 , @xmath52 , and @xmath53 , but it increases to @xmath107 magnitudes in @xmath49 and @xmath50 . the departure from the expected symmetry around @xmath104 in the pan - starrs data is even more pronounced than in crts , which , because of the smaller number of pan - starrs points , may indicate that stochastic variability is prevalent . some measurements in the same filter differ by as much as @xmath108 magnitudes at the same orbital phase . there may be some significance to the fact that the peak in the apparent heating effect is displaced toward orbital phases earlier than 0.75 in both crts and pan - starss . an earlier peak implies that the trailing side of the companion , seen while it is moving away from the observer , is hotter than the leading side . this may indicate where the intrabinary shock is closest to the surface of the companion and illuminating it with the greatest geometrical covering factor . in his prediction for `` hidden '' millisecond pulsars of type ii , @xcite argued that the wind ablated from the companion would be displaced in the direction of orbital motion ( leading ) due to the coriolis force . this would force the shock away from the companion in the leading direction , possibly leaving the more efficient heating to the trailing side . any more detailed interpretation of the optical modulation is hampered by the evident intrinsic variability in the crts and pan - starrs data on timescales that are not yet fully explored . however , it would be interesting with new data to relate well - sampled optical light curves to the orbital phase distribution of the radio eclipses , to see if both can be explained by the geometry of the ablated wind . in addition to heating , there may be some modulation due to our changing view of the tidally distorted surface of the companion , nearly filling its roche - lobe . such ellipsoidal modulation produces maxima of equal height at @xmath109 and 0.5 , and unequal minima at @xmath110 and 0.75 , the latter one being lower . this effect dominates the light curve of the 5 hr period redback psr j1628@xmath33205 @xcite and others , e.g. , @xcite . however , the maximum ellipsoidal amplitude is only expected to be @xmath108 mag , while the mean light curve of varies by @xmath111 mag . also , the two local minima in the mean crts light curve are not separated by 0.5 in phase , but more like 0.6 . it is possible that ellipsoidal modulation is a small contributor even though it can not explain the gross features of the light curve of . we would like to know the intrinsic colors of the companion star to estimate its spectral type and radius . while the pan - starrs photometry represents the best broad - band spectral information currently available on this object , its sparse observations , together with the substantial variability of the source , both orbital and random , prevents a reliable color from being obtained from those exposures , which were taken more than 16 minutes apart . in contrast , the single - epoch sdss photometry was acquired within 5 minutes , which is why we used the magnitudes transformed from the sdss to calibrate the crts light curve approximately ( see section [ sec : sdss ] ) . the conclusions that follow from this are tentative , and must be tested with detailed temperature measurements and modeling of the orbital light curve . assuming that we are viewing the orbit nearly edge - on ( @xmath112 ) , @xmath113 at @xmath110 represents the unheated side of the companion . at the dm distance of @xmath97 kpc , the interstellar extinction measured from pan - starrs and 2mass photometry is @xmath114 @xcite , so the extinction - corrected apparent magnitude is @xmath115 and the absolute magnitude is @xmath116 , suggesting a main - sequence star of spectral type m4 , which has a radius of @xmath117 and an effective temperature of @xmath118 k. ( if @xmath119 , the phase of minimum light represents a mixture of spectral types both hotter and cooler than m4 . ) but the sdss colors at @xmath43 , near quadrature , correspond to an earlier spectral type , around k6 . therefore , we can assume that the heated side of the companion is , on average , hotter than a k6 star , or @xmath120 . given the pulsar s apparent spin - down luminosity , @xmath121 erg s@xmath64 corrected for proper motion , we can estimate the efficiency of pulsar heating during the bright phase of the optical light curve . assuming an average @xmath122 k for the heated half of the star , its extra luminosity at phase 0.75 can be expressed in terms of the heating efficiency @xmath123 of an assumed isotropic pulsar wind as @xmath124 , where @xmath125 is the orbital separation . ( other authors use a factor of @xmath126 here instead of our @xmath127 . ) for an assumed ns mass of @xmath128 and an orbital inclination of @xmath112 , @xmath129 and @xmath130 . therefore , pulsar heating is only marginally able to power the observed peak luminosity , because the needed efficiency @xmath123 is @xmath131 . there are a number of theoretical reasons that @xmath123 should not exceed 0.5 , as summarized by @xcite . however , @xcite estimated an efficiency in the range @xmath132 for the four systems they modeled , while @xcite found @xmath133 for the black widow psr j1810 + 1744 . @xcite modeled the redback psr j2215 + 5135 with @xmath134 , and @xcite found @xmath133 for the black widow psr j1311@xmath33430 . so more complex physical effects are clearly having an important influence . one way to reduce the inferred efficiency is to assume that the pulsar wind is beamed in the orbital plane in this system , or channeled by the intrinsic magnetic field of the companion . with an estimated @xmath117 from its optical properties , the companion appears to significantly underfill its roche lobe , which has radius @xmath135 @xcite for @xmath136 . this would reduce any ellipsoidal modulation . if we could double the distance to 1.4 kpc , then the companion could be an m2 star with @xmath137 k and fill its roche lobe . but the difficulty of pulsar wind heating would not be ameliorated much because @xmath123 would become only slightly smaller , @xmath138 , for the same @xmath139 used above . however , these conclusions are sensitive to the effective @xmath139 and the inclination @xmath140 , both of which remain to be measured . figure [ fig_tasc ] illustrates the orbital period variations of psr j1048 + 2339 . large orbital period variations are known to occur in other redback systems such as psr j1023 + 0038 @xcite and j2339@xmath30533 @xcite . quasi - periodic changes in the gravitational quadrupole moment of the companion have been proposed to explain observations showing the orbital period to oscillate @xcite . several other effects can contribute to orbital period variations : ( 1 ) energy dissipation via gravitational wave emission , ( 2 ) mass loss , and ( 3 ) doppler shifts due to the shklovskii effect @xcite or galactic rotation . however , they would result in a monotonic change in @xmath141 and therefore can not adequately explain the observed orbital period variations . @xcite relates oscillations in the gravitational quadrupole moment of a non - degenerate companion star to a cycle of magnetic activity similar to the 11-year solar cycle . the changing magnetic field generates a variable torque between the convective outer layer and radiative inner layer of the star . the torque causes angular momentum to be exchanged between the layers in a quasi - periodic manner , causing changes in the differential rotation between the layers and therefore in the shape of the star . spin - up of the outer layer corresponds to the star becoming more oblate and to an increase in the gravitational quadrupole moment . since the total angular momentum in the binary system is conserved , this causes a decrease in the orbital period . @xcite give an overview of the literature and apply the applegate model to the orbital period variations of psr j2339@xmath30533 . we note that since this model relies on convective and radiative layers being present and stars with mass @xmath142 _ @xmath0 are fully convective , it implies that the companion of psr j1048 + 2339 is more massive than the minimum companion mass @xmath143 _ @xmath0and the orbital inclination is @xmath144 . following the same procedure as @xcite , we obtain a modulation of @xmath141 with a period of @xmath145 years and amplitude @xmath146 s@xmath64 . using @xmath143 _ @xmath0 , and estimated companion radius @xmath147 _ @xmath148 ( section [ sec : companion ] ) , we derive the change in the luminosity of the companion that would account for the energy needed for angular momentum transfer , @xmath149 erg s@xmath64 ( eqn . 13 from @xcite ) . based on the estimated companion effective temperature @xmath150 k ( section [ sec : companion ] ) , its total luminosity is @xmath151 erg s@xmath64 . since the energy budget for the angular momentum transfer is only @xmath152 of the companion s intrinsic luminosity , the companion has sufficient energy output to generate the orbital period fluctuations even if the orbital inclination ( assumed to be 90^@xmath153 ) is overestimated , or @xmath154 or @xmath155 are underestimated . the discovery of the redback pulsar j1048 + 2339 in a _ fermi _ unassociated source and the interesting phenomenology in this binary system add a valuable data point for studies of this still not very well understood class of pulsars . our radio observations show extensive eclipses around superior conjunction as well as sporadic `` mini - eclipses '' at other orbital phases and an occasional long - term absence of pulsed radio emission even near inferior conjunction . this indicates that the ionized gas environment in the system is very dynamic and changes significantly on time scales of hours to days . the radio timing solution that best predicts pulse arrival times includes a fixed proper motion from the apop catalog @xcite and five orbital period derivatives . a likely explanation of the rapid orbital period variations is quasi - periodic changes in the gravitational quadrupole moment of the companion due to a magnetic torque mechanism similar to that at play in the 11-year solar cycle . psr j1048 + 2339 has a 20th magnitude optical counterpart detected by sdss , crts , ptf , and pan - starrs , whose intensity modulation matches the pulsar orbital period . its magnitude and colors are consistent with those of an m4 main - sequence star on the `` night '' side . while the optical light curve has the expected minimum at @xmath156 , corresponding to inferior conjunction of the companion , it does not exhibit a well - defined maximum at @xmath157 . this can be explained by heating not directly from the pulsar but from an asymmetric intra - binary shock . alternatively , the companion magnetic field may be channeling the pulsar wind to specific regions on the companion s surface that get heated preferentially . we did not find an x - ray counterpart in an existing short _ swift _ observation of the field containing psr j1048 + 2339 . the two - year span of the radio ephemeris does not allow a confident detection of gamma - ray pulsations . due to the rapid orbital period variations , for the gamma - ray timing analysis we can use only _ fermi _ data taken within the radio ephemeris time limits . however , the h - test value of a potential gamma - ray pulsation detection improves as the span of the radio timing solution grows , indicating that detection of gamma - ray pulsations is likely as we extend and improve the psr j1048 + 2339 timing solution within the next 12 years . j.s.d . was supported by the chief of naval research and a _ fermi _ guest investigator grant . the arecibo observatory is operated by sri international under a cooperative agreement with the national science foundation ( ast-1100968 ) , and in alliance with ana g. mndez - universidad metropolitana , and the universities space research association . the green bank telescope is part of the national radio astronomy observatory , a facility of the national science foundation operated under cooperative agreement by associated universities , inc . the css survey is funded by the national aeronautics and space administration under grant no . nng05gf22 g issued through the science mission directorate near - earth objects observations program . the crts survey is supported by the u.s . national science foundation under grant ast-0909182 . the pan - starrs surveys ( ps1 ) have been made possible through contributions of the institute for astronomy , the university of hawaii , the pan - starrs project office , the max - planck society and its participating institutes , the max planck institute for astronomy , heidelberg and the max planck institute for extraterrestrial physics , garching , the johns hopkins university , durham university , the university of edinburgh , queen s university belfast , the harvard - smithsonian center for astrophysics , the las cumbres observatory global telescope network incorporated , the national central university of taiwan , the space telescope science institute , the national aeronautics and space administration under grant no . nnx08ar22 g issued through the planetary science division of the nasa science mission directorate , the national science foundation under grant no . ast-1238877 , the university of maryland , and eotvos lorand university ( elte ) and the los alamos national laboratory . the _ fermi_-lat collaboration acknowledges generous ongoing support from a number of agencies and institutes that have supported both the development and the operation of the lat as well as scientific data analysis . these include the national aeronautics and space administration ( nasa ) and the department of energy in the united states , the commissariat lenergie atomique and the centre national de la recherche scientifique / institut national de physique nuclaire et de physique des particules in france , the agenzia spaziale italiana and the istituto nazionale di fisica nucleare in italy , the ministry of education , culture , sports , science and technology ( mext ) , high energy accelerator research organization ( kek ) and japan aerospace exploration agency ( jaxa ) in japan , and the k. a. wallenberg foundation , the swedish research council and the swedish national space board in sweden . additional support for science analysis during the operations phase is gratefully acknowledged from the istituto nazionale di astrofisica in italy and the centre national dtudes spatiales in france . ll right ascension , @xmath158 ( j2000 ) & @xmath159 + declination , @xmath160 ( j2000 ) & @xmath161 + proper motion in @xmath162 ( mas yr@xmath64 ) & @xmath163 + proper motion in @xmath164 ( mas yr@xmath64 ) & @xmath165 + spin frequency , @xmath166 ( hz ) & 214.35478538736(7 ) + spin frequency derivative , @xmath167 ( hz s@xmath64 ) & @xmath168 + epoch of timing solution ( mjd ) & 56897.0 + dispersion measure , dm ( pc @xmath169 ) & 16.6544(1 ) + time of passage through ascending node , @xmath170 ( mjd ) & 56637.598177(1 ) + projected semi - major axis , @xmath171 ( lt - s ) & 0.836122(3 ) + eccentricity & 0 + orbital frequency , @xmath172 ( hz ) & @xmath173 + 1st orbital frequency derivative , @xmath174 ( hz s@xmath64 ) & @xmath175 + 2nd orbital frequency derivative , @xmath176 ( hz s@xmath17 ) & @xmath177 + 3rd orbital frequency derivative , @xmath178 ( hz s@xmath179 ) & @xmath180 + 4th orbital frequency derivative , @xmath181 ( hz s@xmath182 ) & @xmath183 + 5th orbital frequency derivative , @xmath184 ( hz s@xmath185 ) & @xmath186 + planetary ephemeris & de421 + time units & tdb + timing span ( mjd ) & 5650857285 + number of observation epochs & 26 + number of points in timing fit & 72 + weighted rms post - fit residual ( @xmath5s ) & 4.2 + [ -5pt ] galactic longitude , @xmath187 ( deg ) & 213.17 + galactic latitude , @xmath188 ( deg ) & 62.139 + mass function ( _ @xmath0 ) & 0.00100 + surface magnetic field , @xmath189 ( g ) & @xmath190 + spin - down luminosity , @xmath191 ( erg s@xmath64 ) & @xmath192 + characteristic age , @xmath193 ( gyr ) & 2.5 + distance , @xmath194 ( kpc ) & 0.7 + . sessions 34 and 41 correspond to arecibo observations at 1430 mhz . arecibo sessions at 327 mhz typically consist of several 20-minute observations made in succession . in some cases , the last observation is shorter if it was aborted at the end of assigned telescope time . [ fig_det],scaledwidth=99.0% ]

much attention has been paid recently to the existence and properties of spin liquids in strongly correlated materials , both experimentally and theoretically@xcite . such phases , first proposed by anderson , have been put forward in the context of high @xmath0 superconductivity as a possible precursor state , but they can occur in a much broader class of situations , both in real materials @xcite and in microscopic models @xcite . their characterization is challenging on the experimental side , it requires generally involved low - temperature techniques to prove the absence of long - range magnetic order down to temperature @xmath1 , and concerning the theory , either in hubbard - like models or in localized spin models , spin liquids usually result from frustrating interactions but compete with a variety of possible ordered phases , from usual antiferromagnetic ( af ) order to valence bond crystals , spin nematics etc.@xcite clear evidence of a spin liquid ground state was found from an exact solution of the kitaev model @xcite of spins @xmath2 on the honeycomb lattice . this liquid ground state is characterized by a gapless spectrum ( with low - energy excitations described in terms of majorana fermions ) and gapped vortices . the kitaev model itself triggered a large variety of studies , focusing on topological properties , ( non)-abelian excitations , extensions to higher dimensions@xcite , effects of magnetic@xcite or non - magnetic impurities and of a magnetic field @xcite . besides its mathematical beauty and the original properties mentioned above , the kitaev model was recently found @xcite to be relevant to orbitally degenerate systems with strong spin - orbit coupling such as layered iridates na@xmath3iro@xmath4 and li@xmath3iro@xmath4 . there , around each ir atom , the octahedral environment [ see fig . [ struc](a ) ] of oxygens results in a configuration with one hole in the @xmath5 manifold ; and the large spin - orbit coupling selects locally a low - energy doublet of states defining a pseudospin @xmath2 ; and finally , superexchange processes between ir ions driven by moderately strong coulomb interactions lead to an effective description by the kitaev - heisenberg model @xcite for pseudospins @xmath2 residing on the honeycomb lattice . the study of this model showed the robustness of the spin liquid phase for finite af heisenberg interactions [ small compared to ferromagnetic ( fm ) kitaev interactions ] , and a quantum phase transition between this liquid phase and a stripe - ordered phase . in na@xmath3iro@xmath4 , where interaction parameters could be close to this transition , af order was found by magnetization and specific heat measurements@xcite and x - ray magnetic scattering@xcite . the question of how vacancies influence the order and the magnetic response in the kitaev - heisenberg model is of both experimental and theoretical interest . indeed the vacancy can act as a probe allowing to measure spin correlations in its vicinity ; and it can also have drastic effects on these correlations . for concreteness , in @xmath6 spin-@xmath2 antiferromagnets with long - range order , a non - magnetic impurity can enhance the staggered magnetization around the vacancy @xcite . in dimerized phases , the impurity is accompanied by a spinon , which is typically confined to the impurity in valence bond crystals and deconfined in resonating valence bond liquids @xcite , although the question of the confinement length can be subtle @xcite . in a nel - ordered phase , the impurity results in a curie- type contribution to the finite-@xmath7 susceptibility @xcite . in the spin - liquid phase of the kitaev model , a magnetic moment is induced around a spinless vacancy ; interestingly , its contribution to the spin susceptibility diverges logarithmically for @xmath8 @xcite . this makes it relevant to investigate the vacancy - induced magnetic response in a model interpolating between @xmath6 and kitaev - like interactions . moreover , in the iridate samples studied up to now , a substantial amount of disorder might be present , e.g. , site - mixing effects with non - magnetic alkaline ions on the hexagonal ir - sublattice @xcite , which should be taken into account for proper understanding of their physical properties . our aim is to study the effect of non - magnetic vacancies ( and of pairs of vacancies ) in the different magnetic phases of the kitaev - heisenberg model , at zero temperature . we adopt similar notations as in ref . , with a parameter @xmath9 interpolating between heisenberg and kitaev models , and a uniform magnetic field @xmath10 with amplitude @xmath11 : h= -2_i , j _ _ i^_j^+(1- ) _ i , j _ i _ j -_i h _ i , [ ham ] where @xmath12 labels simultaneously an axis in spin space , and ( in the interaction terms ) a bond direction of the honeycomb lattice . for convenience we express @xmath13 in terms of pauli matrices @xmath14 . we take as unit cell a _ @xmath15-bond _ [ see fig . [ struc](b , c ) ] , on which kitaev interactions are of the form @xmath16 . each site is labeled by an index @xmath17 , in bijection with @xmath18 where @xmath19 denotes the position of the unit cell and @xmath20 the sublattice index , respectively . elementary translations between neighbor cells are @xmath21 in cartesian coordinates . we consider periodic clusters of @xmath22 sites ( this cluster can be represented with an hexagonal shape , see fig . [ cormag ] ) , @xmath23 or @xmath24 sites ( both with the shape of a parallelogram ) . these clusters respect all spatial symmetries of the model , i.e. translation and @xmath25-rotation - other symmetries combining spin- and spatial rotations @xcite are not used here . in section [ vfs ] , we discuss several features of the model in a magnetic field . we will show how a spin canting mechanism allows one to understand the phase diagram for a field oriented along an easy axis , as well as to understand a directional anisotropy of the static susceptibility . also the perturbations due to heisenberg interactions and due to small magnetic fields on the spin correlation functions are explored . in section [ ces ] , we study the effect of a single non - magnetic vacancy in the system , in various phases of the model , and analyze the magnetization pattern in the vicinity of this vacancy in a small field . in section [ ibv ] , we discuss situations where two vacancies , close enough to each other , effectively interact . we focus mainly on the stripe phase and explain the underlying mechanism of local selection of an ordered pattern by a vacancy pair . section [ scr ] provides a short summary and some concluding discussions . we first consider some basic properties of the kitaev - heisenberg model in a magnetic field , and study the magnetic response to a field oriented along different crystallographic directions . this comparison is important because of the symmetry of the model , which is clearly not @xmath6 invariant due to the kitaev interactions . in the context of iridates , this anisotropy is due to spin - orbit coupling which results in the easy spin axes @xmath26 $ ] , @xmath27 $ ] , @xmath28 $ ] [ corresponding to the octahedral axes , see fig . [ struc](a ) ] , compared to the cubic diagonal @xmath29 $ ] direction . the clearest anisotropic features are seen in the stripe - ordered phase , and we compare this situation to that of the nel phase . we first want to characterize the various phases found in a magnetic field . it is already known @xcite that at zero field , the spin liquid phase extends from @xmath30 to @xmath31 ( the kitaev limit ) ; the stripe - ordered phase , with spins pointing either in @xmath32 , @xmath33 or @xmath15 direction [ for the last case , the spatial structure , i.e. @xmath15-stripe pattern , is shown in fig . [ stripe2](a ) ] , is located between @xmath34 and @xmath35 ; and a nel - ordered phase is found for @xmath36 . to determine whether these phases persist in a magnetic field and to understand the magnetic response of the system , one needs adapted structure factors which act as order parameters . for the nel phase , it reads : s_neel=_r , r _ , _ r , _ r , with @xmath37 being 1 for @xmath38 , and @xmath39 otherwise ; for the stripe - ordered phase , the structure factors corresponding to @xmath32- and @xmath15- stripe patterns are defined as follows : s^x(_x)&=&_r , r,, e^i _ x ( r -r ) _ , ^x_r , ^x_r , , + s^z(_z)&= & _ , r,, e^i _ z ( r - r ) ^z_r , ^z_r , , with @xmath40 shown in fig . [ struc](d ) ; for instance , @xmath41 which is equivalent to the more intuitive @xmath42 ( but which is not in the first brillouin zone ) . both wave vectors account for a phase factor @xmath39 between neighboring unit cells separated by @xmath43 in real space . these structure factors also evidence the symmetries spontaneously broken in each of the 6-fold degenerate ground states in the stripe phase : indeed at zero field , in addition to translational symmetries , the hamiltonian possesses a @xmath44 symmetry - @xmath45 standing for the @xmath46 transformation and @xmath47 ( labeled as @xmath48 in ref . ) for the invariance by a cyclic permutation of the @xmath32 , @xmath33 and @xmath15 spin components coupled to a spatial rotation by @xmath49 . ( color online ) ( a ) schematic picture of an octahedron ( 6 oxygens surrounding a ir@xmath50 ion ) with the octahedral axes @xmath32 , @xmath33 , @xmath15 - which are easy axes on identically - labeled bonds in the kitaev - heisenberg model . ( b ) a hexagon in the ir@xmath50 plane , indicating bonds labeled @xmath32 , @xmath33 , @xmath15 according to their local easy axis regarding kitaev interactions . ( c ) honeycomb ( direct ) lattice with the labeling @xmath51 of bonds according to their in - plane orientation . a unit cell ( centered on a @xmath15-bond ) is represented in the dashed rectangle , as well as elementary translations @xmath52 . ( d ) first brillouin zone , with the position of stripe - order wave vectors @xmath53 ( @xmath12).,title="fig:",width=132 ] ( color online ) ( a ) schematic picture of an octahedron ( 6 oxygens surrounding a ir@xmath50 ion ) with the octahedral axes @xmath32 , @xmath33 , @xmath15 - which are easy axes on identically - labeled bonds in the kitaev - heisenberg model . ( b ) a hexagon in the ir@xmath50 plane , indicating bonds labeled @xmath32 , @xmath33 , @xmath15 according to their local easy axis regarding kitaev interactions . ( c ) honeycomb ( direct ) lattice with the labeling @xmath51 of bonds according to their in - plane orientation . a unit cell ( centered on a @xmath15-bond ) is represented in the dashed rectangle , as well as elementary translations @xmath52 . ( d ) first brillouin zone , with the position of stripe - order wave vectors @xmath53 ( @xmath12).,title="fig:",width=132 ] + ( color online ) ( a ) schematic picture of an octahedron ( 6 oxygens surrounding a ir@xmath50 ion ) with the octahedral axes @xmath32 , @xmath33 , @xmath15 - which are easy axes on identically - labeled bonds in the kitaev - heisenberg model . ( b ) a hexagon in the ir@xmath50 plane , indicating bonds labeled @xmath32 , @xmath33 , @xmath15 according to their local easy axis regarding kitaev interactions . ( c ) honeycomb ( direct ) lattice with the labeling @xmath51 of bonds according to their in - plane orientation . a unit cell ( centered on a @xmath15-bond ) is represented in the dashed rectangle , as well as elementary translations @xmath52 . ( d ) first brillouin zone , with the position of stripe - order wave vectors @xmath53 ( @xmath12).,title="fig:",width=136 ] ( color online ) ( a ) schematic picture of an octahedron ( 6 oxygens surrounding a ir@xmath50 ion ) with the octahedral axes @xmath32 , @xmath33 , @xmath15 - which are easy axes on identically - labeled bonds in the kitaev - heisenberg model . ( b ) a hexagon in the ir@xmath50 plane , indicating bonds labeled @xmath32 , @xmath33 , @xmath15 according to their local easy axis regarding kitaev interactions . ( c ) honeycomb ( direct ) lattice with the labeling @xmath51 of bonds according to their in - plane orientation . a unit cell ( centered on a @xmath15-bond ) is represented in the dashed rectangle , as well as elementary translations @xmath52 . ( d ) first brillouin zone , with the position of stripe - order wave vectors @xmath53 ( @xmath12).,title="fig:",width=136 ] + ( color online ) ( a ) representation of a @xmath15-stripe pattern ( stripe pattern with spins oriented along the @xmath54 $ ] axis - neighboring spins are aligned on @xmath15 bonds and antialigned on other bonds . ( b ) schematic representation , at the level of a single hexagon , of a canted stripe pattern : here a @xmath15-stripe pattern , with spins tilted of an angle @xmath55 towards a field along [ 100 ] direction.,title="fig:",width=158 ] ( color online ) ( a ) representation of a @xmath15-stripe pattern ( stripe pattern with spins oriented along the @xmath54 $ ] axis - neighboring spins are aligned on @xmath15 bonds and antialigned on other bonds . ( b ) schematic representation , at the level of a single hexagon , of a canted stripe pattern : here a @xmath15-stripe pattern , with spins tilted of an angle @xmath55 towards a field along [ 100 ] direction.,title="fig:",width=158 ] + both af ( nel and stripe ) ordered phases of the model persist under moderate fields , which tilt spins by an angle proportional to the small field amplitude ( _ canted nel phase _ @xcite or _ canted stripe phase _ @xcite ) . to allow for this tilting , the untilted spins ( for @xmath56 infinitesimal ) should not be collinear with the field @xcite , and the staggered magnetization of the selected ordered pattern has an angle with @xmath57 as close to @xmath58 as possible . we first discuss the effect of a small field in the nel phase , which allows to precise what we will understand in the following by ( an)isotropy in the magnetic response . in the pure heisenberg model , both classical and quantum cases , spontaneous breaking of @xmath6 symmetry in the thermodynamic limit leads to an anisotropic zero - field susceptibility . on the other hand , the ground state of a finite cluster is known to be a singlet , which does not break @xmath6 symmetry . in order to perform a classical treatment of the heisenberg model in absence of spontaneous symmetry breaking , one can consider for instance a @xmath6 symmetric combination of classical nel states - such a state can mimic , from an experimental viewpoint , the response of a polycristalline sample . the magnetic response of this state is isotropic in the sense that a small field results in a magnetization of amplitude @xmath59 along the field , and @xmath59 does not depend on the field direction . when considering the kitaev - heisenberg model for small finite @xmath60 , with the classical approach above , one can find for any field direction a nel pattern with spins perpendicular to it ; within this approximation , the magnetic response is again isotropic , despite the absence of @xmath6 symmetry due to small kitaev interactions . however , the combination of quantum fluctuations and small kitaev interactions have been found to select one of the cubic axes as an easy spin axis and open thereby a small spin gap @xcite ; this will lead to small anisotropies in the spin susceptibility . in the stripe phase , even at the classical level only six ordered patterns are allowed at zero field . for a generic field @xmath57 with @xmath61 , @xmath62 and @xmath63 taking three distinct values , the three allowed stripe orientations are energetically differentiated , depending on the field direction . for instance , in a small field @xmath64 $ ] , the two @xmath15-stripe patterns are favored ( along with the @xmath33-stripe patterns ) , since they allow for the field to tilt spins by a small angle @xmath55 [ see fig . [ stripe2](b ) ] , such that the uniform magnetization develops an @xmath32-component . this effect is evidenced by structure factors characteristic of the ordered phase - in the case of fig . [ diag ] ( with @xmath65 ) , in a small @xmath66$]-oriented field the @xmath32-stripe pattern becomes more favorable than the @xmath15-stripe pattern : @xmath67 jumps from approximatively @xmath68 at @xmath69 to @xmath2 at small finite @xmath63 , while @xmath70 ( not shown ) drops to almost @xmath71 . as for the magnetization per site @xmath72 , it grows linearly with @xmath63 in this regime . the tilting angle @xmath55 can be determined simply in the classical approach , where the energy per unit cell as function of @xmath55 , for given @xmath60 and @xmath63 , is : e_s(h_z,)=-(1+)-2h_z+4(1-)^2 , and its minimization w.r.t . @xmath55 gives a magnetization per spin @xmath73 , in relatively good agreement with the linear behavior observed at small field ( see fig . [ diag ] for the case @xmath65 ) ; also , the predicted @xmath60-dependence of the ( uniform ) susceptibility per spin , ^z=()_h=0= , [ chz ] is in good agreement with the numerical results in fig . [ khiz1 ] . note that the direction of the field is relevant in the stripe phase : a field @xmath74 $ ] differentiates stripe patterns , and selects two of them where spins are exactly orthogonal to the field as @xmath75 ; in contrast , in a @xmath76 $ ] field the three stripe directions remain equivalent ( thus @xmath77 for @xmath78 ) ; also , none of the uncanted stripe patterns is orthogonal to a @xmath29$]-field , so the magnetic response for this field direction necessarily differs from the response to a @xmath66$]-field . ( color online ) magnetization @xmath59 ( per site ) and @xmath79stripe structure factor @xmath67 versus @xmath63 for various values of @xmath60 and @xmath80 . dotted lines : perturbative results for @xmath59 in the polarized phase . inset : phase diagram in the @xmath81 plane ; continuous transition lines are interpolations of points from @xmath22 data and dashed lines are perturbative estimates.,width=283 ] ( color online ) evolution of correlations @xmath82 with @xmath60 on a @xmath22 cluster at @xmath69 , for various relative positions @xmath83 with @xmath17 and @xmath84 indicated in the inset . the red dot indicates the value @xmath85 of the @xmath86 correlation at @xmath31 , which coincides with an analytical result from ref . .,width=294 ] ( color online ) susceptibility per spin @xmath87 , evaluated as @xmath88 for small @xmath89 , and its equivalent @xmath90}$ ] for a @xmath29$]-oriented field , on clusters @xmath22 or @xmath23 . shown also the quantity @xmath91 for @xmath23.,title="fig:",width=294 ] + we now turn to the spin liquid regime , corresponding to the weakly perturbed kitaev model , and describe the effect of perturbations to the system . our main focus is on the @xmath92 correlations . in the pure kitaev model ( @xmath31 , @xmath69 ) , these correlations are non - zero only when @xmath93 are nearest neighbors on a @xmath15-bond @xcite , where they take a value @xmath94 which we can see in fig . [ diag2 ] ( the finite - size correction to this quantity , i.e. to the ground state energy , is negligible even for the @xmath22 sites cluster ) . the gapless nature of excitations persists in a small axial field @xmath63 ; in contrast , a @xmath29$]-oriented field opens a gap resulting in a topologically ordered phase @xcite . yet even in the first case , the field causes peculiar @xmath92 correlations with spatial oscillations perpendicular to @xmath15-bonds and a power - law decaying envelope , as shown for @xmath31 in ref . . some of these trends appear in the spin - spin correlations displayed in fig . [ cormag](a ) . due to the small size , one can not identify a power - law behavior ; but one notices that correlations decrease with distance , much more slowly in the direction parallel to @xmath15 bonds than in the direction perpendicular to them ( i.e. parallel to @xmath95 ) . given the periodic boundary conditions on the @xmath22 lattice , this is in agreement with the period of three unit cells for spatial oscillations predicted in ref . . the liquid phase is stable with respect to small heisenberg interactions ( for @xmath96 at zero field @xcite ) . these interactions induce correlations between further neighbors , of different type than those induced by a small field [ compare fig . [ cormag](a ) and ( b ) ] . the spatial alternation of sign on fig . [ cormag](b ) reminds of the @xmath15-stripe pattern , with the difference that here these correlations are short - ranged . a proof of this last point has been given in ref . for @xmath69 , and indeed absolute values @xmath97 of correlations shown in fig . [ diag2 ] seem to decrease rapidly possibly exponentially with distance , at least in the spatial direction orthogonal to @xmath15 bonds . on nearest neighbors , these correlations are proportional to @xmath98 ( not too close to @xmath35 ) . with both perturbations at play , i.e. when both @xmath98 and @xmath63 are finite and small , their effects should add up ; therefore , and since the correlations induced by heisenberg terms decay exponentially , one should observe a correlation pattern with a power - law tail as in the case of @xmath99 , but modulated as in a @xmath15-stripe pattern for short distances , at least for @xmath100 . eventually , for any value of @xmath60 , a sufficiently high field drives the system into a _ polarized phase _ with spins parallel to @xmath101 . in the case of a @xmath66$]-field , by considering transverse interactions @xmath102 and @xmath103 as perturbations [ the unperturbed ground state being fully polarized for @xmath104 , within second order perturbation theory one finds m_0 = _ z 1- , in good agreement with numerical data for @xmath105 ( see dashed curves in fig . [ diag ] ) . when a magnetic field @xmath63 is applied to a system which is initially stripe- or nel ordered , aside from the spin canting in the af ordered state , the excitation energy of a competing polarized state decreases with @xmath63 so that this state becomes the ground state above a critical field @xmath106 . this field can be evaluated by estimating the energies ( both per unit cell ) @xmath107 of the canted stripe phase and @xmath108 of the polarized phase , as a function of field - then the equation @xmath109 determines @xmath106 . for both quantities , we add to the classical value a term ( last terms in both lines of eq . [ ecantpol ] ) accounting for quantum fluctuations , considered in second order perturbation theory , and evaluated respectively from the uncanted stripe- or the fully polarized state : e_s(h_z)&=&-(1+)- - , + e_fz(h_z)&=&-2h_z+3 - 5- . [ ecantpol ] we compare this estimate of @xmath106 with numerical estimates obtained from the @xmath63-dependence of the stripe structure factor @xmath67 and the magnetization @xmath59 for @xmath22 and @xmath23 clusters both quantities show very weak size - dependence ( see fig . [ diag ] ) and their derivatives with respect to @xmath63 are maximized for values very close to each other . for @xmath65 , this numerical estimation gives @xmath110 compared to a value @xmath111 found from the criterion @xmath109 . both types of estimates are shown as a function of @xmath60 , in the phase diagram in the inset of fig . [ diag ] . for @xmath112 , the magnetization curve @xmath72 on a finite cluster at fixed @xmath60 ( @xmath113 in fig . [ diag ] ) shows a sharp increase in a field range centered on @xmath114 ; this value is used to locate the transition to the polarized phase ( our precision on @xmath115 is limited here by larger finite size effects in the liquid state than in ordered phases ) . this can be understood by modeling the energy per unit cell for the liquid phase by e_l=3(1 - 3)^^_^(=1,h=0)-^z h_z^2 , and looking for the field value such that @xmath116 . this gives a value @xmath117 , which increases slightly with @xmath98 . similarly , the position of the stripe - liquid transition , which at zero field occurs at @xmath35 , is almost insensitive to a small field @xmath63 . the schematic phase diagram in fig . [ diag ] has a similar structure as its counterpart for a @xmath29$]-oriented field @xcite , except that we do not see a multicritical point at @xmath118 . this may be connected with the absence of @xmath48 symmetry in the hamiltonian eq . [ ham ] in a @xmath66$]-field ; moreover , the nature of the spin liquid for @xmath112 in a small axial field differs from that of the topologically ordered phase in a small @xmath29$]-field , mainly in the structure of the low - energy spectrum . we analyze now in more detail the zero - field susceptibility : it has unusual properties due to a special spin symmetry of the model , which is lower than @xmath6 . in the stripe phase ( @xmath119 ) , if we consider the general case of a @xmath120$]-oriented field , that is @xmath57 parallel to @xmath121 , the corresponding susceptibility @xmath122}$ ] behaves as @xmath123 , where @xmath124 is a rational number which can be found from the classical picture of a spin - flop mechanism . one can see in fig . [ khiz1 ] that susceptibilities along different axes follow well this type of law in the stripe phase . the coefficient @xmath124 is maximized for a @xmath66$]-field ( @xmath125 , see eq . [ chz ] ) , but also for a @xmath126$]-field , and minimized for a @xmath29$]-field ( @xmath127 ) . note here that the @xmath29 $ ] axis corresponds , in the context of susceptibility measurements in ref . , to the @xmath128 axis of na@xmath3iro@xmath4 , i.e. perpendicular to the ir - honeycomb layers [ see fig . [ struc](b ) ] . concerning the susceptibility in the @xmath129 plane , this model predicts that in the stripe phase @xmath130 takes a value between @xmath131}=\frac{1}{4(1-\alpha)}$ ] and @xmath132}=\frac{1}{5(1-\alpha)}$ ] . an average over the field orientations in the @xmath129 plane gives @xmath133 . no such simple expressions for the susceptibility are available for the liquid phase ; but there , despite large finite - size effects ( compared to the stripe phase ) one notices that for a given axis susceptibilities tend to increase with @xmath60 , with again @xmath90 } \le \chi^{[001]}$ ] . as for the nel phase , a classical approach does not predict there any anisotropy in the static susceptibility ( see related discussion in section [ phas ] ) , and the anisotropy resulting from quantum fluctuations at small finite @xmath60 should be smaller than in the stripe phase but the nel phase is obtained for values of @xmath60 likely unrealistic for na@xmath3iro@xmath4 @xcite and also not supported by experiment@xcite . the experimental data of ref . reveal an anisotropy in the static susceptibility as well , but the measured susceptibility is significantly larger along the @xmath128 axis than in the @xmath129 plane , in contrast to our model prediction @xmath134 . this suggests that other factors could contribute to the susceptibility anisotropy@xcite . for instance , anisotropic contributions are expected to arise from the van vleck susceptibility @xmath135 , and also via the g - value anisotropy in the ground - state kramers doublet , i.e. , @xmath136 , due to lattice distortions lowering the octahedral symmetry . these effects involve the inclusion of higher , excited spin - orbital , states . nevertheless , we believe that the kitaev - heisenberg model describes some characteristic features of layered iridates : for instance , recent susceptibility measurements@xcite on polycrystals of li@xmath3iro@xmath4 and of na@xmath3iro@xmath4 showed that the zero - temperature susceptibility was about twice higher for the first ( li ) compound than for the second . assuming that both compounds can be described by the kitaev - heisenberg model and are in the stripe phase , the value of @xmath60 for the li compound should be significantly larger than for the na compound - probably @xmath137 if one compares the paramagnetic curie temperatures of both compounds and follows theoretical results of ref . . this is actually in agreement with the law @xmath138 which we find for any field direction ( and thus for polycrystalline samples ) : since by assumption @xmath139 , the observed ratio@xcite @xmath140 would imply @xmath141 . this suggests that the li - compound is quite close to the stripe - liquid transition . in the following , we shall confine our study to the kitaev - heisenberg model and explore the role of non - magnetic vacancies , unavoidably present in real compounds , and which , in general , can significantly alter the properties of frustrated magnetic systems @xcite . we now study the effect of a single non - magnetic vacancy on the magnetic response of the system , especially at low - field . as previously , we will address separately the different phases of the model . ( color online ) ( a , b ) correlation functions @xmath92 with respect to site @xmath17 indicated by a black circle . ( c , d ) induced magnetization patterns @xmath142 due to the presence of a vacancy , on the @xmath22 cluster . circle areas are proportional to the quantities shown . parameters are respectively @xmath143 in ( a ) and ( c ) , while @xmath144 in ( b ) and ( d ) [ a small symmetry - breaking field @xmath145 was added in ( d)].,title="fig : " ] ( color online ) ( a , b ) correlation functions @xmath92 with respect to site @xmath17 indicated by a black circle . ( c , d ) induced magnetization patterns @xmath142 due to the presence of a vacancy , on the @xmath22 cluster . circle areas are proportional to the quantities shown . parameters are respectively @xmath143 in ( a ) and ( c ) , while @xmath144 in ( b ) and ( d ) [ a small symmetry - breaking field @xmath145 was added in ( d)].,title="fig:",width=145 ] + ( color online ) ( a , b ) correlation functions @xmath92 with respect to site @xmath17 indicated by a black circle . ( c , d ) induced magnetization patterns @xmath142 due to the presence of a vacancy , on the @xmath22 cluster . circle areas are proportional to the quantities shown . parameters are respectively @xmath143 in ( a ) and ( c ) , while @xmath144 in ( b ) and ( d ) [ a small symmetry - breaking field @xmath145 was added in ( d)].,title="fig:",width=137 ] ( color online ) ( a , b ) correlation functions @xmath92 with respect to site @xmath17 indicated by a black circle . ( c , d ) induced magnetization patterns @xmath142 due to the presence of a vacancy , on the @xmath22 cluster . circle areas are proportional to the quantities shown . parameters are respectively @xmath143 in ( a ) and ( c ) , while @xmath144 in ( b ) and ( d ) [ a small symmetry - breaking field @xmath145 was added in ( d)].,title="fig:",width=141 ] + we start by considering the liquid phase , and choose the field direction along the @xmath66 $ ] axis , for which features characteristic of the kitaev model are the clearest . first , we show that the combination of an infinitesimal field and a spin vacancy induces a peculiar magnetization pattern around the vacancy . the relevant quantity is the induced magnetization : ^z_j = ^z_j -m_0 , which gives , for a given field @xmath63 and site @xmath84 , the change of the magnetization at this site relative to its value ( @xmath59 ) in the vacancy - free system . we relate the vacancy - induced magnetization pattern to the correlation function @xmath146 of the vacancy - free system . the most striking feature of the magnetization patterns shown in fig . [ cormag](c , d ) is the large magnetization on the @xmath15-neighbor of the vacancy ( referred to as site @xmath147 in fig . [ difmz ] from now on ) . the magnetization is significantly larger there than on any other site and only on this site does it remain finite at vanishing field in the kitaev model : @xmath148 for @xmath31 and @xmath149 ( @xmath150 for @xmath22 ) . actually , in the thermodynamic limit ( tl ) the magnetization ( both the total one and @xmath151 ) should behave with field as@xcite @xmath152 ; our results are consistent with this prediction in the sense that @xmath153 decreases with increasing @xmath80 , so that it should vanish in the tl as expected . as for the logarithmic field - dependence , we stress that this feature pertains to the tl , while for finite systems we can only notice weak non - linearities in the magnetization curves , discussed later in more detail . the magnetic response localized on the @xmath15-neighbor of the vacancy is clearly connected with the fact that the correlation function @xmath154 is non - zero only between nearest - neighbors sharing a @xmath15-bond @xcite : if one spin is removed at site @xmath155 , the term @xmath156 obviously disappears from the hamiltonian and thus the spin @xmath147 is easily polarizable by the @xmath15-oriented field ( one can also interpret this behavior in terms of dangling majorana fermions @xcite ) . yet , we already saw in section [ phas ] that a small field induces @xmath154 correlations beyond nearest neighbors , it makes sense to focus in more detail on the magnetization pattern around a vacancy at small finite @xmath63 . [ cormag](c ) shows the variation of magnetization @xmath157 due to the presence of the vacancy . indeed , the spatial variations of this quantity presents similarities with those of correlations in the vacancy - free system : @xmath157 is positive everywhere and decreases with increasing distance to the vacancy . moreover , the spatial oscillations of correlation function appear in the magnetization pattern . indeed , if one compares values of @xmath158 at relative position either @xmath159 or @xmath160 from the vacancy [ see values @xmath161 and @xmath162 respectively in fig . [ cormag](c ) ] , their respective ratio is comparable to that of corresponding correlations in the vacancy - free system . while @xmath163 correlations scale as @xmath164 , the magnetization enhancement @xmath158 on further neighbors of the vacancy grows approximately linearly with @xmath63 ; in contrast , for @xmath165 , it decreases from @xmath153 at @xmath69 towards small negative values in the polarized phase ( see @xmath166 in fig . [ difmz ] ) . although the response observed here is obtained for a finite cluster size , we may expect that in a large system , the spin at site @xmath147 acquires a large magnetization due to the combined effect of the neighboring vacancy and the small field , and in turn polarizes its other neighbors in a way that the @xmath157 reproduces the above discussed correlation function pattern ( fig . [ cormag](a ) ) . + if one considers the effect of a vacancy on the magnetic response of the system in a field @xmath63 , a relevant quantity to consider is @xmath167-(n-1)m_0(h_z)$ ] , with @xmath59 the magnetization _ per site _ of the vacancy - free system , and @xmath168 $ ] is the total magnetization of the system with @xmath169 spins plus a vacant site . this quantity , shown in fig . [ difmz ] , vanishes in the infinite field limit ( where all spins are polarized ) , but also at small field in the tl . in finite systems , the previously discussed finiteness of @xmath153 goes along with a deviation from the expected logarithmic behavior of @xmath170 ; yet , we notice on large enough systems ( @xmath22 ) a non - linearity of the magnetization curve in the small field range @xmath171 , the slope @xmath172 decreasing moderately with increasing small @xmath63 . but when @xmath63 becomes close to the value @xmath115 of the liquid / polarized transition for the undoped system , the slope @xmath173 increases strongly , so that @xmath174 peaks near @xmath175 . the exact position of the peak depends slightly on the cluster size , and the peak amplitude grows with increasing system size ( roughly linearly : compare amplitudes for @xmath24 and @xmath22 on fig . [ difmz ] ) . the existence of this peak can be understood in the following way : not only does the presence of a vacancy cause an increase of the magnetization in the liquid phase , but it shifts slightly the liquid / polarized transition towards smaller fields @xmath176 ( the difference between both fields being likely proportional to the vacancy concentration in a diluted limit ) , and the peak signals a field range @xmath177 . this behavior can loosely remind of spin polarons formed around mobile holes in a paramagnetic background , so that the temperature of transition to the ferromagnetic ( fm ) phase shifts up with increasing hole density@xcite . here , instead of temperature the parameter driving the transition is the magnetic field , and moreover a vacancy is here not mobile , but its effect is as well to polarize its surroundings more and more when approaching the fm phase ; consequently , at small finite vacancy concentration , the transition field should be slightly smaller than in the vacancy - free case . spin - vacancy induced magnetization @xmath174 as a function of magnetic field @xmath63 , for different @xmath60 and cluster sizes . shown also is a partial contribution ( open circles ) of the spin at site @xmath147 , being the @xmath15-neighbor of the vacancy , for @xmath31 and @xmath22.,width=294 ] an analogy between the spin correlations in the vacancy - free system and the magnetization pattern around a vacancy can also be noticed in the case where the kitaev hamiltonian is perturbed by small heisenberg interactions [ see fig . [ cormag](b ) and fig . [ cormag](d ) ] . note that in fig . [ cormag](d ) , we also included an infinitesimal field @xmath63 , the effect of which is only to break the kramers degeneracy related to the odd number of spins . there , both the correlations and the vacancy - enhanced magnetizations @xmath158 decrease rapidly with distance to the vacancy . moreover the sign of these quantities alternates consistently with the pattern of @xmath15-oriented stripes shown in inset of fig . [ slrstripes ] , and their decrease is much faster when going perpendicularly to stripes than along stripes : for the magnetizations @xmath178 ( or equivalently @xmath158 , since @xmath179 ) , one can again compare in fig . [ cormag](d ) the absolute values at positions @xmath160 ( @xmath180 ) and @xmath159 ( @xmath181 ) from the vacancy . the staggered @xmath15-bond magnetization , which measures the @xmath15-stripe order parameter , m^z(_z)=_r e^i _ z r ^z_r , a+^z_r , b , is shown in fig . [ slrstripes ] . it also evidences that , in the liquid phase , the short - range stripe pattern is induced around the vacancy by small heisenberg interactions . starting from the @xmath31 limit , the magnetizations at further neighbors increase ( in absolute values ) significantly with increasing amplitude of heisenberg interactions , i.e. with decreasing @xmath60 . we expect that the extensive quantity @xmath182 diverges in the tl for @xmath183 . moreover , we note that the vacancy - induced magnetization @xmath174 , for fixed small field @xmath63 , is reduced by the presence of heisenberg interactions . it is significantly smaller for @xmath113 than for @xmath31 , as the tendency to stripe order around the vacancy , at small field , competes with the aligning effect of the field . consequently , the peak in @xmath184 shifts slightly towards higher fields with decreasing @xmath60 : the effect of heisenberg interactions is to increase the position @xmath185 of transition to the polarized phase ( as for @xmath115 in the vacancy - free case , see fig . [ diag ] ) . we briefly comment here on the fact that in the polarized phase ( and unlike in small field ) , the vacancy - induced magnetization change @xmath166 at the @xmath15-neighbor is negative . this can be understood with the perturbative treatment of transverse interactions described in section [ phas ] ; but here , to evaluate the quantum correction to the magnetization at a given site , we consider the off - diagonal ( perturbing ) couplings only on neighboring bonds see ref . for a similar approach . within this approach , the magnetization at this site is @xmath186 ; and the predicted value for @xmath166 is of the same order as seen in fig . [ difmz ] and it decreases in a similar way as @xmath63 increases , sufficiently far away from the transition field to either the liquid or the stripe phase . qualitatively , one can interpret this by the fact that at high field , fluctuations from a fully - polarized state are effectively stronger at the vacancy s @xmath15-neighbor than elsewhere , since the absence of ferromagnetic interaction on the missing @xmath15-bond makes it easier to flip this spin than other spins . with a similar reasoning one can explain that , in the same field range , the magnetization at other nearest neighbors of the impurity is enhanced compared to @xmath59 . if , instead of a field along an easy axis , a field in the @xmath29 $ ] direction is applied , one can assume@xcite that the magnetic response is approximately a linear combination of responses to each field component : thus each neighbor of a vacancy acquires an anomalously large magnetization - the neighbor spin at site @xmath84 , that would be connected to the missing spin by a @xmath187-bond , has a magnetization @xmath188 , and in turn brings a smaller polarization , also parallel to @xmath189 , to other spins in its vicinity - mainly on neighboring sites which are close to the spatial axis in the prolongation of the missing @xmath187-bond . hence , three domains should coexist in the system , each with a different magnetization direction . note that a vector @xmath190 $ ] can be defined and used to characterize the polarization of spins around the vacancy . qualitatively , @xmath191 should point in a direction closer to the easy axis @xmath187 for which @xmath189 is the largest than to other easy axes ; but it can have three non - zero components simultaneously . in the stripe - ordered phase , among the 6 allowed stripe patterns ( 3 possible stripe directions times 2 spin orientations ) , a vacancy in combination with an infinitesimal field @xmath192 selects the @xmath15-stripe pattern for which @xmath193 up to small quantum corrections . this selection is due to the necessary imbalance between the number of spins respectively aligned and antialigned with the field ( see inset of fig . [ slrstripes ] ) . this state is stable versus increasing @xmath63 , until the field reaches a critical value ( or anisotropy field @xmath194 ) , where canted stripe patterns those with stripe directions along @xmath32 or @xmath33 become more favorable . this effect is similar to the local rotation of the staggered magnetization close to a vacancy in @xmath6 antiferromagnets @xcite . here , the main difference is that there is only a finite number of stripe patterns which can not be continuously connected to each other without going out of the ground state manifold ; consequently , the system has to choose between allowed stripe patterns , tilted or not by the field , and @xmath191 is forced to point exactly along one easy axis @xmath187 , corresponding to ( one of ) the most favorable pattern(s ) . to understand the existence of a finite anisotropy field , one has to compare the energy of the @xmath15-stripe pattern ( _ _ untiltable__@xcite in a @xmath66$]-field ) to that of the _ tiltable _ @xmath32- and @xmath33-stripe patterns . the former , thanks to its total magnetization @xmath195 , acquires locally an energy gain @xmath196 with respect to the zero - field value but it has zero susceptibility ( @xmath197 ) thus no contribution @xmath198 to its energy ; in contrary , the field lowers the energy of the latter patterns by an extensive quantity @xmath199 . this comparison gives @xmath200 for @xmath201 , not far from the position of the jump in @xmath202 in fig . [ slrstripes ] . in the vicinity of the stripe - liquid transition , the magnetization pattern induced by a small @xmath66$]-field ( @xmath203 ) around a vacancy can be of special interest : assuming that vacancies are diluted enough such that their mutual interactions can be neglected , the patterns expected on both sides of the transition are clearly distinct from each other . in the stripe phase , all neighboring spins of the vacancy have a large , almost field - independent magnetization , with two of them aligned with the field and the third one antialigned with @xmath57 . in contrast , in the liquid phase only one nearest neighbor spin is significantly polarized ( with a strongly field - dependent polarization ) and aligned with the field ; and the two others are weakly polarized opposite to the field . thus , if one would have a local probe to determine the ( site - resolved ) magnetization in the neighborhood of a vacancy this probe could be @xmath204 nmr for instance the measured pattern could indicate on which side of the transition the system is . we have also seen that a vacancy can block the spin canting mechanism in the stripe phase , under a small field in ( or close to ) the @xmath66 $ ] axis . this phenomenon is absent for a field in the @xmath29 $ ] direction , where all stripe patterns remain equivalent and continuously tiltable by the field . consequently , in a finite system , the situation concerning the susceptibility anisotropy at zero - field : @xmath205}$ ] , is now opposite to the previously discussed case without vacancies where we found @xmath90}<\chi^z$ ] . concerning now the thermodynamic limit with a finite concentration @xmath206 of vacancies , if one assumes that @xmath206 is large enough to influence the magnetic response at low temperatures but small enough so that vacancies behave independently from each other , a similar effect could account for the experimentally observed @xmath207 . however , given that in a small field , within the kitaev model , the magnetization enhancement caused by two vacancies residing on the same sublattice is much larger than twice the magnetization enhancement caused by a single vacancy @xcite , one should take care about cooperative effects between the vacancies . ( color online ) the @xmath15-stripe order parameter @xmath208 computed on the @xmath22 cluster , as a function of @xmath60 at different small fields @xmath63 . right - inset : the magnetization pattern @xmath209 around the vacancy for @xmath201 and @xmath210 . left - inset : total magnetization @xmath170 as a function of a field @xmath63 applied in the liquid ( @xmath113 ) and in the stripe ( @xmath201 ) phases , with a magnetization jump at @xmath211.,title="fig:",width=302 ] + we turn now to the case where several vacancies are present in the system ; we first consider their effect on the magnetization curves ; then , we will focus on their effective interaction in the stripe phase and show how a vacancy pair can select an ordered pattern . it is instructive to compare the effect of vacancies or pairs of vacancies on magnetization curves , for given field direction and system size . to do this comparison we choose a @xmath66$]-field and consider the periodic @xmath22 cluster , but discuss as well implications for the thermodynamic limit . magnetization curves @xmath202 without or with vacancies , in different phases of the kitaev - heisenberg model , are shown in fig . [ magnsl ] . _ in the nel phase _ , although the total magnetization @xmath212 is not a conserved quantity in presence of small kitaev interactions ( @xmath213 ) , the magnetization curve @xmath202 on finite clusters displays a succession of plateaux ( almost flat for @xmath214 ) , as seen in fig . [ magnsl](a ) . they are separated by jumps @xmath215 corresponding to a flip of one spin @xmath2 . at fixed @xmath60 the step - like increase of the magnetization per site @xmath72 disappears only in the tl , the steps becoming smaller and closer to each other as @xmath80 increases . moreover the steps disappear , at fixed cluster size , as @xmath60 gets closer to @xmath216 . within this phase , one is tempted to define an effective spin @xmath2 , @xmath217 , associated to the vacancy , such that @xmath218 corresponds to @xmath217 pointing along @xmath219 ; with two vacancies , their effective @xmath220 s add up if vacancies are on the same sublattice ( e.g. , n.n.n . case ) and cancel each other if on opposite sublattices ( n.n . case ) . from the behavior of @xmath221 one obtains similar conclusions concerning effective couplings between other components of @xmath222 . one could therefore think that an effective interaction between two vacancy spins , if it can be defined , should be @xmath6-invariant , and possibly of heisenberg type @xmath223 ; yet , the gapless nature of the undoped system and the continuous behaviour of magnetization curves in the tl make it unclear how to formulate the effect of vacancies in terms of effective spins . _ in the liquid phase _ , in contrast , the magnetization does not show any plateau , but vacancies cause peculiar features at small fields : we have already seen that , compared to the zero - vacancy curve , the slope of @xmath202 at small field is significantly increased by the presence of one vacancy in the system . if two vacancies are present , their effect depends strongly on their respective sublattices ( see fig . [ magnsl](c ) ) : if they are on opposite sublattices the @xmath202 curve is , as in the vacancy - free case , linear for small @xmath63 [ and @xmath224 is only slightly modified ] ; but if they are on the same sublattice , this slope is greatly increased for a large range of @xmath63 the vacancy - induced increase in @xmath170 is more than twice that of the single - impurity case . again , even though we can not identify a logarithmic behavior @xmath225 as derived in ref . for @xmath31 , we see that vacancies , as soon as they are not equally distributed on both sublattices , act like _ partially - unbound moments _ causing non - linearities in the magnetization properties at low field . _ in the stripe phase _ , as in the liquid phase , kitaev interactions are strong enough to prohibit the existence of plateaux in the magnetization curves @xmath202 , except at small fields in presence of vacancies : with one vacancy , the previously discussed @xmath193 plateau is again suggestive of an effective spin @xmath2 of the vacancy ; and with two vacancies on the same sublattice , the @xmath226 plateau seen on fig . [ magnsl](b ) ( pair p1 ) extends up to a field @xmath227 , that is more than twice the anisotropy field @xmath194 for a single vacancy . this indicates a cooperative effect of both vacancies , which stabilize the @xmath15-stripe pattern more than if they behaved independently from each other . to describe this cooperative behavior , we will , similarly as in section [ sosf ] , analyze the effect of vacancies on the energies of the various stripe patterns . ( color online ) magnetization curves @xmath202 on the cluster @xmath22 with 0 , 1 , or 2 vacancies . the labels @xmath228 correspond to relative positions of 2 vacancies as shown in panel ( c ) . panels ( a),(b ) , and ( c ) correspond to @xmath229 , respectively.,title="fig:",width=306 ] + ( color online ) magnetization curves @xmath202 on the cluster @xmath22 with 0 , 1 , or 2 vacancies . the labels @xmath228 correspond to relative positions of 2 vacancies as shown in panel ( c ) . panels ( a),(b ) , and ( c ) correspond to @xmath229 , respectively.,title="fig:",width=294 ] + ( color online ) structure factors ( a ) @xmath230 and ( b ) @xmath67 at @xmath231 calculated for various respective positions of 2 vacancies @xmath232 , @xmath233 , @xmath234 , indicated in the insets . also shown is the structure factor of the vacancy - free system ( open circles ) , for which @xmath235 . all results are obtained on the @xmath22 cluster.,title="fig:",width=272 ] + ( color online ) structure factors ( a ) @xmath230 and ( b ) @xmath67 at @xmath231 calculated for various respective positions of 2 vacancies @xmath232 , @xmath233 , @xmath234 , indicated in the insets . also shown is the structure factor of the vacancy - free system ( open circles ) , for which @xmath235 . all results are obtained on the @xmath22 cluster.,title="fig:",width=272 ] we focus now on the stripe - ordered phase , and will explore how at @xmath236 , a pair of defects can , depending on their relative position , favor or disfavor one of the three stripe directions . in terms of symmetries , this can be understood as a breaking of @xmath47 symmetry related to the spatial orientation of stripes , while the @xmath45 symmetry ( @xmath237 ) remains unbroken . remarkably , as shown in fig . [ mstripy_2 ] , the selected stripe pattern for a given vacancy pair depends on the coupling constant @xmath60 . + this effect is easy to understand for nearest - neighbor defects on a @xmath238bond ( configuration @xmath232 : see fig.[mstripy_2 ] ) ; classically , the @xmath238stripe pattern has an energy lower by e^z_1 = 2 - 4than the one of other stripe patterns . indeed , in the @xmath15-pattern only one fm bond - here a @xmath15-bond , with a classical contribution to the energy @xmath239 - is lost due to vacancies , instead of two for other patterns ( e.g. two @xmath32-bonds , in a @xmath32-stripe pattern ) , and for @xmath240 fm interactions contribute more than af ones ( whose classical contribution to the energy is @xmath241 ) . for @xmath242 the situation is reversed , which explains that @xmath230 becomes small and @xmath243 . for second neighbors ( case @xmath233 on fig . [ mstripy_2 ] ) , stripe patterns are classically degenerate , but this degeneracy is lifted by quantum fluctuations . we evaluate in second order perturbation theory the contribution to the energy of various bonds in which at least one site is a neighbor of the vacancy ; actually only af bonds give a contribution in this approximation . for instance fluctuations on a bond belonging to the same hexagon as the two impurities yield a contribution @xmath244 $ ] to the energy of the @xmath15-stripe state . collecting contributions from all relevant bonds , the energy of the @xmath15-stripe state ( relative to that of other stripe orientations ) is : e^z_2=2(1 - 2)^2 ( + - ) , which is positive for @xmath60 in the vicinity of @xmath2 , and changes sign at @xmath245 . this explains the drop of the @xmath15-stripe structure factor ( dotted curve on fig . [ mstripy_2](a ) ) , from large values for @xmath246 $ ] , to almost zero for smaller @xmath60 . the smaller values of @xmath247 in the vicinity of @xmath248 are related to the smaller energy scale separating the different stripe patterns : the above expression , with its prefactor @xmath249 , indicates that the 3 patterns become degenerate at @xmath248 . a similar energy comparison between different patterns for third neighbor vacancy pairs ( configuration @xmath234 : see fig . [ mstripy_2 ] ) predicts a change of stripe orientation at @xmath250 between @xmath238stripes ( @xmath251 ) and @xmath32-stripes ( @xmath252 ) , following from the perturbative result : e^z_3 = , whereas this orientational change is seen on @xmath22 at a slightly smaller value @xmath253 . notice that here ( and more generally for pairs of vacancies at further neighbor sites which are not related to each other by any reflection symmetry of the honeycomb lattice ) the three stripe orientations are all non - equivalent to each other . finally we note that at zero - field , the conclusions above hold also for pairs of neighbor vacancies with other relative positions : for instance with next - nearest neighbor vacancies separated by a @xmath32- and a @xmath15-bond and @xmath254 , according to the previous reasoning the favored pattern is a @xmath33-stripe pattern . in presence of a small magnetic field , the @xmath45 symmetry characterizing each stripe orientation ( e.g. , @xmath15-stripes ) is broken . moreover , similar effects occur as in the previously discussed single vacancy case : if the most favorable pattern in an infinitesimal @xmath66 $ ] field ( e.g. , @xmath15-stripe pattern around a vacancy pair in configuration @xmath232 , for @xmath240 ) is untiltable for this field direction , as the field amplitude @xmath63 gets larger than a critical value @xmath255 this pattern becomes less favorable than other patterns which are tiltable . in the limit of large systems @xmath256 , a classical estimate of @xmath255 gives a result @xmath257 which , for @xmath201 and @xmath22 , has a value @xmath258 comparable to the observed position of the magnetization jump . the relevant quantity for the discussion of either low temperature properties or magnetic ordering at small vacancy concentrations is the energy scale @xmath259 , namely the energy which favors a particular stripe pattern in the vicinity of a vacancy pair @xmath260 standing for the number of bonds separating both vacancies . far away from the stripe / liquid transition ( say , for @xmath261 ) @xmath259 decreases rapidly with @xmath260 ( e.g. , @xmath262 for @xmath263 ) . closer to @xmath35 , larger quantum fluctuations make the stripe - orientation selection by @xmath264 neighbor vacancy pairs as efficient as by nearest neighbor vacancy pairs . several consequences can result from this : ( i ) in the simpler case with only two vacancies , the energy lowering of some stripe patterns for @xmath265 also stabilizes these patterns with respect to the spin liquid , which leads to a slight shift of the stripe / liquid transition towards larger @xmath60 possibly @xmath266 to @xmath267 depending on the relative position of vacancies , as can be seen in fig . [ mstripy_2](a ) . ( ii ) more importantly , if one focuses on systems with substantial vacancy concentrations , say , @xmath268 , neighboring vacancy pairs as described in fig . [ mstripy_2 ] are abundant enough to select locally stripe orientations that will frustrate each other at the global scale . ( iii ) as a result , the system could display a glassy behavior , especially in the vicinity of the stripe - liquid transition . the resulting glassy phase would consist of stripe - ordered nanoscale domains , with a vacancy pair in each domain , selecting its stripe orientation . in order to perform a modeling of such a system and to reproduce a possible spin - glass behavior , one should obviously need to consider systems of ( at least ) hundreds of ir atoms , and to take into account the energy cost of domain walls between regions of different stripe orientations . moreover , obviously not all vacancies are grouped into distinct first , second or third nearest neighbor vacancy pairs ; in particular the effect of unpaired vacancies , even at zero external field , could be subtle , although one can imagine single vacancies pinning walls between domains of different stripe orientation . we have considered here several important aspects of the kitaev - heisenberg model , which , despite its simple structure , displays in its @xmath1 phase diagram three distinct phases , two of which are antiferromagnetically ordered while the third one is a spin liquid . we employed numerical methods ( exact diagonalization ) for our investigation of the properties of the model in a magnetic field and the effect of vacancies on the magnetic properties . the numerical methods were complemented by classical and perturbative analysis particularly in the frame of the stripe phase . we showed how the ground state phases evolve in a magnetic field oriented along the @xmath66 $ ] direction ( one of the local easy axes of the kitaev - heisenberg hamiltonian ) . we found that the three phases are robust to small fields . next , we analyzed the influence of the field direction . the magnetic response is qualitatively distinct in the different phases . in the stripe phase , the magnetization can be described by a spin canting mechanism , which also provides an understanding of the anisotropy in the numerically obtained susceptibilities . in the liquid phase , we compared the effect of small magnetic fields and of small heisenberg interactions , each considered as a perturbation of the kitaev hamiltonian . such perturbations are shown to induce specific patterns in the spin - spin correlations beyond nearest neighbors , while at larger fields a transition into the spin - polarized phase occurs . we determined the phase diagram of the model in a @xmath66$]-field , with good agreement between numerical results and analytical estimates . for the latter analysis , we took into account both the spin canting and the quantum fluctuations in the ordered and polarized phases . this phase diagram has similarities with its counterpart for a @xmath29 $ ] field orientation @xcite , however , a major difference between them is that for a @xmath66$]-field we do not see any quantum critical point in the vicinity of the stripe - liquid phase transition . we presume that this is related to the absence of a field - induced gap in the low - energy spectrum of the liquid phase for a @xmath66$]-field@xcite . subsequently , we focused on the effect of non - magnetic vacancies on the spin correlations and the magnetic response of the model . for a single vacancy in the liquid phase , we found that the spatially anisotropic magnetization pattern around the vacancy is related to the anisotropy of spin correlations in the vacancy - free system . up to finite - size effects inherent to our approach , these results are also in agreement with predictions of a non - linear response to a small @xmath66$]-field , whose effect is mostly seen on one specific neighbor of the vacancy . in the stripe phase , a vacancy coupled to a small [ 001]-field breaks both the @xmath45 ( time reversal ) and the @xmath47 ( orientational ) symmetry characterizing the ground state manifold ; a single stripe pattern is selected , which is not continuously tiltable by the field in contrast to the vacancy - free situation . having in mind the description of systems at finite vacancy concentration , we considered the interplay of two vacancies sitting either on nearest or further neighbor sites . again the magnetic response to an easy - axis field depends strongly on the nature of the phase , but it also depends sensitively on the relative position of vacancies . in the liquid phase a non - linear behavior , much stronger than for the single - impurity case , is seen if the two vacancies are on the same sublattice , while the response remains linear if they reside on different sublattices . in the stripe phase , vacancy pairs select a specific stripe orientation even at zero field . this selection mechanism , depending on the value of coupling parameter @xmath60 , can be well understood by a perturbative approach evaluating locally the effect of quantum fluctuations . the response to a small field depends on whether the selected stripe pattern is tiltable or not for this field direction@xcite . although the strongest selection effect is caused by nearest neighbor vacancy pairs for a wide range of @xmath60 , further neighbor pairs have to be considered as well in the vicinity of the stripe - liquid transition . therefore we conclude that the intrinsic randomness of vacancies , in combination with the selection mechanism induced by vacancy pairs , may lead to spin - glass behavior in the stripe regime . there , a vacancy concentration of a few percents should be enough for the system to show a glassy behavior , with a tendency to the formation of stripe - ordered nanodomains around vacancy pairs . finally , we comment on the possible relevance of this work to layered iridates , which motivated the derivation of the kitaev - heisenberg model . the experimental data ( magnetic susceptibility , heat capacity@xcite ) on na@xmath3iro@xmath4 indicate antiferromagnetism at low temperature and a substantial magnetic anisotropy . yet , the angular dependence of the measured susceptibility is qualitatively opposite to that found for a vacancy - free model . we then showed that non - magnetic vacancies in the dilute limit , i.e. , where interactions between vacancies are irrelevant , can reverse locally this tendency by blocking the spin - canting for some field directions . the recent x - ray magnetic scattering data @xcite ruled out the nel state , implying that the kitaev interaction in na@xmath3iro@xmath4 may dominate over a simple heisenberg coupling . in addition to the stripe phase intrinsic to the kitaev - heisenberg model , the so - called zig - zag spin order is also consistent with the data @xcite . an element in favor of a description of layered iridates a@xmath3iro@xmath4 ( a@xmath269li , na ) by the kitaev - heisenberg model is the comparison of the low temperature susceptibilities of these compounds@xcite . using our result @xmath138 for the stripe phase , we found a relation between @xmath60 values in these two compounds , which indicates that the li compound , as the na - based one , is most probably in the stripe phase but much closer to the transition towards the spin liquid . this conclusion is in agreement with previous work@xcite estimating the relative @xmath60 values from a comparison of the paramagnetic curie temperatures . the application of uniaxial pressure on a single crystal of li@xmath3iro@xmath4 might even drive the system into the liquid phase . to describe the experimental situation in layered iridates on a quantitative level , other factors such as lattice distortions have to be considered as well . in particular , the lattice distortions may bring about substantial anisotropy in the ground state @xmath270-factors and in the van vleck contribution to the magnetic susceptibility , originating from transitions to higher lying spin - orbital quartet split by non - cubic ( tetragonal and/or trigonal ) crystal fields . furthermore , these distortions may lead to spin interactions not included in the kitaev - heisenberg model . for intermediate strength of spin - orbit coupling like in iridates , these effects might be essential for a quantitative description of magnetic properties . moreover , our study suggests that a glassy behavior could occur in the vicinity of the spin liquid phase ; in this context , considering the present model at larger scales , with a finite density of vacancies , could help to explore the effect of temperature . a quantitative comparison of our calculated susceptibilities to those measured in ref . would include also the effect of temperature , especially as the measured anisotropy was also large in the paramagnetic phase .

the crab was the first source of gamma rays to be convincingly detected by ground based telescopes @xcite and measurements of its emission spectrum between @xmath8 and @xmath9 by various atmospheric erenkov detectors are now available @xcite . the flux measurement above @xmath10 recently reported by the stacee experiment , using the mirrors of a solar energy research facility to collect erenkov light , is the first detection below 200 gev by a ground based device @xcite . at these energies emission from the crab is steady and generally accepted to come from the nebula , arising from the inverse compton scattering of the synchrotron photons observed at lower energies @xcite . the egret detector on board the compton gamma ray observatory was used to study the crab from @xmath11 to @xmath12 @xcite . the differential energy spectrum measured by egret is well described by the sum of two power laws . below @xmath13 the steep spectrum is attributed to the synchrotron radiation from the nebula , while beyond @xmath13 the spectrum hardens and is dominated by pulsed emission . the detailed origin of the pulsar emission is uncertain . the outer gap @xcite and polar cap @xcite models offer differing pictures . current very high energy measurements create difficulties for some outer gap models @xcite but refining the picture requires observations in the heretofore uncovered @xmath14 region . determining the energy at which pulsed emission is again overtaken by the nebula flux is one of the goals of the present work . while the crab itself is a rather special object , the success of the synchrotron self - compton ( ssc ) model as applied to the nebula has wide implications . on the one hand , this bright source is a test piece for the study of supernova remnants as the acceleration sites of high energy cosmic rays , with at issue the question of whether proton or electron acceleration dominates in a given source . in addition , the ssc mechanism is a cornerstone for the interpretation of the broadband spectra of agns of the blazar class @xcite . the experimental data from the crab which support the ssc picture consist of egret flux measurements up to @xmath12 , with large uncertainties in the region above @xmath15 @xcite , and the extrapolation across more than a decade in energy to the spectra measured by the atmospheric erenkov experiments . clearly , an independent measurement in the intervening @xmath16 region ( @xmath17 ) where the inverse compton peak in the power spectrum is expected to lie would further constrain the parameters of this important model . the minimum energy threshold , @xmath18 , for current ground based imaging atmospheric erenkov experiments is limited to @xmath19 by the rate of accidental triggers due to the night sky light and , in the case of single mirror experiments , by the rate of local muon triggers . the simplest way to reduce the threshold of such an experiment is to increase the available mirror area , @xmath20 , as @xmath21 ; an approach which is being followed by the magic collaboration @xcite . alternatively , an array of smaller telescopes can be used to reach thresholds of @xmath22 as predicted for the veritas @xcite and hess @xcite experiments . these experiments are currently under construction and have not yet started taking data . celeste was designed to reach a very low energy threshold without a large expenditure of time and resources by exploiting the mirrors of an existing structure ; a de - commissioned solar farm in the french pyrenees . an array of 40 such mirrors , used by celeste to sample the arrival time and photon flux of the erenkov wave front at intervals of @xmath23 , provides a total mirror area of @xmath24 . celeste uses techniques similar to those pioneered by the early wavefront sampling experiments asgat @xcite and themistocle @xcite which operated on the same site , but uses a much greater mirror area and more sophisticated trigger logic and data acquisition electronics . unlike the imaging experiments , the wavefront sampling method gives no direct information about the shower morphology , but alternative methods of hadron rejection can be developed using the shape of the wavefront and the distribution of erenkov light on the ground . since their crab detection cited above , stacee has lowered their threshold to @xmath25 and expects to descend to @xmath26 @xcite . the graal experiment also uses a heliostat array but without secondary optics obtains a relatively high threshold of @xmath27 @xcite . in this paper we present the first measurement of the flux from the crab above @xmath28 , as well as an upper limit for pulsed emission , using the celeste heliostat array . we begin with a description of the experiment followed by a summary of the data sample and observation techniques . celeste exploits a new experimental technique so we outline the analysis method in some detail , including the results of extensive monte carlo simulations of the detector and the analysis of data taken in common with the cat imaging erenkov telescope . the gamma ray flux measurement and the pulsed flux upper limit are presented and the implications for the emission models are discussed . further details on these measurements and on the celeste experiment in general are available in @xcite . the celeste experiment is described in full detail in the experiment proposal @xcite and in @xcite . here we outline the most important features and the status of the experiment during the relevant observation period . [ principle ] illustrates the experimental principle . celeste uses 40 heliostats of a former solar electrical plant at the thmis site in the eastern french pyrnes ( n. @xmath29 , e. @xmath30 , altitude @xmath31 ) . each back - silvered heliostat mirror has an area of @xmath32 and moves on an alt - azimuth mount . the heliostats are controlled from the top of a @xmath33 tall tower , located south of the heliostat field , which houses the secondary optics , photomultiplier tubes ( pmts ) and data acquisition system . the alignment of the heliostats has been verified by mapping the images of bright stars using the pmt anode current . the light from all 40 heliostats is reflected to the top of the tower . to separate these signals from each other we use a secondary optical system , as illustrated in fig . [ optics ] . we have chosen to place the photomultiplier assembly on the optical axis to minimize coma aberrations , although this results in a loss of light due to the shadow formed . the spherical mirrors of the secondary optics are divided into six segments on three levels with three different focal lengths in order to reduce this shadowing effect and to produce images of approximately the same size regardless of the heliostat position in the field . one large segment views the farthest heliostats , two others view those at intermediate distance , and three small segments are used for the heliostats at the foot of the tower . at the secondary mirror focus is the entrance face of a solid winston cone glued to a two - inch pmt ( philips xp2282b ) , one for each heliostat . the winston cone determines the surface area of the secondary mirror seen by that pmt , such that the optical field - of - view of each tube is @xmath34 mrad ( full width ) . this field - of - view is slightly smaller than the angular size of air showers in our energy range and helps maximze the ratio of erenkov to night sky light . the single photoelectron ( pe ) pulse width , after pre - amplifiers ( gain=100 , ac - coupled ) and 23 m cables to the counting house , is just under @xmath35 ( full width at half maximum ) . pmt gains are set reasonably low ( @xmath36 ) to avoid damage to the tubes from night sky light , and the electronic gains are such that the amplitude of a single photoelectron in the counting house is @xmath37 on average . these amplitudes were measured _ in situ_. in fact , studies of the average response of each detector to the hadronic background events have enabled us to calibrate the relative efficiency of each heliostat , and the pmt high voltages are now set so as to correct for this ( in the range @xmath38 ) in order to give an even trigger response across the heliostat field . the pmt signals are sent to both the trigger electronics and to the data acquisition system . the trigger is designed to reach the lowest possible threshold . programmable analog delays compensate for the changing optical path lengths as the source direction changes during the observation . the switched - cable delays broaden the pe pulse widths by a full nanosecond for the maximum delay . eight pmt signals are summed in each of five groups as shown in fig . [ trigger ] , and the sums enter a discriminator . programmable logic delays further compensate for the varying path lengths between the trigger groups . the logic delay introduces a deadtime of the order of 5% . a trigger requires the logic coincidence of at least three of the five groups , with an overlap of @xmath39 . the analog sum over eight heliostats provides us with a good signal to noise ratio for the erenkov pulse , while the logic coincidence removes triggers due to afterpulsing in the pmts , local muons or low energy hadronic events illuminating only a few heliostats . each pmt signal is further amplified ( @xmath40 ) and sent to an 8-bit flash adc ( fadc ) circuit ( etep 301c ) that digitizes the signal at a rate of @xmath41 ( 1.06 ns per sample ) . the depth of the fadc memory is 2.2 @xmath42s , and one photoelectron corresponds to 3 digital counts . when a trigger occurs , digitization stops and a window of 100 samples centered at the nominal erenkov pulse arrival time is read out via two vme busses in parallel . readout requires @xmath43 , which for typical raw trigger rate of 25 hz gives an acquisition deadtime fraction of @xmath44 . the trigger also latches a gps clock , which is read out and included in the data stream . in parallel with the erenkov pulse data acquisition , scalers record the single group trigger rates , the final trigger rate , and the readout rate . acquisition deadtime is determined from the latter two . the anode current of each pmt ( @xmath45 ) is also recorded , as is some meteorological information . the observations presented here were taken on clear , moonless nights during the crab season between november 1999 and march 2000 . all the data were taken when the source was within 2.5 hours of transit , that is , with an angle from the zenith , @xmath46 . the observations were made in the on - off tracking mode , in which an observation of the source is followed or preceded by an observation at the same declination offset in right ascension by an appropriate amount ( usually 20 minutes ) . the offset region is then used as a reference to provide a measure of the background of cosmic ray events . it is particularly important in the case of celeste to cover the same elevation and azimuth ranges during the on and the off source observations as the heliostat optical collection efficiencies change appreciably due to the projection of the heliostat surface viewed by the pmts , and ( less importantly ) due to optical aberrations . both of these effects depend upon the heliostat orientation and thus upon the source direction . in addition , matching on and off source observations ensures that the on and off data were taken using exactly the same path through the delay electronics . celeste has a number of options when deciding how to observe a source . the majority of the data here were taken in `` single pointing '' , wherein all the heliostats were aimed at a point @xmath47 upward from the center of the heliostat field towards the source such that the center of their fields of view converged at the expected maximum point of erenkov emission for gamma showers . this method collects the largest number of photons , allowing us to operate with the lowest possible energy threshold . it seems likely , however , that other pointing strategies may provide better sampling across the shower and hence better hadron rejection . with this in mind , a smaller number of runs were taken using `` double pointing '' in which half the heliostats pointed at @xmath48 , and the other half at @xmath49 . a monte carlo study of some different pointing methods is available in @xcite . the observing log is summarised in table [ tab : obslog ] . the trigger logic was set such that 3 groups out of the 5 were required to exceed their discriminator threshold in order to trigger the experiment . the discriminator threshold levels for each of the 5 trigger groups are checked nightly by measuring the trigger rate as a function of discriminator level in order to find the break point between accidental coincidences of random noise pulses and erenkov flashes . we set the discriminators such that the noise triggers contribute less than 1% of the total rate ( fig . [ trigrate ] ) . for more than 90% of the crab data the discriminator level was set to @xmath50 , that is , an average of @xmath51 for each of the 8 heliostats in a group , giving a final trigger rate of @xmath52 . expressing the discriminator level for the analog sum of 8 heliostats in a group in terms of pe per heliostat implies that the cerenkov pulses for each heliostat are perfectly in time with each other . we have checked this timing by reconstructing the group sums using the fadc data ( although the path to the trigger electronics is not identical to the acquisition path ) and by oscilloscope measurements during observations . for the data in this paper , three channels were as much as 2 ns out of time , while the other 37 channels were less than 1 ns from the average . the three outlying channels have since been corrected , and the group sum pulses are now routinely digitized using additional fadcs . the pmt anode current information and the measured trigger rates of each group are very sensitive to changes in the sky conditions and are used to verify that the atmosphere was stable throughout the on - off pair . any data which showed evidence of poor weather or equipment problems were rejected . the remaining total data set consists of 14.3 hours of on - source exposure . here we outline the important stages in the analysis of celeste data ; data cleaning , shower reconstruction and hadronic background rejection . we also present the results of extensive monte carlo simulations which have been used to derive the analysis techniques and to estimate the sensitivity and threshold of the experiment . celeste has the advantage of being situated on the same site as a well calibrated atmospheric erenkov imaging telescope , cat @xcite . this has allowed us to examine the collection efficiency for a subset of the celeste data and should in the future allow us to cross - calibrate energy , direction , and acceptance between cat and celeste . for each event which triggers celeste , we record a window of @xmath53 around the erenkov pulse for each pmt using fadcs with a sampling period of @xmath54 ( i.e. 100 samples ) . the fadc window is chosen such that the erenkov pulse is expected to arrive in its center . the beginning of the fadc window ( the first 30 samples ) is used to calculate the pedestal level . a small constant voltage offset applied to the unipolar input of each fadc allows fluctuations in the night sky background to be measured . significant differences in the amplitude @xmath55 of these fluctuations can be seen depending upon the brightness of the region of sky viewed by the pmt . the possibility of systematic effects in the data due to differences in night sky background levels between the on and off source regions of the sky is a known problem for atmospheric erenkov experiments . @xcite proposed a method of `` software padding '' , for use with the whipple telescope , in which the noise fluctuations of the adc signals from the darker region of sky are artificially increased to the same level as the brighter region by adding from a randomly sampled gaussian distribution . the effects of night sky light differences can be seen at both the trigger level and in parameter distributions during the analysis procedure . these systematic effects produce a significant difference between the number of events remaining from the on source and off source regions after analysis cuts . the difference can be either positive or negative , depending on which region is the brighter , and a positive difference mimics a real signal . celeste is particularly prone to these problems due to its large mirror area and angular acceptance per pmt which combine to give a night sky light background rate of @xmath56 . the use of fadcs introduces another complication in the case of celeste : if we wish to extract more information than just the integrated charge over the pulse , a simple addition of charge sampled from a night sky background distribution to the measured charge is not sufficient . the effect of additional sky noise on the complete erenkov pulse shape must be accounted for . the only way to equalize the night sky background fluctuations in software then , is to simulate the response of the pmt - fadc electronics chain to an increased rate of single photo - electrons . we model the single pe pulse using events triggered by cosmic ray muons passing through the winston cones , in standard operating conditions except with the tower door closed , blocking outside light . these pulses contain many ( @xmath57 ) pes , generated at the photocathode at almost exactly the same time so to a good approximation the muon pulse shape is the same as that of a single pe , only of greater amplitude . the results agree with those obtained on a test bench with an oscilloscope , and with single pe pulses measured by the fadc s , using much higher pmt gains which change the pmt time response somewhat . by simulating the fadc response to single pes arriving at different rates , we obtain a calibration curve of measured fluctuation against the background rate of pes due to night sky light of the form @xmath58 where @xmath59 is a constant and @xmath60 is the night sky background rate in @xmath61 per ns . this curve can be used to calculate the rate of simulated pes which needs to be added to the darker field in order to equalize the night sky background fluctuations . software padding has been applied to all the on - off pairs used in this analysis but this alone is not sufficient to remove all the biases caused by night sky background differences as a brighter region of sky also causes a slight increase in the amount of near threshold events which trigger the experiment . this can be explained as follows : additional night sky background fluctuations cause showers which would otherwise be below threshold to trigger . they also prevent some events which would otherwise be above threshold from triggering , but because the cosmic ray spectrum is very steep , the former effect is bigger than the latter , and there is a net night sky background dependent increase in the number of triggered events . we have therefore found it necessary to apply a `` software trigger '' at a level higher than the hardware trigger level , in order to remove these additional small events . using the fadc data , we reconstruct the analog sum pulses seen by each of the five trigger groups and then apply the condition : @xmath624 groups @xmath63 per heliostat . this provides us with comparable background data in the on and off fields , and reduces the fraction of events triggered by accidental noise coincidences to less than @xmath64 , but has the effect of increasing the energy threshold of the experiment ( fig . [ ethresh ] ) . to test the performance of the software trigger , we have divided the crab data set into two subsets , based on the sign of the difference in the average pmt currents between the on and off source observations . [ sttest ] shows the difference between the on and off source observations for the distribution of the total charge measured in all the erenkov pulses for these two subsets . a clear bias in the number of small events is apparent in the raw data , with the direction of the bias depending upon the sign of the current difference . after application of the software trigger , the bias has been removed . in order to use the information recorded by the fadcs , it is necessary to select and parameterize the erenkov peaks . this is done by fitting a function of the following form : @xmath65 where @xmath66=time in nanoseconds , @xmath67=peak time and @xmath68=peak amplitude . [ peaks ] shows some examples of fitted peaks and illustrates the effect of saturation in the fadcs . studies using simulated peaks indicate that the peak fitting algorithm can accurately reconstruct the timing and charge information for peaks which have saturated the fadcs up to twice their dynamic range . the fit parameters for each peak are stored for use later in the analysis . only events having at least 10 erenkov peaks with an amplitude greater than 25 digital counts ( @xmath69 p.e . ) are used in the analysis ( @xmath70 ) . imaging erenkov telescopes have become the most powerful instruments at energies greater than @xmath71 due to their efficiency in reducing the hadronic background . typically , it is possible to reject over 99% of the background events while retaining 50% of the gamma ray signal @xcite . at celeste energies , a smaller total number of photons and intrinsic fluctuations in the shower development mean that the differences between the gamma and hadron showers which trigger are less pronounced . in addition the small field of view of celeste , which is necessary to keep the night sky light background at a reasonable level , often truncates the shower , again causing hadron and gamma showers to look alike . these points , and also the fact that the trigger system rejects many hadron showers at the hardware level , mean that hadron rejection at the analysis stage is not very efficient for celeste ; however , small differences do remain , as the gamma showers tend to develop in a more regular manner than the hadron showers . we have written a complete detector simulation package , including a full treatment of the complicated optical system of celeste and a detailed model of the trigger and acquisition electronics , for use with standard air shower simulation packages . using the monte carlo simulations we have investigated various ways of exploiting the fadc timing and charge information to provide hadron rejection . two rather simple parameters have been studied in detail : the group homogeneity , @xmath72 , and the shower axis angle , @xmath73 . the group homogeneity is a measure of the homogeneity of the erenkov light pool at ground level . it is determined from the variance in the amplitude of the five trigger group pulses normalized to the mean amplitude . the trigger group pulses are derived by summing the 8 fadc windows of the heliostats in each group . @xmath74 where @xmath75 are the amplitudes of the 5 reconstructed trigger group pulses . [ sigmagrp ] shows the distribution of @xmath72 for gamma rays and off source data after applying the software trigger and requiring a minimum of 10 erenkov peaks . the gamma rays were simulated over a range of azimuth and zenith angles so as to match the range covered by all the data for the @xmath76 crab data set described in table [ tab : obslog ] . the off source data shown is the sum of all the off source data in this data set . according to this plot , a cut at @xmath77 conserves @xmath78 of the gammas which remain after the software trigger , while rejecting @xmath79 of the remaining hadrons , giving a quality factor q=1.6 where : @xmath80 and @xmath81 and @xmath82 are the fraction of gammas and hadrons conserved by the cut respectively . low energy gamma ray air showers are only a few kilometres long and the majority of the erenkov light is emitted from a small region . the erenkov wavefront is therefore spherical to a good approximation ( fig . [ spherefit ] ) . using the arrival times of the erenkov pulses we are able to reconstruct this wavefront using an analytical @xmath83 minimization procedure @xcite . assuming that the point of emission was at a fixed distance @xmath84 from the site towards the source , the fit gives the position @xmath85 of the shower maximum relative to the tracked point , @xmath86 . simulations indicate that this position is reconstructed with an error of @xmath87 . it is important to know the timing resolution for each detector when making the fit . we have calculated this resolution by studying the response to a nitrogen laser pulse sent to a diffuser mounted at the top of the tower . the same laser was used for a similar purpose by the themistocle experiment on the same site @xcite . the timing resolution is also dependent upon the background night sky light level and on the amplitude of the pulse . this dependency is difficult to test with the laser so we have measured it by generating simulated peaks , adding them to real night sky background data and then comparing the reconstructed peak time with the known injection time of the simulated peak . the resolution reaches @xmath88 ns for peaks well above the night sky noise level , and is worse for larger and smaller peaks due to fadc saturation and relatively larger night sky fluctuations , respectively . using the expected point of maximum emission we can attempt to measure the angle , @xmath73 , of the shower axis relative to the pointing direction , which will be zero in the case of gamma rays originating from a point source at the center of the field of view . to do this we need a second point at ground level , simply calculated by taking the mean position of the heliostats on the ground , weighted by the charge sampled by each detector . more complex algorithms have been tested for calculating the impact parameter , but none has proved more effective than this simple method , which gives a @xmath89 error of @xmath23 according to the simulations . the distribution of @xmath73 for simulated gammas and for real off source data after the software trigger , requiring a minimum of 10 erenkov peaks and @xmath90 is shown in fig . [ theta ] . as expected , the simulated gamma rays concentrate at small values of @xmath73 , with an angular resolution of @xmath91 . unfortunately the hadronic background showers , although simulations suggest that they can trigger the experiment from as far away as @xmath92 from the pointing axis , are reconstructed with an angular spread of only @xmath93 . a cut on @xmath73 alone at @xmath94 predicts a quality factor of only 1.1 after the other cuts have been applied . in addition to the crab nebula , celeste has recently been used to detect gamma ray emission from the tev blazar markarian 421 @xcite . observations made at the same time by the cat experiment allowed us to know the status of this highly variable source . the source was observed in december 1999 in a quiescent state and in january and february 2000 in an active state , with flares reaching a level of @xmath95 according to cat . the results from the celeste analysis show a non - detection for the december period ( a significance of @xmath96 for @xmath97 of on source data ) and a very significant ( @xmath98 for @xmath99 ) detection for the january - febuary observations . these results are noted here as they provide further convincing evidence for the stability of the celeste analysis . the on source star field for the region of mkn 421 contains a star of magnitude 6.1 in the center of the field . this causes the measured average pmt anode currents for the on source fields to be typically 13% higher than for the off field . the sky noise differences in the case of the crab vary by as much as @xmath100 but for all data pairs the dispersion is about @xmath101 , and the mean difference is smaller than our measurement error . the non - detection of mkn 421 in december 1999 implies that the celeste analysis has correctly dealt with the systematic effects in the data due to sky noise differences for this problematic source . we can therefore be confident that the smaller sky noise differences in the case of the crab observations do not pose a problem , and that our result presented in this paper is not significantly biased by systematic effects . our flux determination uses the results of the analysis of the larger of the two data sets listed in table [ tab : obslog ] : the @xmath102 of observations with all heliostats pointing at @xmath103 . the filled circles in figs . [ sigmagrp ] and [ theta ] show the distribution of the excess events in the on source data for @xmath72 and @xmath73 respectively . as predicted for a gamma ray signal , the on source excess concentrates at low values of @xmath72 . table [ tab : crablow ] shows the number of events which remain from the on and off source observations after the pre - analysis and analysis cuts . as discussed in the previous section , the first two cuts ( the software trigger and @xmath70 ) serve only to correct for night sky background differences and to ensure that there is enough information to reconstruct the shower reasonably well . the remaining cuts have been optimized on the simulations in order to reduce the hadronic background and improve the signal to noise ratio . as expected from the simulations , the most effective cut parameter is @xmath104 , with an observed quality factor of 1.4 , lower than the predicted 1.6 ( quality factors calculated after the software trigger and @xmath105 cuts ) . we note that at each stage of the analysis , after the initial pre - analysis cuts , the ratio of excess to background increases , from an initial value of 0.6% , to 5.0% when all cuts are applied . however , we determine the crab flux without using the cut on @xmath73 , as the monte carlo predicts only a small improvement in the significance of the result yet adds another source of error into the flux estimation . after the @xmath104 cut we find an excess of 2727 events , implying a rate of @xmath106 and a final statistical significance of @xmath107 . table [ tab : effall ] shows the cut efficiencies at each stage of the analysis procedure for the off source data , for the real crab on - off source excess , and for the simulated gamma rays . the agreement between the measured excess and the gamma simulations is reasonable , given the large errors on the excess fraction . we have also analysed the other set of crab observations taken in a `` double pointing '' mode , with half the heliostats pointing at @xmath103 and the other half at @xmath108 . the results are shown in table [ tab : crabdouble ] . a statistically significant signal is apparent in this smaller data set , the gamma ray rate being @xmath109 after all cuts , and @xmath110 without the @xmath73 cut . the monte carlo predicted an improvement in sensitivity with this pointing strategy due to its less - biased sampling of the erenkov light distribution at ground level , particularly for those showers with large impact parameters . in consequence , the cut on the homogeneity of the light distribution , @xmath104 , becomes more effective at rejecting the hadronic background . more data is needed to confirm the double pointing crab sensitivity of @xmath111 , compared to @xmath112 for single pointing . double pointing is now the preferred method of operation for celeste . further work is under way in order to determine the optimum pointing altitudes , trigger configurations and analysis methods . atmospheric erenkov telescopes , unlike satellite experiments , can not be calibrated with a test beam . monte carlo simulations of the detector response to air showers are therefore the most important tool for calculating both the detector sensitivity and determining the best analysis strategies . the work presented here has made use of the kaskade shower simulation package @xcite . tests using version 4.5 of the corsika package @xcite indicate an effective surface area for gamma rays @xmath11325% higher than that of the kaskade simulations , regardless of the initial photon energy . the reason for the discrepancy is not yet clear , and an additional systematic error has been included in the flux estimation to reflect this . [ effsurf ] shows the effective surface area of celeste for gamma rays as a function of the initial photon energy at the raw trigger level , after the software trigger , and after the analysis cuts ( @xmath70 and @xmath114 ) , using the kaskade monte carlo . the detector simulation was for 11 km single pointing towards the crab at transit , with a trigger threshold of @xmath51 per heliostat . the curve after cuts can be parametrized as @xmath115 @xmath116 , with @xmath117 in gev . the area is an order of magnitude smaller than for an imaging telescope because convergent viewing restricts the impact parameter at which a gamma shower will be seen by enough heliostats to trigger the experiment . a valuable partial test of our effective area calculations can be made by using those showers which trigger both cat and celeste . approximately 20% of the celeste events , corresponding to around 30% of cat events are common and can be identified as such , with a probability better than 99.9% , by their arrival time measured with gps clocks by the two experiments . during this observing season we have collected 13 hours of common data on the crab . the standard cat analysis @xcite when applied to the full data set results in an excess of 1268 gamma events over a background of 3131 hadrons . the same analysis applied only to the common events produces an excess of 418 gammas over a background of 526 hadrons . from these numbers we see that imposing a celeste trigger increases the signal to noise ratio in the cat data sample by a factor of two , although it does not improve the significance of the result as the data sample is smaller . cat measures the shower impact parameter with better resolution than celeste @xcite . [ commonsurf ] shows this reconstructed impact parameter for simulated data , and for the excess events from the common crab data set . the data are well reproduced by the simulations in terms of both the shape of the distributions and in the predicted fraction of common events . this gives us confidence that the effective surface area for celeste , at least in the energy region of the common cat - celeste events , is understood . the effective area varies with the source position in the sky , as indicated in fig . [ efficiency ] . knowing the azimuth angles under which the crab was observed , we have used the polynomial fit in fig . [ efficiency ] to correct our measured gamma ray rate . in addition , for each run we correct for our acquisition dead time of @xmath118 which is measured during the observations . there is no evidence for time variability in the measured flux of high energy emission from the crab nebula for gamma ray energies above and below the energy range of celeste @xcite . [ sigmacrab ] shows the rate calculated for each of the 41 on - off pairs , after accounting for the varying gamma ray detection efficiency . a constant fit to these points has a positive mean and a @xmath83 value of 30.7 for 40 degrees of freedom , as would be expected for a steady signal , which gives us further confidence in the stability of the celeste analysis . we obtain the corrected measurement of @xmath119 ( statistical uncertainty only ) . knowing the effective surface area as a function of energy we can calculate the expected response of celeste to a typical spectrum of gamma rays . fig . [ ethresh ] shows the energy distributions of simulated events for an input @xmath120 differential gamma ray spectrum , close to the spectral shape for high energy emission from the crab in the celeste energy range @xcite . a useful definition of the energy threshold for atmospheric erenkov detectors is the energy at which the differential gamma ray rate is maximum for a typical source . according to this definition , the energy threshold for celeste at the raw trigger level for a source at the position of the crab at transit is @xmath121 . from the zenith , which is the same position as for the crab at transit . ] the gamma rays have been simulated with the same distribution of azimuth and zenith angles as the 11 km crab observations , increasing the energy threshold to @xmath122 at the raw trigger stage . as mentioned in the previous section , a software trigger is applied during the analysis to correct for night sky background effects in the data . this increases the energy threshold to a level of @xmath123 . further analysis cuts ( @xmath70 and @xmath114 ) reduce the number of gamma rays observed , but do not increase the energy threshold . the systematic errors on our measurement have two different origins . the uncertainty on the energy scale is due principally to errors in the conversion of the measured signal to a flux of erenkov photons , which is a combination of many factors ( photon losses through the optical system , pmt quantum efficiencies , electronic calibration errors ) . we bracket the overall uncertainty arising from the combination of these elements as follows . first , during the celeste prototype studies we measured the night sky background in our wavelength range at themis to be @xmath124 @xcite , in the direction of the crab at transit ( 20 degrees south of zenith , towards the populated valley below the site ) . from this we expect 1 photoelectron per nanosecond per phototube , corresponding to anode currents of @xmath125a , close to the observed range around @xmath126a . studying the fadc pedestal widths used in the padding software also yields values of @xmath127 photoelectron per nanosecond per phototube . we further compare currents measured while aligning the heliostats using star scans with predictions from the optical simulation : the measured values are typically 20 % less than expected . results of studies of the atmospheric extinction using ccd photometry and a lidar will be reported in future work but are not included in the present study . finally , the observed cosmic ray trigger rate is 30% higher than predicted by the monte carlo . from these considerations we believe the energy scale uncertainty to be less than @xmath128% . the corresponding acceptance curves are @xmath129 , leading to an uncertainty on our threshold . the input spectrum assumed in determining the absolute flux ( see discussion below ) has little effect on the threshold . for this analysis then , we quote an energy threshold of @xmath3 . the other principal source of systematic error is the uncertainty on our efficiency for detecting gamma rays . as mentioned earlier , there is an energy independent discrepancy of 25% in the effective surface area as calculated using two different shower generation monte carlos . both monte carlo s use the u.s . standard atmosphere . @xcite recently explored the effects of different atmospheric profiles on the cherenkov light yield , finding a @xmath130 variation for 100 gev gamma rays for midlatitude summer and winter atmospheres , neglecting aerosol variations . we conclude that the uncertainty in the cherenkov light yield for gamma rays in our energy range is 25% . we also assign a systematic error of 10% to the cut efficiencies deduced from the simulations ( table 3 ) . at present the event - by - event energy determination in celeste is poor . to compare our rate measurement with models and with results from other experiments requires convoluting our detector acceptance , @xmath131 ( see fig . [ effsurf ] ) , with an assumed source spectrum . the simplest hypothesis is that of a power law differential flux , @xmath132 . in a @xmath133 representation ( or , equivalently , @xmath134 ) the celeste energy range corresponds to the top of the parabola - like spectral shape attributed to inverse compton production of gamma rays in the nebula @xcite , and @xmath135 is a good approximation . it yields an integral result for celeste of @xmath136 . stacee used this approach , with @xmath137 @xcite . a more realistic hypothesis recognizes that the spectrum deviates from a pure power law . we use a parabola - like spectral shape of the form @xmath138 above @xmath139 we use the values of @xmath140 , and @xmath141 taken from cat @xcite . below @xmath139 we let @xmath141 be a free parameter , but require continuity at 500 gev . we determine @xmath142 such that the convolution with @xmath131 yields our measured rate and thus obtain an integral flux of @xmath143 . we further vary @xmath141 and determine that the range of ( @xmath144 to @xmath145 is consistent with the rate and acceptance uncertainties , including that of the energy scale . repeating the process using the whipple @xcite , or hegra @xcite spectra gives very nearly the same results . we thus determine our flux to be @xmath5 we applied this procedure above 190 gev to compare with stacee and obtain @xmath146 in agreement with their result of @xmath147 . to represent this integral measurement on a differential plot , we use @xmath142 to calculate @xmath134 at our energy threshold . this is shown as a triangle in fig . [ spectrum ] . as above , the error bar is obtained by finding the range of @xmath141 that accomodates the uncertainties on our measurement . the value shown is @xmath148 . [ spectrum ] also shows the imager measurements , as well as the envelope defined by varying the imager fit parameters @xmath140 , and @xmath141 by one standard deviation around their central values . our measurement favors the lower part of the range allowed by the imagers and is compatible with the results from egret . one of the primary goals of the celeste experiment is to investigate the periodic emission from gamma ray pulsars in the cutoff region below 100 . the celeste data include the arrival time of each event measured to a precision of @xmath149s using a time - frequency processor slaved to a global positioning system ( gps ) clock which provides synchronisation every second . this timing information has been used to search for evidence of periodicity in our crab data . in order to verify our periodic analysis procedure we have made observations of the optical emission from the crab pulsar using the celeste heliostats . given the optical flux from the crab pulsar @xcite , we expect a flux of @xmath150 over a night sky background of typically @xmath151 . in standard operation , the pmt anode currents for all forty heliostats are converted to a buffered voltage which is digitized and stored with the data stream . the current - to - voltage conversion integrates the signal over @xmath152 . for the optical pulsar study , three of these current outputs were ac - coupled , in order to subtract the steady component due to the night sky background and the nebula , and sent to a 16-bit adc card readout by a pc at a frequency of @xmath153 . a gps time reference was obtained for the optical data by sending the same pulse every @xmath154 as a trigger to celeste and as data to the adc card . we then tracked the crab pulsar and recorded the current fluctuations during 30 minutes . the synchronised times were converted to the solar system barycenter frame using the jpl de200 ephemeris @xcite . [ optical ] shows the phase histogram using the frequency ephemerides obtained by @xcite . the double - peaked signal from the pulsar is clearly visible . we use the same code to calculate the phase of the air shower events . from the egret pulsar detections , the tev upper limits , and the model predictions it is clear that the search for pulsed gamma ray emission requires as low an energy threshold as possible . to date we have no evidence of a pulsed signal . we present the pulsar search using the same analysis as used to measure the steady emission flux , that is , applying the software trigger , @xmath70 and @xmath114 . although this raises our energy threshold , we take this cautious approach because the efficiency is better understood . the light curves of both the on source data and the off source data remaining after cuts for the @xmath76 crab data set of table [ tab : obslog ] are shown in fig . [ phasograms ] . table [ tab : pulsed ] summarises the contents of the plots as well as the results of the h - test @xcite . the distributions are statistically flat . in order to calculate upper limits for the pulsed emission we assume that the pulse profile is the same as that seen by egret at lower energies with emission concentrated in a main pulse in the phase range 0.94 - 0.04 and a secondary pulse in the range 0.32 - 0.43 @xcite . we use the method of helene to determine an upper limit of @xmath155 pulsed events at the 99% confidence level @xcite . this corresponds to 12% of the observed steady signal . we include the detector acceptance as follows . we take the double power law fit of the total spectrum measured by egret @xcite , and attenuate the sum with an exponential cutoff , @xmath156e^{-e / e_0},\ ] ] in units of @xmath157 photons @xmath158s@xmath159mev@xmath159 . we convolute this spectrum with the acceptance after cuts shown in fig . [ ethresh ] , and find that for @xmath160 gev we would expect @xmath161 events . including the 30% uncertainty in the energy determination degrades this value to @xmath162 gev . [ pulsedspec ] shows @xmath163 , where we have placed a point at the energy threshold obtained for our steady signal to guide the eye . we note that our limit is not directly comparable to that obtained by the stacee @xcite group since they used the larger acceptance corresponding to their measured steady spectrum for comparison with the prediction of tev pulsed emission . our hypothesis of an attenuated egret spectrum restricts our acceptance to the low energy range of fig.[ethresh ] , yet our upper limit still provides the most constraining measurement so far on the position of the cutoff point . in the future , improved trigger electronics and observing and analysis strategies optimized for pulsar observations should allow us to increase our acceptance at low energy . the radiation from the crab nebula is dominated by non - thermal emission which is believed to be generated by synchrotron radiation from highly relativistic electrons with energies up to @xmath164 . the electrons are accelerated at the shock front where a relativistic wind of charged particles emerging from the pulsar meets the surrounding nebula @xcite . recent high resolution x - ray observations by the chandra observatory have shown an inner ring of x - ray emission which may correspond to the position of this shock @xcite . @xcite and @xcite have described the electrons in terms of two populations of different energies . the first , generated over the whole lifetime of the nebula and covering energies up to @xmath22 , produces synchrotron radiation from radio wavelengths to the far infra - red while the second , more recently accelerated population , with energies @xmath165 produces synchrotron emission from the infra - red up to @xmath1131 gev . it was first suggested by @xcite ( also @xcite and @xcite ) that the synchrotron self - compton mechanism could give rise to radiation from the crab above @xmath15 . this process , in which inverse compton scattering of the synchrotron photons by the relativistic electrons boosts the photons up to much higher energies , has been modelled by various workers , most recently @xcite , @xcite and @xcite . while the synchrotron photons are the most important component , photons due to infra - red emission from dust and to the microwave background will also be upscattered and contribute significantly to the high energy emission . [ spectrum ] shows the result of this work along with the measurements from egret and three atmospheric cherenkov imaging telescopes . the shape of the inverse compton spectrum is relatively insensitive to the model parameters , but the absolute flux depends strongly upon the magnetic field strength in the emitting region , which in turn depends upon @xmath166 , the ratio of the magnetic field strength to particle energy density in the pulsar wind . @xcite have proposed that an additional component due to bremsstrahlung radiation from the relativistic electrons in dense filaments of nebular gas may provide an increased flux in the @xmath167 range , which could account for a possible discrepancy between the models and the egret points around @xmath15 . the uncertainties are still large but the celeste measurement does not seem to point towards such an effect . the calibration of such a complex instrument as celeste is a large project in itself . our measurement errors are currently dominated by systematic effects which should decrease as this work proceeds , the most important being to improve our determination of the energy scale . the crab pulsar is a source of 33 pulsed radiation from radio wavelengths to gev gamma ray energies . periodic emission is observed by egret up to energies of @xmath12 @xcite . despite early claims @xcite , no pulsed emission has been detected by the present generation of ground based atmospheric erenkov experiments . the previous best upper limits are at @xmath8 , from the whipple @xcite and cat @xcite groups , and the limit at @xmath168 by stacee @xcite . two general classes of models have been proposed to describe the pulsed gamma ray emission from the high energy pulsars observed by egret . in the polar cap models @xcite electrons accelerated from the neutron star surface at the magnetic pole emit by curvature radiation or magnetic inverse compton scattering , triggering photon - pair cascades in the pulsar magnetosphere from which the observed radiation emerges . outer gap @xcite models place the emission region in the outer magnetosphere where electrons are accelerated across charge depleted regions near the light cylinder . both models predict a cutoff in the pulsed emission below @xmath2 , and the exact position for the cutoff can be used to discriminate between them . @xcite treat the electrodynamics of the outer gap from first principles . the free parameter in their model is the current density at the gap boundaries , which in turn depends on the distance of the gap from the light cylinder . our upper limit excludes the hypothesis that the current density vanishes at the gap surface , since the model predicts a gamma ray flux extending to 60 gev in that case . [ pulsedspec ] includes the prediction of their model for the case of a small current density at the inner boundary , and a null current at the outer boundary . [ pulsedspec ] also shows the predictions of a polar cap model , along with the egret measurements and higher energy upper limits . the celeste upper limit constrains the high energy emission more strongly than the previous whipple measurement , but increased sensitivity at lower energy is still needed to favor a particular model for the emission processes . we have presented the first detection by the atmospheric erenkov technique of a gamma ray source , the crab nebula , at energies below 100 using the celeste experiment . the measured flux is compatible with most emission models . no periodic signal has been detected but our upper limit allows us to constrain further the cutoff point for emission from the pulsar . as our uncertainties decrease we will be able to determine the energy range in which the nebula and pulsar contributions are comparable . the data reported on in this paper were collected during the first observation season with a fully operational 40 heliostat array . in single pointing mode we now have a sensitivity to the crab of @xmath169 . a smaller dataset obtained with heliostat double pointing appears to confirm monte carlo predictions of improved sensitivity , yielding @xmath111 , although more data is required for confirmation . it seems likely that _ a posteriori _ optimization of our hadron rejection cuts , along with the development of new analysis techniques , will enable us to improve our sensitivity in the future . celeste is currently being upgraded by the addition of another 13 heliostats , bringing the total to 53 , allowing greater flexibility in pointing strategies . funding was provided by the in2p3 of the french cnrs and by the grant agency of the czech republic . we gratefully acknowledge the support of the regional council of languedoc - roussillon and of electricit de france . we thank dr . b. lott , dr . r. lessard , dr . f. aharonian , dr a. harding and dr k. hirotani for useful discussions . f. piron provided the cat data used herein . aharonian , f.a . & atoyan , a.m. , 1995 , astropart phys , 3 , 275 aharonian , f. , et al . , 1999 , a&a , 346 , 913 aharonian , f.a . , et al . , 2000 , apj , 539 , 317 arqueros , f. , et al . , 2001 , astropart . submitted atoyan , a.m. & aharonian , f.a . , 1996 , mnras , 278 , 525 baillon , p. , et al . , 1993 , phys . , 1 , 341 barrau , a. , et al . , 1998 , instrum . meth . , a416 , 278 bernlhr , k. 2000 , astropart . bhat , p.n . , et al . , 1986 , nature , 319 , 127 le bohec , s. , et al . , 1998 , nucl methods a , 416 , 425 bradbury , s.m . , et al . , 1999 , in the 26th international cosmic ray conference , ed . d. kieda , m. salomon , b. dingus ( salt lake city ) , 5 , 280 cawley , m.f . , 1993 , in towards a major atmospheric erenkov detector - ii , ed . lamb ( calgary ) , p176 cheng , k.s . , ho , c. & ruderman , m. , 1986 , apj , 300 , 500 covault , c. et al . in the 27th international cosmic ray conference , ( hamburg , 2001 ) in press . daugherty , j.k . & harding , a.k . , 1982 , apj , 252 , 357 daugherty , j.k . & harding , a.k . , 1996 , apj , 458 , 278 dermer , c. d. & schlickeiser , r. , 1993 , apj , 416 , 458 dowthwaite , j.c . , et al . , 1984 , , 286 , l35 fierro , j.m .. et al . , 1998 , apj , 494 , 734 ghisellini , g. & maraschi , l. , 1996 , in asp conf 110 , blazar continuum variability , ed . h. r. miller , j. r. webb , & j. c. noble ( san francisco : asp ) , 436 gibson , a.i . , et al . , 1982 , nature , 296 , 833 giebels , b. , et al . , 1998 , nucl . instr . meth . a 412 , 329 , with additional details in the doctoral thesis available at goret , p. , et al . , 1993 , a&a , 270 , 401 gould , r.j . , 1965 , phys.rev.lett . , 15 , 577 grindlay , j.e . & hoffman , j.a . , 1971 , ap . letters , 8 , l209 heck , d. , et al . , 1998 , report fzkas 6019 , forschungszentrum karlsruhe helene , o. , 1983 , nucl . methods , 212 , 319 herault , n. , 2000 , reconstruction des paramtres des gerbes de gamma et contribution lanalyse des donnes dans lexprience celeste , thse de doctorat , universit louis pasteur de strasbourg , http://www.lal.in2p3.fr/presentation/bibliotheque/publications/theses00.html hillas , a.m. et al . , 1998 , apj , 503 , 744 hirotani , k. & shibata , s. , 2000 , preprint ( astro - ph/0011488 ) hirotani , k. & shibata , s. , 2001 , preprint ( astro - ph/0101498 ) holder , j. , et al . , 2000 in heidelberg gamma ray symposium , ed . aharonian & h. voelk , 635 de jager , o.c . & harding , a.k . , 1992 , apj , 396 , 161 de jager , o.c . , 1994 , apj , 436 , 239 de jager , o.c . , et al . , 1996 , apj , 457 , 253 kennel , c.f . & coroniti , f.v . , 1984 , apj , 283 , 694 and 710 kertzmann , m.p . & sembroski , 1994 , nucl . methods a , 343 , 629 kohnle , a. , et al . , 1999 , in the 26th international cosmic ray conference , ed . d. kieda , m. salomon & b. dingus ( salt lake city ) , 5 , 239 lessard , r.w . , et al . , 2000 , apj , 531 , 942 lyne , a.g . , pritchard , r.s . & roberts , m. at http://www.jb.man/ac.uk/ pulsar / crab.html marcowith , a. , henri , g. & pelletier , g. 1995 , mnras , 277 , 681 masterson , c. , et al . , in proc . gamma-2001 conference ( baltimore ) , ( in press ) martinez , m. , et al . , 1999 , in the 26th international cosmic ray conference , ed . d. kieda , m. salomon , b. dingus ( salt lake city ) , 5 , 219 musquere , a. , et al . , 1999 , in the 26th international cosmic ray conference , d. kieda , m. salomon , b. dingus ( salt lake city ) , 3 , 460 de naurois , m. , 2000 , reconversion dune centrale solaire pour lastronomie @xmath170 . premire observation de la nbuleuse du crabe et du blazar markarian 421 entre @xmath171 et @xmath172 , thse de doctorat , universit paris vi , http://polywww.in2p3.fr/celeste/these/these.html oser , s. , et al . , 2001 , apj , 547 , 949 punch , m. , et al . , 1992 , nature 358 , 477 . percival , j.w . et al . , 1993 , apj , 407 , 276 piron , f. , 2000 , etude des proprits spectrales et de la variabilit de lmission gamma suprieure 250 gev des noyaux actifs de galaxies de type blazar observs dans le cadre de lexprience cat , thse de doctorat , universit paris xi , http://lpnp90.in2p3.fr/ cat / papers / these_piron.ps.gz ramanamurthy , p.v . , et al . , 1995 , apj , 450 , 791 rees , m.j . & gunn , j.e . , 1974 , mnras , 167 , 1 reposeur , t. et al . , to be submitted to nucl . ( 2001 ) rieke , g.h . & weekes , t.c . , 1969 , apj , 155 , 429 romani , r.w . & yadigaroglu , i - a . , 1995 , apj , 438 , 314 romani , r.w . , 1996 , apj , 470 , 469 smith , d.a . , et al . , celeste experimental proposal , http://polywww.in2p3.fr/celeste/public/cxp.ps.gz standish , e.m . , jr . , 1982 , a&a , 114 , 297 sturner , s.j . , dermer , c. d. & michel , f.c . , 1995 , apj , 445 , 736 vacanti , g. et al . , 1991 , apj , 377 , 467 weekes , t. c. et al . , 1989 , apj , 342 , 379 weisskopf , m.c . , et al . , 2000 , apj 536 , l81 . ccccccc raw trig . & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & + software trig . & @xmath178 & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath183 + @xmath70 & @xmath184 & @xmath185 & @xmath186 & @xmath187 & @xmath182 & @xmath188 + @xmath114 & @xmath189 & @xmath190 & @xmath191 & @xmath192 & @xmath193 & @xmath194 + @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 & @xmath201 + lrrrr software trigger ( s.t . ) & & @xmath202 & & @xmath203 + @xmath70 & & @xmath204 & @xmath205 & @xmath206 + @xmath114 & & @xmath207 & @xmath208 & @xmath209 + @xmath195 & & @xmath210 & @xmath211 & @xmath212 + all cuts & & @xmath213 & & @xmath214 + all cuts , after s.t . & & @xmath215 & @xmath216 & @xmath217 + @xmath70 , @xmath114 , after s.t . & & @xmath218 & @xmath219 & @xmath220 + ccccccc raw trig . & @xmath221 & @xmath222 & @xmath223 & @xmath224 & @xmath225 & + software trig . & @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 + @xmath70 & @xmath232 & @xmath233 & @xmath234 & @xmath229 & @xmath230 & @xmath235 + @xmath114 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 + @xmath195 & @xmath242 & @xmath243 & @xmath244 & @xmath241 & @xmath245 & @xmath246 + cc total number of on events & 67022 + total number of off events & 64295 + + pulsed phase fraction & 0.21 + number of on events in expected phase windows & 14062 + number of on events outside expected phase windows & 52960 + significance for the pulsed phase domain & @xmath247 + + value of the h - test for on source events & 2.60 + value of the h - test for off source events & 1.17 + upper limit at the 99% confidence level for h - test & @xmath248 + + upper limit at the 99% confidence level using helene method & @xmath249 +

as the large hadron collider ( lhc ) continues to successfully probe the nature of electroweak symmetry breaking ( a recent breakthrough being the discovery of a higgs candidate @xcite ) , a solution to the gauge hierarchy problem , if it exists , has evaded our efforts so far . an elegant candidate for such a solution and one which addresses several other open issues as well , _ e.g. _ dark matter and gauge coupling unification , is supersymmetry ( susy ) . because of its theoretical appeal , it is important to understand all the guises that susy may adopt in order to recognize it if it is produced at the lhc . an important open issue in susy , which has strong ramifications for its lhc phenomenology , as well as its cosmology , is the gauge invariance of lepton and baryon number violating interactions . aside from introducing many new unknown parameters , these interactions also lead to rapid proton decay . the most common solution to address this issue is the imposing of a discrete symmetry , r - parity , defined as @xmath9 ( see @xcite for a review ) . this forbids all tree - level lepton and baryon number violating terms and also causes the lightest supersymmetric particle ( lsp ) to be stable and therefore neutral . this lsp can then play the role of dark matter and would manifest itself at colliders as missing energy . however , proton decay requires the elimination of only the lepton number or baryon number violating terms and from a theoretical perspective , r - parity can be _ ad hoc_. furthermore , an open - mindedness to possible signals at the lhc should push us to consider alternatives . finally , when r - parity is violated the stringent constraints on superpartner masses , derived from negative collider searches for missing energy events at the tevatron and the lhc @xcite , can be relaxed . a systematic study of all possible r - parity violating terms and their effects on phenomenology is an arduous task and furthermore , one would like a mechanism for understanding why proton decay is significantly suppressed . a well - motivated solution to both of these issues are models which can predict the fate of r - parity . a natural framework for this endeavor is in the context of @xmath10 symmetries , see @xcite for early examples can be also used to construct models where the rpv couplings , arising from effective operators , are intrinsically small @xcite . ] . while some @xmath11 models predict r - parity conservation @xcite , the most minimal ones ( in terms of particle content ) require r - parity violation ( rpv ) @xcite . even some non - minimal models prefer rpv from considerations of the renormalization group evolution of the soft masses @xcite . these and other models of spontaneous r - parity violation , such as @xcite , have the common feature that they can be described , in an effective field theory way , by bilinear r - parity violation ( brpv ) : the only r - parity violating terms are the mixings between the lepton doublets and up - type higgs doublet in the superpotential . this makes brpv a powerful tool for studying possible signatures of spontaneous rpv . furthermore , proton decay is highly suppressed and it is important to mention that a gravitino lsp can be a dark matter candidate in such models @xcite . once r - parity is broken the lsp is no longer stable and therefore astrophysical constraints on its nature do not apply @xcite . accordingly , from a purely phenomenological point of view any superpartner can be the lsp , and studies of the different possibilities in brpv models ( and of sleptons and sneutrinos in general models @xcite ) and their relation to neutrino masses and mixings have been carried out in e.g. @xcite . in this paper we extend upon previous results by considering sneutrino lsp decays in brpv into `` non - conventional '' modes : pairs of heavy standard model ( sm ) states ( @xmath12 , @xmath13 , @xmath14 , @xmath15 ) , invisible modes ( @xmath16 ) and different - flavor charged lepton final states ( @xmath17 , @xmath18 ) . such states , to our knowledge , had not received much attention before despite the fact that they can dominate sneutrino decays and can yield unique and unanticipated susy signals . specifically , neutrino masses force rpv to be small so its only effect is on the decay of the lsp . therefore , for a sneutrino lsp , every susy event will eventually decay into two sneutrinos which could then decay into one of the aforementioned states . as we will show , this typically translates into into different - flavor charged leptons or heavy sm modes , depending on the sneutrino generation . the main goal of this paper is to study the sneutrino decays into these `` non - conventional '' modes and to show that they can dominate over the traditionally considered @xmath19 final state . we find that the latter dominate roughly when the brpv term , @xmath20 is smaller than the vacuum expectation value ( vev ) of the sneutrino , @xmath21 , for a given flavor of sneutrino , @xmath22 . in the case when brpv is the only source of neutrino masses , this possibility can only hold for one sneutrino flavor , however , when this assumption is relaxed , it can hold true for all three generations . therefore , if two or more generations of sneutrinos decay via rpv , it might be possible to rule out brpv as the sole generator of neutrino masses . the rest of the paper is organized as follows . in section [ sec : generalities ] we discuss the generalities of the brpv model , in particular those related with the neutral scalar sector . in section [ sec : sneutrino - mix - analyr ] we derive formulas for brpv induced mixings . in section [ sec : sneutrino - decays - analyticalr ] we write the relevant couplings for brpv sneutrino decays , give analytical formulas for the different partial decay widths , analyze the constraints on parameter space enforced by neutrino data and present our results . in section [ sec : conclusions ] we summarize and present our conclusions . in what follows we will briefly describe the main features of the bilinear r - parity breaking model , in particular those related with the neutral scalar sector . we shall closely follow the notation used in @xcite and assume that all parameters are real . throughout the text matrices will be denoted in bold - face . in addition to the mssm r - parity conserving superpotential ( where we have suppressed relevant indices ) : @xmath23 the bilinear r - parity breaking model also contains the following terms : @xmath24 where @xmath25 is the @xmath26 completely antisymmetric tensor , @xmath27 runs over the sm fermion generations and @xmath20 is the r - parity and lepton number breaking bilinear parameter with units of mass . consistency then requires a new set of soft susy breaking terms in the scalar potential , namely @xmath28 neglecting soft flavor mixing , the scalar potential relevant for neutral scalars is @xmath29 with @xmath30 and @xmath31 . electroweak symmetry is broken once the higgs and slepton acquire a vev , @xmath32 and @xmath33 , with @xmath34 gev and @xmath35 . the doublets are parameterized as @xmath36 with the neutral components given by @xmath37 here we introduce the notation @xmath38 to differentiate the cp - even sneutrinos from the cp - odd . in the basis @xmath39 the linear part of the neutral scalar potential can be written as @xmath40 where the @xmath41 s are the so - called tadpoles . at tree - level these are ( see eqs . ( [ eq : full - ssb - scalar - potential ] ) and ( [ eq : vlinear ] ) ) @xmath42 where @xmath43 and @xmath44 . the minimization of the potential , @xmath45 , requires the tadpoles to vanish . thus , the vevs can be determined from the system of equations in ( [ eq : tree - lvel - tadpoles ] ) by imposing @xmath46 ( @xmath47 ) . in particular , considering only leading order brpv terms , the sneutrino vevs can be written as @xmath48 where @xmath49 ( the tree level sneutrino mass ) and @xmath50 . without conserved r - parity , there are no quantum numbers to distinguish the leptons and sleptons from the gauginos and higgsinos and higgs bosons respectively . therefore , in brpv several mixings between supersymmetric and non supersymmetric particles exist : ( @xmath22 ) neutralinos mix with neutrinos , ( @xmath51 ) charginos mix with charged leptons , ( @xmath52 ) higgs bosons mix with the sneutrinos and ( @xmath53 ) charged higss bosons mix with charged sleptons . these mixings are important for calculating lsp decays , especially mixings of type ( @xmath22 ) as they allow to fix via experimental neutrino data the size of the brpv parameters . since mixings of type ( @xmath53 ) are not of interest for sneutrino decays , in what follows we will only discuss analytical approximations for mixings of type ( @xmath22)-(@xmath52 ) . in the basis @xmath54 the neutral fermion mass matrix can be written as @xmath55 with @xmath56 where @xmath57 is the neutralino mass matrix : @xmath58 and @xmath59 and @xmath60 are the soft masses for the bino and wino respectively . @xmath61 is the @xmath62 neutralino - neutrino mixing matrix given by @xmath63 in the weyl mass eigenstate basis , defined as . ] @xmath64 the mass matrix becomes @xmath65 due to the smallness of the brpv parameters , at order @xmath20 , @xmath66 can be block diagonalized by decomposing the diagonalizing matrix @xmath67 as follows @xcite : @xmath68 the matrix @xmath69 block diagonalizes @xmath66 to the form @xmath70 , where @xmath71 is the tree - level light neutrino mass matrix . the mixing parameters @xmath72 can thus be determined to be @xmath73 where @xmath74 ( with @xmath75 , @xmath76 being the weak mixing angle ) and @xmath77 finally the block diagonal mixing matrices @xmath78 and @xmath79 in ( [ eq : n - matrix ] ) diagonlize the neutralino and neutrino effective mass matrix , which reads @xmath80 since this matrix has two vanishing eigenvalues it can be diagonalized by only two rotation matrices , namely @xmath81 with @xmath82 for sneutrino decays the relevant part of the @xmath83 mixing turns out to be the @xmath84 block , that from eqs . ( [ eq : neutralino - neu - mixing - parameters ] ) and ( [ eq : leptonic - mix - mat ] ) can be written as @xcite @xmath85 where @xmath86 and @xmath87 with @xmath88 . the coefficients @xmath89 are given by @xmath90 with @xmath91 . taking into account eqs . ( [ eq : leptonic - mix - mat ] ) and ( [ eq : mixing - angles ] ) , explicitly @xmath92 ( @xmath93 ) and @xmath94 are given by @xmath95 in the bases @xmath96 the chargino charged lepton mass matrix is determined by the following lagrangian @xmath97 where @xmath98 and in the basis in which the charged lepton mass matrix is diagonal @xmath99 can be written as @xmath100 the block diagonal matrices correspond to the mssm chargino and charged lepton mass matrices whereas the off - diagonal mass matrices read @xmath101 defining the weyl mass eigenstates as . ] @xmath102 the diagonal mass matrix @xmath103 is obtained through the biunitary transformation : @xmath104 approximate analytical expressions for the mixing matrices @xmath105 have been discussed in @xcite . here we describe the method for completeness . the off - diagonal block matrix @xmath106 , being proportional to the charged lepton yukawa couplings , can be neglected , and due to the smallness of the brpv parameters the mixing matrices @xmath105 can be written according to @xmath107 where , in first approximation in the brpv parameters , the matrices @xmath108 block - diagonalize the mass matrix , namely @xmath109 and the matrices @xmath110 diagonalize in turn the chargino mass matrix , with the rotation angles given by @xmath111 from equation ( [ eq : block - diag - mc ] ) the matrices @xmath112 are found to be @xmath113 which implies @xmath114 is suppressed with respect @xmath115 by a factor @xmath116 and thus can be neglected ( @xmath117 ) . explicitly @xmath115 can be written in terms of the brpv parameters and the coefficients entering in the chargino mass matrix : @xmath118 in the basis @xmath119 the mass matrix of the cp - even neutral scalars @xmath120 is determined by the following quadratic terms @xmath121 in what follows we will discuss approximate analytical formulas for the @xmath122 mixing . to our knowledge , this is the first time that they are explicitly given . the entries of the mass matrix in ( [ eq : cp - even - mm ] ) involve the parameters @xmath123 , @xmath124 , the soft susy breaking coefficients @xmath125 and @xmath126 , and the r - parity breaking parameters @xmath20 , @xmath21 and @xmath127 . we use the minimization conditions of the scalar potential ( @xmath128 ) to remove the parameters @xmath127 . in doing so the matrix @xmath129 can be written in terms of the cp - odd neutral scalar mass @xmath130 , the sneutrino masses @xmath131 , @xmath132 and the brpv parameters @xmath20 and @xmath21 , namely @xmath133 where the following relations have been used @xmath134 note that phenomenologically consistency requires the inclusion of the one - loop correction in the @xmath135 entry , which in the limit of no stop mixing is @xmath136 see @xcite for a more complete expression . in the absence of brpv , @xmath129 corresponds to the neutral cp - even higgs mass matrix of the r - parity conserving mssm , which can be diagonalized via the rotation matrix @xmath137 the elements of the @xmath138 right - lower block sneutrino mass matrix in ( [ eq : cp - even - mm ] ) are given by @xmath139 finally , for the @xmath122 mixing @xmath140 block we have @xmath141 in the mass eigenstate basis defined as @xmath142 the lagrangian in ( [ eq : cp - even - mm ] ) becomes @xmath143 where @xmath144 is diagonal . assuming real parameters @xmath145 can be parameterized as @xmath146 where the @xmath147 are @xmath148 rotation matrices . if the @xmath149 mixing is small as expected due to the smallness of the brpv parameters required by neutrino data a perturbative diagonalization of the mass matrix in ( [ eq : cp - even - mm ] ) can be done . by neglecting the brpv parameters in @xmath129 and @xmath150 the rotation matrix reduces to @xmath151 when acting on @xmath152 the matrix @xmath153 diagonalizes the @xmath154 block @xmath129 according to @xmath155 ( where @xmath156 and @xmath157 are the light and heavy cp - even higgs bosons ) and modifies the brpv mixing matrix @xmath158 : @xmath159 for the mixing angle we have @xmath160 . the matrices @xmath161 eliminate the first row entries in ( [ eq : brpv - mm - modified ] ) and up to @xmath162 leave the block - diagonal matrices @xmath163 and @xmath164 diagonal . the rotation angles are given by @xmath165 the matrices @xmath166 instead eliminate the second row elements in ( [ eq : brpv - mm - modified ] ) leaving again , up to order @xmath167 , the block - diagonal matrices diagonal . the rotation angles in this case read @xmath168 with these results at hand and neglecting terms of @xmath169 the full rotation matrix in ( [ eq : rot - mat - approx - s0 ] ) can be written as @xmath170 where @xmath171 and @xmath172 account for the sneutrino and cp - even neutral higgs components of the mass eigenstates , and read @xmath173 with the results of section [ sec : sneutrino - mix - analyr ] we are now in a position to discuss approximate formulas for sneutrino decays . from now on we will consider only cp - even sneutrino decays , @xmath174 , and so will drop the superscript @xmath175 . possible tree - level two - body sneutrino final states include fermionic modes @xmath176 , @xmath177 and @xmath178 ; electroweak gauge bosons modes @xmath179 and @xmath180 and higgs bosons modes @xmath181 , @xmath182 , @xmath183 and @xmath184 . the goal of this paper and section is not a full study of the parameter space and all decays but rather to show that these heavy states can dominate the traditionally considered decay to @xmath19 and to identify the relevant parameter space for this dominance . taking into account our approximate results for the chargino - right - handed lepton mixing , @xmath185 , mixing with right - handed leptons is zero ( @xmath186 , see eq . ( [ eq : xir - xil ] ) ) , the full lagrangian for sneutrino decays into charged leptons is reduced to @xcite @xmath187 where the mixing matrices are given by eqs . ( [ eq : xil - explicitly ] ) and ( [ eq : full - rot - s0-final ] ) . due to the smallness of the brpv induced mixing we take @xmath188 . note that while the first term in ( [ eq : charged - lepton - modes - lag ] ) necessarily leads to two same - flavor opposite - sign charged lepton final states the second term can lead to different - flavor signatures . the corresponding feynman diagrams for these processes are depicted in figure [ fig : sneutrino - decays](@xmath189,@xmath190 ) . being proportional to sm charged lepton yukawa couplings these decays are dominated by final states involving @xmath191 s . sneutrino decays are induced by sneutrino - higgs mixing while in figure @xmath190 by chargino - charged lepton or neutralino - neutrino mixing , depending on whether the final states involves charged lepton or neutrinos . the open circles with a cross inside indicate a brpv mixing insertion.__,width=377,height=132 ] for sneutrino decays into neutrinos ( invisible decays @xmath192 ) , the relevant interactions are given by @xmath193 where the coupling reads @xmath194 where @xmath195 and @xmath196 the elements of the neutral fermion mixing matrix correspond to the entries of the matrix given in eq . ( [ eq : n - matrix ] ) . these interactions induce invisible sneutrino decays as the ones shown in figure [ fig : sneutrino - decays]@xmath190 . in the up and down quark mass eigenstate basis the sneutrino - quark - quark interactions are dictated by @xmath197 as for the charged lepton final states , these decays are controlled by sm quark yukawa couplings and thus are dominated by @xmath198 and @xmath15 , the last one if kinematically allowed . the lagrangian for gauge boson final states is given by @xmath199 with @xmath200 and @xmath201 . for higgs final states we write the lagrangian involving @xmath156 and @xmath157 and @xmath202 decays and so do not present the lagrangians that govern these interactions . ] : @xmath203 where the fully symmetric coupling @xmath204 is given by @xmath205\,,\ ] ] with @xmath206 and @xmath207 the interactions in ( [ eq : lagrangian - scalars ] ) for @xmath208 lead to decays of type @xmath189 in figure [ fig : sneutrino - decays2 ] while for @xmath209 to those shown in figure [ fig : sneutrino - decays2]@xmath190 . fermionic final states are dominated by third generation quark and charged leptons . due to the structure of the higgs - sneutrino mixing , @xmath210 , @xmath211 and @xmath15 final states are possible independently of the sneutrino flavor , whereas @xmath212 final states are only sizable for tau and mu sneutrinos . neglecting the final state masses , the partial decay widths for @xmath212 , @xmath210 and @xmath211 decays can be written as @xmath213 ^ 2\,,\\ \label{eq : bbar - ttbar - partial - widths2 } \gamma(\tilde\nu_i\to b\bar b)&= \frac{3 m_b^2\,g_f}{4\sqrt{2}\pi\,c_\beta^2 } m_{\tilde\nu_i}(\boldsymbol{r^{s^0}})^2_{(i+2)1}\,,\end{aligned}\ ] ] where @xmath214 is the fermi constant . for invisible decay modes the partial decay width , summing over lepton flavors , can be written according to @xmath215 with @xmath216 given by ( [ invisible - coupling ] ) . for @xmath15 final states the phase space factors are relevant , accordingly the corresponding decay width reads @xmath217 for gauge boson final states the partial decay width is given by @xmath218 where @xmath219 , @xmath220 and the amplitude @xmath221 reads @xmath222 for @xmath223 final states the partial decay width is given by @xcite @xmath224 with @xmath225 and @xmath226 the integration variables @xmath227 lie in the ranges @xmath228 $ ] and @xmath229 $ ] . the parameter @xmath230 ( @xmath231 being the total decay width of the gauge boson @xmath232 ) allows a smooth transition in the threshold region where the off - shell gauge boson becomes on - shell ( @xmath233 ) , the calculation of @xmath234 thus requires numerical integration over the variables @xmath227 in the transition region . outside that region , i.e. for @xmath235 it can be written according to @xcite @xmath236 where @xmath237 is a kinematic function given by @xmath238 finally for cp - even higgs bosons final states the partial decay width can be written as @xmath239 where @xmath240 , @xmath241 and the dimensionful coupling @xmath204 is given in ( [ eq : gijk - coupling ] ) . for the particular case @xmath242 ( @xmath14 and @xmath243 final states ) the width reduces to @xmath244 neutrino data plays an important role in brpv , since brpv parameters contribute to neutrino masses . therefore , even if brpv is not solely responsible for neutrino masses , _ @xcite , one must not saturate neutrino masses via brpv . therefore it would be worthwhile to discuss this correlation briefly here . we proceed by assuming brpv as the sole source of neutrino masses and later consider relaxing this assumption . at tree - level , brpv allows for only one massive neutrino , as mentioned earlier . one - loop contributions then leave only one generation massless . furthermore , for the approximations made in this paper , the tree - level mass is larger than the one - loop mass . the upshot of all this is that only the so - called normal hierarchy of neutrino masses is allowed . recent neutrino data @xcite then specifies the following neutrino masses : @xmath245 in terms of the model parameters @xmath246 the experimental values of the atmospheric and reactor angles yield two additional constraints given by eq . ( [ eq : mixing - angles ] ) . therefore , the atmospheric sector entirely fixes the @xmath247 parameters . ( @xmath27 ) parameters and sneutrino vevs as required for explaining neutrino data for the r - parity conserving parameters discussed in the text . the horizontal ( vertical ) solid lines in the right - panel plot indicate the values where @xmath248 is dominated by @xmath249 ( @xmath250 ) , i.e. where @xmath251 ( @xmath252).__,title="fig:",width=283,height=245 ] ( @xmath27 ) parameters and sneutrino vevs as required for explaining neutrino data for the r - parity conserving parameters discussed in the text . the horizontal ( vertical ) solid lines in the right - panel plot indicate the values where @xmath248 is dominated by @xmath249 ( @xmath250 ) , i.e. where @xmath251 ( @xmath252).__,title="fig:",width=291,height=253 ] constraints on the @xmath253 parameters arise from the solar sector . as long as the one - loop contribution to neutrino masses is smaller than the tree - level one ( an assumption we use throughout ) , @xmath254 is accurately determined by @xmath255 one - loop diagrams @xcite : @xmath256 where @xmath257 stands for sbottom mixing , @xmath258 are defined in eq . ( [ eq : epsilon - tilde - parameters ] ) and @xmath259 ( with @xmath260 is a scalar passarino - veltman function @xcite ) . note that due to @xmath261 , @xmath262 moreover the solar mixing angle can be written as @xcite @xmath263 thus , eqs . ( [ eq : solar - mass - scale ] ) and ( [ eq : solar - mixing ] ) provide two constraints and determine , from eqs . ( [ eq : epsilon - tilde - parameters ] ) , @xmath264 ( @xmath265 or @xmath266 ) as a function of @xmath250 ( @xmath267 or @xmath268 ) . once the @xmath269 s and @xmath253 s are fixed the sneutrino vevs are automatically fixed as well through @xmath270 see eq . ( [ eq : alignment - vector ] ) . this in turn fixes all the relevant mixings for sneutrino decays ( @xmath271 , @xmath272 , @xmath273 ) once the r - parity conserving supersymmetric parameters are specified . with this knowledge in hand , we start to explore the consequences of the neutrino sector on the brpv parameters . we note that our results have been verified using spheno @xcite ] . figure [ fig : epsilons - vevs ] shows typical values for @xmath264 and sneutrino vevs as a function of @xmath250 . the plots were obtained by fixing @xmath274 , @xmath275 and @xmath276 where we have assumed a conservative lower bound on the sneutrino mass of 100 gev , see @xcite for more details . we will use this parameter point throughout the paper . neutrino observable @xmath277 are varied in their 3@xmath278 experimental range @xcite . we further assume that @xmath250 is positive . for these assumptions we see from the left - hand side that @xmath279 and @xmath280 have a lower bound and from the right - handed side that this lower bound is larger than @xmath281 and @xmath282 respectively . on the other hand , since @xmath250 is undetermined , @xmath249 dominates it up to about @xmath283 . while allowing negative values for the parameters changes this picture , it leaves one important qualitative property the same : * only in one generation , @xmath284 , can the sneutrino vev , @xmath285 , be larger than the bilinear mixing term , @xmath286 , when brpv is the sole contributor to neutrino masses . * this will have important consequences in the next section . also of note in the left - panel plot , @xmath250 obeys only an upper bound determined by the condition @xmath287 , which in terms of the bilinear r - parity violating parameters translates into @xmath288 . in contrast the @xmath264 parameters , due to solar neutrino physics constraints , are forced to lie in a `` narrow '' range and are such that a region where @xmath289 exist . consequently , while @xmath290 are mostly determined by @xmath264 , @xmath248 is controlled by @xmath249 in the region where @xmath291 , as demonstrated by the horizontal solid lines in figure [ fig : epsilons - vevs ] ( right - panel ) which correspond to @xmath251 . the bilinear r - parity breaking parameters selected as described above satisfy neutrino data , and thus lead to brpv models that can account for neutrino masses and mixings . however , it might be that these parameters are not sufficiently large to account for the neutrino mass scales . in that case their contribution to the atmospheric and solar masses are still determined by eqs . ( [ eq : m - atmospheric ] ) and ( [ eq : solar - mass - scale ] ) but are such that , for example , @xmath292 and @xmath293 . figure [ fig : vevs - noatmospheric - nosolar ] shows the results for an illustrative case where the brpv model fits the atmospheric mass scale as well as the atmospheric and reactor angles in their @xmath294 experimental range , but the contribution to @xmath295 is subdominant . note that we plot only @xmath296 , as we have found that @xmath297 . this result as well as @xmath298 in the region of small @xmath265 are due to the constraints arising from fitting @xmath299 and @xmath300 . fits the experimental @xmath294 range but @xmath254 falls below the measured value ( see the text for more details).__,width=340,height=253 ] in summary , if data is not explained by the brpv parameters all the @xmath253 s can be small mainly due to the absence of the solar data constraint , thus implying that , in this case , a region where the three sneutrino vevs are large ( @xmath301 ) exist . we are finally ready to address the main aim of this paper : showing that sneutrino lsp decays into heavy sm pairs , invisible modes and different lepton flavor final states _ i.e. _ @xmath302 ; @xmath16 and @xmath303 ( @xmath304 ) can dominate over the traditionally considered @xmath211 mode , which was typically considered as the dominate mode @xcite . we begin with a very general study ignoring all neutrino constrains ( susy parameters can always be chosen in such a way so that brpv does not saturate the neutrino masses ) and exploring in the @xmath20-@xmath21 parameter space where the aforementioned modes dominate the @xmath19 decays of @xmath305 . without neutrino constraints , these decay properties are independent of generation . we scan over the following values : @xmath306 and @xmath16 with the @xmath211 final state . it has been assumed that brpv does not saturate neutrino masses . blue points indicate where @xmath307 ( left - hand side plot ) or @xmath308 ( right - hand side plot ) while the red points the opposite . the invisible mode involves a sum over final state lepton flavors , and so its importance is determined by the relative size between @xmath20 and any of the three sneutrino vevs . the vertical axes in the right - hand side plot corresponds to the largest sneutrino vev obtained when randomly scanning the parameter space . these plots therefore show an important general result : the @xmath309 and invisible modes have larger branching fractions than the @xmath211 mode only when the sneutrino vev is much larger than its corresponding bilinear @xmath253 parameter.__,title="fig:",width=283,height=245 ] and @xmath16 with the @xmath211 final state . it has been assumed that brpv does not saturate neutrino masses . blue points indicate where @xmath307 ( left - hand side plot ) or @xmath308 ( right - hand side plot ) while the red points the opposite . the invisible mode involves a sum over final state lepton flavors , and so its importance is determined by the relative size between @xmath20 and any of the three sneutrino vevs . the vertical axes in the right - hand side plot corresponds to the largest sneutrino vev obtained when randomly scanning the parameter space . these plots therefore show an important general result : the @xmath309 and invisible modes have larger branching fractions than the @xmath211 mode only when the sneutrino vev is much larger than its corresponding bilinear @xmath253 parameter.__,title="fig:",width=283,height=245 ] and @xmath123 below 1000 gev but large enough so that the sneutrino is the lsp . we also use @xmath310 gev as suggested by recent lhc results . for @xmath12 and @xmath16 modes the results are displayed in figure [ fig : sneu - ww - bb ] where blue dots indicate the points at which these channels have partial decay widths much larger than the @xmath211 mode , while the red points show the opposite . it is striking that the ratio of the partial widths are relatively independent of r - parity conserving susy parameters and depend only on the brpv parameters . from this figure , one can conclude that the @xmath309 and the @xmath16 partial widths dominate the @xmath19 width only when the sneutrino vev is much larger than the bilinear mixing parameter and , as will be shown below , this condition holds for the other heavy sm final states . couple this with the main result of the previous section : satisfying neutrino masses solely through brpv means that @xmath311 can only be satisfied in one generation , indicates that heavy final states as well as invisible modes can only have branching fractions larger than the @xmath211 mode for one generation of sneutrinos . we now turn to the discussion of the different lepton flavor final states . these modes ( @xmath312 ) , being induced by chargino charged lepton mixing , are mainly controlled by the bilinear mixing parameters for the flavor of the final lepton which differs from the sneutrino flavor ( see eq . ( [ eq : xil - explicitly ] ) and ( [ eq : bbar - ttbar - partial - widths ] ) ) _ i.e. _ @xmath313 not @xmath314 . the @xmath211 mode , on the other hand , is governed by @xmath314 ( same flavor as the sneutrino ) if @xmath315 , or @xmath316 , if @xmath317 . therefore , when @xmath315 ( region where @xmath318 modes dominate @xmath319 ) , neutrino constraints imply all three @xmath253s are of the same order ( see fig . [ fig : epsilons - vevs ] ) . therefore the different lepton flavor final states become subdominant to the @xmath318 modes due to charged lepton yukawa suppression . in contrast , when @xmath317 ( @xmath319 modes dominate @xmath318 modes ) the different lepton flavor final states can have larger partial widths than the @xmath211 mode provided @xmath320 ^ 2 $ ] exceeds the corresponding yukawa suppression @xmath321 , which turns out to be possible only for second and third generation sneutrinos as shown in fig . [ fig : epsilons - vevs ] , where it can be seen that @xmath320 ^ 2 $ ] can be at most @xmath322 ( note that @xmath323 , @xmath324 while @xmath325 ) . as in the case of heavy sm and invisible modes , these different lepton flavor final states can dominant over the @xmath211 final state only in a single sneutrino generation . we continue by considering the specific susy point discussed in the last section and examining first and third generation sneutrino decays into @xmath19 , @xmath326 , @xmath327 and different lepton flavor modes assuming that brpv is solely responsible for neutrino masses . we have also calculated the partial decay widths for the @xmath14 and @xmath13 final states , finding that they have a similar dependence on @xmath328 and are of the same order as the @xmath12 mode ( the @xmath329 channel is suppressed due to the off - shell top for this sneutrino mass ) . the results are displayed in figure [ fig : width - snu1-snu3 ] versus @xmath328 , where it can be seen that @xmath330 exhibits the usual bahavior quoted in the literature , that is to say , a dominant @xmath211 mode followed by a slightly yukawa suppressed @xmath331 mode and strongly suppressed @xmath16 , @xmath319 , @xmath332 and @xmath333 final states . the huge supression of the different lepton flavor mode , @xmath334 , in the small @xmath250 region is due to the combined effect of the electron yukawa suppression and the smallness of @xmath335 . the @xmath332 final state being controlled by @xmath336 involves only the electron yukawa suppression , and so it has a larger partial decay width than the @xmath337 mode . the situation for @xmath338 , however , is quite different . in the region of large @xmath250 it features a @xmath211 dominant mode and apart from the @xmath339 mode it behaves like @xmath330 . in the region of small @xmath250 , in contrast , the `` non - conventional '' modes @xmath340 ( @xmath341 ) , @xmath16 , @xmath12 , @xmath342 and @xmath343 have the largest partial widths by far , greatly surpassing the @xmath211 final state . in particular , the different lepton flavor channels exceed this mode by more than two - orders of magnitude , thus drastically changing the sneutrino lsp phenomenology . some words are in order as regards the size of the @xmath331 partial width in @xmath338 decays . this decay mode is a unequivocal signal of brpv models , but its size is strongly dependent on the nature of the bilinear parameters . in r - parity violating models satisfying the conditions @xmath344 and @xmath345 at certain high energy scale , the bilinears can be rotated away . they are , however , subsequently induced at the tev scale by rge running effects @xcite therefore inducing the @xmath331 mode , but with a quite suppressed partial width ( typically several orders of magnitude smaller than that of the @xmath211 channel ) . in a mssm based analysis , as the one performed here , where all the brpv parameters are treated as free parameters subject only to phenomenological constraints this is not necessarily the case , as demonstrated in the right - hand side plot in fig . [ fig : width - snu1-snu3 ] . ( left ) and @xmath346 ( right ) to @xmath211 , @xmath319 , @xmath331 , @xmath327 , @xmath347 , @xmath348 and @xmath349 final states as a function of @xmath350 . brpv parameters have been fixed by using neutrino oscillation data ( see the text for more details ) . the remaining heavy sm modes , @xmath351 and @xmath343 , have the same behavior and are of the same order than the @xmath319 mode ( are not shown to avoid crowding the plots ) . decays of @xmath352 are not shown since they are very similar to those of @xmath353.__,title="fig:",width=283,height=245 ] ( left ) and @xmath346 ( right ) to @xmath211 , @xmath319 , @xmath331 , @xmath327 , @xmath347 , @xmath348 and @xmath349 final states as a function of @xmath350 . brpv parameters have been fixed by using neutrino oscillation data ( see the text for more details ) . the remaining heavy sm modes , @xmath351 and @xmath343 , have the same behavior and are of the same order than the @xmath319 mode ( are not shown to avoid crowding the plots ) . decays of @xmath352 are not shown since they are very similar to those of @xmath353.__,title="fig:",width=283,height=253 ] figure [ fig : width - snu1-snu3 ] reinforces the arguments given in the previous paragraphs . reiterating : even if bilinear r - parity violation accounts for neutrino data , the @xmath211 channel is not necessarily the dominant decay mode . we have checked that for third and second generation sneutrino lsps the different lepton flavor final states dominate ( @xmath341 , @xmath354 for @xmath338 and @xmath355 , @xmath356 for @xmath352 ) while first generation sneutrinos decays are driven by invisible and heavy sm modes ( @xmath351 and @xmath319 ) . as already pointed out and exemplified for @xmath338 in fig . [ fig : width - snu1-snu3 ] , these decays can dominate only in a single sneutrino generation . so , observing at least two different sneutrino flavors decaying to the aforementioned modes ( something that turns out to be viable if the sneutrino mass splittings do not allow the heavier sneutrinos to decay into the lsp ) will prove that brpv does not account for neutrino data , because this would require @xmath357 for those generations , which is not possible if neutrino physics is determined solely by brpv , see fig . [ fig : vevs - noatmospheric - nosolar ] . focusing further on those final states of @xmath358 decays ( excluding the @xmath343 mode since it behaves like the @xmath319 mode and is smaller by an order of 1/2 ) , we study @xmath358 decays as a function of @xmath359 in figure [ fig : brxxoverbb - mnu ] allowing a @xmath358 heavy enough so that the @xmath329 channel is also open . all the r - parity conserving parameters have been fixed according to the supersymmetric point used in the previous section . the r - parity breaking parameters have been fixed by adjusting neutrino observables to their best fit point values @xcite and @xmath360 . we have checked that the values are insensitive to changes in @xmath124 in the range [ 2,30 ] . off - shell calculations of the higgs and top states have been neglected since the different lepton flavor final states dominant over the full sneutrino mass range . both plots support our claim : even if brpv is the sole generator of neutrino masses a sneutrino lsp does not necessarily decay predominantly into @xmath211 final states . and @xmath340 ) , invisible ( @xmath327 ) , @xmath319 , @xmath351 and @xmath15 final states normalized to @xmath361 as a function of @xmath362 . right plot : decay branching ratios for dominant ( excluding @xmath343 ) and @xmath211 modes as a function of @xmath362 . the branching ratios have been calculated in the small @xmath250 region ( @xmath363 ) and for bilinear r - parity breaking parameters fixed by neutrino parameters according to their best fit point values @xcite . both plots show the main result of this paper : sneutrino lsp decays are not necessarily dominated by the @xmath211 mode.__,title="fig:",width=283,height=245 ] and @xmath340 ) , invisible ( @xmath327 ) , @xmath319 , @xmath351 and @xmath15 final states normalized to @xmath361 as a function of @xmath362 . right plot : decay branching ratios for dominant ( excluding @xmath343 ) and @xmath211 modes as a function of @xmath362 . the branching ratios have been calculated in the small @xmath250 region ( @xmath363 ) and for bilinear r - parity breaking parameters fixed by neutrino parameters according to their best fit point values @xcite . both plots show the main result of this paper : sneutrino lsp decays are not necessarily dominated by the @xmath211 mode.__,title="fig:",width=283,height=245 ] one can extrapolate from figure 6 the behavior of the r - parity violating sneutrino decays for the first two generations , in the analogous region @xmath364 . in these cases , only the leptonic different flavor states shift and they do so by a factor of approximately @xmath365 for @xmath366 and @xmath367 for @xmath368 . therefore , @xmath352 decays will still be dominated by different flavor leptonic final states , while @xmath369 will decay predominately into heavy sm states and neutrinos . relaxing the assumption that bilinear r - parity violation is the sole generator of neutrino masses ( as in the second case discussed in section [ sec : constraints - neu - physics ] ) will not drastically change the results in this section . this is because the relative importance of the ratios @xmath370 is determined by @xmath371 . however , a quantity that becomes relevant in this case is the sneutrino decay length @xmath372 , since the size of the bilinear r - parity breaking parameters can be small enough to cause the sneutrino to decay outside of the detector . neglecting the lorentz boost factor , we have found that as long as the r - parity violating parameters fit the neutrino data , in general @xmath372 is well below 1 mm ( inline with @xcite ) . we have also checked that if @xmath373 the decay length is generically below @xmath374 cm . note that values of @xmath375 in the range @xmath376 $ ] might be in conflict with neutrino data when the contributions from the mechanism responsible for neutrino masses are taken into account ( for example a standard seesaw ) . once the contributions to the atmospheric and solar mass scales fall below @xmath377 the decay length can eventually exceed 1 m almost independently of the sneutrino mass ( as expected due to the large partial decay widths the different lepton flavor modes have ) . figure [ fig : decay - length ] shows the decay length for @xmath378 . , where @xmath379 have been taken according to their best fit point values @xcite.__,width=340,height=253 ] thus far , this section has focused on sneutrino decays . at this point , we briefly comment on production . if the sneutrino is indeed the lsp , all produced supersymmetric particles produced will cascade decay down to it . the dominate production depends on the relative masses , with strong production @xcite dominating unless the gluinos and squarks are too heavy , in which case relatively light sneutrinos can still be pair produced through electroweak processes . regardless , the parameter region explored in this paper yields compelling signals since the dominate different flavor leptonic final states would have little background if the intermediate sneutrino can be reconstructed . meanwhile , the @xmath211 mode corresponding to the parameter space typically considered in the literature would suffer from larger qcd background and would therefore be more difficult to detect . furthermore , if the brpv contributions to neutrino masses are subdominant to some other mechanism , sneutrino decays could involve large displaced vertices ( see fig . [ fig : decay - length ] ) thus rendering its identification even more plausible . brpv is a good effective theory for models of spontaneous r - parity violation , in which proton decay as well as the number of new parameters is under control compared to explicit r - parity violation . in the presence of brpv , sneutrino lsp decays can in some cases resemble a heavy higgs . however , as we have discussed , sneutrinos ( third and second generation ) can dominantly decay to different - flavor charged leptons final states or heavy sm modes ( first generation ) , thus completely altering the `` conventional '' sneutrino lsp phenomenology driven by the @xmath211 mode . we have studied these decays into @xmath12 , @xmath343 , @xmath14 , @xmath15 , @xmath16 and @xmath303 ( @xmath304 ) and found that long as @xmath317 , they have larger branching ratios than the @xmath198 mode . as discussed in section [ sec : constraints - neu - physics ] , for models where the brpv parameters are fixed by neutrino data , this is possible only for a single sneutrino flavor , due to the constraints from the solar sector . therefore , if 2 or more sneutrino generations are degenerate enough to decay via brpv into these states one could rule out brpv as the sole source of neutrino masses . for models where the bilinear r - parity breaking parameters do not contribute significantly to neutrino masses , the constraint @xmath380 for all flavors is viable . accordingly , in these brpv models large sneutrino lsp branching fractions into either @xmath303 or @xmath12 , @xmath343 , @xmath14 and @xmath15 are possible . in this case , however , special attention has to be paid to the sneutrino decay length @xmath372 . we calculated @xmath372 , finding that as long as the brpv contributions to neutrino masses are not below @xmath381 the decay length is generically below @xmath374 cm . regardless of these considerations , sneutrino decays into different - flavor charged leptons or heavy sm final states ( @xmath303 or @xmath12 , @xmath343 , @xmath14 and @xmath15 ) are an interesting phenomenological possibility that has not received much attention before . because brpv only affects the decays of the lsp , assuming a sneutrino lsp means that susy events could cascade decay to either different flavor charged leptons or two pairs of heavy sm states , which constitutes a unique and novel susy signature . we want to thank martin hirsch for his comments in the early stage of this project and werner porod for spheno support . ss would also like to thank pavel fileviez perez for comments . das is supported by a belgian fnrs postdoctoral fellowship . dr has been partially supported by sostenibilidad - udea and colciencias through the grant number 111556934918 . 99 g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * ( 2012 ) 1 [ arxiv:1207.7214 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * ( 2012 ) 30 [ arxiv:1207.7235 [ hep - ex ] ] . r. barbier , c. berat , m. besancon , m. chemtob , a. deandrea , e. dudas , p. fayet and s. lavignac _ et al . _ , phys . rept . * 420 * , 1 ( 2005 ) [ hep - ph/0406039 ] . s. chatrchyan _ et al . _ [ cms collaboration ] , arxiv:1206.3949 [ hep - ex ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 709 * , 137 ( 2012 ) [ arxiv:1110.6189 [ hep - ex ] ] . l. m. carpenter , d. e. kaplan and e. -j . rhee , phys . lett . * 99 * , 211801 ( 2007 ) [ hep - ph/0607204 ] . l. m. carpenter , d. e. kaplan and e. j. rhee , arxiv:0804.1581 [ hep - ph ] . p. w. graham , d. e. kaplan , s. rajendran and p. saraswat , jhep * 1207 * , 149 ( 2012 ) [ arxiv:1204.6038 [ hep - ph ] ] . c. s. aulakh and r. n. mohapatra , phys . b * 119 * ( 1982 ) 136 . m. j. hayashi and a. murayama , phys . b * 153 * ( 1985 ) 251 . r. n. mohapatra , phys . * 56 * ( 1986 ) 561 . j. m. mira , e. nardi , d. a. restrepo and j. w. f. valle , phys . b * 492 * , 81 ( 2000 ) [ hep - ph/0007266 ] . d. aristizabal sierra , d. restrepo and o. zapata , phys . d * 80 * , 055010 ( 2009 ) [ arxiv:0907.0682 [ hep - ph ] ] . c. s. aulakh , a. melfo , a. rasin and g. senjanovic , phys . b * 459 * ( 1999 ) 557 [ hep - ph/9902409 ] . c. s. aulakh , b. bajc , a. melfo , a. rasin and g. senjanovic , nucl . b * 597 * ( 2001 ) 89 [ hep - ph/0004031 ] . p. fileviez perez and s. spinner , phys . b * 673 * ( 2009 ) 251 [ arxiv:0811.3424 [ hep - ph ] ] . v. barger , p. fileviez perez and s. spinner , phys . * 102 * ( 2009 ) 181802 [ arxiv:0812.3661 [ hep - ph ] ] . m. ambroso and b. ovrut , jhep * 0910 * ( 2009 ) 011 [ arxiv:0904.4509 [ hep - th ] ] ; m. ambroso and b. a. ovrut , int . j. mod . phys . a * 25 * ( 2010 ) 2631 [ arxiv:0910.1129 [ hep - th ] ] . p. fileviez perez and s. spinner , phys . d * 83 * ( 2011 ) 035004 [ arxiv:1005.4930 [ hep - ph ] ] . a. masiero and j. w. f. valle , phys . b * 251 * , 273 ( 1990 ) ; w. buchmuller , l. covi , k. hamaguchi , a. ibarra and t. yanagida , jhep * 0703 * , 037 ( 2007 ) [ hep - ph/0702184 [ hep - ph ] ] . f. takayama and m. yamaguchi , `` gravitino dark matter without r - parity , '' phys . b * 485 * , 388 ( 2000 ) [ hep - ph/0005214 ] . s. dimopoulos , d. eichler , r. esmailzadeh and g. d. starkman , phys . d * 41 * , 2388 ( 1990 ) . a. bartl , m. hirsch , t. kernreiter , w. porod and j. w. f. valle , jhep * 0311 * , 005 ( 2003 ) [ hep - ph/0306071 ] . d. aristizabal sierra , m. hirsch and w. porod , jhep * 0509 * , 033 ( 2005 ) [ hep - ph/0409241 ] . w. porod , m. hirsch , j. romao and j. w. f. valle , phys . d * 63 * , 115004 ( 2001 ) [ hep - ph/0011248 ] . m. hirsch and w. porod , phys . d * 68 * , 115007 ( 2003 ) [ hep - ph/0307364 ] . s. roy , b. mukhopadhyaya and , phys . d * 55 * , 7020 ( 1997 ) [ hep - ph/9612447 ] . b. mukhopadhyaya , s. roy , f. vissani and , phys . b * 443 * , 191 ( 1998 ) [ hep - ph/9808265 ] . s. bar - shalom , g. eilam and b. mele , phys . d * 64 * , 095008 ( 2001 ) [ hep - ph/0106053 ] . y. grossman , h. e. haber and , phys . d * 59 * , 093008 ( 1999 ) [ hep - ph/9810536 ] . m. hirsch , m. a. diaz , w. porod , j. c. romao and j. w. f. valle , phys . d * 62 * , 113008 ( 2000 ) [ erratum - ibid . d * 65 * , 119901 ( 2002 ) ] [ hep - ph/0004115 ] . m. a. diaz , m. hirsch , w. porod , j. c. romao and j. w. f. valle , phys . d * 68 * , 013009 ( 2003 ) [ erratum - ibid . d * 71 * , 059904 ( 2005 ) ] [ hep - ph/0302021 ] . m. nowakowski and a. pilaftsis , nucl . b * 461 * , 19 ( 1996 ) [ hep - ph/9508271 ] . m. hirsch and j. w. f. valle , nucl . b * 557 * , 60 ( 1999 ) [ hep - ph/9812463 ] . m. drees and m. m. nojiri , phys . d * 45 * , 2482 ( 1992 ) . f. de campos , m. a. garcia - jareno , a. s. joshipura , j. rosiek , j. w. f. valle , nucl . b * 451 * , 3 ( 1995 ) [ hep - ph/9502237 ] . a. djouadi , j. kalinowski and p. m. zerwas , z. phys . c * 70 * , 435 ( 1996 ) [ hep - ph/9511342 ] . a. djouadi , phys . rept . * 457 * , 1 ( 2008 ) [ hep - ph/0503172 ] . e. ma , mod . a * 17 * , 1259 ( 2002 ) [ hep - ph/0205025 ] . d. aristizabal sierra , m. hirsch , j. w. f. valle and a. villanova del moral , phys . d * 68 * , 033006 ( 2003 ) [ hep - ph/0304141 ] . d. k. ghosh , g. senjanovic and y. zhang , phys . b * 698 * , 420 ( 2011 ) [ arxiv:1010.3968 [ hep - ph ] ] . v. barger , p. fileviez perez and s. spinner , phys . b * 696 * , 509 ( 2011 ) [ arxiv:1010.4023 [ hep - ph ] ] . d. v. forero , m. tortola and j. w. f. valle , phys . d * 86 * , 073012 ( 2012 ) [ arxiv:1205.4018 [ hep - ph ] ] . m. c. gonzalez - garcia , m. maltoni , j. salvado and t. schwetz , arxiv:1209.3023 [ hep - ph ] . g. l. fogli , e. lisi , a. marrone , d. montanino , a. palazzo and a. m. rotunno , phys . d * 86 * , 013012 ( 2012 ) [ arxiv:1205.5254 [ hep - ph ] ] . g. passarino and m. j. g. veltman , nucl . b * 160 * , 151 ( 1979 ) . w. porod , comput . commun . * 153 * , 275 ( 2003 ) [ hep - ph/0301101 ] ; w. porod and f. staub , comput . commun . * 183 * , 2458 ( 2012 ) [ arxiv:1104.1573 [ hep - ph ] ] . j. abdallah _ et al . _ [ delphi collaboration ] , eur . j. c * 31 * , 421 ( 2003 ) [ hep - ex/0311019 ] . b. de carlos and p. l. white , phys . d * 55 * , 4222 ( 1997 ) [ hep - ph/9609443 ] . b. de carlos and p. l. white , phys . rev . d * 54 * , 3427 ( 1996 ) [ hep - ph/9602381 ] . e. nardi , phys . d * 55 * , 5772 ( 1997 ) [ hep - ph/9610540 ] . m. kramer , a. kulesza , r. van der leeuw , m. mangano , s. padhi , t. plehn , x. portell and , arxiv:1206.2892 [ hep - ph ] .

the current successes of cold dark matter cosmologies , and in particular the @xmath1cdm models , are well known . these hierarchical models begin with the microwave background fluctuations and proceed forward to reproduce the structure in the ly@xmath2 forest at @xmath3 ( e.g. , mandelbaum et al . 2003 ) and continue to evolve the dark matter structure distribution to arrive at results consistent with the power spectrum of the distribution of galaxies at the present epoch ( e.g. , tegmark et al . 2004 ) , the fraction of baryons in rich clusters ( white et al . 1993 ) , and other notable results . these models follow the evolution of dark matter , and the accompanying galaxies are assumed to form as gas flows into the ever growing dark matter condensations ( e.g. , white and rees 1978 ) . the most detailed results describing this structure evolution have come from highly sophisticated numerical simulations that follow the growth of mass fluctuations in the early universe , such as the millennium simulation ( springel et al . 2005 ) . however , there are some well known difficulties with these models in that they predict too many faint and too many bright galaxies ( e.g. , benson et al . 2003 ) , and in addition they predict that the largest galaxies should be blue and star forming . these are serious difficulties , and they have been addressed through the use of various assumptions about physical processes that work on scales not resolved by the numerical simulations . these `` semi - analytic '' models have addressed the problems at both the dwarf galaxy and the brightest galaxy scales ; the issues addressed in this paper involve only the problems associated with the evolution of the brightest galaxies . the problems there are a direct result of the hierarchical nature of the models ; minute fluctuations in the very early universe grow due to continuing dark matter and baryonic infall , and in the context of the models this infall should continue to the present epoch , possibly leading to significant star formation now in the most massive galaxies . thus in this picture the most massive galaxies would be blue due to the large population of young stars , whereas in fact the largest galaxies are generally seen as `` red and dead '' at the present epoch , implying that the most active epoch of star formation occurred earlier ( e.g. , madau et al . 1996 , cowie et al . 1996 ) . to address this problem , one of the most commonly used assumptions in semi - analytic models is that of `` feedback '' from the central regions of the galaxy , where energy in some form is produced in sufficient quantity that it can heat the inflowing gas , perhaps stopping or reversing the inflow , and in any case providing enough energy to suppress star formation . once this has been done , the subject galaxy will then passively evolve , and its aging stellar population will become redder with time until the galaxy colors match the observations . motivated by the observation of active galactic nuclei ( agn ) in many massive red galaxies , several agn related driving mechanisms for this feedback have been suggested . radiatively driven winds , agn driven shocks and mass outflows , qsos , and radio sources associated with agns have all been proposed ( e.g. , sijacki et al . 2007 ) . in some instances qsos have been referred to as `` high luminosity '' agn feedback and radio - loud agn ( or `` radio agn '' ) have been called `` low luminosity '' agn feedback . this may not be a meaningful distinction , since many qsos may well be examples of powerful radio sources that are directed toward the observer , with their collimated emission doppler boosted to very high luminosities . the use of agns , and in particular radio agns , as sources of the energy required to suppress star formation in feedback models has appeal because estimates for the total energies of these objects derived from energy equipartition arguments have existed for some time . these minimum energies ( from @xmath4 ergs ) are , in an integral sense , adequate to provide the needed heating for agn feedback to work . moreover , recent estimates of the total energies present in some radio sources tend to lie factors of 10 higher than the estimates based on the equipartition energy present in relativistic electrons and fields ( brzan et al . 2008 ) . thus the total energies available in radio agn make these objects viable candidates for powering feedback and for suppressing star formation in massive galaxies , and radio agn have been incorporated into semi - analytic models that are based on cosmological simulations such as the millennium run ( e.g. , croton et al . 2006 , bower et al . 2006 , okamoto et al . for example , croton et al . ( 2006 ) show that a form of radio agn energy input can not only reproduce the observed truncation of the bright end of the luminosity function , but that it can also possibly suppress accretion and give rise to an old , red population of elliptical galaxies . radio agn feedback has also been used to explain the presence of old `` red and dead '' galaxies in voids ( croton & farrar 2008 ) , where environmental effects are much less pronounced . finally , radio agn feedback has also been suggested by several authors as a means of suppressing cooling or providing reheating in the cores of rich clusters of galaxies , thereby solving the `` cooling flow '' problem and at the same time accounting for the colors of brightest cluster galaxies ( e.g , fabian et al . 2000 , brggen & kaiser 2002 , ruszkowski et al . 2004 , mcnamara & nulsen 2007 ) . while the integrated total energies of radio sources are adequate , and sometimes more than adequate ( e.g. , best et al . 2006 ) to provide the needed feedback in semi - analytic models , the key question of `` does it really work ? '' remains . having enough total energy is only the first of several conditions that need to be met . in order for radio agn feedback to work , this energy contained in the outflowing radio sources must be transferred to the surrounding medium . in order to suppress star formation , this energy transfer has to be fairly efficient , and equally important , for radio agn feedback it has to be turned into heat , and it has to be distributed over most of the volume of gas in and around the halo of the parent galaxy . finally , all this has to happen in a time less than the local cooling time or dynamical infall time . until all of these conditions can be met , the effectiveness of radio agn feedback in solving significant difficulties with current cosmological models will remain in question . this paper provides a calculation of one way this energy transfer process can work . section ii presents an overview of the physical processes involved in radio agn feedback , while section iii describes a calculation of the energy transfer during the interaction of a radio source with its environment that covers up to 5 orders of magnitude in scale . the results of this calculation are given in section iv , and the conclusions are found in section v. a fundamental property of radio agn that has a major impact on the issue of feedback is the bipolar nature of the energy outflow . except for low energy `` compact radio sources '' , all radio agn show this morphology . the degree of collimation can be moderate , with opening angles of the flow of order twenty degrees or so , or it can be extreme , as in the case of the very energetic and highly collimated flows associated with high luminosity radio sources and quasars . this two - sided collimated outflow is the principal obstacle to radio agn feedback being an efficient and effective mechanism for energizing and heating ( or blowing away ) the gas in the central regions of the parent galaxy , since the heating and suppression of star formation must occur throughout the central region and not just along the two narrow cones in the vicinity of the radio jets . thus for feedback to be effective , some processes must be found that transfers energy from the jet to the surrounding medium , converts that energy to heat , and distributes it over most of the volume in the central regions of the parent galaxy . some possible mechanisms for this energy transfer have been previously suggested , such as shock waves from highly supersonic outflows ( brggen et al . 2007 , graham et al . 2008 ) or sound wave dissipation from less energetic events ( sanders & fabian 2007 , 2008 ) . these will be considered in more detail later , but in general shock waves are one - time events that rapidly slow and dissipate in the direction transverse to jet motion , while sound waves may or may not carry enough energy to heat large volumes of gas through their eventual dissipation . more importantly , the process of energy deposition by the sound waves in a magnetoionic medium is not clearly understood in this case , especially with regard to the dissipation lengths . clarification of possible feedback processes can be found from the luminosities , morphologies and demographics of the radio agn outflows . most extended radio sources can be placed in one of two categories first proposed by fanaroff & riley ( fanaroff & riley 1974 ) . these two classes , fr - i and fr - ii , are characterized by different luminosities and by very different radio morphologies . high luminosity fr - ii objects are generally characterized by two very highly collimated jets of emission extending over large distances , typically hundreds of kpc , with the jets terminating in regions of high surface brightness ( `` hot spots '' ) surrounded by lower surface brightness lobes which often have a `` swept back '' or `` bow shock '' appearance ( e.g. ferrari 1998 , de young 2002 ) . the morphologies of the lower luminosity fr - i objects are dramatically different , with the highest surface brightness near the nucleus of the parent galaxy and a gradual dimming of brightness along two moderately collimated jets with much wider opening angles than seen in the fr - ii objects . these outflows are also often bent or distorted and display a meandering morphology as the flow moves outward . this pattern of radio emission is very suggestive of the slow , subsonic flow associated with plumes and turbulent flows , and it was suggested many years ago ( e.g. , de young 1981 , bicknell 1984 ) that these radio sources were characterized by significant interaction with the surrounding ism or igm that resulted in mass entrainment and a deceleration of the agn outflow as it moves away from the nucleus . it may also be that the original outflow speeds in the fr - i objects are less than those existing in fr - ii sources , but in any case the current modeling of these objects usually involves deceleration to subsonic flow , presumably mediated by mass entrainment from the ambient medium ( e.g. , laing & bridle 2002 , 2004 ) . the second key point in considering radio agn feedback comes from the demographics of fr - i and fr - ii radio sources . it has been known for some time that the local space density of extended radio sources is dominated by fr - i objects ( e.g. , ledlow & owen 1996 , owen & ledlow 1997 , jackson & wall 1999 , willot et al . 2001 ) , where the local ( @xmath5 ) density of fr - i objects is roughly @xmath6 times that of fr - ii radio galaxies . the local density of fr - i radio sources , and a calibration point for constructing radio luminosity functions ( rlf ) is about 400 gpc@xmath7 ( rigby , best & snellen 2008 ) . due to the rapid decline in the detection of the fainter fr - i objects with distance , the dependence of the fr - i / fr - ii ratio with redshift is less clear , but recent surveys and modeling of the rlf out to moderate redshifts of less than 1 suggest that the space density of fr - i objects clearly dominates in the local universe ( willot et al . 2001 , cara & lister 2008 , donoso et al . thus to a very good approximation the `` generic '' radio agn outflow in the nearby universe can be taken to be an fr - i radio source . this may also be the case at moderate redshifts as well , since the previous picture of a constant fr - i space density with cosmic epoch coupled with a rapidly evolving fr - ii space density may be overly simple ( e.g. , rigby , best & snellen 2008 ) . in any case it is clear that `` radio agn feedback '' for redshifts less than one can be taken as `` fr - i agn feedback '' , and henceforth the focus here will be on fr - i objects as sources of radio agn feedback . a third observational result that constrains the feedback process comes from radio and x - ray observations of radio sources in rich clusters of galaxies . for the first time since their discovery over thirty years ago , it is now possible to carry out calorimetry of some radio source outflows by using estimates of the @xmath8 work done by the outflows in inflating cavities observed in the icm . ( e.g. , brzan et al . 2008 , mcnamara & nulsen 2007 and references therein . ) the role of such cavities as being a signature of radio agn feedback in clusters has been discussed elsewhere , and the relation of this phenomenon to more general fr - i outflows will be discussed at the end of section 4 . the focus here is on the general problem of radio agn feedback , both in and out of clusters and groups of galaxies , and the relevant result from the radio source calorimetry in clusters is that it reveals outflow energies that are generally in excess of the equipartition values that were commonly assumed in the past ( brzan et al . 2008 ) . these @xmath8 values are often an order of magnitude greater than the total energies contained in relativistic electrons and magnetic fields in minimum energy equipartition , and they are consistent with most of the energy residing in the kinetic energy flux of the outflow ( e.g. , de young 2006 ) . with these three observational constraints - the meandering plume - like morphology of fr - i radio sources , the dominance of fr - i objects in the nearby radio agn population , and the existence of much more energy in the outflows than is required to account for the synchrotron radio emission , the most likely paths for radio agn feedback become more clearly defined . first , if these outflows are truly subsonic over most of their length , and the morphological data are very suggestive of this , then heating by shock waves is not likely to be a significant energy transfer mechanism in these objects , nor is momentum transfer likely to form a nearly spherical outflowing wind that removes the ambient gas . second , because there is strong evidence that agn outflows are relativistic on very small ( @xmath9 parsec ) scales , the fr - i morphology strongly suggests significant deceleration of the flow . the uniform appearance of the outflows from the nucleus outward without any clear discontinuities together with the spreading of the outflow from the nucleus outward implies that this deceleration process begins near the nucleus and continues more or less uniformly throughout the outflow region within the parent galaxy . deceleration of the flow means a loss of momentum from the jet via a transfer of momentum from the jet to another medium , with the most obvious candidate being the ambient gas in the ism of the host galaxy . the most natural and most obvious candidate for this deceleration process is thus entrainment of , and transfer of momentum to , the ambient medium at the interface between the outflowing agn material and the ism and galactic halo gas . in addition to continuous deceleration from entrainment , deceleration of the outflow from its original speed near the galactic nucleus could entail the production of shock waves and subsequent heating , since vlbi observations of agn suggest very rapid outflow speeds on the parsec and subparsec scales in the innermost regions . however , the propagation of a decelerating supersonic jet into the circumnuclear gas will produce a driven bow - shock configuration that , for uniform flow , is a one - time event . in addition , except for the small angular region near the head of the jet , such shocks are generally weak , oblique , and decelerating . nonetheless , they probably occur around the initial outflow regions of the jets , and as such they could produce localized heating in the innermost regions of the parent galaxy , especially if the agn outflow involves multiple outbursts . this process may be at work in outflows in rich clusters such as 3c 84 , where deceleration is most dramatic due to the very high ambient densities . because the total energy involved in these shocks is much less than the kinetic energy of the outflow , and because the focus here is not on agn in rich clusters , the emphasis here will be on the transport of energy by the outflow itself , with most of the flow deceleration being due to continuous entrainment of the ambient medium . this entrainment process is mediated through the non - linear development of surface instabilities , most notably the kelvin - helmholtz ( kh ) instability , and it is driven by the large ( @xmath10 erg s@xmath7 ) kinetic energy flux of these outflows derived from the calorimetry described above . analytic approximations , laboratory experiments , and numerical simulations have all shown that this instability in shearing flows is essentially inevitable . incompressible flows , compressible flows , supersonic flows , mhd flows and relativistic shearing flows all show the onset and non - linear growth of the kh instability ( e.g. , chandrasekhar 1961 , brown & roshko 1974 , clemens & mungal 1995 , aloy et al , 1999 , ryu , jones & frank 2000 , perucho et al . 2004 , 2007 ) . in the fully non - linear stage of development the instability evolves into a mixing layer dominated by large scale vortex structures that entrain material from both sides of the layer into the layer itself ( e.g. , dimotakis & brown 1976 ) . the thickness of this layer grows as one moves down the jet from the onset of the non - linear phase of the instability , and the growth of the layer is nearly linear with distance along the jet until the jet is fully infiltrated by the layer . at that point a different evolutionary phase begins , as is described in section 3 . for a plane mixing layer , which is a good approximation to the cylindrical mixing layer in a round jet in its initial stages ( e.g. , dimotakis et al . 1983 , freund et al . 2000 ) , the initial growth is characterized by a nearly constant opening angle that depends upon the relative flow speeds and the fluid densities as @xmath11 where @xmath12 and @xmath13 are the densities in the outflow and the ambient medium respectively , and @xmath14 is the mach number of the outflow relative to the sound speed in the ambient medium ( e.g. , brown & roshko 1974 , dimotakis & brown 1976 , de young 1993 ) . empirical data suggest @xmath15 and @xmath16 . within this growing mixing layer the entrainment proceeds through the initial production of large scale eddies whose diameter is roughly equal to the thickness of the layer , followed by the development of finer scale structures within the eddies as well as merging of the larger scale structures . in other words , the mixing layer becomes fully turbulent . moreover , the thickness of this turbulent mixing layer grows as one moves downstream along the jet , and eventually the turbulence will penetrate throughout the entire volume of the outflow . from eq . 1 , the distance along the jet from the onset of the non - linear mixing layer to the point of complete infiltration of turbulence into the interior of the jet can be approximated by @xmath17 where @xmath18 is the radius of the jet at the onset of the mixing layer . beyond this point the flow within the jet is fully turbulent ; many experimental examples of such turbulent , decelerated jets can be found ( e.g. , dimotakis et al . 1983 , mungal et al . 1992 ) , and within the limits of resolution the outflow from fr - i objects such as 3c 31 is indistinguishable from these experimental results . thus the fr - i demographics , when combined with the evidence for uniform deceleration in their agn outflows , the universality of shear driven surface instabilities that entrain mass and provide momentum transfer through the growth of turbulent mixing layers , and the morphological similarities between fr - i outflows and subsonic turbulent plumes all provide evidence that the most common energy outflow associated with radio agn feedback is that of diverging , subsonic , fully turbulent plumes . this argument is given further support from a comparison of high resolution radio observations of fr - i outflows with experimental results concerning fully turbulent round jets propagating into a uniform medium . there is considerable evidence that such jets are self - similar flows ( e.g. , pope 2000 , hussein et al . 1994 ) , and that these flows all have the same opening angle once self - similarity is established . this angle is about 23 - 24 degrees , and an examination of radio observations of the fr - i objects 3c 31 and 3c 296 ( hardcastle et al . 2005 , laing et al . 2008 ) shows that both the north and south jets in these objects are very well enclosed by an opening angle of about 23 degrees ( @xmath19 ) . once the outflows show sudden divergence or bending this is no longer the case , but in the inner regions of the flow where the ambient conditions are likely to be fairly constant or slowly varying , the agreement with self - similar turbulent jets is remarkable . elements of this picture have been suggested before , and the first numerical simulation of mass entrainment in astrophysical jets was performed many years ago ( de young 1986 ) ; there has since been considerable development of phenomenological models of these flows ( e.g. , laing et al . 1999 , laing & bridle 2002 , 2004 , wang et al . 2009 ) . because magnetic fields are present in both the agn outflows and in the ambient medium , possible inhibition of the mixing layer growth by magnetic effects might occur , especially for `` favorable '' field geometries where @xmath20 is parallel to @xmath21 . however , it is clear that in the context of mhd outflow models , the average kinetic energy density must be much greater than the average magnetic energy density in order for the agn outflow to occur . ( the opposite case of nearly mass - free `` poynting '' jets is discussed in the last section . ) in the mhd case the concern is then that very local b field amplification due to turbulence could lead to local field energies sufficient to damp the development of the mixing layer . in this case , detailed 3d numerical mhd simulations of the development of the k - h instability and its fully non - linear growth ( e.g. , ryu et al . 2000 ) show that the flow remains `` essentially hydrodynamic '' . small scale field amplifications do occur , and at late times the amplitudes are comparable to the flow energy densities , but these areas are very localized and do not inhibit the growth of turbulent flow in the mixing layer on the larger scales of energy injection into the turbulence . it is important to note the difference between these dynamic , driven outflows and the much slower , passive evolution of buoyant " bubbles in an intracluster medium , where large scale icm fields may have an influence on the bubble evolution . this characterization of radio agn feedback as turbulent fr - i outflows provides the initial conditions for finding processes that can transform this outflow into widespread heating of the ambient medium . a major advantage of turbulent flows is that the ultimate fate of fully developed turbulent flows is conversion of the turbulent energy into heat . thus if mass entrainment and the generation of turbulence is the most efficient coupling of the agn outflow to the ambient medium , then the the natural evolution of that turbulence will provide the most effective heating of the gas in the galaxy by radio agn outflows . a central issue is how long this conversion takes . the next section provides a quantitative estimate of this time through a calculation of the evolution of a driven turbulent flow from its onset until its dissipation . the turbulent flow that describes the radio agn outflow can be recast in a form common to many turbulent flows . on some large scale comparable to a defining scale of the problem ( e.g. , the radius of the outflow ) energy is injected in the form of turbulent eddies . this process , as verified by experiment , involves a large scale ingestion of material on both sides of flow boundary in a `` gulping '' mode by the largest eddies ( e.g. , dimotakis & brown 1976 ) , followed by a rapid generation of finer and finer structure and mixing within this large scale as the flow evolves to smaller scales . in terms of the evolution of the energy spectrum of the turbulence , this can be described as energy injection over a small range of wavenumber @xmath22 followed by a loss free cascade of energy to larger wavenumbers . for homogeneous and isotropic hydrodynamic turbulence this process results in the standard power - law kolmogorov spectrum with @xmath23 . ( for mhd turbulence the power - law exponent is less well defined but is generally somewhat smaller . ) the dissipation of the turbulent kinetic energy into heat occurs at and above some wavenumber @xmath24 which generally lies many orders of magnitude above the energy injection scale @xmath25 . the kolmogorov formalism describes an equilibrium state where the energy input at @xmath25 is exactly balanced by the dissipation at @xmath24 . the problem being addressed here is not as simple , since what is needed is a calculation of the time required for the flow to reach the dissipation range at @xmath24 after the injection has begun at @xmath25 . because the evolution of the turbulent flow involves the non - linear interaction and energy transfer among turbulent scales of many different sizes , both upward and downward in @xmath22 space , the evolution must be calculated explicitly . in addition to the time dependent nature of the flow , a major hurdle is the many orders of magnitude in scale between the size of the large eddy injection range ( tens of parsecs or more ) to the energy dissipation scale due to effective viscosities in ionized plasma ( possibly meters or less ) . some simplifying assumptions can be made . if the overall outflow of the fr - i agns is subsonic or transonic relative to the ambient medium , then the flow can be regarded as generally incompressible . in addition , in a frame comoving with the mean local outflow speed the turbulence can be treated as locally homogeneous and weakly isotropic ( in that helicity is allowed ) to a good approximation . under these conditions a technique may be used that calculates the three dimensional time dependent evolution of turbulence , both hydrodynamic and magnetohydrodynamic , over several orders of magnitude in spatial scale . the original method was devised to treat purely hydrodynamical turbulence ( e.g. , orszag 1970 , 1972 ) but was subsequently generalized to include magnetic fields in the mhd approximation ( pouquet et al . 1976 , pouquet et al . 1978 , de young 1980 ) . this approach takes moments of the fourier transforms of the equations of mass , momentum and energy conservation in order to obtain equations that describe the time evolution of second moments of velocity and magnetic field , i.e. , the kinetic energy and magnetic energy , as a function of time and spatial scale . the system of moments is closed by using a quasi - normal approximation that assumes the fourth order moments are related to the second order moments in the same way as is the case for a gaussian distribution of fourier modes . this `` quasi - normal '' approximation permits closure of the system of equations at second order and results in differential equations that describe the temporal and spatial evolution in fourier space of the square of the fluid velocity ( i.e. , kinetic energy per unit mass ) and the magnetic energy per unit mass . the calculation uses a complex form of eddy viscosity that employs a markov cutoff in `` memory '' of the eddy interactions and includes the non - linear transfer of energy among turbulent eddies of different sizes , including triads of eddies with differing wavenumbers . some additional details are found in the appendix . the principal advantages of this eddy damped , quasi - normal markov ( edqnm ) method are its ability to calculate the time evolution of the turbulence over very large spatial ranges and its relatively modest computational requirements , especially when compared to the more straightforward methods of direct numerical simulation . the technique is is wide use , and its accuracy has been verified by comparison with both direct numerical simulations and with laboratory experiments ( e.g. , gomez et al . 2007 , vedula et al . 2005 , turner & pratt 2002 , lesieur & ossia 2000 , staquet & godeferd 1998 ) . details of this technique can be found in orszag ( 1977 ) , lesieur ( 2008 ) , and pouquet et al . ( 1976 ) and de young ( 1980 ) for the mhd generalization . as discussed in the appendix , evolution of the turbulence is followed by solving a series of integro - differential equations of the general form @xmath26dpdq.\ ] ] here @xmath27 is the fourier transformed magnetic or kinetic energy at wavenumber @xmath22 , @xmath28 is a dissipation constant ( either viscosity or magnetic diffusivity ) , @xmath29 is the forcing function or driving term , which in this case arises from the surface instabilities at the jet - ambient medium interface that create the large scale turbulent eddies driving the turbulent flow . the term @xmath30 represents the non - linear transfer terms in the flow which mediate the transfer of magnetic and kinetic energies over all wavenumbers ; i.e. , @xmath30 describes the processes that create the cascade of energy from large scale to small scales , the amplification of magnetic fields by fluid turbulence , and the inverse cascades , if any , of energy from small scales to larger scales . ( which in general is significant only in the presence of flows with net helicity . ) in general all of @xmath27 , @xmath31 , and @xmath30 are time dependent . the explicit form for @xmath30 can be complex because it describes the interaction of hydrodynamic turbulent eddies with each other , the interaction of hydrodynamic turbulence with any magnetic fields and the resulting possible magnetic field amplification , the effects of eddy viscosity , and the self interaction of magnetic turbulent structures with each other . the complete mhd forms of @xmath30 , including the effects of helicity ( * @xmath32 * and its magnetic counterpart * @xmath33 * ) are given in de young ( 1980 ) . in the present case magnetic effects are not of major concern . however , a very small , dynamically unimportant magnetic field is included for completeness ; its initial energy density is @xmath34 that of the turbulent kinetic energy density . it could in principle play a role in the dissipation process at small scales , but to include these effects would require a knowledge of the detailed field reconnection and dissipation mechanisms , which in turn would require a specification of field geometries and strengths as a function of scale length . since these are unknown , the magnetic diffusivity is chosen to be consistent with the hydrodynamic dissipation as described below . magnetic field amplification to equipartition with the hydrodynamic energies will occur on very small scales , but in the absence of any helicity this will have no effect on the evolution of the turbulence on the scales of interest here . the problem is specified by the injection of kinetic turbulent energy at some rate @xmath29 at a scale defined as the energy range , and the evolution of the turbulent energy spectrum is then followed as the energy cascades to smaller and smaller scales . a natural unit for length is that corresponding to the energy injection scale , @xmath35 , and a corresponding time scale comes from the largest eddy turnover time ; i.e. , @xmath36 , where @xmath37 is the mean velocity of the turbulent flow . because the turbulence is driven by shear instabilities in the mixing layer between the agn outflow and the surrounding medium , @xmath37 is comparable to the local jet outflow speed . the calculations here are basically scale free . this is because the actual dissipation scale is very many orders of magnitude smaller than the energy injection scale , and inclusion of the energy scale , which is necessary , means that even with this technique the calculation can not span the scale lengths from energy injection to dissipation . the dissipation scale is given by @xmath38 , where @xmath39 is the reynolds number appropriate to the large scale flow and @xmath25 is the wavenumber corresponding to the energy injection scale . the hydrodynamic reynolds number is @xmath40 , where @xmath41 is the average velocity at scales @xmath25 ; i.e. , @xmath42 . the magnetic reynolds number is just this expression with the magnetic diffusivity @xmath43 substituted for the kinetic viscosity @xmath28 . for a hot ( @xmath44k ) rarefied ( @xmath45 ) plasma @xmath46 ( e.g. , spitzer 1962 ) , and energy injection scales of tens of parsecs and sonic or subsonic flow speeds of @xmath47 km / sec imply ten or more orders of magnitude in scale between the energy injection scale and the dissipation scale . even though the equations of the form ( 2 ) above are solved numerically on a logarithmic scale in @xmath22 space , a solution that spans the complete scale from @xmath25 to @xmath24 is not feasible , especially in view of a causality condition discussed below . thus the actual values of the dissipation constants @xmath28 and @xmath43 never enter into the calculation , and no inherent scales are present . this means that application of the results can be applied to many different spatial and time scales , and this characteristic will be used in estimating the time for the turbulent energy flow to actually reach the regime where energy is dissipated into heating the ambient medium . because the wavenumber range included in the calculations is less than the complete range to @xmath24 , some mechanism has to be used at the upper end @xmath48 of the calculated wavenumber range to simulate the continuation of the loss free cascade of energy to wavenumbers beyond @xmath48 . if this is not done , unphysical reflection of turbulent energy from the upper end of the calculated wavenumber range can occur . a standard method for treating this is to use artificially large values of the dissipation constants @xmath28 and @xmath43 . these are adjusted so that the resulting effective values of @xmath24 are somewhat less than @xmath48 so that no spurious energy reflections are seen . the differential equations of the form given by eq . 2 are solved numerically on a discrete grid in @xmath22 space , and in order to avoid non - physical energy fluctuations , a time step control similar to a courant condition must be used . the value of the time step in the integration is chosen to be less than the minimum of @xmath49 . this ensures the correct behavior at high wavenumbers ; it also illustrates why extending the calculated wavenumber range beyond about 5 orders of magnitude is not practical . because the calculation is scale free , the obvious scaled values to choose are @xmath50 and @xmath51 ; once the physical dimension and speed corresponding to these are chosen , the time step is determined and the problem is well defined . the scale - free nature of the calculation allows a simple estimate of the time required to actually reach the true dissipation range once a calculation has reached equilibrium over the more limited wavenumber range described above . suppose a calculation is made of the evolution of the turbulence spectrum between some @xmath52 and @xmath53 using a @xmath54-function energy input at @xmath52 . let @xmath53 be some large multiple @xmath55 of @xmath52 , about 100 , and suppose the spectrum reaches a steady state at some time @xmath56 , where @xmath57 is measured in units of @xmath58 with @xmath59 ; @xmath60 is roughly the large scale eddy turnover time at @xmath52 , and @xmath61 is the mean flow speed of the turbulence on this scale . hence @xmath56 can be measured in units of the turnover time and thus @xmath62 . because it is scale free , the same calculation also applies to driving turbulence between @xmath53 and @xmath63 with a @xmath54-function at @xmath53 . in this case a steady state is reached in some time @xmath64 where now @xmath57 is measured in units of @xmath65 with @xmath66 . similarly , the calculation can be made between @xmath67 and @xmath68 , and so on until some @xmath69 is reached where @xmath70 . then the total time required for the turbulence to cascade from @xmath52 to the dissipation range is just @xmath71 . but @xmath72 and similarly for the other @xmath73 . but in this scale - free calculation , @xmath74 , and thus @xmath75 , and so @xmath76 . to a first approximation all of the @xmath77 are equal , though it could be argued that as one goes to smaller and smaller scales the relevant values of @xmath77 decrease due to an increasing effective viscosity , especially in the presence of magnetic fields . by setting all @xmath78 a lower limit on the value of @xmath79 is obtained , and this is the most constraining value for the issue of concern here . then @xmath80 , and since @xmath81 the time for the turbulent flow to reach the dissipation range and begin to heat the ambient medium is @xmath82 . % $ $ \ ] ] thus a good approximation of the time for the turbulent cascade to reach the dissipation stage is just the number @xmath83 of large scale eddy turnover times that are required for the establishment of a steady state turbulent cascade over a factor of about @xmath84 in scale and beginning with the scale of the large scale energy injection . i.e. , @xmath85 . in this model the energy injection occurs on the scale of the largest eddies when the outflowing jet first becomes fully turbulent from outer boundary to centerline . in the limiting case the energy injection spectrum could be taken as a delta function centered on @xmath86 , and the energy injection function in eq . 2 would have the form @xmath87 ) . since the goal of the calculation is to determine the time required for the turbulence originating on a scale @xmath25 to cascade down to the dissipation range , the injection spectrum should be some sharply peaked function centered on @xmath25 . a narrow gaussian form may be more cosmetically appealing but will be seen to have little effect on the outcome of the calculation . the value of the constant @xmath88 controls the amplitude of the energy injection spectrum , and because the calculation is looking at the minimum time required for the turbulent cascade to reach the dissipation range , values of @xmath88 near the maximum allowed are of interest . the units of @xmath88 are ( kinetic ) energy per unit mass per unit time , and in the dimensionless units used here , @xmath88 should be some fraction of @xmath89 , where @xmath90 is the value of the jet radius at which the flow becomes fully turbulent throughout its volume . the maximum value of @xmath88 is then equal to one . a `` reasonable '' choice might be to set an average large scale eddy size to half @xmath90 and the similarly for some average rotation speed of those eddies as @xmath91 , which gives a value of @xmath92 in dimensionless units where @xmath93 and @xmath94 . in fact , it will be seen that values of @xmath88 between 1.0 and 0.1 yield roughly the same results . introducing a time variation in @xmath88 is easily done , but such variations could only be less than the maximal value and thus defeat the purpose of finding minimum equilibration times . hence @xmath88 will be kept constant . a major virtue of the formalism used here is that the calculation is carried out with logarithmic intervals in @xmath22 space , and though this introduces some complications to ensure that energy transfer among structures in neighboring regions of wavenumber space is not incomplete , it allows the calculation to cover many orders of magnitude in scale and thus permits the development of a turbulent cascade through the loss - free inertial range . in general the calculations here cover almost four orders of magnitude in scale in three dimensions , which would require a major computational effort in a direct numerical simulation . most calculations used a wavenumber range of @xmath95 $ ] , while more extensive runs extended the upper limit to @xmath96 . the kinetic energy in the edqnm formalism must be positive definite , so the initial energy spectra were set equal to a constant but negligible value of @xmath97 in scaled units at all wavenumbers . this is three to four orders of magnitude below the injection energy and has no effect on the evolution of the energy spectrum resulting from the turbulent mixing layer . the energy levels at the upper and lower ends of the wavenumber range were not fixed after initialization . in practice there is no significant inverse cascade of energy in the absence of helicity , so the value of @xmath98 at the low end of the wavenumber range ( largest scale ) does not evolve from its initial value . the energy at the largest wavenumber @xmath48 begins to fall immediately due to the very rapid energy transfer on the smallest scales , but this value always remains non - zero . the major free parameters in these calculations are the shape and amplitude of the energy injection spectrum @xmath99 . other parameters that can be varied are the wavenumber range , the length of time covered by the calculation , the value of the viscosity to test for any influence on the spectral evolution , and the density of points in wavenumber space . for most of the calculations the shape of @xmath99 was taken to be a delta function centered at the energy injection scale @xmath25 , as discussed above . in some runs this shape was changed to a gaussian distribution in the log of injection energy sharply peaked at @xmath100 , and with @xmath101 in log@xmath22 . the values of @xmath88 ranged from the maximum of 1.0 to 0.01 , with the most common value being 0.25 as described previously . the normalization constant for the gaussian runs was chosen so that the peak of the distribution had the same amplitude at @xmath102 as did the delta function energy injection , thus providing a comparable energy injection rate . the scaled wavenumber range was in most cases from 0.015 to 128 , with a few runs being made over the range @xmath103 $ ] . values of the time evolution of the turbulent energy spectra were calculated at equally spaced logarithmic intervals in the @xmath22 domain ; some runs were repeated with the density of points increased by a factor of two to see what effect , if any , would result from changing the size of @xmath104 . as described above , values of the viscosity @xmath28 were chosen so that an effective @xmath24 lay near the value of @xmath48 . various small changes in @xmath28 were made in some runs to find the smallest value of @xmath28 that would not produce any `` feedback '' from the upper end of the @xmath22 range into the larger scales . such unphysical inverse cascades were immediately apparent by the growth or `` bunching up '' of elevated @xmath98 values at the top end of the wavenumber range . adequately high dissipation to ensure only a direct and unimpeded cascade to smaller scales ( large @xmath22 ) was always shown by a rolloff in values of @xmath98 from a power law at the highest wavenumbers . the timestep was determined for each run via the effective courant condition described earlier ; increases in @xmath105 much above this value immediately result in zero or negative values of @xmath98 at the highest wavenumbers . the objective of these calculations is a determination of the time elapsed from the onset of driven turbulence on some large scale to the transformation of the turbulent energy into heat at the dissipation scale . this first occurs when the turbulent cascade reaches an equilibrium or steady state condition with energy flowing in at @xmath25 and being dissipated at @xmath24 after transit through a loss free cascade to ever smaller scales . establishment of this equilibrium is an asymptotic process , hence the time required for equilibrium will be set at that time when any changes in the energy spectrum per unit time are less than a few percent . this will give a minimum time for establishment of equilibrium and will result in the minimum time for feedback into heat to occur . the approach to equilibrium is slightly complicated because there is no well defined `` front '' of energy marching down to smaller scales into a constant background of very low turbulent energy . instead , as soon as the calculation is begun , energy begins to flow to higher wavenumbers , as it should . this means that the initial constant and very small value of @xmath98 set at the beginning shows a decay at large @xmath22 before the energy flow from the injection region arrives at the smallest scales . hence the shape of the spectrum evolves over the full wavenumber range for all times greater than zero . figure 1 , which shows the early evolution of the turbulent kinetic energy spectrum as a function of time , illustrates this behavior . in this case the energy injection spectrum has a delta function form for @xmath29 at @xmath106 , and time is measured in units of the @xmath106 eddy turnover time , @xmath107 . at early times ( @xmath108 ) the effects of the energy drain at large wavenumber is very evident , since the spectrum in this range is reduced far below the initial constant value of @xmath97 described in section 3.2 . nonetheless , at @xmath109 the flow of energy from the injection region at @xmath106 to smaller scales is clearly seen , with a `` front '' of the energy cascade having arrived as far as @xmath110 . by @xmath111 the effects of energy injection have reached to scales of order @xmath34 smaller than the injection region , but , as figure 1 shows , the spectrum has not yet reached an equilibrium state by @xmath112 . figure 2 shows the late time evolution of the kinetic energy spectrum , and there it is clear that for scales smaller than the injection scale , there is very little evolution in the spectrum at times later than @xmath113 , with little or no evolution past @xmath114 . at larger scales there continues to be a slow growth of turbulent energy on scales of up to 10 times the size of the injection region , consistent with previous calculations . however , this region is of no concern for the present problem , and in addition the applicability of homogeneity on these scales is less certain for the jet structures considered here . the region @xmath115 has no effect on small scale evolution , and in the absence of helicity almost all the injected energy cascades to the small scale regions ( e.g. , de young 1980 ) . figure 3 shows the resultant spectrum at @xmath113 when the wavenumber space is extended to @xmath116 , thus spanning over 5 orders of magnitude in scale . as discussed above , this is a practical maximum for a single calculation due to the time step constraints at the smallest scales . figure 3 also shows a comparison between a @xmath54 function energy injection at @xmath106 and a gaussian energy injection on the same scale with the normalization discussed in the previous section . it is clear that there is no significant difference in the late time spectra that result from the two different energy injection functions . also evident in figure 3 is the establishment of a clear power - law energy spectrum for times greater than @xmath113 . finally , figure 4 shows the time evolution of the total ( kinetic plus magnetic ) energy for the @xmath54 function injection case that covers the range @xmath117 . from this figure it is clear that the turbulence has not reached an equilibrium state for times less than @xmath118 . however , it does seem that equilibrium is established by @xmath114 . the early rapid growth of @xmath119 for times less than ten is due to the constant inflow of injected energy with no loss at the high wavenumber end having yet occurred . the slower growth for @xmath120 reflects the evolution of the spectrum after energy loss begins but is not yet in equilibrium . in addition , after @xmath111 the conversion of kinetic energy into magnetic energy begins to become noticeable at the few percent level . throughout all these calculations , the total turbulent energy is always dominated by the kinetic energy ; the maximum magnetic energy integrated over all scales never exceeds about 70 percent of the kinetic energy . additional calculations were made with variations of other parameters mentioned in the previous section , such as variation of @xmath121 , @xmath104 , @xmath105 , @xmath122 , and @xmath28 . however , these variations did not significantly alter the overall results presented in figures 1 though 4 . in particular , the variation of @xmath121 , which might be thought to have the largest effect , was not particularly notable . calculations were made with values of @xmath121 varying from 0.1 to 1.0 , and while the amplitude of the turbulent energy spectrum increased as @xmath121 was increased , the shapes of the equilibrium spectra were virtually invariant over an order of magnitude change in @xmath121 . this is not surprising given the tendency for the turbulent mhd cascade to evolve toward a common power - law form . moreover , the scaled time required to establish equilibrium was about @xmath123 in all cases , which is consistent with a scale free calculation . variations in @xmath105 and @xmath28 have been discussed above ; these are controlled by the courant condition to preserve positive definite energies and to avoid unphysical inverse cascades of energy at large wavenumber . the effects of changing the upper limit on the wavenumber range are shown in the figures ; a more clear definition of the loss - free power - law energy cascade is seen with larger @xmath122 , but the time required to reach equilibrium is not affected . this is also an expected result , since equilibrium is reached most rapidly at the smallest scales ( largest @xmath22 ) . decreasing the value of @xmath104 and increasing the density of computational points has the expected effect of making the energy transfer among triads of wavenumbers slightly more efficient . a doubling of the density of points in @xmath22 results in maximum change in @xmath124 of 7 percent at the upper end of the wavenumber range . again , the time to reach equilibrium , which is governed by processes on the larger scales , is unaffected . thus a robust conclusion that emerges from these calculations and which is illustrated in figures 2 and 4 is that the turbulent cascade reaches equilibrium over the wavenumber range @xmath125 in about 100 large scale eddy turnover times ; @xmath126 . this result can then be combined with equation 3 for the time required after the onset of energy injection to reach turbulent dissipation of energy into heat ; @xmath127 \\ \approx ( n\ell_{1}/v).\ ] ] thus these results imply that a value of @xmath128 is appropriate for these turbulent jet flows . with this result in hand it is now possible to estimate the time required for radio agn outflows to convert their outflow energy into heat . to do this requires a conversion from the scale free calculation here into the relevant astrophysical quantities . the first scale to establish is the spatial scale that corresponds to @xmath25 where the energy is injected . this is basically the size of the large scale fully developed turbulent eddies which are driving the turbulence , powered by the vorticity generated in the shearing motions of the mixing layer . as described earlier , the mixing layer grows in thickness until it permeates the entire jet , at which point the flow is fully turbulent , and it is at stage that the interior of the jet can be characterized as a homogeneous and isotropic ( in a comoving frame ) turbulent flow . hence the size of the large scale eddies at the point of complete turbulent infiltration or saturation is an appropriate measure of the energy injection scale . given the internal structure of mixing layers , a good approximation for this eddy size is just the radius of the outflowing jet at saturation , since that is where the linearly expanding mixing layers merge together and the potential flow in the core of the jet ceases . experimental data on gaseous turbulent jets also show that the large scale structures at this point are comparable to the jet radius ( dimotakis et al . 1983 ) . in addition , it is at this point that the opening angle of the outflow increases slightly and takes on the common value of @xmath129 seen in fully turbulent round jets . ( some experimental data for turbulent jets also show the creation of larger scale structures that develop in the downstream flow , but these are often attributed to merging or pairing of large eddies formed earlier rather than an indication of an energy injection scale ( e.g. , brown & roshko 1974 , dimotakis et al . in addition , these structures are most apparent in fluid turbulent jets and are less dominant in very high reynolds number gaseous jets ( dimotakis et al . 1983 , dahm & dimotakis 1990 ) . ) determination of the jet radius when it becomes fully turbulent can be estimated from analytic approximations to @xmath130 or from observations of the early stages of fr - i outflows . the latter is probably the more robust method , given the uncertainty in the initial jet densities and outflow speeds . radio observations of the inner regions of 3c 31 and 3c 296 ( hardcastle et al . 2002 , laing et al . 2008 , hardcastle et al . 2005 ) permit an estimation of the jet width in its earliest turbulent phase . in particular , the southern jets in these objects show an increase of the jet opening angle to @xmath131 degrees that marks the transition to self - similar fully turbulent outflow . estimates of the jet radii at these points give values in the the range @xmath132 kpc , and thus a reasonable estimate from these data for the scale corresponding to @xmath133 is about @xmath134 kpc . a consistency check can be made by looking at the corresponding values of @xmath130 and the density ratios and outflow speeds that would be required . for example , in 3c 31 the estimated point where the jets appear fully turbulent is at a distance of about 1.7 kpc from the nucleus , while in the case of 3c 296 this distance is about @xmath135 kpc . using this as a rough measure of @xmath130 , and assuming that the functional form for plane mixing layers given above can be generalized to mixing layers for round jets , given the overall similarity of the mixing layers in the two cases ( e.g. , freund et al . 2000 , dimotakis et al . 1983 ) , then it is possible to see if `` reasonable '' values of outflow densities and speeds near the nucleus result . the jet becomes fully turbulent at a radius @xmath136 , and using @xmath137 and @xmath138 kpc as an average of the above numbers , together with the observed temperatures and inferred number densities for the ambient medium in 3c 31 and 3c 296 at about 1.5 kpc from the their nuclei ( hardcastle et al . 2002 , 2005 ) gives jet flow speeds in the innermost regions of @xmath139 cm s@xmath7 and jet number densities of @xmath140 @xmath141 , this is consistent with models of `` light '' jets which are mildly relativistic in the innermost regions ( e.g. , laing et al . thus different and independent approaches indicate that a value of @xmath142 kpc is a consistent value to take for the scale of the energy injection range , and this value will be taken to be the size corresponding to @xmath25 . once a value of @xmath90 is established , the remaining scales can be fixed by setting a velocity scale , and this is usually chosen to be the turbulent turnover speed of the large scale eddies in the energy range . the flow in the mixing layer is subsonic relative to the ambient medium as argued previously , perhaps transsonic at most in the initial stages . hence the value of the eddy turnover speed @xmath143 will be some fraction of the local sound speed in the ambient medium , and this in turn has been established for some fr - i objects via observations of x - ray emission in the central regions of the parent galaxies . these observations are consistent with thermal bremsstrahlung radiation , and thus a temperature and sound speed for this gas can be established . in particular , chandra observations of 3c 31 and 3c 296 ( hardcastle et al . 2002 , 2005 ) are consistent with gas temperatures of order @xmath144k , with corresponding sound speeds of @xmath145 cm s@xmath7 . thus a an average value of @xmath146 in the energy injection range is consistent with observations and will be used here . this value is significantly less than the above estimate of @xmath147 in the innermost region of the jet , and so it is also consistent with significant deceleration having occurred , especially in the outer layers of the jet , due to mass entrainment via the growing turbulent mixing layer . once these scales are established , a time unit of @xmath148 , the large eddy turnover time , follows naturally , as does the unit of turbulent kinetic energy density per unit mass , @xmath149 . this last quantity is the unit ( per unit wavenumber ) used in the energy spectra shown in figs . 1 - 3 . the previous two sections provide the information needed to establish an estimate of the time required from the establishment of a fully turbulent outflow in an fr - i jet to the onset of conversion of turbulent energy into heat . as argued above , both observations and analytic approximations suggest a large scale eddy size at the energy injection region of @xmath142 kpc , with an accompanying turbulent eddy speed of @xmath150 cm s@xmath7 . this gives a time unit for the turbulence calculations of @xmath151s or @xmath152 yr . the nonlinear turbulence calculation shows the establishment of an equilibrium spectrum in about 100 large scale eddy turnover times , which gives an estimate of the time for the onset of conversion of outflow energy into heat of about 100 million years . these times are comparable to or less than the bremsstrahlung cooling times deduced from x - ray observations of hot gas in and around the host galaxies of fr - i radio sources ( e.g. , hardcastle et al . 2002 , 2005 ; laing et al . 2008 ) . where is this heat deposited ? the above time estimate is a measure of the beginning of the conversion into heat ; it presumably continues for a time comparable to the lifetime of the agn outflow . this injection of heat can not take place within the confines of the outflows that are observed to be well defined jets , since the energy density in these regions is so high that conversion of all the directed kinetic energy into heat at that point would result in pressures high enough to decollimate the jet . hence there is a need to know how far the jets will propagate in the @xmath0 years after they become fully turbulent . the mean outflow speeds in these objects are unknown , but an estimate can be gained from the arguments that the jets are fully turbulent flows . as mentioned before , such jets are self - similar flows , and experimental data show that the on - axis outflow speed in such jets is approximately 10 times the mean outflow speed on the surface of the jet ( e.g. , pope 2000 , hussein et al . 1994 ) . from previous considerations the outflow speed on the surface of the jet is of the order of or less than the local sound speed , and this has been estimated to be @xmath44 cm s@xmath7 in the region of turbulent energy injection . thus in this phase of the jet propagation a mean outflow speed of @xmath0 cm s@xmath7 may be used , which would place the onset of conversion into heat at a distance of 100 kpc or so from the onset of the fully turbulent phase . this is well beyond the collimated outflow region of most fr - i objects , and hence deposition of heat in this region would not be in conflict with the observed properties of the jets . moreover , this region may lie in that portion of the outflow where the agn flow has become much more broadly dispersed and less well collimated . this decollimation process is necessary if radio agn feedback is to work , since the heating must take place over a large volume in order to provide significant suppression of star formation . the 100 kpc estimate is somewhat uncertain , since beyond ten kpc or so from the nucleus the radio morphology of fr - i outflows shows a clear departure from simple self - similar flow . however , it could be that there is very little deceleration beyond this point , and that most of the mass entrainment is completed by this time . the jet morphologies are not inconsistent with this , in which case the outflow speeds near the centerline of the jet remain about the same as they were in the self - similar regime , and the above estimate of @xmath153 kpc retains some validity . this is suggestive , but hardly definitive . there is in addition the fact that centerline outflow speeds in self - similar jets decrease linearly with distance , though the relevant constant in this relation is unknown ; hence there will be some , possibly non - negligible , deceleration of the jet from the onset of self - similarity until the @xmath154 kpc scale is reached . the late stages of the evolution of these flows raises the issue of their relation to the agn outflows seen in the cores of rich galaxy clusters . much work has been done on the role of `` radio bubbles '' in the icm , relic and otherwise , and their possible influence on the evolution of the icm and brightest cluster galaxies ( e.g. , fabian et al . 2002 , brggen & kaiser 2001 , fabian et al . 2002 , de young 2003 , jones & de young 2005 , oneill et al . 2009 ) , and a natural question arises about the relation of these objects to the flows discussed here . in general , most fr - i agn do not show radio `` bubbles '' , nor are they usually found in the cores of rich clusters . for those that are , it it likely that relatively low powered fr - i jets would be rapidly decelerated and confined by the high ambient pressures and could slowly inflate buoyant cavities in the hot icm , as has been suggested by many models . such cavities are usually small for low radio powers , a few tens of kpc , and the jet structures feeding them are even smaller ( e.g. , mcnamara & nulsen 2007 ) . by contrast , the fully turbulent fr - i jets discussed here are themselves tens of kpc in length before the onset of complex large scale flow patterns such as bends or meanders . hence there is likely to be a continuum in radio source morphology for a given radio power as one moves from very high pressure regions in the centers of rich clusters to lower pressure environments found in small clusters , groups , and individual galaxies . the onset of mixing layers , entrainment and deceleration could occur in all these environments , as has been argued here , with the process occurring on smaller scales in more dense and high pressure environments ( cf . however , one important added factor in the dense icm in cluster cores is the presence of non - negligible external magnetic fields . although the kh instability and resultant mixing occurs unabated in mhd flows with weak fields ( ryu , jones & frank 2000 ) , if the average magnetic energy density becomes comparable to the kinetic energy density , then the mixing layer picture may have to be modified , especially if the external fields have long coherence lengths . in particular , expansion of an inflated `` bubble '' into a magnetized medium may not result in the late stage mixing that simple hydrodynamic flows would suggest ( de young 2003 , de young et al . 2008 ) , so in the case of cluster cores the final step in agn feedback may be more difficult to accomplish . this issue is at present still unresolved . though the creation of bubbles and cavities may be the end state of the flow in high pressure environments , in lower pressure regions the final state could be much more diffuse and distorted due to large scale external magnetic fields or to any flows in the circumgalactic gas caused by galaxy orbits or merging events . the detection ( or not ) of such faint extended regions is very much a function of x - ray and radio telescope sensitivity and selection effects . in this context is it very interesting to note that rudnick & brown ( 2009 ) have recently observed very large , diffuse polarized radio structures surrounding the fr - i object 3c 31 . though further observations are needed , such very large structures could be the signature of the decollimated and widely dispersed portions of radio agn outflows where the heat is deposited from fully turbulent jets . a similar widely dispersed radio emitting region has been known for some time to surround the collimated outflows from the fr - i source associated with m87 ( owen et al . thus the issue of where the turbulent energy is actually deposited remains an important unresolved problem . it was argued above that a distance from the nucleus of @xmath153 kpc was not unreasonable for the onset of heat deposition , but also discussed there were the uncertainties associated with agn outflow speeds for distances greater than @xmath154 kpc . for isolated galaxies , heat deposition at @xmath153 kpc is clearly too distant for thermal suppression of star formation in the near galactic halo , yet the presence of coherent flows on scales @xmath155 10 kpc argues for conversion to heat beyond 10 kpc . the picture is further complicated by low surface brightness radio data for some fr - i objects ( e.g. , 3c 296 , m87 ) ; these show the presence of faint radio emission very close to the stellar extent of the galaxy , and the implication here is that agn outflow material from earlier epochs may have been advected or diffused back into the near circumgalactic regions . these data could imply that deposition of heat is possible into inner galactic halo at distances a few tens of kpc from the nucleus , probably on timescales in excess of @xmath156 years . however , it seems clear that if energy deposition is needed for isolated galaxies in the inner regions of the ism , further investigation into the very late time evolution of outflows is needed to see if this process is viable . the object of this paper is an exploration of the feasibility of radio agn feedback . the constraints on this process are that it must heat the interstellar and circumgalactic medium around the host galaxy in a time less than or the order of the ambient gas cooling time and with enough volume coverage so that significant star formation can be suppressed and the host galaxy will appear to be undergoing passive evolution . the issue of radio agn feedback in massive , rich clusters of galaxies is not specifically addressed , as this process has been treated in many other papers . instead , the more general case of radio agn feedback for galaxies in small groups and clusters and even in isolated systems is considered here . the emphasis is to proceed from the basic properties of agn outflows to see if a continuous chain of physical processes can be established that will lead to the required heating in the required time . the analysis has a small number of fairly simple steps . first , observations show that the most common class by far of radio agn in universe out to @xmath157 and perhaps beyond is the class of fr - i radio sources . second , the morphology of these objects strongly suggests that their bipolar outflows are subsonic or transsonic beyond a few kpc from the nucleus . third , given the evidence for relativistic outflows very close to the nucleus together with the nearly universal onset of global surface instabilities in such shearing flows , this slow outward motion strongly suggests that it is due to deceleration of the flow via the formation of non - linear mixing layers and their accompanying mass entrainment on the surface of the jet . fourth , experimental evidence argues convincingly that such turbulent mixing layers will soon penetrate throughout the jet volume , resulting in a fully turbulent flow in the interior of the bipolar outflows . this conclusion is reinforced by the opening angle of the radio jets observed in some nearby fr - i objects ; this angle is completely consistent with the experimental data on fully turbulent self - similar jet outflows . fifth , it is well known that such turbulent flows eventually cascade to smaller and smaller structures until a dissipation range is reached where most of the turbulent kinetic energy is converted into heat . a series of 3d time dependent calculations of the evolution of such turbulent flows is then performed to see if this eventual conversion of collimated outflow energy into heat can take place in a time short enough to be of interest for agn feedback . the answer appears to be yes , with the first onset of heating occurring roughly @xmath156 years after the development of fully turbulent flow ; this time is comparable to or less than the cooling times derived from x - ray observations of gas surrounding nearby fr - i radio galaxies . finally , the question of where this heat is deposited and whether or not the deposition is sufficiently isotropic is briefly addressed . to retain the observed radio source structure , heating probably occurs beyond the extent of the coherent jet outflow . estimates of outflow speeds for fully turbulent jets indicate energy deposition on scales as large as @xmath9 100 kpc , which is well beyond the coherent jet structures for most fr - i objects . isolated galaxies may require heating on smaller scales , and there is some evidence this might be occurring due to late time evolution of the flows . recent observations of very large scale and diffuse radio structures around 3c 31 and earlier observations of m87 suggest that the suggested large scale heat deposition may be taking place . these simple arguments , while suggestive , are not conclusive . it seems likely that the idea of slow and turbulent outflows from fr - i radio agn is either almost certainly right or very wrong , the latter because we have missed the fundamental nature of jet outflows from agn . for example , if these flows are `` poynting flows '' that are completely dominated at all times by magnetic fields and have very few particles ( e.g. , li et al . 2006 ) , then the nature of the interaction with the surrounding medium might be very different , with little or no role for mhd or hydrodynamic turbulence . however , it is not clear how such magnetically dominated outflows can reproduce the morphology observed in the fr - i radio sources . even if turbulent jets are the correct description , there are details of their evolution suggested by observations that may mean departures from the isotropic turbulence in their interiors that is used here . deflections , bending , or propagation into decreasing density gradients or into inhomogeneities may all cause departures from this condition and perhaps changes in the time required for turbulence to decay into heat . the magnitude of such departures is currently unknown , and they may be small enough that the above treatment remains a good approximation , especially in the interior of the jet . finally , the large scale distribution of energy from radio agn outflows remains a largely unsolved problem . the observational hints are suggestive , but more observations of faint and diffuse radio emission around fr - i outflows are needed , as are additional calculations of the very late stages of these outflows . portions of this work were completed during the aspen center for physics astrophysics programs in 2008 and 2009 . thanks are due to bob rosner and to tom jones for illuminating discussions and to jessica moy for bibliographic assistance . presented here are some additional concepts and relations that lie behind the derivation of equation ( 2 ) . most of the basic framework of the edqnm method is given in the main body of the paper , and complete derivations for all aspects of the method , together with more detailed descriptions of the method and its variants , are provided in the references cited . a particularly complete treatment is found in lesieur ( 2008 ) . where @xmath161 is the magnetic field , @xmath162 is the density , @xmath163 is the fluid velocity , @xmath164 is the pressure and @xmath165 is the magnetic susceptibility . here @xmath28 is the kinematic viscosity , @xmath43 is the magnetic diffusivity , and @xmath166 is a forcing term that represents the injection of kinetic energy into the system . in principal the formalism also accommodates the injection of magnetic energy as well as kinetic and magnetic helicities ( @xmath167 and @xmath168 ) . the treatment described here is restricted in nearly incompressible flows that are almost isotropic in that they admit helicity in the flow fields . extensions of the method to anisotropic and quasi compressible flows are described in the references . as mentioned in the text , the edqnm method derives equations that govern the second order moments of velocity and magnetic field which thus provide the evolution of the kinetic and magnetic energy densities . these governing equations actually determine the behavior of the fourier transformed moments ; because fourier components are global averages over configuration space , the equations provide the evolution of energy on a particular scale over the entire flow and thus provide immediately the time evolution of the energy spectra . in addition , operation in fourier space allows in this formalism the integration of the governing equations using logarithmic intervals in wavenumber space and so permits coverage over many orders of magnitude in spatial scale in three dimensions , a task that would be extremely difficult through direct numerical simulation . the general form of equation ( 2 ) in the text can be obtained by first taking the fourier transform of the above navier - stokes equation , which gives , for no magnetic fields and no helicity , with @xmath162 explicit , the general form @xmath169\ ] ] where @xmath170 is the fourier transform of the velocity @xmath171 , @xmath31 is the fourier transform of the forcing ( energy injection ) term and f.t . is the fourier transform operator . the transformed incompressibility condition becomes @xmath172 , and this condition allows elimination of the pressure from the equation of motion . this is because the incompressibility equation shows that @xmath173 is in a plane perpendicular to @xmath174 , and thus so are @xmath175 and @xmath176 , while the pressure gradient @xmath177 is parallel to @xmath174 . as a result , the f.t . of @xmath178 is the projection onto the plane containing @xmath173 of the f.t . of the non - linear term @xmath179 . thus the form of the fluid equation in fourier space becomes , in general , @xmath180 where the vector notation has been dropped . the exact form of this equation can be found in lesieur ( 2008 ) . it can now be seen that a similar equation can be found for the fourier transform of the second moment of eq . a1 , and it is also clear that the result of this will include a third order term in @xmath170 on the right hand side . similarly an equation for the third order moments will include an integral over a term of fourth order moments ; this difficulty is of course a result of the non - linear nature of the flow arising from the @xmath181 term in the momentum equation . closure of this system of moments is accomplished by assuming that the fourth order moments are nearly gaussian ; hence the `` quasi - normal '' approximation . this approximation allows the fourth order moments to be written as sums of products of the second order moments , and because the third order moments can already be written as a function of second order moments , the system is closed at this level . a discussion of the accuracy of this approximation and tests of its rigor are found in lesieur ( 2008 ) and references therein . thus the origin of the overall form of eq ( 2 ) in the text can be seen . in general the integral in this equation will contain many terms that are quadratic in kinetic energy , helicity , and magnetic energy , and numerical solution of this integro - differential equation is necessary . it can be shown ( lesieur 2008 ) that energy is strictly conserved when integration over wavenumber space includes all triads of turbulent eddies whose wavenumbers ( k , p , q ) satisfy the requirement that @xmath182 ; thus interactions with @xmath183 include non - local interactions in wavenumber space . a complete tabulation of all the terms included on the right side of eq ( 1 ) is found in de young ( 1980 ) . in addition to a closure approximation , two other constraints are added to the right side of eq . ( 2 ) . in its simplest form , the quasi - normal approximation can , at late times , lead to negative energies in some portions of the spectra . this has been shown to be due to an eventual excess value building up in the third order moments . these moments have been shown experimentally to saturate , and a modification to the theory incorporates this effect by adding a linear damping term in the equation governing the growth of the third order moments which represents the deformation rate of eddies of size @xmath184 by larger scale eddies . this `` eddy damping '' stabilizes the growth rate of the third order moments and results in the eddy damped quasi - normal ( edqn ) version of the theory . a final modification guarantees the positive definite character of the spectra in all situations , not just at late times . this is the introduction of a markov process to remove excessive `` memory '' in the system on timescales of order of the large eddy turnover time , including removal of any time variations in the eddy damping term . with this step the formalism takes on the final form of the edqnm theory . specific forms for these modifications can be found in de young ( 1980 ) and lesieur ( 2008 ) . aloy , m.a . , ibaez , j.m . , mart , j - m , gmez , j - l . , & mller , e. 1999 , apj , 523 , l125 benson , a.j . , bower , r.g . , frenk , c.s . , lacey , c.g . , baugh , c.m . , & cole , s. 2003 , apj , 599 , 38 best , p.n . , kaiser , c.r . , heckman , t.m . , & kauffmann , g. 2006 , mnras , 368 , l67 bicknell , g.v . 1984 , apj , 286 , 68 brzan , l. , mcnamara , b.r . , nulsen , p. , carilli , c. , & wise , m.w . 2008 , apj , 686 , 859 bower , r.g . 2006 , mnras , 370 , 645 brown , g.l . , & roshko , a. 1974 , j.fl.mech , 64 , 775 brggen , m. , & kaiser , c.r 2001 , mnras , 325 , 676 brggen , m. , & kaiser , c.r 2002 , nature , 418 , 301 brggen , m. , heinz , s. , roediger , e. , ruszkowski , m. , & simionescu , a. 2007 , mnras , 380 , l67 cara , m. , & lister , m.l . 2008 , apj , 674 , 111 chandrasekhar , s. 1961 , hydrodynamic and hydromagnetic stability ( oxford : clarendon ) , chap . 11 clemens , n.t . , & mungal , m.g . 1995 , j.fl.mech , 284 , 171 cowie , l.l . , songaila , a. , hu , e.m . , & cohen , j.g . 1996 , aj , 112 , 839 croton , d.j . 2006 , mnras , 365 , 11 croton , d.j . , & farrar , g. 2008 , mnras , 386 , 2285 dahm , w. , & dimotakis , p.e . 1990 , j .. fl.mech , 217 , 299 de young , d.s . 1980 , apj , 241 , 81 de young , d.s . 1981 , nature , 293 , 43 de young , d.s . 1986 , apj , 307 , 62 de young , d.s . 1993 , apj , 405 , l13 de young , d.s . 2002 , the physics of extragalactic radio sources ( chicago : univ . chicago press ) de young , d.s . 2003 , mnras , 343 , 719 de young , d.s . 2006 , apj , 648 , 200 de young , d.s . , oneill , s.m . , & jones , t.w . 2008 , aspc , 388 , 343 dimotakis , p.e . , & brown , g.l . 1976 , j.fl.mech , 78 , 535 dimotakis , p.e . , miake - lye , r.c . , & papantoniou , d.a . 1983 , phys.fl . , 26 , 3185 donoso , e. , best , p.n . , & kauffmann , g. 2009 , mnras , 392 , 617 fabian , a.c . 2000 , mnras , 318 , l65 fabian , a.c . , celotti , a. , blundell , k.m . , kassim , n.e . , & perley , r.a . 2002 , mnras , 331 , 369 fanaroff , b.l . , & riley , j.m . 1974 , mnras , 167 , 31p ferrari , a. 1998 , araa , 36 , 539 freund , j.b . , lele , s.k . , & moin , p. 2000 , j.fl.mech , 421 , 229 gomez , t. , sagaut , p. , schilling , o. , & zhou , y. 2007 , phys . 19 , 48101 graham , j. , fabian , a.c . , & sanders , j.s . 2008 , mnras , 386 , 278 hardcastle , m.j . , alexander , p. , pooley , g.g . , & riley , j.m . 1997 , mnras , 288 , l1 hardcastle , m.j . , worrall , d.m . , birkinshaw , m. , laing , r.a . , & bridle , a.h . 2002 , mnras , 334 , 182 hardcastle , m.j . , worrall , d.m . , birkinshaw , m. , laing , r.a . , & bridle , a.h . 2005 , mnras , 358 , 843 hussein , h.j . , capp , s. , & george w.k . 1994 , j.fl.mech 258 , 31 jackson , c.a . , & wall , j.v . 1999 , mnras , 304 , 160 jones , t.w . , & de young , d.s . 2005 , apj , 624 , 586 laing , r.a . , parma , p. , de riter , h.r . , & fanti , r. 1999 , mnras , 306 , 513 laing , r.a . , & bridle , a.h . 2002 , mnras , 336 , 328 laing , r.a . , & bridle , a.h . 2004 , mnras , 348 , 1459 laing , r.a . , bridle , a.h . , parma , p. , feretti , l. , giovannini , g. , murgia , m. , & perley , r.a . 2008 , mnras , 386 , 657 ledlow , m.j . , & owen , f.n . 1996 , aj , 112 , 9 lesieur , m. , & ossia , s. 2000 , j. turb . , 1 , 7 lesieur , m. 2008 , turbulence in fluids , ( springer ) , 4th ed . li , h. , lapenta , g. , finn , j.m . , li , s. , & colgate , s.a . 2006 , apj . 643 , 92 madau , p. , ferguson , h.c . , dickinson , m.e . , giavalisco , m. , steidel , c.c . , & fruchter , a 1996 , mnras , 283 , 1388 mandelbaum , r. , mcdonald , p. , seljak , u. , & cen , r. 2003 , mnras , 344 , 776 mcnamara , b.r . , & nulsen , p.e.j . 2007 , araa , 45 , 117 mungal , m.g . , lozano , a. , & van cruyningen 1992 , exp.fl , 12 , 141 okamoto , t. , nemmen , r. , rodrigo , s. , & bower , r.g . 2008 , mnras , 385 , 161 oneill , s.m . , de young , d.s . , & jones , t.w . 2009 , apj , 694 , 1317 orszag , s.a . 1970 , j.fl.mech , 41 , 363 orszag , s.a . , & patterson , g.s . 1972 , phys.rev.lett , 28 , 76 owen , f.n . , & ledlow , m.j . 1997 , apjs , 108 , 41 owen , f.n . , eilek , j.a . , & kassim , n.e . 2000 , apj , 543 , 611 perucho , m. , hanasz , m. , mart , j - m . , & miralles , j - a . 2007 , phys.rev.e , 75 , 56312 pope , s.b . 2000 , turbulent flows ( cambridge : cambridge univ press ) pouquet , a. , frisch , u. , & lorat , j. 1976 , j.fl.mech 77 , 321 pouquet , a. , & patterson , g.s . 1978 , j.fl.mech , 85 , 305 rigby , e.e . , best , p.n . , & snellen , i. 2008 , mnras , 385 , 310 ruszkowski , m. , brggen , m. , & begelman , m.c . 2004 , apj , 615 , 675 rudnick , l. , & brown , s. 2009 , apj , 137 , 145 ryu , d. , jones , t.w . , & frank , a. 2000 , apj , 545 , 475 sanders , j.s . , & fabian , a.c . 2007 , mnras , 381 , 1381 sanders , j.s . , & fabian , a.c . 2008 , mnras , 390 , l93 sijacki , d. , springel , v. , di matteo , t. , & hernquist , l. 2007 , mnras , 380 , 877 spitzer , l. 1962 , the physics of fully ionized gases ( new york : wiley - interscience ) springel , v. et al . 2005 , nature , 435 , 629 staquet , c. , & godeferd , f.s . 1998 , j.fl.mech 360 , 295 tegmark , m. et al . , 2004 , phys.rev.d , 69 , 103501 turner , l. , & pratt , j. 2002 , j. phys . a , 35 , 781 vedula , p. , moser , r.d . , & zandonade , p.s . 2005 , phys . 17 , 55106 wang , w. , kaiser , c.r . , laing , r. , alexander , p. , pavlovski , g. , & knigge , c. 2009 , mnras , 397 , 1113 white , s.d.m . , & rees , m.j . 1978 , mnras , 183 , 341 white , s.d.m . , navarro , j.f . , evrard , a.e . , & frenk , c. s. 1993 , nature , 366 , 429 willott , c.j . , rawlings , s. , blundell , k.m . , lacy , m. , & eales , s. 2001 , mnras , 322 , 536

this paper describes a geometrical framework for the study of conformally invariant fields in conformally flat spaces in four dimensions . applications to scalar and electromagnetic fields are made . in particular , we show that the equation @xmath0 in @xmath1 , @xmath2 , where @xmath3 is the laplace - beltrami operator , leads under very moderate assumptions to both the maxwell equations and the conformal eastwood - singer gauge conditions @xcite on @xmath4d conformally flat spaces . in addition , our proof of this result allows us to propose a new fiber bundle in which , broadly speaking , the maxwell equations on a conformally flat space are converted into constrained scalar equations on minkowski space . this drastically simplifies the practical calculations . the basic geometrical idea is to build four - dimensional spaces as the intersection in @xmath1 , of a surface and the five - dimensional null cone ( invariant under the linear conformal group so@xmath5 ) . the metric on such a space is induced from that of @xmath1 . in particular , for any given robertson - walker ( rw ) metric , one can always find a surface such that the induced metric is the rw metric . now , to each point of a space obtained in this way corresponds a half line of the cone , hence all spaces can be realized as subsets of the set of the half lines : the cone modulo the dilations . in addition , in the intersection of two such subsets the spaces are related through a weyl rescaling . as a special case , the minkowski space can be obtained in that scheme . this allows us to build a particular atlas of the cone modulo the dilations . in effect , copies of that minkowski space can be obtained by displacing the surface intersecting the cone thanks to the action of o@xmath6 . this generates a covering of the set of the half lines . one can go a step further by introducing coordinates of @xmath7 with the specific property that their restriction to a particular minkowski space of the covering are the usual minkowskian cartesian coordinates . each space of the covering together with its corresponding set of minkowskian cartesian coordinates can be turned into a local chart of the cone modulo the dilations , which is thus endowed with a minkowskian atlas . this atlas is one of the key ingredients to perform simple calculations . in particular , it allows us to use the local weyl rescaling to handle global problems . the study of conformal fields , as the electromagnetic field , viewed as restrictions to four dimensional spaces of fields on @xmath7 can be traced back to the seminal paper of dirac @xcite . there , he introduces the `` six cone formalism '' which sets the conditions , mainly the homogeneity of the fields on @xmath7 , to obtain conformal minkowskian fields . this approach takes into account the so@xmath5 symmetry of the minkowskian equations under consideration from the beginning . this formalism has been extended in group theoretical context by mack and salam @xcite almost @xmath8 years later and in study of conformal generalizations of qed on minkowski and anti - de sitter spaces in the mid 80 s by many authors ( see @xcite and references herein ) . the fact that the conformal group so@xmath5 includes as subgroups , besides the poincar and the anti - de sitter groups , the de sitter group was our starting point in the study of the relation between conformal scalar field in the de sitter and minkowski space @xcite,@xcite . the generalization to the electromagnetic field in a conformal gauge on the de sitter space was tackled in @xcite . there following @xcite we used a method of auxiliary fields to obtain a two - point function in a conformal gauge which reduces to the eastwood - singer gauge @xcite . these works was concerned by free quantum fields , but part of the methods used also applied in the case of classical fields with sources . in that context , the tools previously build , especially the use of minkowskian charts on the set of half lines , allowed us to reproduce , with simplified calculations , the result obtained by higuchi and cheong @xcite for the problem of two charges in de sitter space . the geometrical framework we build up in the present work encompass and generalizes the method we developed in these previous studies . it extends the formalism to a class of conformally flat spaces which contains in particular the spaces endowed with a rw metric . it provides an atlas which permits a global use of the weyl relation to minkowski space . last but not least , it also provides a deeper view on the geometrical nature of the objects involved . this article is organized as follows . the sec . [ sec - geometry ] is devoted to the geometrical framework . we first gives a global view and motivate the main definitions . then we turn to the geometrical formulation and consequences on tensors fields of the assumptions of transversality and homogeneity . this allows us to see how the weyl rescaling emerges in this context . the action of the so@xmath5 group on tensor fields is then discussed . the construction of the minkowskian atlas , and the expression of some properties in it follows . this section ends by the proof that all the rw metrics can be obtained in the present formalism . in sec . [ sec - scalarfield ] , we discuss the conformal scalar fields , we show how both equations and fields on @xmath7 and on a conformally flat space are related . the sec . [ sec - maxwell ] contains the proof of the proposition that the maxwell equations and the eastwood - singer gauge condition are obtained from homogeneous one form fields satisfying the laplace - beltrami in @xmath7 and some transversality requirement . although the result is purely geometrical , the proof makes use of calculations performed in the minkowskian atlas . this make apparent the practical calculation method inherited from this geometrical framework . some generalizations are discussed in sec . [ sec - somecom ] . some properties of homogeneous tensors are reminded in appendix [ sec - homogtensor ] . appendix [ sec - appcoordmink ] resembles a few additional comments on some particular coordinate systems for the sake of completeness . definitions and conventions are collected in appendix [ sec - appendicedef ] . for convenience , the conventions for indices are repeated here : @xmath9 the space @xmath7 is provided with the metric @xmath10 . a point of @xmath7 is denoted by @xmath11 or @xmath12 , the letter @xmath12 refers most often to the linear structure of @xmath7 . this section is intended to give the reader a global view of the formalism . we motivate the introduction of key structures as the cone modulo the dilations or the bundle @xmath13 . we give their definitions and their main properties , leaving the proofs to the forthcoming sections . for convenience a reminder of definitions and conventions is given in appendix [ sec - appendicedef ] . the main geometrical construction is pictured in fig . [ fig - globalpicture ] . a four dimensional manifold @xmath14 is obtained as the intersection of the five dimensional null cone @xmath15 of @xmath7 and a surface @xmath16 , in which @xmath17 is homogeneous of degree one . on a manifold @xmath14 each point is intercepted by a single half line of @xmath15 . conversely , thanks to the homogeneity of @xmath17 , each half line intercepts at most once the manifold @xmath14 . as a consequence each manifold @xmath14 can be realized as a subset @xmath18 of a single set : the set @xmath19 of the half lines of @xmath15 , namely the cone modulo the dilations . this construction can be translated in relations between spaces which are collected in the following diagram @xmath20 } \xymatrix { \setr^6 & \\ x_f \ar[u]^{l_f } & x'_f\subset \mathcal{c ' } \ar[ul]_{l'_f } \ar[l]^{\lambda_f } } \ ] ] in which @xmath21 & \mapsto x_f : = [ x ] \cap x_f , \nonumber\end{aligned}\ ] ] where @xmath22 $ ] is the half line of @xmath15 ( that is a point of @xmath19 ) which contains the point @xmath11 , @xmath23 is the canonical injection , and @xmath24&\mapsto x_f : = [ x ] \cap x_f,\nonumber\end{aligned}\ ] ] which is a diffeomorphism . note that , thanks to the homogeneity of @xmath17 one has @xmath25 , and thus @xmath26 now , a set of half lines may cross several manifolds @xmath14 , in other words the realizations @xmath18 on @xmath19 of these manifolds have a non - empty intersection , in that case one can define some common coordinate systems ( at least locally ) . this property allows us to drastically simplify many practical calculations , this can be explained as follows . firstly , as we will prove in sec . [ sec - minkatlas ] , the cone modulo the dilations as a manifold can be covered by a collection of realizations @xmath18 of minkowski spaces , each endowed with a cartesian system of coordinates . we will call minkowskian atlas this covering together with these coordinates , and minkowskian charts the elements of this atlas . secondly , we will prove in sec . [ sec - weylrelation ] that in their intersection two realizations @xmath18 are related through a weyl rescaling . finally , due to the existence of the above covering all the manifolds @xmath14 are weyl related to a minkowski space , that is are conformally flat . in addition , and ( this is the point for the simplifications ) a conformal equation reads under its minkowskian form in a minkowskian chart . amongst the spaces obtained within this scheme the space @xmath27 , which is obtained as the intersection of the cone with the @xmath28-sphere @xmath29 , provides a realization of the abstract manifold @xmath19 itself . indeed , @xmath29 is intercepted only once by each half line of @xmath15 ( fig . [ fig - globalpicture ] ) . we will use this realization of @xmath19 , which is included in the einstein space , for drawing conformal diagrams in sec . [ sec - minkatlas ] and some practical calculations ( sec . [ sec - scalarfield ] ) . other important structures , more specifically related to the maxwell field , have to be introduced . in the scheme we use , which is inspired by the dirac s `` six cone formalism '' @xcite , the maxwell field @xmath30 , a one form field on @xmath14 , is obtained from a one form field @xmath31 on @xmath7 homogeneous of degree zero . we remind ( see appendix [ sec - homogtensor ] ) that the cartesian components of @xmath31 are then homogeneous of degree -1 . in sec . [ sec - maxwell ] we will prove the following property : for @xmath31 homogeneous of degree zero , @xmath32 homogeneous of degree @xmath33 satisfying @xmath34 the field @xmath35 and the current @xmath36 defined on @xmath14 satisfy the maxwell equations and the eastwood - singer gauge condition @xcite : @xmath37 where @xmath38 is the laplace - beltrami operator on @xmath14 . note that for maximally symmetric spaces the eastwood - singer gauge condition reduces to the more familiar form @xmath39 with a constant ricci scalar @xmath40 . note also that the two conditions appearing in ( [ eq - box6aetdieze - overviewformalism ] ) mean that the vector fields @xmath41 and @xmath42 are tangent respectively to @xmath15 and to @xmath14 . beside the property by itself its proof makes use of a new geometrical structure : the fiber bundle @xmath43 it is obtained in restricting the base space of the cotangent bundle @xmath44 to @xmath14 . this object allows us to solve some problems in a more simple way than by tackling them directly on @xmath14 . it also describes more accurately the transition between the fields on @xmath7 and those on @xmath14 . more precisely , although under the assumptions made in ( [ eq - box6aetdieze - overviewformalism ] ) the maxwell field @xmath30 is directly obtained as the pullback of @xmath31 . the proof of this result involves an intermediate step in which the equations @xmath45 are , in a sense which will be made precise in sec . [ sec - maxwell ] , restricted to the six conformal scalar equations @xmath46 where @xmath47 is the scalar laplace - beltrami operator on @xmath14 and in which both the @xmath48 s and the @xmath49 s belongs to the set @xmath50 of the sections of @xmath51 . these equations reduce ( when the two constraints of ( [ eq - box6aetdieze - overviewformalism ] ) are applied ) to the maxwell equations and the eastwood - singer condition ( [ eq - maes - overviewformalism ] ) . however they are more easier to handle . even more , when quantizing the electromagnetic field , these equations account directly for the gauge fixing : part of the @xmath52 s components leads to the maxwell field , the others carry constraints @xcite . the field @xmath48 is obtained from the field @xmath31 by restricting the base of @xmath44 to @xmath14 and the field @xmath30 from the field @xmath48 by further restricting the fiber @xmath53 to @xmath54 . the relations between @xmath55 and @xmath30 ( also true for @xmath56 and @xmath57 ) are collected in the following diagram @xmath58_{\widetilde{l_f } } \ar[dr]^{l^*_f } & \\ a^f \in \mathcal{b}_f^ * \ar[r]_{{r_f } } & \omega_1(x_f ) \ni a^f } \ ] ] in which @xmath59 such that : @xmath60 , @xmath61 = a(x ) [ v]$ ] , @xmath62 such that : @xmath63 , @xmath64 = a(x)[u ] $ ] . this is a condition on tensor fields which will appear very often . roughly speaking , for a tensor @xmath65 fulfilling this relation , the map @xmath66 has a good behavior with respect to group invariance , weyl relation , equations ... this condition reads as follows : a tensor @xmath67 fulfills the transversality condition if and only if , for any @xmath68 and for any @xmath69 , @xmath70as soon as one of the arguments @xmath71 is equal to the dilation field @xmath72 , @xmath73 . note that , when @xmath74 , one often finds the little bit stronger condition @xmath75 which , for @xmath76 , reduces to @xmath77 ( from which the name `` transversality condition '' originates after dirac s paper @xcite , altough in the context of mechanics this kind of condition is usualy termed `` horizontality '' ) , equivalently in an index - free notation : @xmath78 . this condition will eventually be required for any field considered here . for quantum fields , as usual when a gauge condition is present , the implementation of the constraint at the quantum level ( in a gupta - bleuler quantization scheme ) is done after the quantization . in this section , we consider two manifolds , @xmath14 and @xmath79 defined , as in sec . [ sec - overviewformalism ] , as the intersections of @xmath15 and the manifolds @xmath80 . let @xmath67 be an homogeneous tensor field of degree @xmath81 , we want to compare the fields @xmath82 and @xmath83 . for this purpose we realize both fields on @xmath19 and obtain a weyl relation between @xmath84 and @xmath85 . let us show that if @xmath65 fulfills the transversality condition then , on @xmath86 , the following property holds : @xmath87\right ) = \left(k^f_h([x])\right)^{d(t ) } \left(t'^h\right)\left([x]\right).\ ] ] where we defined @xmath88 ) : = \frac{h(x)}{f(x)}.\ ] ] thus the function @xmath89 appears as the conformal factor . for further references we also define the related one form on @xmath19 @xmath90 note that , no reference to a metric structure is made in these two definitions . in order to prove ( [ eq - weyltenseurshomogenes ] ) , we begin with showing that , for @xmath91 and @xmath92}(\mathcal{c'})$ ] one has @xmath93)\left [ v'\right ] = h(x)l'_{h*}([x])\left[v'\right ] + n,\ ] ] where @xmath94 is a null vector which belongs to a half - line @xmath22 $ ] . we first remark that @xmath95)= \frac{f(x)}{h(x)}l'_f([x])=k_f^h([x])l'_f([x]).\ ] ] then , differentiating the rightmost term , one obtains , for any @xmath68 and @xmath96}(\mathcal{c'})$ ] , @xmath97)[v']= l'_{k_f^h*}([x])[v']l'_f([x])+k_f^h([x])l'_{f*}([x])[v'].\ ] ] the first term of the r.h.s . of this expression is proportional to @xmath11 , and as a consequence , belongs to the half - line @xmath22 $ ] . the expression ( [ eq - lf*etlh * ] ) follows at once . now , let us consider @xmath98}(\mathcal{c'})$ ] . then using in succession the definitions ( [ eq - deflpf])-([eq - deflamdaf ] ) , the homogeneity of @xmath65 , the property ( [ eq - lf*etlh * ] ) and finally the transversality condition for @xmath65 ( sec . [ sec - transvcond ] ) , one has @xmath99)\left[v'_1,\ldots , v'_p\right]\\ & = \left(l'^*_f t\right)\left([x]\right)\left[v'_1,\ldots , v'_p\right]\\ & = t\left(\frac{x}{f(x)}\right)\left[l'_{f*}v'_1,\ldots , l'_{f*}v'_p\right]\\ & = \left(\frac{1}{f(x)}\right)^{d(t ) - p}\!t(x)\left[l'_{f*}v'_1,\ldots , l'_{f*}v'_p\right]\\ & = \left(\frac{1}{f(x)}\right)^{d(t)}\!\!\!t(x)\left[f(x)l'_{f*}v'_1,\ldots , f(x)l'_{f*}v'_p\right]\\ & = \left(\frac{1}{f(x)}\right)^{d(t)}\!\!\!t(x)\left[h(x)l'_{h*}v'_1 + n_1,\ldots , h(x)l'_{h*}v'_p + n_p\right]\\ & = \left(\frac{1}{f(x)}\right)^{d(t)}\!\!\!t(x)\left[h(x)l'_{h*}v'_1,\ldots , h(x)l'_{h*}v'_p \right]\\ & = \left(\frac{h(x)}{f(x)}\right)^{d(t)}\!\!\!t'^h([x])\left[v'_1,\ldots , v'_p\right],\\\end{aligned}\ ] ] which is the announced result . finally , thanks to the isomorphism @xmath100 between the @xmath14 s and the @xmath18 and using the notation @xmath101 for the tensors on the @xmath18 s instead of @xmath102 s , one can recast the above relation ( [ eq - weyltenseurshomogenes ] ) under the more familiar form @xmath103 in which @xmath104 stands for @xmath105)=t^f(x_f)$ ] . in particular , between metrics this relation specializes to @xmath106 which makes apparent that the eq . ( [ eq - weyltenseurshomogenes ] ) is in fact a weyl relation in the usual sense . in this section we specify the action of so@xmath5 on the various objects defined above . in particular , we define the action of so@xmath5 on the sections @xmath48 of the fiber bundle @xmath107 and show that the transversality condition ensures the so@xmath5 invariance of the construction . the natural action @xmath108 of so@xmath5 on @xmath7 yields an action @xmath109 on @xmath14 defined through : @xmath110 setting @xmath111 for any @xmath112 , one obtains the action of the conformal group on the space time : @xmath113 let us recall that , for a scalar field @xmath114 homogeneous of degree @xmath115 the operator @xmath116 is defined through : @xmath117 also , the group acts on @xmath114 through the natural representation @xmath118 . the field @xmath119 is defined on @xmath14 and we now define the representation @xmath120 of the conformal group on it . we just impose that the operator @xmath116 intertwines @xmath120 and the natural representation @xmath121 on @xmath114:@xmath122 that is to say , using that @xmath123 for any @xmath124 , @xmath125 which is the well - known action of the conformal group on conformal pseudo scalar fields of weight @xmath115 . the group acts on a 1-form @xmath31 of @xmath7 through @xmath126 where @xmath127 stands for the spinorial action on the 6 components of @xmath31 . we are going to define a representation @xmath128 of so@xmath5 on the 1-forms of @xmath14 as well as a representation @xmath129 on the sections of @xmath51 . moreover we will prove that the map @xmath130 intertwines these representations as soon as the form @xmath31 fulfills the transversality condition : @xmath131 which reads @xmath132= ( l_g^f)^ca^f(x)[u],\forall x\in x_f,\ \forall u\in t_x(x_f),\ ] ] as soon as @xmath133 . we first build the representation @xmath129 on the field @xmath48 in a very similar way as for the scalar field . we just impose that @xmath134 intertwines the representations @xmath129 and @xmath135 and obtain straightforwardly : @xmath136=(\omega_g^f(g^{-1}x))^{r-1}g\cdot a^f(l_{g^{-1}}^fx)[v],\ ] ] for any @xmath137 , which is _ a fortiori _ true for any @xmath138 . we now define the representation @xmath128 of the group on the forms @xmath30 . as a result , we can not define a representation on the @xmath30 in the same way as above . as we will see , one must impose a condition on @xmath31 . in place of this construction , we define directly the representation @xmath128 through @xmath139= ( \omega_g^f(g^{-1}x))^ra^f(l_{g^-1}^fx)[(l_{g^{-1}}^f)_*u],\ ] ] for @xmath124 and @xmath138 . we begin with calculating @xmath140 on any @xmath138 using the leibniz rule : @xmath141=\frac{g\cdot u}{f(gx)}-g x\left(\frac{\partial f(g x)[u]}{f(gx)^2}\right).\ ] ] the crucial remark is that the second term of the r.h.s . belongs to a line of the cone . this term will be denoted as @xmath142 in the following . at this time we can set down the representation @xmath128 , putting @xmath143 for the readability : @xmath144&=\omega^ra^f(l^f_{g^{-1}}x)\left[\frac{1}{\omega } g^{-1}\cdot u+f\right]\\&=\omega^ra(l^f_{g^{-1}}x)\left[\frac{1}{\omega } g^{-1}\cdot u+f\right]\\ & = \omega^ra(l^f_{g^{-1}}x)\left[\frac{1}{\omega } g^{-1}\cdot u\right]\\ & = ( \omega_g^f(g^{-1}x))^{r-1}g\cdot a^f(l^f_{g^{-1}}x)[u ] , \end{aligned}\ ] ] the before last equality being due to the transversality condition . the result ( [ eq - intertwine ] ) follows immediately from ( [ eq - rho_tilde ] ) and ( [ eq - rho ] ) . as a consequence , the map @xmath145 intertwines the representations @xmath128 and @xmath146 as soon as @xmath31 fulfills the transversality condition . note that the identification @xmath147 allows to realize the group action on @xmath18 as well . as explained in sec . [ sec - overviewformalism ] all the spaces @xmath14 are realized as subsets @xmath18 of @xmath19 through the one to one map @xmath100 . then , one can use the weyl relation ( [ eq - weylmetricxfxh - short ] ) in order to simplify practical calculations in problems involving conformally invariant equations . as well known , these calculations can be much simpler than those performed on @xmath14 especially when the second space @xmath79 is a minkowskian one . unfortunately , one can not in general recover the space @xmath14 with only one minkowskian space . however , we will prove that the covering of the whole @xmath19 is possible using four minkowskian spaces . these can be endowed with coordinates systems in order to form an atlas . in addition , these systems can be chosen to be cartesian minkowskian coordinates to makes calculations simpler . now , the formalism we develop use @xmath7 as framework , in practical calculations the coordinates basis on which fields can be expanded are chosen on @xmath7 . as a consequence , the cartesian minkowskian coordinates of the atlas have to be related to coordinates in @xmath7 . in the sequel , we will call minkowskian systems both the systems in @xmath7 and those deduced from them in the atlas of @xmath19 . this atlas will also be named minkowskian as well as charts that compose it . we will prove that one minkowskian plane @xmath148 together with one minkowskian coordinate system , the @xmath94-system , are sufficient to generate the whole minkowskian atlas . note that , some general comments in relation with such kind of coordinates are made in appendix [ sec - appcoordmink ] . first let us introduce a useful graphical representation of @xmath19 as conformal diagram . remind that @xmath19 can be realized as the intersection @xmath27 of the @xmath28-sphere , obtained through @xmath149 , @xmath150 , with the cone @xmath15 ( see fig . [ fig - globalpicture ] ) . points of @xmath27 and @xmath19 are then identified through the map @xmath151 . this space @xmath27 is naturally endowed with a very convenient global coordinate system @xmath152 obtained by setting @xmath153 in the hyper - spherical coordinate system @xmath154 defined through @xmath155 in which @xmath156 , \varphi\in[0,2\pi]$ ] and @xmath157 correspond to the usual spherical coordinates on @xmath158 . the condition @xmath159 gives a system on the cone , @xmath160 gives a system on @xmath27 . now , since as a result of the considerations of the sec . [ sec - weylrelation ] a space @xmath14 is related to @xmath27 through a weyl rescaling , the subsets @xmath18 can be conformally pictured in the plane @xmath161 . in the sequel we will call `` @xmath161-diagram '' this conformal mapping ( see fig . ( 2 ) for an example . ) . one may note that the metric element on @xmath27 in the @xmath152 system is @xmath162 this form shows that @xmath27 is indeed included ( @xmath163 belongs to a subset of @xmath164 ) in the static einstein space of positive curvature . now , let us consider the space @xmath165 , @xmath166 . as shown in @xcite this space is a minkowski space . the coordinate system @xmath167 defined by @xmath168 ( whose inverse is given for convenience in appendix [ sec - appcoordmink ] ) provides on @xmath169 a chart , called @xmath94-chart in the sequel . the restriction to the cone is obtained for @xmath170 and that for @xmath148 for @xmath171 . the metric induced from @xmath7 on @xmath172 is , in the @xmath173-coordinate basis , @xmath174 , the usual form of the minkowski metric . the points where the above system becomes singular , that is on the subset @xmath175 , correspond to points at infinity in the minkowski space @xmath165 . they are mapped to the boundaries of the @xmath94-chart on the conformal @xmath176-diagram in fig . ( [ fig - northandsouthchart ] ) . similar considerations apply to the space @xmath177 defined by @xmath178 and the coordinate system @xmath179 is obtained by replacing @xmath94 by @xmath29 in ( [ eq - coordnord ] ) . a set of minkowskian charts covering @xmath19 is finally obtained by moving the two surfaces @xmath148 , @xmath180 with some elements of so@xmath5 . more precisely , we obtain new minkowski spaces through the action of the one parameter subgroup of so@xmath5 generated by @xmath181 . the manifolds @xmath182 ( resp . @xmath183 ) defined through @xmath184 are minkowski spaces . corresponding to each case , one can define a coordinate system @xmath185 analogous to @xmath186 , replacing @xmath187 by @xmath188 in ( [ eq - coordnord ] ) and similarly for @xmath29 in place of @xmath94 . as a consequence , we obtain an atlas of @xmath19 , in the usual sense , in which the systems of coordinates make the weyl relations very transparent . now every @xmath18 space is an open subset of @xmath19 and can be endowed with the above atlas . as a final remark , the space @xmath180 can be obtained from the space @xmath148 through the transformation @xmath189 which is also an element of o@xmath6 . as a consequence , all the elements of our minkowskian atlas can be obtained from @xmath148 through a transformation of o@xmath6 that is to say through an ( inoffensive ) cartesian isometric change of variables in @xmath7 . thus , the whole atlas is generated from @xmath148 and the @xmath94-coordinates as announced in the beginning of this section . we note that , for @xmath124 , @xmath190 also , in @xmath94-coordinates the dilation field reads @xmath191 . now , a one - form @xmath74 homogeneous of degree zero , is expanded in the coordinate basis @xmath192 as @xmath193 the last term is for convenience divided by @xmath194 in order that @xmath195 be of the same degree of homogeneity ( zero ) as the other components . one has @xmath196 in particular , the transversality condition reads @xmath197 . the same result applies in others minkowskian coordinates . here we prove that for any robertson - walker ( rw ) metric @xmath198 one can find a function @xmath199 such that the metric induced from @xmath7 on the conformally flat space @xmath200 be the robertson - walker metric @xmath198 . in other words , each robertson walker space can be realized at least locally as a space @xmath201 . following ibison @xcite the set called @xmath202 of conformally flat metrics defined through : @xmath203 @xmath204 being a conformal factor and @xmath205 the cartesian minkowskian coordinates , includes the rw metrics . the notation @xmath206 for the arguments of the function @xmath207 indicates that it depends separately of the time coordinates and of the space coordinates and that the space coordinates appears only through the radius @xmath208 . let us assume now that @xmath207 for some rw metric is given in a minkowskian coordinate system , say the system @xmath209 ( [ eq - coordnord ] ) . the function @xmath210 is invariant under the subgroup so@xmath211 of so@xmath5 generated by @xmath212 . the equation @xmath213 defines a so@xmath211 invariant space endowed with a metric which satisfies ( [ eq - rtmetric ] ) , thus a rw space . in this section , we consider scalar fields on @xmath7 , homogeneous of degree @xmath115 . these fields yield scalar fields on @xmath14 . we show that these fields fulfill the scalar conformal equation as soon as the original field fulfill the laplace - beltrami equation on @xmath7 . let @xmath114 and @xmath32 two scalar fields on @xmath7 homogeneous of degree -1 and -3 respectively , and suppose that @xmath214 we claim that @xmath215 where @xmath38 is the laplace - beltrami operator and @xmath40 the scalar curvature for the space @xmath14 . we consider @xmath27 the realization of the cone up to the dilation defined sec . [ sec - minkatlas ] , on which all @xmath14 spaces realize as subsets . setting @xmath216 and @xmath217 in the hyperspherical coordinates ( [ eq - coordalpbetsurc ] ) and using the homogeneity of @xmath114 , the equation ( [ eq - equationscalairer6 ] ) reads @xmath218 the restriction to @xmath27 ( @xmath219 ) then gives @xmath220 which is the desired result . we obtain the similar result for the other hyper surfaces @xmath14 by using the weyl correspondence between the spaces @xmath14 . note that although straightforward the change of variables leading to the equation ( [ eq - equationscalairecoordr1r2 ] ) is rather cumbersome . in this section we deal with one - forms @xmath31 and @xmath32 of @xmath7 satisfying @xmath221 and such that @xmath31 is homogeneous of degree 0 and @xmath32 homogeneous of degree @xmath33 ( remind that this implies that @xmath222 and @xmath223 are homogeneous of degree @xmath224 and @xmath225 respectively ) . we first show how in the present formalism the equation @xmath221 leads to a set of six copies of the equation of the conformal scalar field . then , we give the proof that @xmath226 satisfies the maxwell equations as soon as @xmath31 fulfills the transversality condition , and that an additional condition on @xmath32 allows us to recover a conformal gauge condition . suppose that @xmath51 is endowed with a coordinate system @xmath228 where @xmath229 is any coordinate system of @xmath14 and @xmath230 a cartesian system of coordinates on @xmath231 . in such a coordinate system the equation on @xmath48 reads @xmath232 where @xmath47 is the scalar laplace - beltrami on @xmath14 and @xmath40 the ricci scalar . this is a straightforward consequence of the fact that , on @xmath7 in cartesian coordinates one has @xmath233 : since in a cartesian coordinate system @xmath234 of @xmath7 the equation @xmath235 reduces to six copies ( one per components of @xmath31 and @xmath32 ) of the scalar equation ( [ eq - equationscalairer6 ] ) , each components @xmath222 and @xmath223 being homogeneous of degree @xmath236 and @xmath237 , one can thus apply the result ( [ eq - equationscalairexpf ] ) of the sec . [ sec - scalarfield ] . two comments are in order concerning equations ( [ eq - equationsurfibre ] ) . in first , they do not present any gauge ambiguity and we proved in @xcite that they allow a quantization of the maxwell field on de sitter space using a gupta - bleuler scheme where the condition of transversality is translated to a condition on states after quantization . in second , they can be used to solve classical propagation problems for the maxwell field : for instance , we considered in @xcite the two - charges problem for the de sitter space . in this section , we implicitly identify the spaces @xmath14 and @xmath18 , writing all the objects on the common manifold @xmath238 . we now prove the statement of sec . [ sec - overviewformalism ] : for @xmath31 homogeneous of degree zero , @xmath32 homogeneous of degree @xmath33 satisfying @xmath239 the field @xmath35 and the current @xmath240 defined on @xmath14 satisfy the maxwell equations and the eastwood - singer gauge condition @xcite : @xmath241 we first prove the result on the minkowskian chart @xmath165 , after what , using the o@xmath6 invariance of the hypothesis and of the conclusion , and the properties of the minkowskian charts , we obtain this result on all the minkowskian charts . in the following , for readability , we note @xmath242 , for @xmath243 and @xmath244 . we first consider the equation @xmath45 , and apply ( [ eq - equationsurfibre ] ) on the minkowski space @xmath165 defined by @xmath187 ( used to build the n - chart ) , this yields the system @xmath245 where @xmath246 is the scalar laplace - beltrami operator on @xmath247 , @xmath248 and @xmath249 . as a second step we express in the @xmath94-chart of @xmath19 ( [ eq - coordnord ] ) the above system of equations ( [ eq - equationscalaire - an ] ) . this leads to the system already obtained in @xcite ( with slightly different notations ) @xmath250 then , we express the two constraints appearing in ( [ eq - box6aetdieze - maxwell ] ) in the @xmath94-chart . the first one is the transversality condition applied to the field @xmath31 , using the formula of ( [ eq - transvcondcn ] ) one has @xmath251 the second condition @xmath252 implies the transversality of @xmath32 since @xmath253 , thus @xmath254 in addition , @xmath252 rewrites @xmath255 = 0 $ ] , which in the @xmath94-coordinates on the cone ( @xmath256 ) reads @xmath257 thanks to the homogeneity of @xmath17 this equation rewrites @xmath258 then , using the definition ( [ eq - defw ] ) , it becomes @xmath259 where the dot refer to the minkowskian metric @xmath260 . using the constraint ( [ eq - aplus=0]-[eq - jcw ] ) , the system ( [ eq - systauxmink ] ) becomes @xmath261 owning to the map @xmath262 the above system leads to @xmath263 on the minkowski space @xmath165 . note that the map @xmath262 becomes obvious in the n - coordinates ( [ eq - coordnord ] ) : for all @xmath264 one has @xmath265 . for the sake of argument , let us introduce a self explanatory symbolical notation for the two operators appearing in the l.h.s . of the maxwell equations and the eastwood - singer gauge condition . with them the above system reads : @xmath266\right)_\mu = j^{{\scriptscriptstyle n}}_\mu\\ & es_{{\scriptscriptstyle n}}[a^{{\scriptscriptstyle n } } ] = j^{{\scriptscriptstyle n } } \cdot w^{(f , f_{{\scriptscriptstyle n } } ) } , \end{aligned } \right .\ ] ] this result is available for the @xmath94-chart . nevertheless , the o@xmath6 invariance of hypothesis and conclusion and the fact that any minkowskian chart can be deduced from the @xmath94-chart by mean of an o@xmath6 transformation , proves that this result is true for any minkowskian chart of our atlas . now , using the same symbolical notation for the two operators appearing in the l.h.s of ( [ eq - maes - maxwell ] ) , and following @xcite to apply a usual ( local ) weyl transformation between @xmath165 and @xmath14 to each eqs . of the system ( [ eq - systmink - w ] ) , one obtains @xmath267\right)_\mu = j^f_\mu\\ & es_f[a^f ] + w^{(f , f_{{\scriptscriptstyle n } } ) } \cdot \left(m_f [ a^f]\right)= j^{{\scriptscriptstyle f } } \cdot w^{(f , f_{{\scriptscriptstyle n } } ) } , \end{aligned } \right .\ ] ] where the dots now refer to the metric @xmath268 and where we have taken into account the homogeneity of both the electromagnetic and the current one - form fields . they have respectively a degree of @xmath224 and of @xmath225 which correspond to conformal weights of zero and @xmath33 for the components of the fields : @xmath269 , @xmath270 . finally , the second equation simplifies and we obtain the announced result on the @xmath94-chart . then , using the above remark on o@xmath6 invariance , we obtain the result on the whole space . note that the maxwell equations are obtained independently of the condition @xmath252 , which is used only to obtain the eastwood - singer gauge condition . note also that , in absence of source ( @xmath271 ) , the eastwood - singer gauge condition is automatically fulfilled as soon as the transversality condition on @xmath31 is fulfilled . it is important to point out that from a physical perspective the initial conditions and currents are given in the space @xmath14 . in particular , the minkowskian currents appearing in ( [ eq - systmink - w ] ) are defined through the equation @xmath272 in order to satisfy the weyl relation . they have in general no physical meaning in the minkowski space . besides the results presented in the previous sections , the formalism depicted in the present paper explicits the geometrical nature of the various objects used in our previous works . this allows us to consider some straightforward generalizations , let us comment briefly about the classical and quantum situations . the method used for the classical propagation problem considered in @xcite extends naturally to the conformally flat spaces @xmath14 . to summarize , given a set of initial conditions an currents in @xmath14 the problem of finding the @xmath30 solution of the maxwell equations in the eastwood - singer gauge ( or in a gauge contained in it ) in @xmath14 amounts to use in the minkowskian atlas the propagation formula established in @xcite , which uses only the minkowskian scalar green s function . the quantization scheme used in @xcite for the free electromagnetic field on de sitter space can be transposed here for the most part . the definition of an so@xmath5-invariant scalar product on the space of the solutions of ( [ eq - equationsurfibre ] ) and the obtention ( from the known mode solutions of ( [ eq - equationsurfibre ] ) ) of a reproducing kernel for the @xmath273 s with respect to that product can be reproduced almost verbatim . the result is : @xmath274 , where @xmath275 is the scalar two - point function on @xmath14 . the construction of the fock space , including the determination of the physical subspace is formally identical . the general form of a covariant two - point function , a bi - tensor on a specific @xmath14 , requires more developments ( note that , such a function has been proposed recently in minkowskian coordinates @xcite ) . this could be the object of future investigations . the authors thank c. cachot for valuables discussions and suggestions about the secs.([sec - minkatlas ] ) and ( [ sec - rw ] ) . for reference , we recall here some properties of an homogeneous @xmath276-tensors @xmath277 of @xmath7 . the set @xmath7 , together with the scalar product @xmath278 @xmath279 , is naturally endowed with a structure of linear space . a point in @xmath7 is located by a vector @xmath12 whose components in cartesian coordinates are @xmath234 . the linear structure allows us to identify the tangent spaces at any point through @xmath280 . this makes sense for expressions as @xmath281 since the tangent space at @xmath12 is identified with the tangent space at @xmath282 . let us consider the dilatation map in @xmath7 @xmath283 @xmath284 being a positive real number . since @xmath285 this is also the result of a push - forward on the vector @xmath12 . in other words for a @xmath276-tensor field @xmath277 of the pullback reads @xmath287 . now , an homogeneous @xmath276-tensor field @xmath277 of degree @xmath288 is defined through the relation @xmath289 thus , homogeneous @xmath276-tensor field @xmath277 of degree @xmath115 satisfy @xmath290 the homogeneity of a tensor field can be related to that of its components in some coordinate basis whose coordinates @xmath291 @xmath292 are each one homogeneous functions of degree @xmath293 . precisely , let us show that : for @xmath294 a @xmath276-tensor field of @xmath7 , homogeneous of degree @xmath295 one has @xmath296 in particular for @xmath297 @xmath298 this result is obvious in the particular case @xmath297 since the eq . ( [ eq - tenshomoliendefpullback ] ) reads in components @xmath299 now , moving to the more general coordinate basis @xmath291 on has @xmath300 where @xmath301 , is homogeneous of degree @xmath302 . taking homogeneity into account in the r.h.s . of the above expression leads to @xmath303 from which the result follows . the system appearing in sec . [ sec - minkatlas ] has already been used in the literature ( see for instance @xcite ) , it is reminiscent of systems called polyspherical systems ( see for instance @xcite ) . one may note that they are framed in such a way that the `` extremal''coordinates @xmath304 are functions of the constraints ( defining the cone @xmath15 and the space @xmath148 respectively ) whereas the `` central '' coordinates @xmath205 are those on the background space @xmath14 . cartesian minkowskian coordinates can be obtained on @xmath14 setting @xmath305 instead of @xmath306 , with this choice conformally invariant equations will appears on their minkowskian form on @xmath14 . for the de sitter space , one may verify that the @xmath94 and @xmath29 minkowski charts correspond to the stereographic projections from the north and south poles from which their names originates . finally , we note for convenience that the inverse of system ( [ eq - coordnord ] ) reads @xmath307 we summarize here the conventions , main structures and maps used in this paper . they are discussed in sec . [ sec - overviewformalism ] . here are the conventions for indices : @xmath9 the coefficients of the metric @xmath308 of @xmath309 are denoted @xmath310 . the definitions of spaces and maps reads : @xmath19 denotes the set of the half lines of the cone @xmath15 , namely the cone modulo the dilations . one has @xmath314 \in \mathcal{c'}$ ] iff @xmath314=\{u \in \mathcal{c } : \exists \lambda > 0 , u = \lambda y\}$ ] the laplace - beltrami operators on @xmath1 and @xmath14 are denoted respectively by @xmath3 and @xmath38 . the operator @xmath327 is the scalar laplace - beltrami operator on @xmath14 : it acts on its argument as if it were a scalar . aaa m. eastwood and m. singer , phys . 107a * , 73 ( 1985 ) . dirac , ann . math . * 37 * , 429 ( 1936 ) . g. mack and a. salam , ann . phys . ( n.y . ) * 53 * , 174 ( 1969 ) . f. bayen , m. flato , c. fronsdal , and a. haidari , phys . rev . d * 32 * , 2673 ( 1985 ) . e. huguet , j. queva , and j. renaud , phys . d * 73 * , 084025 ( 2006 ) . e. huguet , j. queva , and j. renaud , phys . d * 77 * , 044025 ( 2008 ) . s. faci , e. huguet and j. renaud , phys . d * 84 * , 124050 ( 2011 ) . a. higuchi and l. y. cheong , phys . d * 78 * , 084031 ( 2008 ) . s. faci , e. huguet , j. queva , and j. renaud , phys . d * 80 * , 124005 ( 2009 ) . b. binegard , c. fronsdal , and w. heidenreich , j. math * 24 * , 2828 ( 1983 ) . m. ibison , j. math . phys . * 48 * , 122501 ( 2007 ) . s. faci , arxiv:1205.3804 . h.a . kastrup , annalen phys . * 17 * , 631 ( 2008 ) .

the high mass x - ray binary ( hmxb ) system was discovered in 1969 april during a balloon experiment @xcite . the system consists of an accreting neutron star ( ns ) fed by the surrounding stellar wind of the b type emission line companion wray 977 @xcite . a recent luminosity estimate derived from atmospheric models puts its distance at @xmath0kpc @xcite , the value utilized in this paper . the orbital period was established to be @xmath1days @xcite using _ ariel _ 5 observations and was refined with the _ burst and transient source experiment ( batse ) _ to @xmath2days with an eccentricity of @xmath3 @xcite . @xcite discussed a possible orbital evolution and determined an orbital period of @xmath4d , assuming no change in orbital period . @xcite determined that the mass of the companion was in the range @xmath5 and the radius of wray977 was @xmath662@xmath7 , obtained by fitting atmosphere models . the x - ray flux is highly variable throughout an individual binary orbit but follows a distinct pattern when averaged over multiple orbits ( see figure [ fig : orbitlc ] ) . shortly before the periastron passage , the x - ray luminosity increases drastically in the energy band above @xmath8kev , as seen in _ rossi x - ray timing explorer ( rxte)_/ all sky monitor ( asm ) data @xcite . the ns passes closest to the companion at a distance of @xmath9 @xcite . shortly after the periastron passage , @xmath10 , the x - ray luminosity dips for a short period of time . @xcite demonstrated that neither a simple spherical wind model nor a circumstellar disk model around wray 977 are sufficient to describe the observed variations in the folded _ rxte_/asm data . an additional stream component is able to account for the sudden increase in x - ray luminosity , as the ns passes trough the stream shortly before periastron and accretes more material . this model also explains a slightly higher x - ray luminosity around @xmath11 , when the ns passes through the accretion stream a second time . pulsations with a period of @xmath12s were discovered in the _ ariel-5 _ observations @xcite , making one of the slowest known pulsars . the pulse period has varied drastically throughout the last @xmath13 years @xcite . prior to 1984 , the pulse period stayed relatively constant at 695s@xmath14700s and then spun up between 1985 and 1990 to @xmath15s . from 1993 until the beginning of 2008 , the change in the spin reversed again , showing a decline . _ fermi_/galactic burst monitor ( gbm ) data have revealed that experienced another spin reversal and briefly spun - up with the pulse period decreasing from @xmath16s to @xmath17s between may 2008 and october 2010 . since october 2010 , the pulse period has shown only small variations around @xmath17s . most recently , @xcite discovered a peculiar 1ks dip in the luminosity of , where the pulsations disappeared for one spin cycle during the dip . several such dips have been previously observed in vela x-1 @xcite , where it is assumed that the accretion on the ns was interrupted for a short period of time . the pulse phase average spectrum of is described using a power law with a high energy cutoff . the continuum does not show a strong variation in the intrinsic parameters ( @xmath18 , and @xmath19 ) throughout the orbit @xcite , as seen in two data sets from _ rxte _ , taken in 1996 and 2000 , sampling most phases of the binary orbit . one of the major characteristics of the x - ray spectrum of is the high and strongly variable column density of its line - of - sight absorber throughout the orbit , indicative of a clumpy stellar wind ( @xmath20 @xmath21 ) . in addition to the high column density , a very bright fe k@xmath22 emission line can be observed . this line has shown a strong correlation with the observed luminosity , indicating that the line is produced by local clumpy matter surrounding the neutron star @xcite . @xcite used the _ rxte _ data set from 2000 to perform phase resolved spectroscopy and showed that an absorbed and partially covered pulsar continuum ( power law with fermi - dirac cutoff ) as well as a reflected and absorbed pulsar continuum were consistent with the data . a cyclotron resonance scattering feature ( crsf ) at @xmath23kev was first discovered with _ ginga _ @xcite . @xcite found systematic deviations from a power law continuum at @xmath13 and @xmath24kev in _ bepposax _ , where the former could not be confirmed as a crsf due to the proximity of the continuum cutoff . @xcite excluded the existence of a crsf at @xmath13kev and showed that the crsf centroid energy varies between 3038kev over the pulse rotation of the ns . furthermore , they showed that the crsf centroid energy and width are correlated . we report on two observations of performed with the _ suzaku _ satellite in mid 2008 and early 2009 . this paper is structured as follows : section 2 discusses the observations and data reduction . section 3 shows the phase averaged results . section 4 discusses the pulse profiles and phase resolved spectra . sections 5 and 6 discuss the results and conclusions , respectively . _ suzaku _ observed on 2008 august 25 with an exposure time of @xmath25ks ( obsid 403044010 ; hereafter obs . 1 ) . the observation was cut short by a set of target of opportunity observations and was continued on 2009 january 5 , acquiring an additional @xmath26ks exposure time ( obsid 403044020 ; obs . 2 ) . both main instruments , the x - ray imaging spectrometer ( xis ; * ? ? ? * ) and the hard x - ray detector ( hxd ; * ? ? ? * ) were used in these observations . the two observations correspond to orbital phases of 0.19 ( obs . 1 ) and 0.38 ( obs . 2 ) , where obs . 1 falls into the lowest flux part of the binary orbit ( see figure [ fig : orbitlc ] ) . the @xmath27kev absorbed flux was @xmath28erg@xmath21s@xmath29 for obs . 1 and @xmath30erg@xmath21s@xmath29 for obs . 2 ( see table 1 for details ) . both observations were performed using the hxd nominal pointing to enhance the sensitivity of the hxd detectors . the xis detectors consist of two front illuminated ( fi ) ccds xis 0 and 3 , and one back illuminated ( bi ) ccd , xis 1 . all three instruments are sensitive between @xmath31kev and @xmath32kev , although the two fi cameras have a higher sensitivity above @xmath33kev , and the bi chip is more sensitive below @xmath33kev . to minimize possible pile - up , the xis instruments were operated with the 1/4 window option with a readout time of 2s . data were taken in both @xmath34 and @xmath35 editing modes , which were extracted individually with the _ suzaku _ ftools version 16 as part of heasoft 6.9 . the unfiltered xis data were reprocessed with the most recent calibration files available and then screened with the standard selection criteria as described by the _ suzaku _ abc guide . the response matrices ( rmfs ) and effective areas ( arfs ) were weighted according to the exposure times of the different editing modes . the xis data were then grouped with the number of channels per energy bin corresponding to the half width half maximum of the spectral resolution , i.e. grouped by 8 , 12 , 14 , 16 , 18 , 20 , and 22 channels starting at 0.5 , 1 , 2 , 3 , 4 , 5 , 6 , and 7kev , respectively ( m. a. nowak , 2010 , private communication ) . the xis spectral data were used in the energy range of @xmath27kev for reasons described below . the hxd consists of two non - imaging instruments : the pin silicon diodes ( sensitive between 1260kev ) and the gso / bgo phoswich counters ( sensitive above @xmath36kev ) . to determine the pin background , the _ suzaku _ hxd team provides the tuned pin non x - ray background ( nxb ) for each individual observation . in addition , the cosmic x - ray background ( cxb ) was simulated following the example in the abc guide and both backgrounds were added together . the pin data were grouped by a factor of 5 below and by a factor of 10 above 50kev for obs . 1 . in obs . 2 the channels were grouped by 3 throughout the whole energy range . gso data were extracted and binned following the _ suzaku _ abc guide . the pin data energy range of 1560kev and the gso data energy range of @xmath37kev were used for the phase averaged analysis . due to the short exposure times , no gso data were extracted in obs . 1 and in the phase resolved analysis of obs . 2 . is a source with very large and highly variable photoelectric absorption in the line of sight as well as a very strong fe k@xmath22 emission line . for spectral modeling , the calculated responses of the xis detectors showed a ` leakage ' emission level @xcite , which stemmed from the instrument characteristics . that is , a small fraction of an event charge cloud is registered at energies below the peak energy , creating a low energy tail in the spectrum . this ` tail ' is about three orders of magnitude weaker than the intensity of the peak count rate of the measured photon . the response to the fe line is therefore described as a gaussian plus a constant at lower energies , approximately three orders of magnitude below the peak value and extending to below 1kev . due to the strong fe k@xmath22 line and the very strong absorption in this source , the constant level response of the fe line extends above the measured flux below 2kev . this results in significant residuals below @xmath38 kev , which seem to be stronger for the bi spectrum then for that of the fi ccd . figure [ fig : badspec ] shows the phase averaged spectrum from the second observation of and the best fit models using the heasoft generated xis responses as detailed in section 3 . together with the _ suzaku _ guest observer facility ( gof ) , which provided for us an experimental response parameter file where the ` low energy tail ' is ignored , we studied this behavior in more detail . this new response for all three instruments was applied to the individual xis spectra . due to the fact that the new response completely excluded the missing ` tail ' , we now observed an excess of the data with respect to the model below @xmath39kev . the introduction of additional power law and/or black body components could improve the fits at lower energies . due to the fact that the true level of the tail is not known , these additional components can not be interpreted physically , and we decided to use the original response matrix and confine the selected xis energy range to above 2kev . phase averaged broad band spectra in the 260kev ( obs . 1 ) and 290kev ( obs . 2 ) energy ranges were obtained . from the technical description of the xis instruments , it is known that small discrepancies , e.g. , in fitted power law slope , have been observed between fi and bi xis instruments . a difference of @xmath40 in the power law index has been observed in calibration data and was also observed here . when modeling the fi and bi instruments with a common power law index , the residuals of the bi instrument show a systematic deviation . however , only the fi or bi instruments could be modeled together with the hxd instruments . due to this fact and the higher sensitivity of the fi xis instruments above 2kev , we concentrated our discussion on the results obtained with the two fi instruments . each data set , fi and bi , was modeled individually with the hxd data to compare the differences in the continuum . we found that when using the fi and bi instruments individually with the hxd instruments , in both observations the best fit spectral values are consistent with each other within error bars , indicating that the usage of the hxd reduced the observed discrepancies in the power law parameters . the continuum model for pulsars can so far only be modeled with empirical models , consisting of a ` powerlaw ` with a cutoff at higher energies , which is typical for this kind of source @xcite . three empirical models are widely used : the simple high energy cutoff ( ` highecut ` ) and the fermi - dirac cutoff ( ` fdcut ` , * ? ? ? * ) both are used in combination with a simple power law component . in addition , the negative - positive exponential powerlaw model ` npex ` @xcite is a slightly more complicated model including the power law component . for the data analyzed in this work , the best results have been obtained with the ` fdcut ` model : @xmath41 where @xmath42 is the normalization at 1kev , @xmath43 is the cutoff energy , and @xmath19 is the folding energy of the fermi - dirac cutoff . for the obs . 1 and obs . 2 fi data sets the best fit @xmath44 s were 260 and 542 with 245 and 267 degrees of freedom ( dof ) , respectively ( see figure [ fig : spec010 ] and [ fig : spec020 ] ) . replacing the smoother fermi - dirac cutoff with the ` highecut ` model in obs . 2,resulted in slightly worse residuals ( @xmath45 ) compared to the best fit results of the fd - cutoff model ( figure [ fig : spec010 ] d ) and an emission line - like residual at @xmath46kev , which is most likely due to the sharp break of the power law at the cutoff energy in this model . in comparison , the npex model shows even worse residuals ( @xmath47 ) and only results in reasonable fits when the exponential curvature at higher energies is independent in the partially and fully covered component . the best fit cutoff energy is very close to that of the observed crsf feature ( see below ) and these two parameters are rather strongly correlated ( see figure [ fig : cont ] ) . to avoid a degeneracy of the crsf values due to a changing cutoff energy , @xmath43 was frozen in all the fits to the best fit value from the obs . 2 fi spectrum ( 29.2kev ) . previous observations of indicated the existence of clumps in the stellar wind @xcite , which were modeled using partial covering absorption in addition to fully covered photoelectric absorption of the smooth stellar wind . in the present analysis , the low energy portion of the spectrum also required a partial covering component ( figure [ fig : spec010]b and [ fig : spec020]b ) , which was modeled using the ` tbnew ` model ( wilms et al . , 2011 , in prep . ) , an updated version of the existing ` tbabs ` model @xcite . in addition , a non - relativistic , optically - thin compton scattering component ` cabs ` was included , as is necessary for column densities @xmath48 @xmath21 , where the plasma becomes compton thick and part of the emission is scattered out of the line of sight . the @xmath49 values from ` tbabs ` and from the ` cabs ` component were set equal and treated as one model component for the smooth stellar wind ( @xmath50 = ` tbabs1*cabs1 ` ) and a second component for the clumpy partial coverer ( @xmath51 = ` tbabs2*cabs2 ` ) . the @xmath50 and @xmath51 column densities were left independent of each other . for spectral fitting the abundances of @xcite and the cross - sections @xcite were used in all data sets . both observations show residuals to the continuum in the @xmath52kev energy range , which we interpret as the previously observed crsf @xcite . modeling these residuals with an absorption line with a gaussian optical depth ( ` gabs ` ) improved the residuals significantly in both observations ( see figures [ fig : spec010 ] and [ fig : spec020 ] ) . the best fit values of @xmath53kev and @xmath54kev for obs . 1 and obs . 2 , respectively , are consistent within 90% confidence intervals . the widths of the crsfs , @xmath55kev ( obs . 1 ) and @xmath56kev ( obs . 2 ) , are also consistent within errors . an increase of @xmath57 with higher centroid energy , as previously observed in _ data @xcite , could not be detected , although the best fit values hint at such a behavior . table [ tab1 ] shows the best fit values for the continuum with a ` gabs ` crsf line where the given errors are 90% confidence values . to calculate the significance of the crsf in the first and weaker observation , the null hypothesis approach was applied , where 10,000 spectra were created with monte carlo simulations using the best fit parameters without the crsf . gaussian uncertainties were used for the individual model parameter . each model was fitted with and without the crsf line to compare how much an inclusion of the line actually improves the fit . in @xmath58 of all fits , no crsf feature were observed with a larger optical depth than the observed lower limits of @xmath59 , making the existence of the feature significant with @xmath60 . including the crsf in the real data , the best fit improved by a @xmath61 . a similar improvement of @xmath44 was only observed in @xmath62 ( @xmath63 ) of the simulated spectra , concluding that the observed line in the first observation is indeed real . for the second observation , the line was even more pronounced , where the exclusion of the crsf increased the @xmath44 by @xmath64 . lcccc parameter & & + & fi & bi & fi & bi + @xmath65 @xmath21 ] & @xmath67 & 16.6 ( frozen ) & @xmath68 & 20.9 ( frozen ) + & 1.55 ( frozen ) & 1.55 ( frozen ) & @xmath69 & @xmath70 + & 1.17 ( frozen ) & 1.17 ( frozen ) & @xmath71 & @xmath72 + @xmath73 @xmath21 ] & @xmath74 & 76.9 ( frozen ) & @xmath75 & 28.4 ( frozen ) + @xmath76 & @xmath77 & @xmath78 & @xmath79 & @xmath80 + @xmath81 @xmath82 ^ 1 $ ] fully covered & @xmath83 & @xmath84 & @xmath85 & @xmath86 + @xmath87 @xmath82 ^ 1 $ ] part . covered & @xmath88 & @xmath89 & @xmath90 & @xmath91 + @xmath43 [ kev ] & 29.2 ( frozen ) & 29.2 ( frozen ) & @xmath92 & 29.2 ( frozen ) + @xmath19 [ kev ] & @xmath93 & @xmath94 & @xmath95 & @xmath96 + @xmath97 [ kev ] & @xmath98 & @xmath99 & @xmath54 & @xmath100 + @xmath57 [ kev ] & @xmath101 & @xmath102 & @xmath56 & @xmath103 + @xmath104 & @xmath105 & @xmath106 & @xmath107 & @xmath108 + @xmath109 [ kev ] & @xmath110 & @xmath111 & @xmath112 & @xmath113 + @xmath114 ^ 2 $ ] & @xmath115 & @xmath116 & @xmath117 & @xmath118 + eqw@xmath119 [ ev ] & @xmath120 & @xmath121&@xmath122 & @xmath123 + @xmath124 [ kev ] & @xmath125 & @xmath126 & @xmath127 & @xmath128 + @xmath129 ^ 2 $ ] & @xmath130 & @xmath131 & @xmath132 & @xmath133 + eqw@xmath134 [ ev ] & @xmath135 & @xmath136 & @xmath137 & @xmath138 + @xmath139 [ kev ] & @xmath140 & @xmath141 & @xmath142 & @xmath143 + @xmath144 ^ 2 $ ] & @xmath145 & @xmath146 & @xmath147 & @xmath148 + eqw@xmath149 [ ev ] & @xmath150 & @xmath151 & @xmath152 & @xmath153 + @xmath154 [ kev ] & @xmath155 & @xmath156 & @xmath157&@xmath158 + @xmath159 [ ev ] & @xmath160 & @xmath161 & @xmath162 & @xmath163 + @xmath164 ^ 2 $ ] & @xmath165 & @xmath166 & @xmath167 & @xmath168 + eqw@xmath169 [ ev ] & @xmath170 & @xmath171 & @xmath172 & @xmath173 + @xmath174 [ kev ] & @xmath175 & @xmath176 & @xmath177 & @xmath178 + @xmath179 ^ 2 $ ] & @xmath180 & @xmath181 & @xmath182 & @xmath183 + eqw@xmath184 [ ev ] & @xmath185 & @xmath186 & @xmath187 & @xmath188 + @xmath189 [ kev ] & @xmath14 & @xmath14 & @xmath190 & @xmath191 + @xmath192 ^ 2 $ ] & @xmath14 & @xmath14 & @xmath193 & @xmath194 + eqw@xmath195 [ ev ] & @xmath14 & @xmath14 & @xmath196 & @xmath197 + flux@xmath198 absorb . & @xmath199&@xmath200 & @xmath201 & @xmath202 + flux@xmath198 unabsorb . & @xmath203 & @xmath204 & @xmath205 & @xmath206 + @xmath207 & @xmath208 / & / @xmath209 / & @xmath210 & / @xmath211 + @xmath44/dof & 260 / 245 & 145 / 130 & 542 / 267 & 328 / 150 + + the crsf can alternatively be described with the lorenzian shaped ` cyclabs ` xspec model @xcite . ` cyclabs ` is described by the centroid energy @xmath97 , the width @xmath57 , and the resonance depth @xmath104 , similar to the ` gabs ` parameters . the best fit continuum parameters are consistent with the best fit values determined with the ` gabs ` component . the ` cutoff ` energy with a best fit value of @xmath212kev is consistent with the value obtained with the ` gabs ` model . the observed centroid energies of @xmath213kev and @xmath214kev are of the order of @xmath215 lower than the energies obtained with the ` gabs ` model . this discrepancy stems from a different calculation of the line centroid energy and is described in detail in @xcite . the width @xmath216kev and @xmath217kev for obs . 1 and obs . 2 , respectively , is bigger than with the ` gabs ` model . the @xmath44 / dof values of 260 / 245 and 548 / 267 for the two observations using the ` cyclabs ` model and thus were not a better fit when compared to the ` gabs ` model . for the final discussion we use the values determined by the ` gabs ` model . several emission features were observed in the residuals of the fits in both observations and were subsequentially modeled with gaussian emission lines ( see figures [ fig : spec010 ] and [ fig : spec020 ] , inlay ) . the width of each line was set equal to that of the fe k@xmath22 emission line , while the intensities were left to vary independently . energies were loosely constrained around the expected values of neutral material to avoid runaway of the line energies . a constant ( ` const ` ) was applied , taking small instrumental differences in the overall flux normalization into account . the constant was fixed at 1 for xis 0 and was left free for the other instruments . @xmath218 , @xmath219 and @xmath220 in table 1 are the cross calibration constants with respect to xis0 for fi fits and with respect to xis1 for the bi fits . the final model had the form : ` const`*`n_\text{h,1 } ` ( ` pl1`+ ` n_\text{h,2}`*`pl2 ` ) * ` fdcut`*`gabs`+ @xmath221`gaussians ` , where the power law indices of ` pl1 ` and ` pl2 ` are set to be equal to each other . the individual power law normalizations were independent and were used to calculate the fraction of the partial covering . table [ tab1 ] summarizes the best fit values for both observations and for the individual fi / bi data sets . for the interpretation we will concentrate on the fi data for both observations , for reasons mentioned above . the best fit spectral parameters were generally consistent with previous observations , with the exception of the cutoff energy . with a value of @xmath92kev , @xmath43 was significantly higher than the @xmath13kev value measured in previous _ rxte _ @xcite and _ bepposax _ @xcite observations , although these observations used slightly different spectral models for the cutoff energy . @xcite used also the here applied fermi dirac cutoff , resulting in cutoff energies of @xmath222kev . the folding energy for obs . 1 , @xmath223kev , is consistent with values obtained for a similar orbital phase with _ @xcite . in obs . 2 , the value of @xmath224kev is consistent with the _ rxte _ data for the pre - periastron flare and the periastron passage @xcite . the observed column densities for the absorption due to the smooth stellar wind ( @xmath50 ) were consistent between both observations . a larger difference could be observed in the column density associated with the absorption due to the clumped wind ( @xmath51 ) , where the best fit value of obs . 1 is almost a factor three larger than that of obs . 2 . in addition to the h column density in the ` tbnew ` model , the relative ca and fe abundances were also left independent in obs . 2 . the best fit values are measured from the ca and fe k edges at @xmath225kev and @xmath226kev , respectively , and were slightly higher than solar abundances : @xmath69 for ca and @xmath71 for fe . for the first observation , the ca and fe abundances could not be well constrained and were fixed to 1.55 for ca and 1.17 for fe . the partial covering fractions can be calculated from the measured normalization values of the two power law components , @xmath227 and @xmath228 : @xmath229 the factor is almost unity in both observations due to the dominance of the @xmath228 value . similar large covering fractions have been observed in _ xmm - newton _ data @xcite and in the _ rxte _ data @xcite , indicating that the covering fraction does not change significantly for different parts of the orbit . the spectrum softened slightly from @xmath230 for obs . 1 to @xmath79 for obs . 2 . the fe k@xmath22 line was detected at @xmath157kev ( obs . 2 ) , which is , given the instrumental gain systematics , consistent with neutral fe . in addition , the following lines have also been observed in both observations : s k@xmath22 line , ar k@xmath22 line , ca k@xmath22 line , fe k@xmath231 line , and the ni k@xmath22 line , all with energies consistent with neutral material ( e.g. * ? ? ? * ) . note that the ni k@xmath22 line was not detected in the fainter first observation . the observed line intensities are summarized in table [ tab1 ] with 90% confidence errors . the values are consistent between the fi and bi instruments . the widths of the lines were set to be equal to the fe k@xmath22 width within each observation , which was found to be @xmath160ev for obs . 1 and had an upper limit of 5ev in obs . 2 ( fi values ) . the compton shoulder to the fe k@xmath22 line , as observed by the _ observatory @xcite , was not significantly detected in either observation . the observation of @xcite was performed in the pre - periastron phase , where the luminosity was higher than in the two _ suzaku _ observations . a compton shoulder has also been observed with _ xmm - newton _ in data taken during another pre - periastron flare @xcite . barycentric and binary corrected light curves for different energy bands were extracted for both observations using the orbital parameters from @xcite with an updated orbital period and periastron time @xmath232 determined by @xcite . due to the strong variability of the pulse period on short time scales , individually determined pulse periods were used for each observation . an accurate pulse period could be established for obs . 2 , but not for obs . 1 due to its short duration of 10ks . for the second observation the calculated pulse period had a value of @xmath233s . fermi_/gbm instrument measures the pulse period of on a regular basis , resulting in values of @xmath234s for obs . 1 and @xmath235s for obs . 2 ( finger , 2011 , priv . comm ) . to create the pulse profiles , the _ fermi_/gbm - provided pulse periods and epoch times for each observation were used . we studied the energy dependent pulse profiles for obs . 1 and obs . 2 . both pulse profiles are very similar , showing a double peak shape where the main peak ( p1 ) is broader below 10kev . the second peak ( p2 ) stays rather constant in width , but increases its relative intensity toward higher energies . as an example , the pulse profile for obs . 2 in different energy bands is shown in figure [ fig : prof020 ] . comparing these pulse profiles with previous _ rxte _ and _ bepposax _ data showed that the general shape is consistent through all parts of the orbit . the data of the longer obs . 2 were divided into 10 equally spaced phase bins and individual spectra were extracted for the xis and pin instruments . figure [ fig : prof020 ] shows the individual phase bins in the lowest panel . this division resulted in an exposure time in each phase bin of @xmath236ks for each individual instrument . no gso data were used in this analysis due to the reduced exposure time per phase bin . for spectral analysis , the phase averaged model was applied to all phase bins , resulting in best fit values summarized in table [ tab2 ] , as well as figure [ fig : phasespec1 ] . again the cutoff energy was frozen to the phase averaged value of 29.2kev to avoid the previously discussed degeneracy with @xmath97 . lccc cccc ccc parameter & pb1 & pb2&pb3 & pb4&pb5&pb6&pb7 & pb8&pb9&pb10 + @xmath65/@xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath246 & @xmath247 + @xmath73/@xmath237 & @xmath248 & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 & @xmath256 & @xmath257 + @xmath76 & @xmath258 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 & @xmath265 & @xmath266 & @xmath267 + @xmath268 ^ 1 $ ] ful . @xmath269 & @xmath270 & @xmath271&@xmath272&@xmath273&@xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 + @xmath279 ^ 1 $ ] par @xmath280 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 & @xmath289 + @xmath19 [ kev ] & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 & @xmath296 & @xmath297 & @xmath298 & @xmath299 + @xmath97 [ kev ] & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 + @xmath57 [ kev ] & @xmath310 & @xmath311 & @xmath312 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 + @xmath104 & @xmath320 & @xmath321 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 & @xmath327 & @xmath328 & @xmath329 + @xmath330 ^ 2 $ ] & @xmath331 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 & @xmath337 & @xmath338 & @xmath339 & @xmath340 + flux@xmath341 & @xmath342&@xmath343 & @xmath344 & @xmath345 & @xmath346 & @xmath347&@xmath348&@xmath349 & @xmath350 & @xmath351 + c@xmath352/c@xmath353 & 0.94 / 1.15 & 0.94 / 1.15 & 0.95 / 1.19 & 0.94 / 1.21 & 0.94 / 1.21 & 0.94 / 1.19 & 0.95 / 1.16 & 0.95 / 1.14 & 0.94 / 1.10 & 0.94 / 1.11 + @xmath44/dof & 233 / 243 & 288 / 243 & 353 / 243 & 312 / 243 & 253 /243 & 303 / 243 & 283 / 243 & 292 / 243 & 264 / 243 & 247 / 243 + @xmath44/dof no crsf & 241 / 246 & 307 / 246 & 391 / 246 & 341 / 246 & 273 / 246 & 305 / 246 & 291 / 246 & 344 / 246 & 297 / 246 & 270 / 246 + the two absorbing components , @xmath50 from the smooth stellar wind and @xmath51 from the partial coverer vary between 20 and 40@xmath354 @xmath21 , but show an anti - correlated trend throughout the first peak . although the error bars are rather big , one can see an indication that @xmath50 follows the flux in p1 , whereas the @xmath51 value dips at the same time . in contrast to @xcite , the power law index @xmath76 is varying strongly throughout the pulse . the values are 0.91.3 throughout the first peak and drop suddenly to 0.60.8 for the second peak ( figure [ fig : phasespec1 ] ) . measuring the second peak to be significantly harder than the first peak supports the behavior seen in the pulse profile , where the intensity of p2 increases at higher energies , becoming similar to the intensity in p1 . power law normalizations are relatively small and badly constrained for the first power law component . the partially covered power law normalization seems to follow the flux , showing a higher value throughout the first peak . the folding energy does not vary significantly throughout the orbit and shows values between 5 and 6kev ( not shown in figure 7 ) the crsf energy varies between 3040kev , similar to the values observed in the _ rxte _ data @xcite . table 2 shows that the addition of the crsf in the individual spectra improved the @xmath44 for most of the phase bins , except in phase bin 6 , which falls in the gap between both p1 and p2 . phase bins 1 and 7 , also minima in the pulse profile , only show a small improvement in the best fit when the crsf was included . the crsf energy is not very well constrained in phase bins 1 and 6 due to the lack of sufficient statistics . for the geometrical discussion below , these two phase bins are ignored . in all other phase bins the addition of the crsf improves the fit . the energy changes very smoothly throughout the pulse and does not follow explicitly the observed flux in each phase bin . the crsf energy increases during the main peak and reaches the maximum at the falling flank of p1 . during p2 the crsf energy decreases until it reaches a minimum at the dip between p2 and p1 . to estimate the significance of the crsf in the phase resolved analysis , again the null hypothesis method was applied in two out of the 8 remaining phase bins , which are good representatives of all data points . phase bin 2 is a good example for a shallow crsf , whereas phase bin 5 is an example for a phase bin with a higher flux . similarly as with the phase averaged data , 10,000 spectra were simulated and modeled without and with the crsf . the interpretation of the observed feature as being due to stochastic fluctuations with a depth similar to the observed values could be ruled out with a probability of over @xmath355 ( @xmath60 ) in both phase bins . this leads to the conclusion that the crsf can be observed with sufficient significance to allow the interpretation below . in both observations very strong absorption can be observed . @xmath50 , which is the absorption due to the smooth stellar wind , is consistent within both observations . @xmath51 , the absorption from the partial coverer , is significantly stronger in the first observation , establishing the existence of an additional component in the line of sight . @xcite used _ rxte_/asm and pca data to show that a stellar wind and a stream model component can describe the observed count rate throughout the orbit and that the increase of the column density at orbital phase @xmath356 is mainly due to the stream component . the findings in both _ suzaku _ observations are supportive of this theoretical picture . shows very strong luminosity variability on very short time scales @xcite , where the column density increases to up to 50% in periods of lower activity . rxte _ and _ xmm - newton _ data showed a much higher column density in the pre - periastron flare than in the two data points observed with _ suzaku_. one possible explanation is that the wind is indeed so variable and clumpy that these variations are just not properly predictable , especially in times of high activity , such as the pre - periastron flare . @xcite investigated _ rxte_/pca @xmath49 values of the absorbed component from archival observations for different parts of the orbit and found no increased column density in the pre - periastron flare . observations throughout one full binary orbit could help to understand the variations in @xmath49 on time scales of days . the observed luminosity dependence of the power law index @xmath76 and the folding energy is similar to other sources , such as v0332 + 53 @xcite and 4u0115 + 63 @xcite . the power law index @xmath76 shows a hardening for the first observation , whereas the folding energy @xmath19 is slightly higher when compared to obs . 2 . @xcite observed the @xmath19 variation in her x-1 phase resolved _ heao-1 _ data and concluded that the parameter is dependent on the viewing angle of the accretion column and can be used to describe the plasma temperature of the system . in the case of the smaller @xmath19 value in obs . 2 could indicate a lower plasma temperature which can be interpreted that the x - ray emission region is further up in the accretion column @xcite ( see also section 5.2 ) . the softening of the power law index @xmath76 with increased luminosity is also in agreement with the basic model of the accretion column that the plasma temperature decreases with increased height . with increased luminosity , the rate of accreted material would increase and the amount of soft photons created by the lateral walls of the relatively taller column would increase , leading to a softer spectrum , as is observed . for the second observation the abundances for fe and ca were left independent and a slight overabundance was observed . taking into consideration that the abundances used in @xcite are derived from the interstellar medium ( ism ) a small overabundance from an evolved star may be expected . in both observations the crsf was clearly observed and the ` gabs ` absorption component improved the overall fit significantly . strong magnetic fields ( @xmath357 g ) exist close to the ns magnetic poles , where photons at energies close to the landau levels are resonantly scattered from the line of sight and result in an absorption line - like feature in the spectrum . this feature provides a direct method to measure the magnetic field strength close to the ns surface , where the fundamental centroid energy can be described as @xmath358 where @xmath359 is the gravitational redshift at the scattering site , with @xmath360 for typical ns . using the @xmath97 values obtained in the phase averaged spectra , the calculated magnetic field has a strength of @xmath361 g ( obs . 1 ) and @xmath362 g ( obs . 2 ) which is consistent within errors for both observations . the crsf parameters are consistent with previous results using _ rxte _ data and are lower than the observed _ bepposax _ @xcite and _ integral _ @xcite values of @xmath363kev . these observations were obtained during pre - periastron outburst , where the luminosity of the source is much higher than in the two _ suzaku _ observations presented in this paper . a luminosity dependence of the phase - averaged crsf centroid energy is observed in multiple other sources , where an anti - correlation between crsf and luminosity was observed in v0332 + 53 and 4u0115 + 63 @xcite , and a positive correlation was observed in her x-1 @xcite . @xcite have confirmed , from pulse to pulse variability studies , such correlations for v0332 + 53 , 4u0115 + 63 and her x-1 , and also found an anti - correlation for a0535 + 26 . although the crsf values are consistent with a single value in both _ suzaku _ observations , a hint of an anti - correlation can be seen in the data . compared with the _ bepposax _ and _ integral _ data , the crsf centroid energy might have a positive correlation . parallel to the crsf - luminosity correlation , @xcite also observed in pulse to pulse data of four sources that the power law index @xmath76 shows an opposite luminosity dependence than the @xmath97 . in the _ suzaku _ observations discussed here , @xmath76 is observed to soften with increased luminosity , another indication that a possible negative crsf - luminosity correlation exists . the two opposite scenarios seem to depend on the luminosity , i.e. , if the observed luminosity is below or above a critical luminosity ( cl ) which can be derived from the eddington luminosity of a system @xcite . the cl is of order @xmath364 erg s@xmath29 , but depends also on the accretion geometry ( spherical or disc ) as well as the height and diameter of the accretion column . above the cl , the infalling proton density becomes so large that the protons begin to interact and decelerate , creating a radiation pressure dominated shock region above the magnetic pole . this region of increased density is most likely the region where the crsf is created @xcite . with increasing luminosity , the shock region moves higher up in the accretion column , where a smaller local magnetic field value results in a lower observed crsf centroid energy , as is observed in v0332 + 53 or 4u0115 + 63 . the observed crsf - luminosity dependence as well as the @xmath76-luminosity dependence in is very similar to the scenario described here . below the cl , the accreting matter slows down via hydrodynamical shock , ` coulomb friction ' or nuclear collision @xcite . with increasing accretion rate , the luminosity increases and the deceleration region is pushed closer to the ns surface , where the magnetic fields are higher . this would result in a positive crsf - luminosity correlation , as observed in her x-1 and a0535 + 26 . with a distance of 3kpc , the intrinsic unabsorbed @xmath27kev luminosity of @xmath365ergs@xmath29 is significantly below the typical cl of @xmath364ergs@xmath29 . although it has been observed that sources with similar luminosities can show the opposite correlation @xcite due to individual critical luminosity values , the large luminosity difference compared to cl puts this source in the same regime as her x-1 and a0535 + 26 and at odds with the notion of a well - defined cl . the observed variation in @xmath76 and possibly in @xmath97 would then stem from another mechanism . the existence of multiple fluorescence emission lines , especially at lower energies , where the absorption is very dominant , indicates that the source of the emission originates from a region where the column density is not very large , i.e. the outer layers of the stellar wind in our line of sight @xcite . if the emission lines would be embedded deeper in the stellar wind , the lines , especially at lower energies would have to be significantly stronger , to be detected at all . on the other hand , if the line emitting region is in the outer layers of the wind , the incident soft x - ray flux would be drastically reduced , making the equivalent width very large , as observed in the low energy emission lines ( table 1 ) . a possible explanation would be that the emission region is very large for the lines , and maybe spread over the entire surface of the stellar wind . the most dominant line emission stems from the fe k@xmath22 transition at @xmath366kev . the observed energies in both observations are consistent with the emission in neutral material . the ratio of the intensities of fe k@xmath231/fe k@xmath22 for both observations , @xmath367 for obs . 1 and @xmath368 for obs . 2 , is consistent with the fluorescing material being neutral or only slightly ionized @xcite . the equivalent widths ( eqw ) of the two fe k@xmath22 lines show that the intensity relative to the continuum is reduced by a factor of @xmath38 in the second observation . based on emission line widths , @xcite estimated in _ asca _ data that the fe emission originates within @xmath369 cm of the continuum emission source . @xcite use _ xmm - newton _ pulse by pulse observations to observe a correlation with increased flux of the source , showing that the fe emission region is not far from the x - ray source . in the _ suzaku _ observation , however , the fe k@xmath22 line flux does not directly follow the observed luminosity . the increase in absorbed @xmath27kev luminosity is a factor of @xmath8 , whereas the increase in the fluorescence line intensities is only @xmath370 . this difference in intensity change indicates that the distance between continuum and fluorescence emission lines is large , especially when compared to the proposed @xmath371 lt - s from @xcite . in the 6 months time between the two observations , however , the overall emission geometry could have changed , accounting for the difference in the fluorescence intensities . if the emission originates from a confined region in the stellar wind , observations taken during two different orbital phases would each yield a different distance to this region . in addition to the strong fe k@xmath22 line , many other emission lines are observed in the xis spectra in both observations . the observed intensities of these lines show the same behavior , where the flux for the second observation is @xmath372 or less compared to the first observation , although the observed uncertainties for the different lines , especially for s k@xmath22 and ar k@xmath22 , are very large . furthermore , the s k@xmath22 line falls into an energy regime where calibration uncertainties have been previously observed ( e.g. , * ? ? ? * ) , so that a detailed analysis is not possible at this time . historically , shows very strong pulse to pulse variations and an intensive flaring behavior on very short time scales @xcite , e.g. , throughout the pre - periastron flare . the observations obtained here do not show such flaring behavior , and only show the regular pulsation throughout the observation . a pulse period could only be established for obs . 2 of the _ suzaku _ data , which was consistent with _ fermi_/gbm data . with the gbm pulse period for the first observation , pulse profiles for both observations could be produced for different energy bands . the two peaked pulse profiles show a similar behavior for both observations , where the second peak is getting strong at higher energies . this behavior is in contrast to many similar sources , such as 4u 0115 + 63 @xcite , v 0332 + 53 @xcite , 4u1909 + 07 @xcite , and 1a1118@xmath1461 @xcite , where the pulse profile turns into a single peak profile at higher energies . a reduction in the intensity of the second pulse has also been observed in _ bepposax _ data for different parts of the orbit @xcite . the phase resolved analysis of the second observation showed strong variations in several spectral parameters throughout the pulse profile . as previously observed by @xcite , the crsf centroid energy varies by @xmath373 where the highest energy is detected at the peak and the falling flank of p1 . such behavior has been observed in multiple other sources , e.g. , vela x-1 @xcite , 4u 0352 + 309 @xcite , and her x-1 @xcite . a model for the crsf variations is based upon the change in the viewing angle throughout the pulse and thus different heights of the accretion column are probed , yielding a different local magnetic field observed for each phase bin . a simple approach to derive a possible geometry for the neutron star and the magnetic field uses the variation in the observed magnetic field throughout the pulse phase . in this case the very smooth and sinusoidal variation ( figure 8) can be modeled as a simple dipole where the total magnetic moment @xmath374 is calculated from the phase averaged crsf energy value . the variation of the crsf energy is then fitted by changing the geometrical angles of the system until a best fit converges . appendix a discusses the model in detail , introducing the three free parameters : @xmath375 , the viewing angle between line of sight and the ns spin axis , the inclination angle @xmath22 of the magnetic moment with respect to the spin axis , and the angle @xmath231 , indicating the lag of the @xmath376-field plane with respect to the ephemeris , i.e. , the observed shift in pulse phase ( see also figure 9 ) the best fit values for @xmath231 were either @xmath377 or @xmath378 , depending on the starting values of the fit ( see appendix a ) , where the latter value is equivalent to the former when the dipole polarity is flipped . @xmath22 and @xmath375 angles showed a very strong interdependence , and the best fit @xmath379 pairs were : @xmath380 , @xmath381 , @xmath382 , and @xmath383 . all best fit values have a @xmath384 with 7 dof . these pairs do show a degeneracy of @xmath385 and @xmath386 , which can be explained with a geometrical symmetry , when calculating the magnetic field for each individual phase bin . the first and last two pairs can be each treated as the same geometry , rotated by 180 degrees . figure [ fig : magnet ] shows the calculated magnetic field values from the crsf centroid energies and the theoretical values for @xmath376 from the best fit angles for a simple dipole model with @xmath387 , @xmath388 and @xmath389 . the phase resolved energy and width of the crsf show a strong correlation , where the width varies in phase with the crsf energy ( figure . @xcite observed a similar correlation , where the magnetic pole was observed under different viewing angles @xmath390 , where @xmath390 is the angle between the line of sight and the magnetic axis . @xcite showed that the anisotropic velocity field of the electrons in the accretion column leads to a fractional line width of : @xmath391 the @xmath392 ratios vary in the 0.16 0.24 range , corresponding to variations of @xmath393 throughout the pulse phase . the two outliers with large errors in phase bins 1 and 6 were ignored . the variation of @xmath390 can be estimated as @xmath394 and @xmath395 throughout one pulse phase , where the magnetic pole rotates around the ns spin axis , tilted by @xmath375 with respect to the line of sight . using the average ratio of 0.2 , we calculated the values of @xmath396 for the two geometries with @xmath397 ( @xmath398 ) and @xmath387 ( @xmath399 ) . for each geometry , we used the variation of the @xmath392 ratio ( @xmath400 ) to calculate the variation in @xmath390 , resulting in values of @xmath401 and @xmath402 . these variations are smaller then the values expected from the the geometrical discussion , but show a similar behavior , where the variation of the angle is larger for the smaller value of @xmath403 . from equation 4 , we were also able to estimate a possible plasma temperature for the two geometries from the calculated values of @xmath396 . the plasma temperature of @xmath404kev for @xmath405 is very similar to the observed folding energy @xmath19 , which is an indication of the plasma temperature ( * ? ? ? * and references therein ) . the value of @xmath406kev for the @xmath387 geometry is much larger and is not consistent with the observed @xmath19 . assuming that the ns spin axis is aligned with the inclination of the binary system , a value of @xmath405 would put the inclination at @xmath407 , a value which is only marginally above the upper limit of @xmath408 , as determined by @xcite . the strong fe line was detected in all 10 phase bins . the measured line flux did not change significantly throughout the pulse phase , while the @xmath27kev flux varied by more than a factor of 2 . from this we conclude that the distance to the fe fluorescence region is greater than @xmath12 lt - s ( @xmath409 cm ) , from the ns . this is in conflict with the conclusion of @xcite that the emission region must be closer than @xmath369 cm from the ns , based upon the width of the emission lines measured by asca . their assumption is that the emission region is close to the ns and that the material free falls onto the ns surface , encountering much faster velocities , which caused the observed broadening of the lines . the smaller measured fe k@xmath22 line width in the phase averaged spectra is more consistent with the assumption that the line broadening is due to the terminal velocity of @xmath410 km of the line driven wind @xcite . we presented results of two _ suzaku _ observations of taken shortly after the periastron passage . the spectra were modeled with a partially and fully covered power law with a fermi - dirac cutoff and a crsf at @xmath46kev . the column density of the smooth stellar component only marginally changed and the clumpy wind component significantly changed between both observations . flux dependencies of the crsf and @xmath76 were not significant , but do hint at similar correlations such as those observed in v0332 + 53 and 4u0115 + 63 , although the observed luminosity is significantly below the calculated critical luminosity of this system . they are in the realm where one might expect the energy of the crsf to decrease with decreasing flux . the variations in the pulse profiles and the crsf throughout the pulse phase have a signature of a dipole magnetic field . using a simple dipole model we calculated the expected magnetic field values for different pulse phases and were able to extract a set of geometrical angles , loosely constraining the dipole geometry in the ns . model constraints derived using the observed ratio of @xmath57 to @xmath97 , together with calculated plasma temperatures , favor a solution wherein the spin axis is tilted @xmath411 to our line of sight . ss acknowledges support by nasa contract nas5 - 30720 and grant nnx08ax83 g . ff is supported by dlr grant 50 or 0808 and via a daad fellowship . ic acknowledges financial support from the french space agency cnes through cnrs . this research has made use of data obtained from the suzaku satellite , a collaborative mission between the space agencies of japan ( jaxa ) and the usa ( nasa ) the variability of the crsf energy throughout the pulse phase can be used to understand the geometry of the neutron star . we assume a simple dipole magnetic field around a spherical neutron star , however we want to emphasize that these assumptions are very rudimentary , and effects such as the displacement of the dipole field from the neutron star center are not taken into account here . the total magnetic moment @xmath374 of the neutron star can be calculated with : @xmath412 using the average magnetic field from the phase averaged crsf centroid energy ( @xmath413 g ) and a typical neutron star radius of @xmath414 km . the value of the last part , including the azimuthal angle @xmath390 of the magnetic pole was estimated to the average value of 1.54 . the total average magnetic moment is divided into three components based on the cartesian coordinate system in figure [ fig : nsmodel ] : @xmath415 where @xmath22 is the angle between the dipole field and the spin axis and @xmath231 is the rotation angle from the ephemeris , i.e. the x - axis . next , the vector @xmath416 was calculated for each individual phase bin @xmath417 used in the phase resolved analysis . @xmath418 the individual components of @xmath416 are the projections of each phase bin on the three axes , as seen from the line of sight of the observer . the angle @xmath375 corresponds to the angle between the line of sight and the spin axis . for each of the individual phase bins , the magnetic field was calculated using the position vector @xmath416 indicating the position of the magnetic pole and the vector @xmath419 , which indicates the magnetic moment : @xmath420 the calculated magnetic field values were then fitted to the measured magnetic field , varying the three angles @xmath375 , @xmath22 , and @xmath231 until the best fit converged . to avoid wrong solutions due to local minima , the starting parameter of all three angles were varied in steps of @xmath421 before the fit was started . the best fit solutions resulted in the values that are discussed in detail in section 5.5 .

for decades , type ia supernovae ( sn ia ) have been interpreted as thermonuclear explosions that result from either accretion onto a degenerate white dwarf or via coalescence of two degenerate stars ; see @xcite for a thorough review . models that begin fusion subsonically and then turn into a detonation are best at matching observations of typical sn ia @xcite . synthesized intermediate - mass elements ( imes ) like silicon and sulfur in sn ia spectra are the result of c and o burning at low densities due to the pre - expansion of the white dwarf during the deflagration phase . the deflagration turns into a detonation , and the combination of the two produces a significant amount of @xmath4ni , the decay of which powers the light curve . sn ia show a range of energies and spectral characteristics , but the majority of them are quite homogeneous when compared with other supernova types . the relative uniformity amongst sn ia explosions has allowed their exploitation as cosmic standard candles , and observed correlations between light - curve shape , peak luminosity @xcite , and color @xcite have enabled the precise measurement of cosmologically interesting distances . it is with these measurements , taken over a range of redshifts , that estimates of the hubble parameter ( @xmath5 ) and cosmic matter density have been refined @xcite , and evidence for dark energy was revealed @xcite . the sloan digital sky survey - ii ( sdss - ii ) supernova survey was designed to further improve sn ia as distance indicators @xcite . covering a 300 deg@xmath6 area with a rapid cadence , the survey discovered and spectroscopically confirmed roughly 500 sn ia over a range of redshifts out to @xmath7 . @xcite , @xcite , and @xcite used the first year of sdss - ii data to constrain the dark energy equation - of - state parameter @xmath8 , and analyzed the systematic errors limiting the cosmological measurements . identifying the progenitors of sn ia and better understanding their explosion mechanism may improve their reliability as distance indicators . one approach is to study events that do not conform to the general sn ia homogeneity in their spectra or luminosity . subclasses of sn 1991t - like ( hereafter referred to as 91 t ; * ? ? ? * ; * ? ? ? * ) and sn 1991bg - like ( hereafter 91bg ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) events were quickly recognized as being peculiar , although much of the spectroscopic diversity is now known to be caused by a range of photospheric temperatures @xcite . recently , additional subclasses of sn ia have been identified . for example , sn 2002ic @xcite and sn 2005gj @xcite are luminous objects that show hydrogen emission lines in their spectra , unlike normal sn ia . it is suspected they may be sn ia interacting with dense circumstellar material , although a core - collapse scenario has also been proposed @xcite . sn 2002cx ( hereafter 02cx ) was especially peculiar @xcite . it showed a hot ( 91t - like ) spectrum at early times , but it cooled quickly after maximum brightness and its expansion velocities were well below typical for a sn ia . 02cx deviated significantly from the @xcite relation between peak luminosity and decline rate ; it was subluminous for its light - curve shape by @xmath9 mag in the @xmath1 and @xmath10 bands . the well - observed sn 2005hk ( hereafter 05hk ) was spectroscopically very similar to 02cx both exhibited photospheric velocities of roughly 7000 km s@xmath3 . photometrically , 05hk was faint on the phillips relation by @xmath111.0 mag in the optical , but @xmath12 mag more luminous than 02cx @xcite . @xcite have identified several other 02cx - like objects and postulated that their extreme subluminosity suggests that they constitute a class of pure thermonuclear deflagrations : that is , the fusion front moving through the white dwarf fails to make the transition to supersonic burning @xcite and is unable to generate the large amounts of radioactive nickel observed in typical sn ia @xcite . the thermonuclear burning of carbon and oxygen at moderate densities will create imes such as silicon , sulfur , and calcium that dominate the spectrum . though typical sn ia experience this phase only briefly , @xcite reasoned that 02cx - like objects may burn completely via this mechanism . ( for a list of 02cx - like events , see table 9 of @xcite . ) which characteristics separate delayed detonations from pure deflagrations ? simulations of pure deflagration events predict that rayleigh - taylor instabilities will form clumps of radioactive nickel and mix unburned material toward the center @xcite . late - time spectra of 02cx - like events show many permitted iron transitions broken up into thin spikes @xcite , indicative of the lower velocities that characterize 02cx - like supernovae . these spectra also show narrow forbidden lines of ca ii and fe ii , but are quite distinct from normal sn ia at late epochs , whose spectra are dominated by relatively broad , blended fe ii and fe iii lines . spectra of 05hk @xmath13 200 days past maximum have been analyzed for emission from unburned c i and o i , but their existence has not been confirmed @xcite . an extremely subluminous transient , sn 2008 ha ( hereafter 08 ha ; * ? ? ? * ) , appears to be an additional member of the 02cx class , although @xcite have proposed that the extreme nature of 08 ha ( and other 02cx - like objects ) is better matched by the core collapse of a massive star where most of the synthesized radioactive elements fall back to a black hole . with the discovery of si ii and s ii in the early - time spectra of 08 ha @xcite , that proposition becomes less likely . here , we report photometry and spectroscopy of sn 2007qd ( hereafter 07qd ) . its observed properties place it in the 02cx subclass of supernovae , specifically as a member intermediate to 02cx and 08 ha , linking these objects . we present the photometric and spectroscopic observations of 07qd and compare its unique properties with a range of other sn . we estimate the rise time , luminosity distance , and peak luminosity of this peculiar object , and we calculate its kinetic energy . furthermore , we model the spectra to determine the velocity and composition of the ejecta and compare these with the findings of other similar sn to identify common traits . with this information we predict the synthesized @xmath4ni mass and determine whether 07qd shares more in common with a thermonuclear deflagration or a core - collapse scenario . 07qd @xcite was discovered during the sdss - ii supernova survey using the sdss camera on the 2.5 m telescope @xcite at apache point observatory . the sn was located amidst a spiral arm at @xmath14 , @xmath15 ( j2000.0 ) , a projected distance of 10.6 kpc from the nucleus of the sbb / sbc host galaxy sdss j020932.73 - 005959.8 centered at @xmath16 , @xmath17 @xcite . the redshift of the host is @xmath18 , as measured from the sdss galaxy redshift survey @xcite . figure [ f : finderchart ] shows an image of 07qd and its location in its host galaxy . the foreground milky way extinction in the direction of 07qd is @xmath19 mag as calculated from the dust maps of @xcite . the photometry was calibrated in the standard sdss @xmath20 photometric system @xcite . the flux from the supernova was estimated by using the scene modeling technique @xcite from individual calibrated images and without spatial resampling . figure [ f : lightcurves2 ] shows the sdss - ii light curves in the @xmath20 bands ; the data are listed in table [ t : phot_data ] . the time of peak bolometric flux for 07qd is not well defined , but probably occurred within a span of 2 days around maximum apparent @xmath21-band magnitude @xmath22 mag on mjd 54405.39 . we base this on both the available @xmath23 , @xmath21 , and @xmath24 data detected on that date , and the fact that the @xmath23 and @xmath21 fluxes were falling while the @xmath24 , @xmath25 , and @xmath26 fluxes were still rising . @xcite noted that the spectrum of 07qd was similar to that of 05hk , so we compared their light curves . while the maximum @xmath21-band apparent magnitude of 05hk was @xmath27 , it achieved a peak @xmath1-band absolute magnitude of @xmath28 . this is rather dim for a typical sn ia ( @xmath29 mag ) , but @xmath30 mag brighter than 07qd ( see 3.1 ) . featured in figure [ f : lightcurves2 ] are the 05hk light curves corrected to the observed frame of 07qd . in addition , the brightness of 05hk has been reduced by 2.4 mag to better match the peak flux of 07qd and enable their shapes to be compared . ignoring possible differences in dust extinction , 05hk is more luminous than 07qd in all of the sdss bands , and 05hk both rose and declined more slowly in @xmath23 , @xmath21 , and @xmath24 than 07qd . the light - curve widths of the two supernovae are similar at near - infrared wavelengths ( sdss @xmath25 , @xmath26 ) , though 07qd shows a very slow decline rate at red wavelengths . 05hk also appears to peak later than 07qd as the bandpass becomes progressively redder . we compared the light curve of 07qd with those of additional supernovae to identify any possibly similarity to the type ic or ii classes . as 07qd rose rather quickly , we compared it to other events which shared similar rise times . we chose the fast - evolving sn 1994i @xcite , a well - observed prototype for sn ic not associated with a gamma - ray burst , and sn 1999gi @xcite , a type iip ( plateau ) sn whose absolute @xmath1 magnitude at peak was similar to that of 07qd ( described in detail in 3.1 ) . figure [ f : sn_types ] exhibits the shapes of these light curves in the @xmath24 band . though the steep rise implied by the 07qd data may be similar to those of a sn iip or sn ic , the decline of 07qd and 05hk fails to represent the steep drop of sn 1994i or the extended plateau of sn 1999gi . in addition , the rise of 07qd is far faster than that of 05hk and is more similar to that of 08 ha @xcite . figure [ f : colorcomp_gr ] compares the color evolution of 07qd to that of 05hk , and reveals the former to be more blue near maximum brightness . the colors of normal sn ia are fairly well established @xcite and are used to estimate the reddening caused by dust in the host galaxy . however , the intrinsic colors of peculiar 02cx - like events are uncertain , making the estimation of host - galaxy extinction problematic . 07qd is bluer than 05hk , having lower @xmath31 colors of @xmath32 mag and @xmath33 mag at 0.4 and 1.3 days past @xmath1 maximum , respectively . this suggests that dust extinction is not the major cause of the low luminosity of 07qd compared with 05hk . additionally , 07qd appears to decline slower in the @xmath25 and @xmath26 light curves . the 3.58 m telescopio nazionale galileo ( tng ) in the canary islands observed 07qd at 3 days after @xmath1 maximum ( see 3.1 for the determination of @xmath34 ) . the spectrum is a composite of three 30-min exposures obtained with dolores ( device optimized for low resolution ) , a low - resolution spectrograph and camera permanently installed at the nasmyth b focus of the tng . it is equipped with a @xmath35 pixel e2v 4240 thinned , back - illuminated , deep - depleted , astro - bb coated ccd with a pixel size of 13.5 @xmath36 m and a field of view of @xmath37 with a @xmath38 scale . the spectra were observed with the low - resolution blue grism ( lr - b , dispersion 2.52 @xmath39 ) , covering the 36737401 range at an airmass of 1.27 . a slit of @xmath40 width , equal to the average seeing , was used for the observations and was aligned with the supernova and galaxy core at a position angle of @xmath41 . the hobby - eberly telescope ( het ) at the mcdonald observatory in texas collected spectra of 20-min exposure times utilizing the marcario low resolution spectrograph ( lrs ; * ? ? ? * ) at 8 and 15 days past maximum . at the prime focus , the lrs employed a @xmath42 plate scale with a @xmath43 wide by @xmath44 long slit and covered the 40759586 range , though low signal - to - noise ratios severely limit visibility past 8000 . we obtained low - resolution spectra at 10 days past maximum brightness with the low resolution imaging spectrometer ( lris ; * ? ? ? * ) on the 10 m keck i telescope on mauna kea , hawaii . the keck measurement was able to cover bluer wavelengths than the other observations , spanning the 30738800 range . for all spectra , standard ccd processing and spectrum extraction were performed with iraf . the data were extracted using the optimal algorithm of @xcite . low - order polynomial fits to calibration - lamp spectra were used to establish the wavelength scale , and small adjustments derived from night - sky lines in the object frames were applied . we employed our own idl routines to flux calibrate the data and remove telluric lines using the well - exposed continua of spectrophotometric standard stars @xcite . a journal of the spectroscopic observations is given in table [ t : spec_sched ] , and the resulting extracted and reduced spectra are detailed in figure [ f : qd_evolve ] . we infer a rise time of @xmath45 days based on the nondetections in the sdss - ii data , much shorter than that of typical sn ia @xcite . we began our analysis by applying the light - curve fitters mlcs2k2 @xcite and salt2 @xcite to the 07qd , 05hk , 02cx , and 08 ha data using the snana platform @xcite . the resulting fits were of very poor quality , with @xmath46 per degree of freedom , suggesting the inability of these algorithms , trained on normal events , to fit the colors and light - curve shapes of 02cx - like peculiar supernovae . we decided to take a simple approach and estimate the @xmath1-band light - curve stretch @xcite versus peak brightness for the peculiar events and some more normal sn ia . we used snana @xcite to convert from @xmath20 to the standard @xmath47 for sdss - ii sn ia with @xmath48 , verifying the uncertainty with the conversions of @xcite . snana applies the mlcs2k2 k - correction algorithm , which is based on a color - matched normal sn ia spectral energy distribution . this may introduce small errors due to differences in the spectral features between 07qd and typical sn ia . for 07qd we find a maximum absolute magnitude ( after correcting only for milky way galaxy extinction ) of @xmath49 mag . we estimate that the time of peak @xmath1 brightness occurred on mjd @xmath50 , though the true value could be smaller due to the lack of data on the rise . the slopes of the blue light curves of 07qd flatten out rapidly after 10 days . consequently , @xmath51 mag for 07qd , or roughly the same as that of 05hk @xcite . as seen in figure [ f : lightcurves2 ] , 07qd clearly fades more quickly than does 05hk , suggesting that in this case , the measurement of @xmath52 does not compare well with other sn ia . based on our estimation of the rise time , figure [ f : stretch73 ] compares the absolute magnitude of 07qd and its @xmath1-band stretch with those of 02cx , 05hk , 08 ha , and the normal sdss - ii sn ia ( see @xcite for a list of these sn ) . the low - redshift set of sdss - ii contains a handful of 91bg - like events with stretch parameters @xmath53 , but 07qd and 08 ha have narrower light curves and are much fainter . there appears to be a sequence connecting the bright , peculiar events 05hk and 02cx to the extreme 08 ha with the intermediate 07qd in figure [ f : stretch73 ] . to examine the bolometric luminosities , a blackbody was fit to the @xmath20 spectral energy distribution ( sed ) as measured at @xmath1 maximum brightness . ultraviolet measurements shortward of @xmath23 were not available for 07qd . 05hk and 08 ha experienced severe photospheric line blanketing at @xmath54 @xcite , which , along with our keck spectrum ( 10 days past maximum ) , suggests that the luminosity of 07qd drops off sharply at that point . using a hubble constant of @xmath55 km @xmath56 @xmath57 @xcite and our measured redshift , a luminosity distance of 177 mpc ( @xmath58 mag ) was computed . integrating our sed ( corrected for milky way extinction ) and applying our luminosity distance , we calculated the peak quasi - bolometric luminosity to be @xmath59 ergs @xmath56 . assuming `` arnett s rule '' @xcite as applied by @xcite , we estimate the synthesized @xmath4ni mass using the ratio of maximum luminosity to the luminosity produced by one solar mass of @xmath4ni , @xmath60 where @xmath61 days and @xmath62 days are the e - folding times for @xmath63 and @xmath64 , respectively . thus , for our maximum possible @xmath65 day rise time , we estimate an upper limit of @xmath66 m@xmath67 , which is nearly a factor of 5 larger than that of 08 ha [ @xmath68 m@xmath67 ; @xcite ] , but still much smaller than that of 05hk [ @xmath69 ; @xcite ] . here , we assume that the efficiency of radioactive energy deposition @xmath70 is unity ( see discussion in @xcite ) . the ejecta mass may be estimated by utilizing the treatments of @xcite and @xcite , assuming an opacity of 0.1 @xmath71 @xmath72 , as per an fe ii spectrum in the same vein as @xcite . comparing 07qd to a typical sn ia of kinetic energy @xmath73 ergs , rise time 17 days @xcite , and photospheric velocity 10,000 km s@xmath3 , we derived an ejecta mass of 0.15 m@xmath74 . this could be even lower if the rise time of 07qd is under 10 days or using higher estimates of standard sn ia rise times . based on a photospheric velocity 3 days past maximum brightness of 2800 km s@xmath3 ( see 3.2.1 ) , this translates to a kinetic energy of @xmath75 ergs . this is similar to the kinetic - energy estimate of 08 ha by @xcite , but roughly an order of magnitude lower than the @xmath76 ergs of 05hk based on its photospheric velocity @xcite . typical sn ia contain @xmath4ni masses between 0.1 and 1.0 m@xmath74 and ejecta masses ranging from 0.5 m@xmath74 for sn 1991bg to the canonical 1.4 m@xmath74 @xcite . both of these values are significantly larger than our findings for 07qd . we used synow @xcite , a parameterized supernova spectrum synthesis code , to fit our spectra to profiles of various ions at specific velocities , excitation temperatures , and opacities . these ions are simulated in an expanding photosphere of a chosen blackbody temperature , and the resulting spectra can be compared with data . we systematically fit our spectra with ions commonly seen in sn ia at similar epochs ( fe ii , co ii , si ii ; see @xcite and @xcite for lists of expected ions and isotopic yields ) , and attempted to simulate hydrogen and helium to rule out the possibility of a type ii or ib sn . the photospheric velocity was fixed at each epoch by the best fit for the fe ii lines , and subsequent elements were fit either residing in that photosphere or at detached velocities . after confirming the presence or absence of these species , we tried to fit elements uncommon to normal sn ia in order to fit any remaining line profiles . we also attempted to use na i d host and interstellar absorption to measure dust extinction though the procedures of @xcite . the low signal - to - noise ratio of our spectra , combined with features at various velocities scattered throughout the continuum , created too much uncertainty in these equivalent widths to fix an extinction value . figure [ f:3d_decomposed ] and table [ t : qd_3d ] present the results of our synow fit to the spectrum 3 days after maximum brightness , as well as a decomposition of the fit to show the influence of each ion . we found from the fe ii lines a photospheric velocity of 2800 km @xmath77 , which is extraordinarily low . typical sn ia photospheric velocities at this epoch are often in excess of 10,000 km @xmath56 @xcite . the low velocities found for most of the regions makes it highly unlikely that the strong feature at 6300 is from hydrogen or helium either in the photosphere or detached . the best fit to the 6300 absorption was si ii , since the velocity of fe ii was constrained when fitting the @xmath78 , @xmath79 , and @xmath80 features as well as others in the bluer spectral regions . si ii fit the absorption best when at a velocity of 800 km s@xmath3 lower than that of the iron - group ions . also prominent in the @xmath81 region is a broad primary o i line . the secondary signatures of o i are weakly detected at @xmath82 , but masked at @xmath80 due to the strong si ii feature nearby . si iii and c iii were implemented in a manner similar to that used by @xcite to help co ii shape the area around @xmath83 . c ii , also present in the maximum - light spectrum of 08 ha @xcite , was necessary for an absorption feature at @xmath84 . our synow fits suggest the presence of mg i , ti i , cr i , ca ii , and al i. @xcite do not predict these species in a thermonuclear explosion when si iii is strong . it is likely that the features we attribute to these ions are due to other unidentified species or different velocities for existing ones , though we can not definitively rule out these unusual identifications . it should also be noted that c i , o ii , ni ii , and co iii , predicted by @xcite , may be added with little effect on the overall fit . we were unable to fit the absorption between 3800 and 3900 at this photospheric velocity with anything other than k ii , though its inclusion would introduce other discrepancies to the fit . the synow fit in figure [ f : qd_synow_10d ] is derived from parameters used to fit 02cx at 12 days past maximum @xcite , except the velocities and blackbody temperature have been reduced to accommodate the unique nature of 07qd . this fit was chosen due to its consistency in representing 08 ha , 02cx , and 05hk at similar epochs , as is seen in figure [ f : qd_hk_cx ] . the similarity is especially apparent with 08 ha , which shares most absorption features with 07qd . the relative shift between these two spectra was measured to be @xmath85 km @xmath77 ; the photospheric velocity of 08 ha at this epoch is 2000 km s@xmath3 , while the best - fit synow photospheric velocity of 07qd is @xmath86 km s@xmath3 with a blackbody temperature of 8000 k. though @xcite found an excellent match to the 14-day spectrum of 08 ha with a low photospheric velocity of only 600 km s@xmath3 , we attempted our fits with the fe ii estimated velocity of 2000 km s@xmath3 . much of the continuum is dominated by narrow fe ii lines , with co ii playing a large role below 5000 as well , consistent with most iron - core sn @xcite . the modified synow fit also continues to include a prominent si ii feature at 6300 necessitated by the inability of fe ii to fit it without losing other features . this si ii resides within the photosphere and enhances the spectrum at @xmath87 , but less so at @xmath88 and @xmath89 . the carbon and vanadium measured at 3 days past maximum are not prominent at any ionization stage . their existence can not be ruled out , in part due to the influence of the dominant fe ii lines @xcite . the cr ii in the 02cx fit by @xcite is also not necessary , as there is insufficient data to examine regions below 3750 at this and later epochs . we were unable to identify any ionization stage of ti at this epoch , despite the possibility of observing ti i at 3 days past maximum . @xcite identified ti ii at the same epoch in 08 ha , but its addition here does little to the overall fit . other ions include sc ii ( for the absorption at 5450 ) and ca ii ( for the h&k lines ) . the secondary absorption lines s ii are no longer apparent at this epoch , as sc ii is able to effectively represent the 54005700 region without aid . the signal - to - noise ratio of the data appears to be lower than it actually is , due to the low - velocity nature of this particular photosphere . much of the absorption blending in the regions blueward of 4500 proved difficult to fit with synow models , especially where p - cygni profiles are no longer apparent ; it is suspected to be a consequence of the narrow absorption features of several iron - group elements . h@xmath70 and helium ions were tested in synow fits , though no match could be found . we expect and observe host - galaxy h@xmath70 emission at 6563 , though just blueward of this feature we see a small absorption that we identify with c ii . the nature of the trough at @xmath90 is unclear at this time ; very few ions are capable of modeling it without severely affecting the fit elsewhere . he i was of particular interest in the synow fit to the 08 ha spectrum 13 days past maximum conducted by @xcite and to sn 2007j , another 02cx - like sn @xcite , but the na i d line dominates over the predicted he i line . at 10 days past maximum , all other he i features fail to match the spectrum . using these synow fits to the individual epochs , the spectral evolution of 07qd given in figure [ f : qd_evolve ] provides a detailed picture of a developing photosphere . examining the contributions near @xmath91 and @xmath92 present at 3 days but absent at 10 days , fe iii has either greatly fallen in opacity or recombined into fe ii , which has increased in influence . na i and s ii have also decreased in intensity , but remain crucial to the region between 5000 and 6000 . the si ii feature endures through two weeks past maximum , but its strength has lessened , becoming roughly equal to that of fe ii by day 15 . much of the spectrum in the blue region was not measured at subsequent epochs , though extended red wavelengths are given , revealing probable o i and o ii signatures . it is apparent , however , that the early - time spectra beyond 8000 demonstrate the presence of the ca ii near - ir feature , though the ca ii h&k near - uv lines persist . blackbody temperatures have fallen to 8000 k at 8 days past maximum and to 6000 k at 13 days , further intensifying the influence of fe ii profiles over the continuum . it remains to be seen whether the other unidentified line profiles can be remedied with more exotic ions . [ f : qd_hk_cx ] shows the spectra of four 02cx - like sn ia compared at similar epochs . 07qd clearly bears the strongest resemblance to 08 ha ; very few features differ . 02cx and 05hk exhibit much faster photospheres than 07qd and 08 ha , and are likewise shifted toward the blue . the faster photosphere and higher excitation temperatures used in the synow fit of 02cx @xcite broaden many of the fe ii lines , effectively masking the primary si ii absorption , but other major velocity features remain consistent . it should be noted , however , that synow s highly parameterized fitting routine permits many degrees of freedom . in order to independently identify these imes outside of models , we directly compared spectra of 07qd with those of other confirmed sn ia . of special interest is the spectral relation of 07qd to sn 2004eo ( hereafter 04eo ; * ? ? ? similar to the prototypical sn 1992a @xcite , 04eo is characterized as a fast - declining sn ia with slow photospheric velocities . however , it retains an absolute @xmath1 maximum of @xmath93 mag ( within the range of `` normal '' sn ia ) and still fits the phillips ( 1993 ) relation . spectroscopically , 04eo contains strong si ii and s ii lines , typical of sn ia . when the spectra of 07qd are blueshifted to coincide with the photospheric velocity of 04eo , many common features become apparent , especially the si ii and s ii lines . figure [ f : qdvseo ] shows a normalized comparison between early - time spectra of 07qd and 04eo that highlights both the presence and the relative strengths of these lines . similarly , we may compare the maximum - light spectra of 07qd , 08 ha , and 05hk , as seen in figure [ f : qdvshk_3d ] . the spectra of both 07qd and 08 ha , with photospheres blueshifted by 2000 km s@xmath3 and 1750 km s@xmath3 ( respectively ) , mirror many of the major features . although the si ii and s ii profiles are noticeably stronger in 07qd and 08 ha , their presence in 05hk is clear . the similarity between 07qd and 05hk below 5000 is striking identical structures in this region are indicative of similar iron - group contributions . the greatly different photospheric velocities apparent in 07qd and 05hk are examined along with 02cx and 08 ha in figure [ f : vphot_bmax ] , suggesting a relation between the luminosities and expansion velocity in these peculiar cases . we used the fourier cross - correlation subroutine ( fxcor ) in iraf to determine the average offset of the similar features in these spectra , namely the fe ii and si ii lines . the average photospheric velocities of 07qd are faster than those of 08 ha at all epochs when both have measured spectra , but lower than those of any other 02cx - like event . we compare at 1012 days past maximum because it is here that we have spectra of all four of these 02cx - like sn . to first order , we extrapolated the photospheric velocity of 07qd at maximum to be roughly 200 km s@xmath3 greater than that of 08 ha , implying a larger kinetic energy by a factor of 1.14 ( assuming the same ejected masses for both sn ) . since our calculation in 3.1 assumes nickel mass is directly related to luminosity and varies with the rise time by a @xmath94 change in @xmath4ni mass per day , the rise - time dependence is small . as a result , the plot of the photospheric velocity versus peak luminosity suggests that the @xmath4ni yield ( via maximum brightness and rise time ) and kinetic energy per unit mass ( via photospheric velocity ) are correlated in 02cx - like events . through fxcor , the average rate of decline in the photospheric velocity is found to be @xmath95 km s@xmath3 day@xmath3 for the 6355 si ii line and @xmath96 km s@xmath3 day@xmath3 for the other absorption features ( dominated by fe ii ) . both of these values group the velocity evolution of 07qd amongst the faint sn ia described by @xcite , and they are smaller in magnitude than those of 08 ha ( @xmath97 km s@xmath3 day@xmath3 for si ii and @xmath98 km s@xmath3 day@xmath3 for all other ions ) . it is difficult to determine the si ii velocities after maximum brightness for 05hk and 02cx since they are quickly blended with fe ii lines . @xcite found for both 05hk and 02cx that the velocities of the 4555 fe ii line and the ca ii h&k lines remained constant at roughly 6000 km s@xmath3 for several weeks , implying that the overall expansions were quite slow even for typical sn ia . if we assume a constant deceleration through late times , the 700 km s@xmath3 fe ii lines found 227 days past maximum in 02cx imply a 23 km s@xmath3 day@xmath3 decline @xcite . the host galaxy in figure [ f : finderchart ] shows a prominent bar and two major spiral arms . the low inclination allows minor spiral arms to be distinguishable , one of which contains the location of 07qd . the classifications of sbb or sbc suggest ongoing star formation , but 02cx - like sn ia appear to span a wide variety of galactic morphologies @xcite . the keck - i spectrum also included the host galaxy s core in the long slit , allowing a measurement of the oxygen abundance . we used both the optimized method recommended in @xcite and the @xmath99/[\mathrm{o~ii}]$ ] diagnostic to arrive at an oxygen abundance in the host galaxy of @xmath100 . this is equivalent to the sun s oxygen abundance of @xmath101 @xcite , but it exceeds that of the host galaxy of 08 ha , measured to be @xmath102 by @xcite using n2 and o3n2 diagnostics ( see @xcite ) . the striking similarity of the + 10 day ( relative to @xmath1 maximum ) spectrum of 07qd to that of 08 ha suggest that they are indeed similar explosions . since 07qd is also spectroscopically linked to 05hk and thus to 02cx as well ( see figure [ f : qdvshk_3d ] ) , these four peculiar events range in peak luminosity by 4 mag , but constitute a single spectral class . @xcite argued that 08 ha and possibly other 02cx - like objects are actually core - collapse sn ; their low luminosity is the result of either the collapse of a @xmath103 m@xmath104 star directly to a black hole , or electron capture in the core of a 78 m@xmath104 star . however , the clear presence of imes ( notably si ii @xmath1056355 and s ii @xmath1065968 , 6359 ) in 07qd suggests low - density thermonuclear burning and not a core - collapse sn ib or sn ic explosion ( see @xcite for a review of sn spectroscopic classification ) . except for the photospheric velocities , the spectra of 07qd are similar to those of sn 1992a - like sn ia , further strengthening the arguments that it is a thermonuclear runaway . the presence of these imes mirrors the findings of @xcite . sn ibc have been observed with traces of si ii @xcite or s ii @xcite . however , these lines are quite strong in 07qd , suggesting that it is a sn ia rather than a core - collapse event . additionally , the light curve of 07qd , though displaying a sharp rise that is unusual for a typical sn ia , does not show the fast decline typical of a low - luminosity sn ic or the `` plateau '' of a sn iip that achieves similar peak luminosity . the energy released by 07qd is substantially lower than that of normal sn ia , but the predicted nickel mass serves as an intermediate example between 08 ha and 05hk , further linking the two . when compared with other 02cx - like sn , spectra of 07qd mirror that of 08 ha and display striking relationships to others . these comparisons also imply that strong si ii is possible in 02cx and 05hk , though disguised " by fe ii blending due to their higher photospheric velocities . the fast evolution of 07qd suggests that these imes are only detectable for a brief time , and become masked by fe ii or recombine as the photosphere slows and cools . the presence of carbon and oxygen ions in the photosphere echoes the results of deflagration models including those of @xcite , suggesting the presence of unburned white dwarf material and supporting that this class stems from such a progenitor . also of interest is the `` sn .ia '' model , which concerns doubly degenerate white dwarf pairs undergoing a helium flash strong enough to produce @xmath4ni and absolute @xmath10 magnitudes as low as @xmath107 mag @xcite . @xcite used this model to predict the luminosities and rise times that we observe , but also spectra dominated by the he - burning products ca and ti ( detected but not prominent ) . combined with the low observed expansion velocities , we find 07qd to be an unlikely sn .ia candidate . our estimated radioactive nickel yield for 07qd ( @xmath108 m@xmath109 ) is extremely small , and it is difficult to understand how such a weak thermonuclear explosion could disrupt a chandrasekhar - mass white dwarf . @xcite found that a minimum of @xmath110 ergs of nuclear energy is required to unbind a chandrasekhar - mass white dwarf , and that can be achieved by producing 0.3 m@xmath74 of @xmath4ni . this amount is more than ten times the @xmath4ni yield we estimate for 07qd . of course , nonradioactive elements might dominate the nuclear energy term ; for example , the production of 0.37 m@xmath74 of imes could also disrupt the star . however , the synthesized elements in three - dimensional pure - deflagration models discussed by @xcite are dominated by the iron peak , especially @xmath4ni , over imes . it is also possible that the synthesis of nonradioactive @xmath111ni or other stable iron - group elements may supply at least some of the kinetic energy . to estimate the nickel yield , we have applied `` arnett s rule '' @xcite , which allows us to use the peak luminosity as an indicator of the radioactive energy deposition . the rule assumes that nearly all of the energy produced by nickel and cobalt decays is trapped and converted to ultraviolet , optical , and infrared radiation . calculations by @xcite indicate that complete trapping is an excellent approximation for the first month after the explosion when the nickel distribution is concentrated toward the center . deflagration models suggest that rayleigh - taylor instabilities cause hot nickel bubbles to rise as cooler unburned material sinks . @xcite find fairly uniform distributions of @xmath4ni in radially averaged three - dimensional pure - deflagration simulations . this leaves more radioactive nickel in the outer layers of the ejecta than predicted in detonation models and implies that some of the radioactive energy will be lost . additionally , @xcite suggest that highly asymmetric deflagrations could both unbind a white dwarf and provide the mechanism through which heavy iron - group elements may be transported to the surface without a detonation . the ejected mass estimated for 07qd is very low , as is the case for 08 ha @xcite , perhaps suggesting a sub - chandrasekhar mass explosion . however , even normal sn ia events such as sn 1992a @xcite appear to have ejected less than a chandrasekhar mass of material based on similar assumptions . 91bg - like events also give significantly low ejected masses from the methods currently being applied . we suspect that some important physics is being ignored in the ejected mass calculations and the results should be considered ambiguous . the presence of cobalt and high photospheric temperatures in the early - time spectra support the conjecture that radioactive elements were mixed up to the outer layers of 07qd . but @xcite and @xcite show that a uniform @xmath4ni distribution should only result in a 10% to 30% error in the nickel - mass estimate from arnett s rule , not the order of magnitude needed to reconcile the energetics . an `` inverted '' sn ia , where most of the nickel is in the outer layers of the ejecta ( assume @xcite @xmath112 values are @xmath113 ) , would result in inefficient @xmath114-ray deposition and extreme violation of arnett s rule . such a @xmath4ni distribution , however , is likely to result in the ejecta turning optically thin to @xmath114-rays very early . @xcite used the characteristic time for sn ia to become optically thin to @xmath114-rays , @xmath115 , as a way to estimate the ejected mass and kinetic energy . they found that @xmath115 ranges between 25 and 35 days for normal sn ia , coinciding with a range of ejecta and nickel masses . @xcite showed that for a fixed ejecta mass , opacity , and density structure , @xmath116 , so pushing radioactive nickel away from the center ( smaller nickel concentration parameter @xmath112 ) will shorten the time it takes for the ejecta to become optically thin . figure [ f : bol_luminosity ] shows the quasi - bolometric luminosity curve ( constructed from @xmath20 photometry ) for 07qd and displays radioactive energy deposition curves parameterized by the fiducial time , @xmath115 . for high optical depths , the curves should obey arnett s rule and pass through the light - curve maximum . the models that become optically thin to @xmath114-rays soon after explosion have the freedom to be shifted to higher nickel yields as the radioactive energy deposition will match the luminosity after maximum light . the best match to the observations is for @xmath117 days , suggesting the ejecta are efficiently trapping radioactive energy a month after maximum brightness . the optically thin curves decline much faster than the data , implying that arnett s rule applies even for this weak explosion . 02cx and 05hk , measured more than 200 days past maximum , continued to exhibit narrow p - cygni profiles in their spectra @xcite . 08 ha appears to be consistent with this @xcite , so 07qd may be relatively opaque to optical photons and @xmath114-rays for some time . \(1 ) 07qd was spectroscopically similar to both 08 ha and 05hk . strong lines of fe ii and co ii are present in spectra of all three objects , while fe iii and ime features are most visible at early epochs , though their optical depths decline quickly . findings ( 1 ) and ( 2 ) point toward thermonuclear burning during the explosion of 07qd , while points ( 3 ) and ( 4 ) emphasize that 07qd completes a sequence of 02cx - like sn ia both energetically and spectroscopically . the composition and velocity of the ejecta support the picture of a deflagration . however , deflagration models such as @xmath118 used by @xcite do not predict the short rise time and small nickel mass seen in 07qd or 08 ha . the small amounts of synthesized radioactive nickel argue against the unbinding of a chandrasekhar - mass progenitor , but if the fusion is dominated by imes at the expense of nickel , the explosions of massive white dwarfs are still possible . other possible scenarios are as follows . \(a ) the explosions are thermonuclear with sub - chandrasekhar masses . this would decrease the progenitors gravitational binding energies , thus lowering the ejecta masses and internal energies needed to disrupt the stars . \(b ) arnett s rule does not apply to this class of objects , implying that the @xmath4ni mass is significantly underestimated . the bolometric light curve of 07qd , however , suggests that arnett s rule is operational to some degree . the low velocities and energies present in 08 ha and 07qd enable the analysis of many aspects of 02cx - like sn , otherwise hidden by the fe ii blending present in 05hk and 02cx . though the velocities and energies span a wide range , together they constitute a well - defined group of peculiar sn ia . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation ( nsf ) , the u.s . department of energy ( doe ) , the national aeronautics and space administration ( nasa ) , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . some of the data presented herein were obtained at the w. m. keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california , and nasa ; it was made possible by the generous financial support of the w. m. keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community ; we are most fortunate to have the opportunity to conduct observations from this mountain . we thank the keck staff for their assistance . we are grateful for the financial support of the university of notre dame , nasa / stsci grant hst - go-10893.01-a to c.m.m . , the nsf , and the doe , specifically nsf grant ast0908886 and doe grant de - fg02 - 08er41563 to a.v.f . support for this research at rutgers university was provided by doe grant de - fg02 - 08er41562 and nsf award ast0847157 to s.w.j . we thank brian hayden , joe gallagher , and weidong li for discussions and their help in the production of this paper , jerod parrent and the online supernova spectrum archive ( suspect ) , and david jeffery along with the suspend database . adelman - mccarthy , j. k. , et al . 2008 , apjs , 175 , 297 aldering , g. , et al . 2006 , , 650 , 510 arnett , w. d. 1982 , , 253 , 785 arnett , w. d. , branch , d. , & wheeler , j. c. 1985 , nature , 314 , 337 baron , e. , lentz , e. j. , & hauschildt , p. h. 2003 , , 588 , l29 bassett , b. , et al . 2007 , cbet , 1137 benetti , s. , cappellaro , e. , turatto , m. , taubenberger , s. , harutyunyan , a. , & valenti , s. 2006 , , 653 , l129 benetti , s. , et al . 2005 , , 623 , 1011 bildsten , l. , shen , k. j. , weinberg , n. n. , & nelemans , g. 2007 , , 662 , l95 blinnikov , s. i. , rpke , f. k. , sorokina , e. i. , gieseler , m. , reinecke , m. , travaglio , c. , hillebrandt , w. , & stritzinger , m. 2006 , , 453 , 229 branch , d. 1981 , , 248 , 1076 branch , d. , baron , e. , thomas , r. c. , kasen , d. , li , w. , & filippenko , a. v. 2004 , , 116 , 903 brown , g. e. , lee , c .- h . , & moreno mndez , e. 2007 , , 671 , l41 caffau , e. , ludwig , h .- g . , steffen , m. , ayres , t. r. , bonifacio , p. , cayrel , r. , freytag , b. , & plez , b. 2008 , , 488 , 1031 calder , a. c. , plewa , t. , vladimirova , n. , lamb , d. q. , & truran , j. w. 2004 , arxiv : astro - ph/0405162 cappellaro , e. , mazzali , p. a. , benetti , s. , danziger , i. j. , turatto , m. , della valle , m. , & patat , f. 1997 , , 328 , 203 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 chornock , r. , filippenko , a. v. , branch , d. , foley , r. j. , jha , s. , & li , w. 2006 , , 118 , 722 filippenko , a. v. 1997 , , 35 , 309 filippenko , a. v. 2003 , in from twilight to highlight : the physics of supernovae , ed . w. hillebrandt & b. leibundgut ( berlin : springer - verlag ) , 171 filippenko , a. v. , et al . 1992a , , 104 , 1543 filippenko , a. v. , et al . 1992b , , 384 , l15 fisher , a. , branch , d. , nugent , p. , & baron , e. 1997 , , 481 , l89 fisher , a. 2000 , phd . thesis , university of oklahoma foley , r. j. , et al . 2003 , , 115 , 1220 foley , r. j. , et al . 2009 , aj , 138 , 376 foley , r. j. , brown , p. j. , rest , a. , challis , p. j. , kirshner , r. p. , & wood - vasey , w. m. 2010 , , 708 , l61 freedman , w. l. , et al . 2001 , , 553 , 47 frieman , j. a. , et al . 2008 , , 135 , 338 fukugita , m. , et al . 1996 , , 111 , 1748 gamezo , v. n. , khokhlov , a. m. , & oran , e. s. 2004 , in cosmic explosions in three dimensions , ed . p. hflich , p. kumar , & j. c. wheeler ( cambridge : cambridge univ . press ) , 121 garnavich , p. m. , et al . 1998 , , 493 , l53 garnavich , p. m. , et al . 2004 , , 613 , 1120 goldhaber , g. , et al . 2001 , , 558 , 359 gunn , j. e. , et al . 1998 , , 116 , 3040 gunn , j. e. , et al . 2006 , , 131 , 2332 guy , j. , et al . 2007 , , 466 , 11 hamuy , m. , phillips , m. m. , maza , j. , suntzeff , n. b. , schommer , r. a. , & avils , r. 1995 , , 109 , 1 hamuy , m. , phillips , m. m. , suntzeff , n. b. , schommer , r. a. , maza , j. , smith , r. c. , lira , p. , & aviles , r. 1996 , , 112 , 2438 hamuy , m. , et al . 1996a , , 112 , 2391 hamuy , m. , et al . 1996b , , 112 , 2398 hamuy , m. , et al . 2003 , , 424 , 651 harkness , r. p. 1991 , in sn 1987a and other supernovae , ed . i. j. danziger & k. kjr ( garching : eso ) , 447 hatano , k. , branch , d. , fisher , a. , millard , j. , & baron , e. 1999 , , 121 , 233 hayden , b. t. , et al . 2010 , submitted ( arxiv:1001.3428 ) hill , g. j. , et al . 1998 , spie conf . , 3355 , p. 424 holtzman , j. a. , et al . 2008 , , 136 , 2306 horne , k. 1986 , , 98 , 609 howell , d. a. , et al . 2009 , , 691 , 661 jeffery , d. j. 1999 , arxiv : astro - ph/9907015 jha , s. , riess , a. g. , & kirshner , r. p. 2007 , , 659 , 122 jha , s. , et al . 2004 , , 132 , 189 kasen , d. 2006 , , 649 , 939 kewley , l. j. , & dopita , m. a. 2002 , , 142 , 35 kessler , r. , et al . 2009a , , 185 , 32 kessler , r. , et al . 2009b , , 121 , 1028 khokhlov , a. m. 1991 , , 245 , 114 lampeitl , h. , et al . 2009 , , 1782 leibundgut , b. , et al . 1993 , , 105 , 301 leonard , d. c. , et al . 2002 , , 124 , 2490 li , w. , et al . 2003 , , 115 , 453 livio , m. 2000 , in type ia supernovae : observations and theory , ed . j. c. niemeyer & j. w. truran ( cambridge : cambridge univ . press ) , 33 lupton , r. h. , gunn , j. e. , & szalay , a. s. 1999 , , 118 , 1406 maeda , k. , roepke , f. k. , fink , m. , hillebrandt , w. , travaglio , c. , & thielemann , f. k. 2010 , submitted ( arxiv:1002.2153 ) munari , u. , & zwitter , t. 1997 , , 318 , 269 nomoto , k. , et al . 2000 , in gamma - ray bursts , 5th huntsville symposium , 526 , 622 nugent , p. , phillips , m. , baron , e. , branch , d. , & hauschildt , p. 1995 , , 455 , l147 oke , j. b. , et al . 1995 , , 107 , 375 pastorello , a. , et al . 2007 , , 377 , 1531 perlmutter , s. , et al . 1999 , , 517 , 565 pettini , m. , & pagel , b. e. j. 2004 , , 348 , l59 phillips , m. m. 1993 , , 413 , l105 phillips , m. m. , et al . 1992 , , 103 , 1632 phillips , m. m. , et al . 1999 , , 118 , 1766 phillips , m. m. , et al . 2007 , , 119 , 360 pinto , p. a. , & eastman , r. g. 2000a , , 530 , 744 pinto , p. a. , & eastman , r. g. 2000b , , 530 , 757 prieto , j. , garnavich , p. , depoy , d. , marshall , j. , eastman , j. , & frank , s. 2005 , cbet , 302 , 1 pskovskii , yu . 21 , 675 reinecke , m. , hillebrandt , w. , & niemeyer , j. c. 2002 , , 391 , 1167 richmond , m. w. , et al . 1996 , , 111 , 327 riess , a. g. , press , w. h. , & kirshner , r. p. 1995 , , 438 , l17 riess , a. g. , press , w. h. , & kirshner , r. p. 1996 , , 473 , 88 riess , a. g. , et al . 1998 , , 116 , 1009 riess , a. g. , et al . 2009 , 699 , 539 sahu , d. k. , et al . 2008 , , 680 , 580 sako , m. , et al . 2008 , , 135 , 348 sauer , d. n. , mazzali , p. a. , deng , j. , valenti , s. , nomoto , k. , & filippenko , a. v. 2006 , , 369 , 1939 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 schmidt , w. , & niemeyer , j. c. 2006 , , 446 , 627 shen , k. j. , kasen , d. , weinberg , n. n. , bildsten , l. , & scannapieco , e. 2010 , arxiv:1002.2258 smith , j. a. , et al . 2002 , , 123 , 2121 sollerman , j. , et al . 2009 , , 703 , 1374 stritzinger , m. , et al . 2006 , , 450 , 241 travaglio , c. , hillebrandt , w. , reinecke , m. , & thielemann , f .- k . 2004 , , 425 , 1029 turatto , m. , et al . 1996 , , 283 , 1 valenti , s. , et al . 2008 , , 673 , l155 valenti , s. , et al . 2009 , , 459 , 674 wade , r. a. , & horne , k. 1988 , , 324 , 411 woosley , s. e. , kasen , d. , blinnikov , s. , & sorokina , e. 2007 , , 662 , 487 yokoo , t. , arimoto , j. , matsumoto , k. , takahashi , a. , & sadakane , k. 1994 , , 46 , l191 york , d. g. , et al . 2000 , , 120 , 1579 cccccc 54346.41 & -0.010 @xmath119 2.726 & 1.620 @xmath119 1.230 & 0.880 @xmath119 1.235 & -0.730 @xmath119 1.281 & -1.100 @xmath119 3.851 + 54348.41 & -1.780 @xmath119 1.614 & 0.610 @xmath119 0.648 & 1.520 @xmath119 0.698 & 0.200 @xmath119 1.019 & 1.600 @xmath119 3.929 + 54355.42 & 1.500 @xmath119 2.146 & -0.470 @xmath119 0.530 & 0.410 @xmath119 1.111 & -0.370 @xmath119 1.813 & 5.870 @xmath119 5.868 + 54358.37 & 4.310 @xmath119 1.878 & 0.870 @xmath119 0.530 & -0.000 @xmath119 0.707 & -1.480 @xmath119 1.159 & -1.940 @xmath119 4.186 + 54365.40 & 0.000 @xmath119 0.971 & -0.180 @xmath119 0.315 & 0.000 @xmath119 0.496 & -1.500 @xmath119 0.719 & -1.470 @xmath119 3.225 + 54381.42 & 1.400 @xmath119 1.165 & 0.010 @xmath119 0.362 & -0.090 @xmath119 0.590 & -0.340 @xmath119 0.790 & -4.300 @xmath119 3.010 + 54384.43 & 1.550 @xmath119 1.396 & 0.130 @xmath119 0.393 & -0.580 @xmath119 0.557 & -2.860 @xmath119 0.817 & -5.780 @xmath119 3.438 + 54386.41 & 0.260 @xmath119 1.478 & 0.080 @xmath119 0.489 & 0.230 @xmath119 0.634 & 0.670 @xmath119 0.846 & -1.960 @xmath119 4.057 + 54388.42 & 0.240 @xmath119 1.431 & 0.270 @xmath119 0.399 & 0.060 @xmath119 0.587 & -0.250 @xmath119 0.779 & 2.090 @xmath119 3.177 + 54392.42 & -0.840 @xmath119 1.554 & -0.420 @xmath119 0.556 & 1.160 @xmath119 0.837 & 0.120 @xmath119 1.228 & -1.630 @xmath119 4.251 + 54393.42 & 2.100 @xmath119 1.731 & -0.020 @xmath119 0.564 & -0.670 @xmath119 0.546 & -0.050 @xmath119 0.850 & 3.650 @xmath119 4.373 + 54396.29 & -7.950 @xmath119 7.556 & -2.870 @xmath119 3.713 & 0.720 @xmath119 2.329 & - & -3.160 @xmath119 7.688 + 54405.39 & 13.620 @xmath119 2.416 & 19.170 @xmath119 1.007 & 16.510 @xmath119 0.990 & 15.750 @xmath119 1.281 & 7.640 @xmath119 4.463 + 54406.33 & 6.930 @xmath119 1.466 & 17.310 @xmath119 0.638 & 17.080 @xmath119 0.709 & 15.930 @xmath119 0.912 & 17.020 @xmath119 2.989 + 54412.35 & 2.470 @xmath119 1.702 & 9.760 @xmath119 0.514 & 15.720 @xmath119 0.899 & 13.850 @xmath119 1.204 & 15.080 @xmath119 4.166 + 54416.32 & 0.920 @xmath119 1.072 & 6.350 @xmath119 0.388 & 14.140 @xmath119 0.574 & 13.620 @xmath119 0.769 & 17.010 @xmath119 3.035 + 54421.33 & 3.480 @xmath119 1.736 & 5.250 @xmath119 0.502 & 13.290 @xmath119 0.797 & 11.980 @xmath119 1.110 & 12.360 @xmath119 4.757 + 54423.31 & 1.060 @xmath119 1.924 & 4.200 @xmath119 0.829 & 12.460 @xmath119 1.058 & 13.620 @xmath119 1.197 & 19.620 @xmath119 3.800 + 54433.33 & - & 5.430 @xmath119 1.501 & 8.780 @xmath119 2.494 & 9.740 @xmath119 2.832 & 5.980 @xmath119 6.150 + c c c c c c tng & 5 nov . 2007 & 02:06:10 & + 3 & 3 @xmath120 1800 & 36737401 + het & 10 nov . 2007 & 04:23:12 & + 8 & 1200 & 40759586 + keck & 12 nov . 2007 & 12:39:01 & + 10 & 1500 & 30738800 + het & 17 nov . 2007 & 03:58:29 & + 15 & 1200 & 40749586 + r c c c c c na i & 0.3 & 4.5 & 8.0 & 4 & 10 + na i & 0.5 & 2.8 & 4.5 & 1 & 10 + na i & 0.1 & 8.0 & 10 . & 2 & 10 + ca ii & 10 & 2.8 & 3.0 & 2 & 12 + s ii & 2.7 & 2.8 & 3.3 & 2 & 10 + si ii & 3.0 & 2.0 & @xmath121 & 1 & 8 + si iii & 2.0 & 2.8 & @xmath121 & 1 & 10 + fe iii & 0.7 & 2.8 & @xmath121 & 1 & 10 + fe ii & 1.0 & 2.8 & @xmath121 & 1 & 10 + co ii & 3.0 & 2.8 & @xmath121 & 1 & 10 + c ii & 0.02 & 2.8 & @xmath121 & 1 & 10 + c iii & 0.5 & 5.0 & 8.0 & 2 & 10 + o i & 9.5 & 2.8 & 6.5 & 3 & 10 + o iii & 1.8 & 2.8 & @xmath121 & 1 & 10 + mg i & 0.5 & 2.8 & @xmath121 & 2 & 12 + ti i & 1.0 & 2.8 & @xmath121 & 1 & 10 + cr i & 2.0 & 2.8 & 4.0 & 2 & 10 + ca i & 2.0 & 5.0 & @xmath121 & 1 & 10 + al i & 3.0 & 2.8 & @xmath121 & 1 & 10 + r c c c c c co ii & 10 & 2.8 & 3.8 & 1 & 8 + fe ii & 10 & 2.8 & 3.8 & 1 & 8 + ca ii & 35 & 2.8 & 4.8 & 1 & 8 + na i & 0.4 & 2.8 & 4.8 & 1 & 8 + na i & 0.3 & 4.0 & 7.0 & 8 & 8 + na i & 0.3 & 7.0 & 11.0 & 3 & 8 + si ii & 3.0 & 2.8 & 3.8 & 1 & 8 + sc ii & 1.0 & 2.8 & 5.8 & 1 & 10 + o i & 1.0 & 2.8 & @xmath121 & 1 & 8 + o ii & 0.2 & 3.5 & @xmath121 & 1 & 8 + o iii & 2.0 & 2.8 & @xmath121 & 1 & 8 +

entanglement plays an important role in both quantum foundations and quantum information . recently however , an overlap between these two fields has emerged which goes beyond this somewhat obvious reason for common interest . for instance , there is mounting evidence that contextuality plays a role in any enhancement of quantum over classical communication protocols @xcite . a second example is work on the `` _ _ r - p _ _ '' framework for analyzing the information processing power of general ( not necessarily quantum ) probabilistic theories @xcite - a body of work to which this paper is closely related . finally we mention work by one of us @xcite showing that one particular class of ontological models @xcite ( or ` hidden variable theories ' ) of quantum mechanics has a computational power slightly greater than that of regular quantum computation ( if access to values of the hidden variables is granted ) . as in this last example , we will be considering ontological models . such models reproduce the predictions of quantum mechanics by using ( positive , normalized ) probability distributions over a space of ( sometimes , though not always , hidden ) variables to describe the state of a system . unlike quantum states in operational quantum mechanics , these variables are presumed to correspond directly to `` properties of reality '' in some fashion - hence we term the variables the _ ontic states_. measurements in these models are described by positive valued _ indicator functions _ - functions which simply determine the probability of a system in a given ontic state yielding the particular measurement outcome associated with that indicator function . in this way ontological models differ from the many well known quasi - probabilty representations of quantum mechanics . in those representations , even if the states are represented by positive , normalized probability distributions ( e.g. the q - function ) , the measurements will be represented by indicator functions which can be negative ( or greater than 1 ) . such negativity is difficult to interpret if one is interested in describing properties of a reality underpinning quantum mechanics . in fact spekkens has shown that such negativity arises in these representations precisely because they are assumed non - contextual @xcite . in particular , this assumption of non - contextuality is made in the work on the _ r - p _ formalism mentioned above , and is the reason the @xmath0-vectors ( those vectors associated with measurements ) generally have to contain negative elements . similar considerations apply to the work on discrete wigner distributions @xcite . a comprehensive discussion of the ontological model framework was undertaken in @xcite . for our purposes we will not need the majority of the formalism developed in that paper . however , let us mention in passing a few of the conclusions from it . firstly , there exist a wide variety of ontological models , both deterministic and indeterministic , including ones which , unlike the examples given in @xcite , do not provide greater computational power than that of regular quantum computation . secondly , for some models it is necessary to consider not just ontic states of the system under consideration , but also those of the preparation and measurement devices . thirdly , the manner in which such models exhibit contextuality is varied , but one feature ( which in some sense subsumes contextuality ) that is common to all models is a property we term _ deficiency_. loosely speaking , deficiency breaks a symmetry between preparations and measurements in quantum mechanics . it is the property that the set of ontic states which a system prepared in quantum state @xmath1 may _ actually _ be in , is strictly smaller than the set of ontic states which would reveal the measurement outcome @xmath2 with certainty . in this paper , we initiate an investigation into how ontological models can be used to reproduce quantum mechanical statistical predictions for a discrete set of preparations of , and measurements upon , a given system . we start in section [ sec : data_tables ] by discussing a method of representing empirical data in matrix form . in section [ ofs ] we then introduce a matrix factorization of this data , equivalent to an ontological model over a finite number of ontic states . in sections [ sec : model1 ] , [ sec : model2 ] and [ sec : model3 ] we present three factorizations which can describe any such empirical data . we then discuss in section [ detfromindet ] how to turn an indeterministic ontological model into a deterministic one . this raises the issue of how such models manifest the necessary contextuality , and in section [ sec : contextuality ] we discuss this and provide an explicit example . in the second half of the paper , section [ oc ] , we consider the issue of finding ontological models with as few ontic states as possible - a topic with potential application to the classical simulation of quantum systems . we begin with a common @xcite , although somewhat idealistic , formalization of the process of performing an experiment resulting in probabilistic data - data for which we then seek an explanation . the experiment consists of a set of preparation procedures , and a set of measurement procedures . the preparation procedures are defined in terms of different macroscopic ( and therefore distinguishable ) configurations ( settings ) of an apparatus , and these are labeled @xmath3 , where @xmath4 ( @xmath5 standing for `` states '' ) . likewise the different measurement procedures are macroscopically distinguishable , and can be labeled @xmath6 , where @xmath7 . each measurement procedure has some number @xmath8 of outcomes , which , by padding with null outcomes if necessary , we take to be the same for each . we can then label the occurrence of the @xmath9 outcome , when the measurement procedure is @xmath6 , as @xmath10 . we imagine that a large number of experiments are performed , and the probabilities that a given preparation yields a given measurement outcome estimated and tabulated , e.g : @xmath11.\end{array}\ ] ] in this example @xmath12 , @xmath13 and @xmath14 , so that there are @xmath15 rows in the matrix . the sum of every @xmath8 elements within each column must be 1 , since one of the outcomes must obtain for every measurement . for clarity we have drawn a horizontal line underscoring the separation between distinct measurements . we will call such a matrix of values a data table , and generally denote it @xmath16 . it will often be convenient to consider @xmath16 a three - dimensional array , where the particular measurement being performed lies along the third dimension . for the above example we can do so by defining @xmath17\end{array},\\ d^{(2)}&= & \begin{array}{r@{\extracolsep{\fill}}l } & \begin{array}{ccccc } & \:\:\:\mathcal{p}^{(1 ) } & \mathcal{p}^{(2 ) } & \mathcal{p}^{(3 ) } & \mathcal{p}^{(4 ) } \\ \end{array}\\ \begin{array}{c } \mathcal{m}^{(2)}_1\\ \mathcal{m}^{(2)}_2\\ \mathcal{m}^{(2)}_3\\ \end{array } & \left[\begin{array}{cccc } 0.35 \:\ : & 0.72 \:\ : & ... \:\ : & ... \:\:\\ 0.60 \:\ : & 0.21 \:\ : & ... \:\ : & ... \:\:\\ 0.05 \:\ : & 0.06 \:\ : & ... \:\ : & ... \:\:\\ \end{array}\right]\end{array}.\end{aligned}\ ] ] more generally the data table can be taken as specified in terms of @xmath18 , where @xmath19 , @xmath20 , @xmath21 . now in quantum mechanics we generally associate a density matrix @xmath22 with a preparation @xmath23 , and a povm element @xmath24 with a measurement outcome @xmath25 . since we will frequently be considering data tables for which the table entries are presumed to be given by quantum mechanical expectations - i.e. @xmath26 , it is often convenient to label the rows and columns via the appropriate quantum operators . in fact , we will almost always only be considering procedures which correspond to pure quantum states @xmath27 and to sets of sharp ( projective ) measurements @xmath28 ( also known as pvm s ) . in this case we have @xmath29 and it will be convenient to adopt a more quantum notation along the lines of : @xmath30.\end{array}\ ] ] at this stage it is pertinent to consider the following : [ prop1 ] every possible data table can be realized via standard quantum expectation values of projective measurements on pure states . at first sight this proposition may appear incorrect . consider , for example , the following data table : @xmath31.\end{array}\ ] ] as there are two possible outcomes for each measurement , one might surmise that the data was generated by sharp measurements on a qu**b**it prepared in either @xmath32 or @xmath33 . in that case the table is clearly impossible - the first measurement implies the two states are orthogonal , and the second implies there is a two outcome measurement which yields one of the states with certainty , but which is unbiased with respect to the second state . however , consider if the system was quantum mechanically actually a three dimensional qu**tr**it with basis states @xmath34 . in this case it is easy to verify that choosing @xmath35 , the two sets of measurements could be @xmath36 and @xmath37 , where @xmath38 . that is , the states are sharp and the measurements are projective . the point is that there is no way of determining from the raw data table itself any presumed ( quantum ) hilbert space dimension of the systems under investigation . we will use the general notation @xmath39 to denote the hilbert space dimension of such a quantum representation of a data table - for this example @xmath40 . proposition 1 asserts that _ every _ data table can be so constructed - that is , by pure quantum states and projective operators . that this is possible can be seen as follows . since we are not restricted in @xmath39 , let us imagine that every preparation procedure ( column of the data table ) is represented by a state which is orthogonal to all the other states . that is , each state lies in its own subspace of the total hilbert space . clearly , if we can choose measurements so as to reproduce any desired statistics for one such state then , since the states are all orthogonal , we can piece together measurement operators for the different states in a direct sum to produce suitable measurement operators for the whole table . focussing on one state then , imagine the measurement under consideration has outcomes occurring with probabilities @xmath41 . all we need to establish is that , for a given @xmath1 , there is a way of choosing projection operators such that @xmath42 . one way of constructing suitable operators is to work in the basis where @xmath43^t$ ] . let @xmath44 be the row vector @xmath45 $ ] , and construct the unitary matrix which has @xmath44 as its first row and the remaining @xmath46 rows any orthogonal basis of the support of @xmath47 . then the columns of this unitary matrix form a set of orthogonal states , and projectors onto these states satisfy @xmath42 . this procedure is extremely wasteful in terms of @xmath39 , the hilbert space dimension used to represent generic data . this raises the following : [ hilbspacedim ] is there an efficient procedure for finding the smallest hilbert space dimension required to represent an arbitrary data table in terms of pure states and projective measurements ? although we do not have an answer to this problem , we note that being able to solve it would allow one to ( approximately ) solve a quantum one way communication complexity problem . thus open problem 1 may benefit from studies in this field . to see the relation , note that by sending @xmath48 copies of systems prepared according to a procedure @xmath3 , a party alice can have another party bob estimate the probability of the @xmath49 outcome of some measurement @xmath50 occurring . the fixed accuracy of bob s estimation is determined by the size of @xmath48 . thus , if the systems realizing the preparations and measurements consist of @xmath51 qubits then alice needs to send bob @xmath52 qubits for him to obtain his estimation of the desired probability . but suppose that we consider @xmath3 and @xmath50 to be specified by binary strings @xmath53 and @xmath54 and denote the probability that bob is attempting to estimate by @xmath55 . then requiring bob to perform the estimation of @xmath55 with alice sending as few qubits as possible is a quantum one way communication complexity problem . thus the efficient procedure required by open problem 1 would allow us to minimize @xmath51 and thus ( given that @xmath48 is a constant overhead determined only by the accuracy of bob s estimation ) allow us to solve the associated quantum one way communication complexity problem . the lesson of proposition [ prop1 ] is that , at its most basic level , the probabilistic structure of quantum mechanics is without physical content . to obtain such content extra assumptions or restrictions must be imposed . for example , assuming that measurements on distinct systems are described by separable ( tensor products of ) projection operators , the hilbert space structure we build to yield @xmath16 can not be guaranteed to reproduce all possible _ correlated _ probability distributions between multiple systems . these precautions are well known but bear repeating , because in this paper we will be looking at representing a data table via a different probabilistic formalism , a formalism essentially that of classical probability theory . it is well understood that the ontological model formalism allows for an arbitrary data table to be represented - therefore our goal is to investigate more than simply whether such models can reproduce empirical observations . in fact we are primarily interested in using these models to obtain a deeper understanding of the restrictions which occur on any proposed realistic explanation for quantum mechanics . as such we will be looking at the structure of such models when they are used to model data tables corresponding to hypothetical experiments which in quantum mechanics would be described by pure state preparations and rank one projective measurements on a system of a known hilbert space dimension . therefore we will generally have the case that @xmath56 , and open problem [ hilbspacedim ] will not be of concern to us . our primary goal is to examine factorizations of the data matrix @xmath57 , @xmath19 into the product of a @xmath58 _ measurement matrix _ @xmath59 and an @xmath60 _ preparation matrix _ @xmath61 ( both positive - real valued ) : @xmath62 the columns of the preparation matrix @xmath61 are normalized probability distributions i.e. @xmath61 is column - stochastic ; it satisfies @xmath63 , and @xmath64 for all @xmath65 . the interpretation of @xmath61 is that its @xmath9 column is a ( classical ) probability distribution , over @xmath66 distinct _ ontic states _ , which corresponds to the preparation procedure @xmath23 of @xmath16 . that is , we imagine that when a system is prepared according to @xmath23 , it really is prepared in one of the @xmath66 ontic states , and our ignorance of which particular ontic state the system is in is represented by the probability distribution in column @xmath67 of @xmath61 . each row of the measurement matrix corresponds to an _ indicator function _ over the @xmath66 ontic states . that is , @xmath68 is the probability of obtaining measurement outcome @xmath69 , if the actual ontic state of the system is @xmath65 , given that measurement @xmath70 is being performed . besides ( [ mpequalsd ] ) there is one extra constraint on @xmath71 , due to our desire to be able to interpret an ontological factorization as a model of the probabilistic data arising from a realistic framework . specifically , regardless of which ontic state a system is in , one of the measurement outcomes _ must _ pertain . for instance , if we consider the @xmath72 measurement performed on a system actually in ontic state @xmath65 , then one of the @xmath8 outcomes must occur . as such , we need , @xmath73 for each of the @xmath74 . thus the @xmath59 are also necessarily column - stochastic . this is actually an important extra constraint - without it one can reproduce quantum statistics with non - contextual ontological models not having negative indicator functions - which has been shown not to be possible by spekkens @xcite . ( they would have a decidedly weird ontology however , as they would require a strange non - separable conspiracy between preparation and measurement devices . ) with the above constraints in mind , we can define an ontological factorization more formally as : a data table @xmath16 possesses an ontological factorization ( of ) over @xmath66 ontic states if there exists an @xmath60 , column - stochastic matrix @xmath61 , and @xmath8 matrices @xmath59 , each @xmath75 dimensional and column - stochastic , which satisfy @xmath76 for all @xmath77 . [ def : of ] if the entries of @xmath71 are all 0 or 1 , then the ontological factorization is deterministic , otherwise it is indeterministic . we now briefly present a trivial of , which is easily performed for any data table . this will be rendered much less trivial in subsection [ detfromindet ] . specifically , in this of of the data table @xmath16 we choose @xmath78 this of has @xmath79 . that is , the number of ontic states is equal to the number of preparation procedures ( quantum states ) used in constructing the data table . in terms of an ontological model , this of implies that preparing a system in a quantum state @xmath80 is equivalent to preparing the system in a single ( determined ) ontic state ( since each column of @xmath61 has only one non - zero entry ) . in the terminology of @xcite this model is @xmath81-ontic ( and @xmath81-complete ) . measurements in this model are represented by indeterministic indicator functions - the model simply specifies what the probability is that a system in a given ontic state yields each outcome of any specified measurement . this ontological model , when extrapolated to a `` continuum limit '' ( i.e. constructed for a data table of _ all _ measurements and preparation procedures for a quantum system of some fixed hilbert space dimension ) , becomes the model of beltrammati - bugajski @xcite , which was presented as example number @xmath82 in @xcite . in that model the ontological state space is simply the complex projective space , the quantum state is represented by a dirac delta function distribution , and the indicator functions over the space are indeterministic , with probabilistic weights defined by the born rule . model 1 formally encapsulates the commonly expressed viewpoint that the quantum state _ is _ the state of reality . we now turn to a deterministic of which can be constructed for any data table . this of has @xmath83 ontic states . it is convenient to index the ontic states ( that is , the rows of @xmath61 and columns of @xmath71 ) by an @xmath84-tuple of integers of the form @xmath85 where each of the @xmath86 ranges over @xmath87 . we then assign probabilities over the ontic states to create @xmath61 as follows : @xmath88 alternatively we could specify that the @xmath9 column of @xmath61 , denoted @xmath89 , is formed from the @xmath9 column of @xmath16 as follows:@xmath90 consider summing over index @xmath91 : @xmath92 since @xmath93 is column - stochastic . we see that in general summing over an index simply removes all terms from the product which contain that index . summing over all row indices @xmath86 in this way we confirm that @xmath61 is column - stochastic as required . we now consider how to construct row @xmath69 of @xmath59 , which corresponds to the indicator function for @xmath94 . we assign the value 1 to every entry in this row for which the column index @xmath85 has @xmath95 . more formally , @xmath96 to see that this works as desired , take , for example , @xmath97 , so we are dealing with the first row of @xmath59 . we compute the inner product of this row with the matrix @xmath61 : @xmath98 where we have used the observation in eq . ( [ peqn ] ) multiple times . thus we see that we reproduce the required statistics of the data table . an example may help illustrate the idea . consider the following data table with @xmath14 and @xmath13 : @xmath99.\ ] ] following the construction above , the matrices @xmath71 and @xmath61 are given by @xmath100 there are two points to make about this model . firstly , unlike model 1 , the columns of the matrix @xmath61 from model 2 are not orthogonal ( disjoint ) . that is , the probability distributions corresponding to non - orthogonal states overlap . such models are called @xmath81-epistemic in @xcite . in such models , knowing the ontic state of the system does not allow one to infer with certainty what quantum state would describe the same preparation . unfortunately , this is only true in the case of a finite data table , @xmath101 . in the continuum limit of this model , the probability distributions of non - orthogonal states become disjoint . to prove this rigorously requires a diversion into the infinite tensor product structures first considered by von neumann @xcite , and we will not do so here . we point out , however , that even though it is @xmath81-ontic in the continuum limit , it is not , in the terminology of @xcite , @xmath81-complete . ( a @xmath81-complete model would have only one ontic state consistent with any given quantum state . ) finding models which are @xmath81-epistemic in the continuum limit is not easy , some attempts were given in @xcite , and we believe barrett has made some recent progress @xcite . because there is a continuum of quantum states , one way to ensure an ontological model representation of all states and measurements in a hilbert space of dimension @xmath39 is @xmath81-epistemic is simply to find a model which makes use of only a finite number of ontic states . hardy @xcite has shown that this is not possible - as the number of states and measurements increases , the number of ontic states must increase . however , hardy s analysis relies only on the fact that we must be able to divide the ontic state space into disjoint sets ( to represent orthogonal measurements ) , and that as we go to the continuum limit there are an infinite number of orthogonal measurements that need to be represented in this manner . or those yielding outcome @xmath102 ( i.e. deficiency ) . however these are easily incorporated and the conclusions remain unchanged . ] now the number of ways of dividing @xmath66 ontic states into disjoint sets grows exponentially . it would therefore seem that the possibility remains open that the number of ontic states required for an indeterministic model might grow only _ logarithmically _ as the size of the data table . but in fact this is not possible . to see this , consider the simplest case of of s of data tables having @xmath103 , which we can show must satisfy the tight bound @xmath104 . clearly models 1 and 2 set an upper bound of @xmath105 on the required number of ontic states , by providing deterministic and indeterministic factorizations having @xmath79 and @xmath106 respectively . to give the tight bound claimed , it remains to show that for given @xmath5 and @xmath84 there always exists a data table requiring an ontological factorization having @xmath107 if @xmath108 or @xmath109 if @xmath110 . we will give a general construction for a data table that lower bounds @xmath66 and then show how it applies in the case of each of these inequalities between @xmath5 and @xmath84 . first note that we can write the data tables we consider as including only the probabilities for the first outcome of each measurement ( since @xmath103 the probabilities of the second outcomes are then trivially determined ) . thus we can consider a data table with some parameters @xmath111 and @xmath112 having columns made up from all @xmath111-bit binary strings . for example , if @xmath113 and @xmath114 then the data table thus constructed would be , @xmath115 . \end{array } \label{table_upper_bound}\ ] ] but an of of any data table of this form must employ at least @xmath116 ontic states . this follows since the outcomes of measurements @xmath117 and @xmath118 are completely determined given any of the @xmath119 preparations shown ( i.e. the relevant entries of @xmath16 are all @xmath120 ) ; therefore any ontic states in the supports of these preparations ( i.e. assigned a non - zero value in the relevant column of an of s @xmath61 matrix ) must be deterministic ( take @xmath120 values ) with respect to these three measurements . thus in the example given in ( [ table_upper_bound ] ) , we must use at least @xmath119 ontic states to deal with the @xmath119 distinct possibilities for assigning outcomes to @xmath121 , @xmath122 and @xmath123 . clearly , by the same reasoning , such a data table constructed with @xmath116 preparations will require @xmath124 . similarly , any data tables having general parameters @xmath5 and @xmath84 and containing _ sub_-tables of the form ( [ table_upper_bound ] ) will require ofs with at least @xmath116 ontic states . thus it only remains to determine the largest such sub - table - i.e. largest value of @xmath111 - that can fit in data tables with parameters subjected to each of the constraints @xmath108 and @xmath125 . if we have @xmath108 then we can find a sub - table of the form ( [ table_upper_bound ] ) having @xmath126 and @xmath127 , giving @xmath128 . if however , @xmath125 , then the largest sub - table of the form ( [ table_upper_bound ] ) that we can find will have @xmath129 and @xmath127 , yielding @xmath130 . thus in both cases the lower bound @xmath131 holds . combining this with the trivial upper bound provided by models 1 and 2 gives the tight bound @xmath104 , showing that as the size of a data table is increased , the number of ontic states must grow faster than a logarithmic dependence on the parameters @xmath5 and @xmath84 . the general question of how many ontic states are required to represent any particular _ given _ data table is taken up in section iv . the second point to make about model 2 is that the number of ontic states grows exponentially with the number of measurements being performed . however , this need not always be the case , as can be seen for models in which @xmath66 depends on @xmath84 by invoking caratheodory s theorem to yield a lower bound on @xmath66 . ) , and that the entries of @xmath16 are generated by convex combinations of these strings ( with coefficients given by the columns of @xmath61 ) . ] in the next section we explicitly demonstrate such a lower bound by introducing our final model ; a deterministic ontological factorization having @xmath132 . when @xmath108 . but note that in this case we have @xmath133 . ] our final ontological factorization is tailored to employ a number of ontic states that does not increase exponentially with the number of measurements to be described . this model is best illustrated by an example of its construction . consider the data table from ( [ model2d ] ) ( the method we outline is trivially adapted to an arbitrary data table ) . we deal with each column of @xmath16 separately . suppose that we are attempting to reproduce the second column of @xmath16 , corresponding to some preparation procedure @xmath135 ; @xmath136\ ] ] we begin by finding the smallest entry in the chosen column , which in this case takes a value of @xmath137 , and is associated with the second outcome of the first measurement . we then move this entry into another column vector , to be added to what remains of the original . we also remove the same value of @xmath137 from one of the second column entries associated with the remaining measurement - restricting our choice only to ensure that the chosen entry is greater than @xmath137 . referring to the second column of @xmath16 shown above , we see that there are two such entries to choose from , associated with the first and second outcomes of the second measurement and both taking the value @xmath138 . we choose the entry associated with the first outcome , and thus obtain ; @xmath136 \rightarrow \frac{1}{3}\left [ \begin{array}{c } 0 \\ 1 \\ 0 \\\hline 1 \\ 0 \\ 0\end{array}\right ] + \left [ \begin{array}{c } \frac{2}{3 } \\ 0 \\ 0 \\\hline \frac{1}{6 } \\ \frac{1}{2 } \\ 0\end{array}\right]\ ] ] the new binary column vector we have generated provides a valid set of deterministic measurement outcomes for an ontic state ( weighted by @xmath137 ) . in particular , note that by subtracting the same value from a single entry associated with each measurement we have ensured that the binary vector satisfies the stochasticity requirement of ( [ stochasticitym_constraint ] ) . we can repeatedly apply this procedure to the ` remainder ' vector @xmath139 $ ] on the right hand side of the above expression , until only column vectors with binary entries remain . this gives ; @xmath140 & \rightarrow & \frac{1}{3}\left [ \begin{array}{c } 0 \\ 1 \\ 0 \\\hline 1 \\ 0 \\ 0\end{array}\right ] + \left [ \begin{array}{c } \frac{2}{3 } \\ 0 \\ 0 \\\hline \frac{1}{6 } \\ \frac{1}{2 } \\ 0\end{array}\right ] \nonumber\\ & \rightarrow & \frac{1}{3}\left [ \begin{array}{c } 0 \\ 1 \\ 0 \\\hline 1 \\ 0 \\ 0\end{array}\right ] + \frac{1}{6}\left [ \begin{array}{c } 1 \\ 0 \\ 0 \\\hline 1 \\ 0 \\ 0\end{array}\right ] + \left [ \begin{array}{c } \frac{1}{2 } \\ 0 \\ 0 \\\hline 0 \\ \frac{1}{2 } \\ 0\end{array}\right ] \nonumber\\ & = & \frac{1}{3}\left [ \begin{array}{c } 0 \\ 1 \\ 0 \\\hline 1 \\ 0 \\ 0\end{array}\right ] + \frac{1}{6}\left [ \begin{array}{c } 1 \\ 0 \\ 0 \\\hline 1 \\ 0 \\ 0\end{array}\right ] + \frac{1}{2}\left [ \begin{array}{c } 1 \\ 0 \\ 0 \\\hline 0 \\ 1 \\ 0\end{array}\right ] \nonumber\end{aligned}\ ] ] the three @xmath120 vectors generated by this procedure can be used as valid columns of an @xmath71 matrix , and we can take the associated probabilistic weightings to form a @xmath61 matrix row corresponding to the preparation @xmath135 . thus by introducing three ontic states we have formed @xmath61 and @xmath71 matrices that can reproduce the the quantum statistics for all measurements from @xmath16 performed on the single preparation @xmath135 . repeating this process to introduce a new set of ontic states for the remaining column of @xmath61 ( corresponding to a preparation @xmath141 ) , yields the following deterministic ontological factorization , note that since we introduce a whole new set of ontic states for each preparation , the entries of different columns of our @xmath61 matrix are bound to be completely disjoint . thus , following the terminology of @xcite , model 3 can be classified as @xmath81-ontic . the procedure outlined above will generally require introducing @xmath143 ontic states for each preparation from @xmath16 , which suggests @xmath144 . however , model 3 often allows us to exceed this value , so long as the procedure is applied carefully , exploiting repeated entries in @xmath16 to reduce the number of ontic states we need to introduce . by this token the example given above has @xmath145 - almost half of @xmath146 . in this section we will examine a process for turning the indeterministic of of model 1 into a deterministic of . first a point of notation : from now on we will often use the two dimensional array versions of the data table and measurement matrices : @xmath147,\;\;\;m=\left ( \begin{array}{c } m^{(1 ) } \\\hline \vdots \\\hline m^{(m ) } \\ \end{array } \right),\ ] ] so that the of becomes simply @xmath148 . the simplest version of the procedure we outline for recovering determinism from indeterminism is most easily illustrated by an example . consider the following data table : @xmath149.\ ] ] the of of model 1 would have @xmath150 and @xmath151 that is , @xmath152 . the first step is to notice that each of the probabilities in the first column of @xmath16 is an integer multiple of @xmath153 , while each in the second column is an integer multiple of @xmath154 . we therefore expand the supports of the probability distributions in @xmath61 from single ontic states to 4 and 3 ontic states respectively : @xmath155 note that the probability distributions are still disjoint ( in the language of @xcite the model is @xmath81-ontic but not @xmath81-complete ) . we now want to choose a matrix @xmath71 with only 0/1 entries . if we choose the first row of @xmath71 ( i.e. the indicator function for the first measurement outcome ) to be @xmath156 then we will certainly produce the first row of @xmath16 . the key is that the disjointness of the two columns of @xmath61 enables us to assign 1 s and 0 s to @xmath71 s first row in an essentially unrestricted fashion - the first 4 elements of this row simply need to pick up enough @xmath153 s from @xmath61 to give the entry 1/2 for @xmath157 , and similarly the last three entries must pick up an appropriate number of the 1/3 s to give the entry 2/3 for @xmath158 . assigning the second row we need to be more careful . one assignment which naively would seem to work , in as much as it would reproduce the data table entries when multiplied by @xmath61 , is @xmath159 however this violates the constraint of column - stochasticity on the @xmath59 s from definition [ def : of ] . such an indicator function would imply that a system in ontic state 1 gives both the first measurement outcome and the second outcome with certainty ! the only choice for this row once we have chosen the first row as above is @xmath160 proceeding in this manner , one possibility for @xmath71 is @xmath161 imagine now that the data table only contains rational fractions as entries . this would certainly be the case if we were considering a table which was assembled from actual experimental data . clearly , generalizing the above procedure will allow us to find @xmath71 and @xmath61 matrices so as to obtain a deterministic of . this does not give us a procedure for obtaining a finite of for _ any _ data table however , since we may wish to consider ones constructed from quantum states and projectors which have irrational valued overlaps . for the general program of research into ontological models there is no requirement of finiteness - in particular the trick of spreading the support of the probability distribution over extra ontic states can readily be performed by spreading the distribution over variables of cardinality @xmath162 . this yields a procedure for transforming any indeterministic ontological model into a deterministic one - and is essentially what bell did @xcite ( see example 4 in @xcite ) to provide a counterexample to ` von - neumann s silly assumption ' . in doing so one has increased the number of ontic states required to describe the system - the extent to which this is necessary , as opposed to merely sufficient , is an interesting question , which to some extent we take up again in section [ oc ] . , @xmath163 , @xmath153 and @xmath164 respectively . the blue lines separate different 4-outcome pvms . ] in the last four sections we have derived a series of bounds on the number of ontic states that ontological factorizations must employ . by a simple extension of the argument given in sec . [ sec : model2 ] we have that data tables with arbitrary parameters @xmath165 require a number of ontic states that is tightly bounded by @xmath166 . however , the explicit construction that we gave in sec . [ sec : model3 ] showed that by letting @xmath66 depend on all three parameters of a data table , the number of ontic states need not depend exponentially on @xmath84 . these bounds clearly have interesting repercussions for the ability to classically simulate quantum systems efficiently . for example , our lower bound @xmath167 shows that one can not precisely reproduce a general data tables quantum statistics by keeping track of a number of ontic states that grows only logarithmically with the number of preparations and measurements considered . extending and refining the bounds we have presented is an interesting direction for future research , which may benefit from existing results in the literature . for example , aaronson has shown in @xcite that the probabilities for a set of @xmath84 two - outcome measurements performed on an @xmath168 dimensional quantum system can be approximately calculated ( to some fixed accuracy ) from a classical string of length @xmath169 . this shows that a data table containing @xmath84 such measurements and some set of preparations associated with an @xmath168 dimensional quantum system has an _ approximate _ ontological factorization that employs @xmath170 ontic states ( wherein each ontic state encodes a possible instance of the string one can construct to encode the required statistics ) . thus , in cases where one is satisfied with _ approximately _ reproducing the entries of a data table , the results in @xcite give a lower bound on @xmath66 which depends pseudo - polynomially on @xmath84 and @xmath168 , but interestingly not on @xmath5 ( the precise number of preparations in the data table ) . in sec . [ sec : matrix_connection ] we will also note how existing results concerning matrix factorization problems might prove useful in deriving more bounds on @xmath66 . by the famous result of kochen and specker @xcite deterministic ontological models must be measurement - outcome contextual . we now consider how measurement outcome contextuality manifests itself in an of of a data table . contextuality needs to be considered when different quantum measurements share a common projector . in such a case a data table might look something like ( with @xmath14 ) : @xmath171 rows 1 and 4 contain the same entries , and so in quantum mechanics they could correspond to measurement outcomes represented by the same projector . initially one might have been tempted to simply exclude the redundant rows . the essence of the kochen - specker theorem is that doing so would prevent us being able to explain the data table in terms of a deterministic ontological model , and so we retain them . it is at this point that we depart from the _ r - p _ formalism . in the _ r - p _ formalism there is a unique @xmath0-vector for each quantum projector . as mentioned above , this presumption of measurement outcome non - contextuality is the reason the @xmath0-vectors require negative entries ( or entries greater than 1 ) . in terms of our of s , the contextuality required by the kochen - specker theorem manifests itself as follows : if two rows of the data table are identical , then one can not necessarily find an of such that the two corresponding rows of @xmath71 are identical . now , if the only constraint on the factorization was that @xmath148 with @xmath71 a @xmath120 matrix , then this would _ not _ be true , since there would be nothing stopping the replacement of one of those rows of @xmath71 by the other . the key issue , as mentioned above , is the extra column - stochasticity constraint on the factorization , which forces at least one measurement outcome to be obtained for any given ontic state . let us demonstrate all this with an example . in @xcite kernaghan provided an example of a set of measurements in a 4 dimensional hilbert space for which a kochen - specker obstruction exists ( see also @xcite ) . these measurements are built from the following 20 states in a hilbert space of dimension 4 ( the normalization factors of @xmath172 or @xmath163 are omitted , and @xmath173 denotes @xmath174 ) : @xmath175 so , for example , the third measurement @xmath176 consists of the projectors @xmath177 . the proof that these projectors can not be assigned a unique truth value ( i.e. independent of the other projectors in the measurement ) is trivial . in each of the odd number ( i.e. 11 ) of measurements , only one projector can ( and must ) be assigned the truth value 1 . this is in direct conflict with the fact that each of the @xmath178 s appears an even number of times in ( [ perestable ] ) . the data table for these 11 measurements and the 20 states is depicted in a color coded fashion in figure [ fig : peresd ] . applying the procedure of section [ detfromindet ] allows us to generate matrices @xmath71 and @xmath61 for a deterministic of over @xmath179 ontic states . the matrices for this of are depicted in figure [ fig : perespm ] . an example of the contextuality of this of can be seen by considering rows @xmath180 and @xmath181 , which correspond to the same measurement projector @xmath182 ( these rows are marked on @xmath71 in figure [ fig : perespm ] by red arrows ) . clearly each of these indicator functions must have a 1 at all of the set of ontic states which @xmath183 has support over ( c.f . the third column in @xmath61 - that is , ontic states 10 through 13 ) . we see , however , that there are some ontic states ( 71 , 75 and 79 for example ) that do not lie in the support of @xmath183 ) , and to which one of rows @xmath180 or @xmath181 assigns a 1 while the other assigns a 0 . ontic states such as 71 , 75 and 79 also show that there are some ontic states which are assigned a value @xmath82 by the row in @xmath71 corresponding to @xmath184 but not by the column in @xmath61 corresponding to @xmath185 . as discussed in @xcite this is a general ( necessary ) feature of all ontological models : there are necessarily some ontic states which pass a test @xmath2 for a system `` being in '' a state @xmath1 , which can never be prepared when the preparation procedure specifies the quantum state to be @xmath1 . this asymmetry of deterministic ontological models between preparations and measurements , which is not present in quantum mechanics , was termed _ deficiency _ in @xcite . the terminology comes from the fact that the support of the probability distribution corresponding to the quantum state is strictly less than the support of the indicator function corresponding to a projection onto that state , as we have seen . we term _ unfaithful _ those ontic states which do not lie in the support of the probability distribution , but which _ do _ pass the test ( since for some measurements they will `` choose '' to give a different outcome ) . note that if an ontological model is therefore to be able to describe preparing the state @xmath1 by performing a measurement of @xmath2 , it is going to be the case that a system in an unfaithful ontic state must be disturbed by the measurement in order to end up in one of the faithful states . so far we have investigated of s without particular regard to how large the number of ontic states @xmath66 is required to be . in this section we consider the problem of trying to either find of s with @xmath66 as small as possible , or to reduce the value of @xmath66 for a given of - a process we call `` ontological compression '' . such a reduction in ontic states clearly has repercussions for the ability to efficiently simulate quantum systems classically - a point we elaborate on in sec . [ sec : class_simulation ] . but , as we shall see , our ontological compression schemes also provide routes towards constructing @xmath81-epistemic models in which we can truly view the quantum description of a system as a state of knowledge . models 1 and 2 gave of s which require @xmath79 and @xmath83 respectively - these values set upper bounds on @xmath66 for indeterministic and deterministic of s respectively . but since deterministic ontological factorizations can be seen as a special case of indeterministic ones then model 2 also provides an upper bound of @xmath186 for _ indeterministic _ factorizations . there are therefore upper bounds on @xmath66 for both deterministic and indeterministic models which depend only on a data tables measurements , and upper bounds for indeterministic models which depend on only @xmath5 . one may therefore wonder whether an upper bound can be found for deterministic models which depends only on @xmath5 . but this is not possible ; a deterministic ontological factorization always requires at least @xmath187 ontic states . to see why , note that each ontic state , @xmath188 , from a deterministic model is associated with a length @xmath15 binary string , given by the @xmath189 column of the models @xmath71 matrix . thus the ontic states of any deterministic ontological factorization can be thought of as column vectors , @xmath190 , defining vertices on a unit hypercube of dimension @xmath15 ; @xmath191^{dm}$ ] . now consider how such a model reproduces the @xmath9 column of the data table @xmath16 ( corresponding to the measurement statistics for some preparation @xmath192 ) in terms of these @xmath193 column vectors . writing the relevant column of @xmath16 as a vector @xmath194^{dm}$ ] containing the outcome probabilities for each measurement , its entries are reproduced by a convex combination of the @xmath190 , @xmath195 where we use @xmath196 to denote the @xmath197 element of the @xmath49 column of the ofs @xmath61 matrix . thus deterministic models can be thought of as reproducing @xmath16 by convexly summing over a set of @xmath66 points in a @xmath15 dimensional space . but at least @xmath187 such points must be convexly summed for us to be able to represent an arbitrary point from @xmath198^{dm}$ ] . thus we require at least @xmath187 columns of @xmath71 , i.e. at least @xmath187 ontic states . we can also set a trivial lower bound on @xmath66 : [ prop3 ] any ontological factorization must be such that @xmath199 . this follows immediately from the fact that @xmath148 , the ranks of @xmath200 can not exceed @xmath66 , and for general matrices @xmath201 . as an example of where this bound can be attained in a deterministic of , consider the data table : @xmath202.\ ] ] this is the data table for a qubit , where the measurements are the projectors onto the eigenbases of the pauli @xmath203 matrices , and the states are simply the eigenstates of these three operators . this data table has the following of : @xmath204 this of has @xmath205 , which is less than both @xmath206 or @xmath207 , and which happens to be the rank of @xmath16 . ( note that this of is equivalent to a single toy - bit of spekkens @xcite . ) the first technique for ontological compression we consider can be applied to any of , but finds particular utility in compressing the deterministic of s generated from model 1 ( discussed in section [ detfromindet ] ) . the key is to consider the possibility that two columns @xmath65 and @xmath208 of @xmath71 , corresponding to ontic states associated with distinct and non - orthogonal quantum states , are identical . if this is the case , then one of the ontic states is redundant , since any row of @xmath61 ( i.e. any quantum state ) which has support at @xmath65 , could just as well reproduce the relevant statistics in @xmath16 by instead having support at @xmath208 . for example , consider the data table given in ( [ toyd ] ) and suppose that we started with a deterministic un - compressed of of the kind discussed in section [ detfromindet ] so that , @xmath209 @xmath210 we can immediately see that several of the columns of ( [ toymbig ] ) are already equal ( specifically , columns @xmath82 and @xmath211 , @xmath212 and @xmath213 , @xmath180 and @xmath214 and @xmath215 and @xmath181 ) . we can therefore compress the pairs of associated ontic states together . in general however we may not be so fortunate , and we will need to manipulate the columns of @xmath71 to try and make as many identical as possible . so long as we respect the constraints imposed by stochasticity and the requirement @xmath216 we are free to alter the entries of @xmath71 as is convenient . in particular , we can change which ontic states a given indicator function ( i.e. row of @xmath71 ) uses to reproduce any quantum states statistics within @xmath16 so long as we respect the requirement that one and only one outcome of each pvm should ever occur . consider the @xmath217 boxes we have drawn on @xmath71 in ( [ toymbig ] ) . these partition rows corresponding to distinct pvm measurements and sets of ontic states ( columns ) associated with different preparations .- ontic . ] in terms of these boxes , the allowed manipulations of @xmath71 correspond to flipping the columns of a box , @xmath218 as an example of how one can use this freedom to compress two ontic states , columns @xmath180 and @xmath219 of ( [ toymbig ] ) can be made equal by simply applying this permutation to the bottom box containing columns @xmath211 and @xmath219 ( shown in red ) . there is however , a subtlety in forcing columns of @xmath71 to be equal in this way . the permutation we apply to make columns @xmath180 and @xmath219 equal will of course also alter column @xmath211 . but initially this column was already equal to column @xmath82 . to retain this initial equality we must apply the same permutation to the bottom box containing columns @xmath82 and @xmath212 ( shown in blue ) . however , this in turn will alter column @xmath212 , which was initially identical to column @xmath213 and so to retain _ this _ equality we must then permute the bottom box containing columns @xmath213 and @xmath119 ( shown in green ) . performing these three permutations we obtain the following new expression for @xmath71 , @xmath220 combined with ( [ toypbig ] ) this reproduces the statistics from @xmath16 . counting identical columns we find that in fact there are now @xmath119 redundant columns in @xmath71 , three more having appeared due to the extra reshuffling we needed to perform in order to share columns @xmath180 and @xmath219 . these redundant columns can be removed , which results in us compressing the four triples @xmath221 , @xmath222 , @xmath223 , @xmath224 into four ontic states . we can then finally update @xmath61 to represent this compression by combining its rows corresponding to each of the four triples of ontic states . the resultant compressed of is in fact precisely the optimally compressed model given in ( [ toymp ] ) . in cases with larger data tables , the compression is not quite so trivial . the kernaghan of for example , requires one to keep track of eighty ontic states . by applying an algorithm for randomly combining ontic states according to a natural generalization of the example given above , we are able to compress the kernaghan of to use only @xmath225 ontic states . this is much lower than the @xmath226 ontic states needed by model 2 and almost half the number required by our deterministic rendering of model 1 . the resulting @xmath71 and @xmath61 matrices are shown in figure [ fig : perespmcompress ] . clearly the columns of the matrix @xmath61 associated with this compressed of are not disjoint , and consequently the of is @xmath81-epistemic . in fact , the approach to compression outlined above will always result in a @xmath81-epistemic model . it should be noted however , that the compression illustrated in figure [ fig : perespmcompress ] is not guaranteed to be optimal , since it does not saturate the bound given in proposition [ prop3 ] . in general , one would expect the amount by which one can compress an of using this approach to depend on the initially chosen @xmath71 ( picked at random for the above examples ) and the order in which one chooses to compress the various ontic states of the model . optimizing these choices is highly non - trivial . the second method for ontological compression we consider is designed to effect a compression for of s based on model 2 . it is also most easily illustrated by an example . let us return to the example of eqs . ( [ model2d ] ) and ( [ model2mp ] ) , reproduced here : @xmath227\\ \\ \\ \\ \end{array } \begin{array}{c } \!=\!\\ \\ \\ \\ \end{array } \begin{array}{c}\left ( \begin{array}{ccccccccc } 1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 & 1 & 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 1 & 1 & 1 \\\hline 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 \\ 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 \\ 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 \\ \end{array } \right)\\ \\ \\ \\ \end{array}\!\!\!\!\left ( \begin{array}{cc } 0 & 1/3 \\ 0 & 1/3 \\ 0 & 0 \\ 1/9 & 1/6 \\ 1/9 & 1/6 \\ 1/9 & 0 \\ 2/9 & 0 \\ 2/9 & 0 \\ 2/9 & 0 \\ \end{array } \right).\ ] ] it is convenient to view the 9 entries corresponding to each of the two columns of @xmath61 as probability distributions over 9 `` ontic state boxes '' as follows : @xmath228 the indicator functions are all of the form @xmath229 that is , the construction of model 2 is such that summing @xmath23 along a column of the boxes yields the probability for the corresponding outcome of the first measurement , summing along a row yields the same for the second measurement . thus it is clear that if we leave the row and column box - sums invariant we can `` shift around '' the probability weightings that the distributions @xmath230 assign to the ontic states . if we then find that both distributions assign a weighting 0 to an ontic state , it can be removed completely and some ontological compression has been achieved . in fact , we already see that @xmath231 , and so this ontic state could be removed . in order to shift around the probability , one simple operation involving 4 ontic states is to map them as follows : @xmath232 since the final state must be a suitable probability distribution , we require @xmath233 . in our example we can use this to implement : @xmath234 in this manner we nullify weighting of both distributions on the middle ontic state ( i.e. @xmath235=0 ) , and it too could be removed : @xmath236 ( where @xmath237 denotes entries to be deleted . ) we have seen that some sort of ontological compression may be possible , depending on specifics of the data tables involved . why might one be interested in ontological compression ? consider a `` family '' of data tables - that is a set of data tables constructed from an increasing number of states and measurements . an example might be data tables constructed from the family of all stabilizer states on @xmath51-qubits , for increasing values of @xmath51 . if it were the case that this data table can be ontologically compressed such that @xmath66 only grows polynomially with @xmath51 , then an efficient classical simulation of that family of states and measurements is clearly possible . if we denote by @xmath238 the data table for all @xmath51-qubit stabilizer states / measurements so that @xmath238 has @xmath239 rows / columns ( and @xmath240 would then be the data table of ( [ toyd ] ) ) , then we are asking whether there exists an ontological factorization of @xmath238 which has @xmath241 . but we can immediately see that this can not be achieved , because any table containing all stabilizer states and measurements will contain a sub - table in which ( for @xmath103 ) @xmath242 preparations are associated with all @xmath51 bit strings over @xmath51 measurements - i.e. a sub - table of the form shown in ( [ table_upper_bound ] ) . since we have already seen that such a table will require @xmath243 , then its existence clearly prohibits us from finding ontological factorization of the stabilizer data table having @xmath241 . actually , a polynomial sized number of ontic states is too strong a requirement for efficiency : classical monte - carlo simulations , for example , can be provably efficient for certain problems even when the number of ontic states grows exponentially . since our ontological models are also essentially classical probability representations , we similarly expect a polynomial requirement to be too strong for them . we conclude this section with a brief discussion of how finding of s is related to other matrix factorization problems . our data tables are generally not square . however , in instances where ( such as in the example of eqs . ( [ toyd ] ) and ( [ toymp ] ) ) we use the same quantum states to specify both the preparation procedures and the measurements , then @xmath16 is symmetric , with non - negative entries , and positive semi - definite ( ie doubly non - negative ) . these conditions are necessary ( though not sufficient ) for the matrix @xmath16 to be completely positive - that is , factorizable into a product of the form @xmath244 , where the elements of @xmath245 are non - negative . completely positive matrices have been much studied ( see e.g. @xcite ) , in particular with regards to finding ( bounds on ) the smallest row - dimension of @xmath245 for which this factorization is possible ( called the `` cp - rank '' of @xmath245 , or sometimes the `` factorization index '' ) . a completely positive matrix factorization ( i.e. one of the form @xmath246 ) is not generally equivalent to our of s . however , we have already seen an example of such a factorization appearing as an of - the matrices @xmath200 of the example in eqs . ( [ toyd ] ) and ( [ toymp ] ) differ only by an overall multiplicative factor of @xmath163 , and so we can see this example as providing a completely - positive matrix factorization of this particular @xmath16 . now it is interesting to note that there is an upper bound and rank 1 operators @xmath247 in a fixed hilbert space dimension @xmath39 , then finding such non - trivial upper bounds on @xmath66 can not be achieved by simply finding a family of data tables of continually increasing rank and using proposition [ prop3 ] . this is because @xmath248 . ( to see this , note that @xmath249 where @xmath250 and @xmath251 denotes the hadamard ( elementwise ) product - the result follows from standard results regarding this product @xcite and from the fact that @xmath252 . ) ] on the cp - rank @xcite which is polynomial in @xmath253 , which , in turn , is polynomial in @xmath39 . the cp - rank is , in the of picture , the number of ontic states @xmath66 . thus it seems plausible there exist some interesting data table families which have polynomial ( in @xmath39 ) sized ofs . there is a second type of matrix factorization which has some connection to the ofs we have been considering : the so - called `` non - negative matrix factorization '' ( nmf ) . in this factorization a matrix of ( non - negative - valued ) image data @xmath254 is factorized into the product of two matrices @xmath255 with non - negative elements : @xmath256 nmf is often considered from the viewpoint of image compression , where the goal is to obtain an approximate factorization where @xmath257 ( @xmath258 ) have as small a number of rows(columns ) as possible . some particularly simple procedures for performing this approximate factorization , guaranteed to converge to a local optima ( with respect to various matrix distances between the ideal @xmath254 and the approximate one ) , were given in @xcite . it was shown recently in @xcite that if one considers optimizing with respect to the kullback - leibler distance , then the local optima preserve the row and column sums of the original matrix . in particular , if we consider the separate nmf factorizations @xmath259 so @xmath57 is column - stochastic , then the results of corollary 2 in @xcite imply that the nmf found will be such that @xmath260 are column stochastic as we require . of course our problem requires finding many such factorizations with the _ same _ @xmath61 for the @xmath84 different @xmath57 s . interestingly , one reason that nmf is useful in image analysis is that it breaks the image up into `` hidden variables '' ( lee and seung s terminology , not ours ! ) such that an image is comprised of pieces which humans recognise as familiar fundamental components ( the ears , nose and mouth of a facial image for example ) . this is unlike the more standard principal component analysis route to `` eigenimage '' compression . in these terms the of we are considering is asking an interesting question regarding factorizing many different images ( the @xmath57 s ) using the same set of fundamental component images ( the @xmath61 ) by using an ontological formalism to describe finite sets of preparations and measurements performed on a quantum system we have illustrated how many features of ontological models are manifested in such discretized scenarios . we have shown how contextuality and deficiency appear in this formalism and have also been able to use the formalism to build a @xmath81-epistemic model of a discrete set of quantum statistics . we have also seen how any indeterministic description of a finite data table can be developed into an equally valid deterministic one and noted that the same technique ( first employed by bell ) can be applied in the continuum limit . we have also discussed how ofs can be used to compress the number of ontic states needed to describe such discretized quantum scenarios and speculated on repercussions which this might have for classical simulations of quantum systems . n. harrigan , t. rudolph and r. spekkens , `` contextuality in the ontological model formalism '' . this paper is completed , but for months has been awaiting dr . spekkens final approval . please email him and ask him for a copy : r.w.spekkens@damtp.cam.ac.uk

spontaneous chiral symmetry breaking is one of the most important properties to understand the low - energy phenomena of qcd in the real world . this chiral symmetry is expected to be restored in qcd at several extreme conditions such as qcd with a large number of massless quarks , large @xmath0 qcd ( see , e.g. , refs . @xcite ) and qcd in hot and/or dense matter ( see , e.g. , ref . @xcite ) . in ref . @xcite , based on the infrared ( ir ) fixed point existing at a two - loop beta function for a large number of massless quarks ( @xmath1 ) @xcite , it was found through the improved ladder schwinger - dyson ( sd ) equation that chiral symmetry restoration takes place for @xmath0 such that @xmath2 , where @xmath3(@xmath4 for @xmath5 ) . then , in ref . @xcite this chiral restoration at @xmath6 was further identified with `` conformal phase transition '' which was characterized by the essential singularity scaling . moreover , such chiral restoration is also observed by other various methods such as lattice simulation @xcite , dispersion relation @xcite , instanton calculus @xcite , effective field theoretical approach @xcite , etc .. more attention has been paid to the property of the phase transition . especially , it is interesting to ask what are the light degrees of freedom near the phase transition point in the large @xmath0 qcd : for example , in the manifestation of the chiral symmetry restoration la the linear sigma model , the scalar bound state becomes a chiral partner of the pseudoscalar bound state and becomes massless at the phase transition point . on the contrary , in the vector manifestation ( vm ) @xcite obtained by the effective field theoretical approach based on the hidden local symmetry model @xcite , it is the vector bound state which becomes massless as a chiral partner of the pseudoscalar bound state . besides , from the viewpoint of the conformal phase transition @xcite , it is natural to suppose that all the existing bound states become massless near the phase transition point when approached from the broken phase ( see ref . @xcite ) . then , it is quite interesting to study which types of the bound states actually exist near the phase transition point , and investigate the critical behavior of their masses directly from qcd . such studies from the first principle will give us a clue to understand the nature of the chiral phase transition in large @xmath0 qcd . a powerful tool to study the bound states of quark and anti - quark directly from qcd is the homogeneous bethe - salpeter ( hbs ) equation in the ( improved ) ladder approximation ( see , for reviews , refs . @xcite ) . when the mass of the quark is regarded as a constant , we can easily solve the hbs equation by using a so - called fictitious eigenvalue method @xcite . however , for consistency with the chiral symmetry , the quark propagator in the hbs equation must be obtained by solving the sd equation with the same kernel as that used in the hbs equation @xcite , and as a result , the quark mass becomes a certain momentum dependent function . then , in order to obtain the masses and the wave functions of the bound states , it is necessary to solve the hbs equation and the sd equation simultaneously . when we try to solve these two equations in real - life ( @xmath7 ) qcd , however , we encounter difficulties . first of all , for the consistency of the solution of the sd equation with qcd in a high energy region , we need to use the running coupling which obeys the evolution determined from qcd @xmath8-function in the high energy region ( see , for reviews , refs . since the running coupling diverges at some infrared scale , @xmath9 , we have to regularize the running coupling in the low energy region , for which there exist several ways ( see , e.g. , refs . even if we fix the infrared regularization in such a way that we can solve the sd equation on the real ( space - like ) axis , another problem arises when we try to solve the hbs equation . since the argument of the quark mass function in the hbs equations for the massive bound states becomes a complex quantity after the wick rotation has been made , we have to solve the sd equation on the complex plane , which requires an analytic continuation of the running coupling . several works such as in refs . @xcite proposed models of running couplings for a general complex variable which are consistent with perturbative qcd for large space - like momentum . however , they still have branch cuts on the complex plane , and it is a complicated task to obtain the solution of the sd equation for a general complex variable . one way to avoide a such complication is solving the inhomogeneous bs equation for vertex functions to obtain the current correlators in the space - like region which we can fit the mass of the relevant bound state to ( see , e.g. , refs . @xcite ) . another way might be replacing the entire running coupling with an ad hoc analytic function ( see , e.g. , ref . anyway , it is impossible to solve the sd equation on the complex plane without modeling the running coupling . in this paper , we point out that the situation dramatically changes when we increase the number of massless quarks . when @xmath0 becomes larger than @xmath10 , the running coupling obtained from the renormalization group equation ( rge ) with two - loop approximation takes a finite value for all the range of the energy region due to the emergence of the non - trivial ir fixed point . then , we need no infrared regularization , and we do not have any ambiguities coming from the regularization scheme which do exist in the case of small @xmath0 . moreover , an explicit solution of the two - loop rge can be written in terms of the lambert @xmath11 function @xcite , and when @xmath0 is close to @xmath6 the solution of the rge has no singularity on the complex plane except for the time - like axis @xcite . consequently , we can solve the sd equation on the complex plane without introducing any models for the running coupling . based on these facts , we solve the hbs equations for the bound states of quark and anti - quark in large @xmath0 qcd with the improved ladder approximation in the landau gauge . the mass function for complex arguments needed in the hbs equation is obtained by solving the sd equation with the same kernel as that used in the hbs equation . we find the solution of the hbs equation in each of the scalar , vector , and axial - vector channels , which implies that the scalar , vector , and axial - vector bound states are actually formed near the phase transition point . our results show that the masses of the scalar , vector , and axial - vector bound states go to zero as the number of quarks @xmath0 approaches to its critical value @xmath6 where the chiral symmetry restoration takes place . this may suggest the existence of a new type of manifestation of chiral symmetry restoration in large @xmath0 qcd other than the linear sigma model like manifestation and a simple version of the vector manifestation proposed in ref . @xcite . this paper is organized as follows . in section [ sec : sdeq ] we numerically solve the sd equation with an approximate form of the running coupling , and study the critical behavior of the nambu - goldstone boson decay constant . in section [ sec : complex_sd ] we solve the sd equation for complex arguments . section [ sec : hbseq ] is devoted to summarizing the numerical method for solving the hbs equation . section [ sec : numerical_calculations ] is the main part of this paper . we first solve the hbs equation for the pseudoscalar bound state to show that the approximation adopted in the present analysis is consistent with the chiral symmetry . we next solve the hbs equation for the scalar , vector , and axial - vector bound states to obtain their masses . finally we give a summary and discussion in section [ sec : discussion_and_conclusion ] . in appendix [ app : positronium ] we solve the hbs equation for the ortho - positronium with a constant electron mass to show the validity of the fictitious eigenvalue method . the bispinor bases for the bound states are listed in appendix [ app : bispinor - bases ] . in appendix [ app : decay_const ] we calculate the coupling constants @xmath12 , @xmath13 , and @xmath14 of the vector , axial - vector , and scalar bound states to the vector current , axial - vector current , and scalar density . we briefly study numerical uncertainties in the present analysis in appendix [ app : uncertainties ] . in this section we numerically solve the schwinger - dyson ( sd ) equation for the quark propagator with the improved ladder approximation in the landau gauge , and show the critical behaviors of the dynamical mass and the decay constant of the nambu - goldstone boson . we also show the behavior of the fermion - antifermion pair condensate @xmath15 near the phase transition point . schwinger - dyson ( sd ) equation is a powerful tool to study the dynamical generation of the fermion mass directly from qcd ( for reviews , see , e.g. , refs . @xcite ) . the sd equation for the full fermion propagator @xmath16 in the improved ladder approximation @xcite is given by ( see fig . [ fig : sdeq ] for a graphical expression ) @xmath17 where @xmath18 is the second casimir invariant , and @xmath19 is the running coupling . the landau gauge is adopted for the gauge boson propagator . the sd equation provides coupled equations for two functions @xmath20 and @xmath21 in the full fermion propagator @xmath22 . when we use a simple ansatz for the running coupling , @xmath23 @xcite , with @xmath24 being the euclidean momenta , we can carry out the angular integration and get @xmath25 in the landau gauge . then the sd equation becomes a self - consistent equation for the mass function @xmath26 . the resultant asymptotic behavior of the dynamical mass @xmath27 is shown to coincide with that obtained by the operator product expansion technique @xcite . however , it was shown in ref . @xcite that the axial ward - takahashi identity is violated in the improved ladder approximation unless the gluon momentum is used as the argument of the running coupling as @xmath28 . in this choice we can not carry out the angle integration analytically since the running coupling depends on the angle factor @xmath29 . furthermore , we would need to introduce a nonlocal gauge fixing @xcite to preserve the condition @xmath30 . in ref . @xcite , however , it was shown that an angle averaged form @xmath31 gives a good approximation . then , in the present analysis we take the argument of the running coupling as @xmath32 after applying this angle approximation and carrying out the angular integration , we can show ( see , e.g. , refs . @xcite ) that @xmath20 always satisfies @xmath33 in the landau gauge . then the sd equation becomes @xmath34 where @xmath35 and @xmath36 . although the choice of arguments in eq . ( [ eq : approx_of_argument ] ) explicitly breaks the chiral symmetry as mentioned above , it will be shown later that the magnitude of the breaking is negligible . in qcd with @xmath0 flavors of massless quarks , the renormalization group equation ( rge ) for the running coupling @xmath37 in the two - loop approximation is given by @xmath38 where @xmath39 from the above beta function we can easily see that , when @xmath40 and @xmath41 , i.e. , @xmath0 takes a value in the range of @xmath42 ( @xmath43 for @xmath44 ) , the theory is asymptotically free and the beta function has a zero , corresponding to an infrared stable fixed point @xcite , at @xmath45 existence of the infrared fixed point implies that the running coupling takes a finite value even in the low energy region . actually , the solution of the two loop rge in eq . ( [ eq : rge_for_alpha ] ) can be explicitly written @xcite in all the energy region as @xmath46^{-1 } , \label{lan w}\ ] ] where @xmath47 with @xmath48 is the lambert @xmath11 function , and @xmath49 is a renormalization group invariant scale defined by @xcite @xmath50 . \label{eq : lambda}\ ] ] we note that , in the present analysis , we fix the value of @xmath49 to compare the theories with a different number of flavors , and that we have no adjustable parameters in the running coupling in eq . ( [ lan w ] ) , accordingly ( see discussion below ) . we show an example of @xmath51 for @xmath52 by the solid line in fig . [ fig : step ] . ) ( dashed line ) for @xmath52 . , height=226 ] the fact that the running coupling is expressed by a certain function as in eq . ( [ lan w ] ) implies that , in the case of large @xmath0 qcd , _ we do not need to introduce any infrared regularizations _ such as the ones adopted in refs . @xcite for studying real - life qcd with small @xmath0 in which the infrared regularization parameter must be chosen in such a way that the running coupling in the infrared region becomes larger than the critical value @xmath53 for realizing the dynamical chiral symmetry breaking @xcite . the running coupling in large @xmath0 qcd takes a certain value in the infrared region for given @xmath0 , so that we can definitely determine , within the framework of the sd equation , whether or not the dynamical chiral symmetry breaking is realized . actually , the value of @xmath54 decreases monotonically with increasing @xmath0 , and the chiral symmetry restores when @xmath0 becomes large enough . in refs . @xcite , it was shown that the phase transition occurs at @xmath55 for @xmath44 ( corresponding to @xmath56 ) . in order to reduce the task of numerical calculations in solving the hbs equation , we modify the shape of the running coupling . since the dynamics in the infrared region governs the chiral symmetry breaking , we adopt the following approximation for the running coupling @xcite : @xmath57 in this approximation the coupling takes the constant value @xmath54 ( the value at the ir fixed point ) below the scale @xmath49 and entirely vanishes in the energy region above this scale . the dashed line in fig . [ fig : step ] represents the approximated form of the running coupling for @xmath58 . in this subsection we briefly explain how we solve the sd equation numerically . we first introduce the infrared ( ir ) cutoff @xmath59 and ultraviolet ( uv ) cutoff @xmath60 as @xmath61 then , we discretize the momentum variable @xmath62 and @xmath63 into @xmath64 points as @xmath65 \ , \ \ \ \big ( i = 0 , 1 , 2 , \cdots , ( n_{sd}-1 ) \big ) \ , \ ] ] where @xmath66 accordingly , the integration over @xmath63 is replaced with a summation as @xmath67 then , the sd equation in eq . ( [ eq : improved_ladder_sd ] ) with the running coupling in eq . ( [ eq : step ] ) is rewritten as @xmath68 this discretized version of the sd equation is solved by the recursion relation : @xmath69 starting from a suitable initial condition ( we choose @xmath70 ) , we update the mass function by the above recursion relation . then , we stop the iteration when the convergence condition @xmath71 ^ 2 \ < \ \varepsilon^2 \lambda^6\ ] ] is satisfied for sufficiently small @xmath72 , and regard this @xmath73 as a solution of eq . ( [ eq : discretized_largenf_sd_1 ] ) . in fig . [ fig : mass_function ] , we show the numerical solution for the mass function @xmath74 . ) for @xmath52 . , height=226 ] here , we took @xmath75 ( @xmath76 ) as an example and used the following parameters : @xmath77 now , let us study the critical behavior of the fermion mass as @xmath0 is varied . note that we can use @xmath54 instead of @xmath0 as an input parameter , because once we choose a value of @xmath0 , the value of @xmath54 is uniquely determined from eq . ( [ eq : alpha_ir ] ) . for example , @xmath78 implies @xmath79 and @xmath80 implies @xmath81 . we solve eq . ( [ eq : discretized_largenf_sd_1 ] ) for various values of @xmath54 and plot the values of @xmath82 in fig . [ fig : sigma_ir ] . for several values of @xmath54 ( indicated by @xmath83 ) . solid line shows the function in eq . ( [ eq : fitting ] ) with the best fitted value @xmath84 . , height=264 ] here , @xmath85 represents the dynamical mass defied by @xmath86 . it should be noticed that @xmath87 is defined in the space - like region which does not represent the pole mass of fermion . as we will show in section [ sec : complex_sd ] , the present fermion propagator does not have any poles and then there are no pole masses of fermion . we compare this result with the analytic solution @xcite : @xmath88 in the above form there is an ambiguity in the prefactor . then , we introduce the function @xmath89 and fit the value of the pre - factor @xmath90 by minimizing @xmath91 in the range of @xmath92 $ ] . the resultant best fitted value of @xmath90 is @xmath93 we plot the function in eq . ( [ eq : fitting ] ) with the best fitted value @xmath84 in fig . [ fig : sigma_ir ] ( solid line ) . this clearly shows that the @xmath54-dependence of the resultant @xmath94 from our numerical calculation is consistent with the analytic result : the dynamical mass function vanishes when @xmath54 reaches the critical value @xmath53 . noting that decreasing @xmath54 corresponds to increasing @xmath0 for fixed @xmath95 as we discussed in the previous subsection , we see that the chiral symmetry restoration actually occurs at @xmath96 @xcite . so far we have solved the sd equation and obtained the mass functions for the various values of @xmath54 . now we can calculate the pseudoscalar meson decay constant @xmath97 at each @xmath54 by using the pagels - stokar formula @xcite : @xmath98\ } { \left(\ x \ + \ \sigma^2(x ) \ \right)^2 } .\ ] ] in fig . [ fig : f_pi ] , we plot the values of @xmath97 for several values of @xmath54 ( indicated by @xmath83 ) . to study the critical behavior of the pseudoscalar meson decay constant we use the function of the form in eq . ( [ eq : fitting ] ) and fit the value of @xmath90 by minimizing @xmath99 for @xmath92 $ ] . the resultant best fitted value of @xmath90 is @xmath100 we plot the fitting function with @xmath101 in fig . [ fig : f_pi ] ( dotted line ) . this shows that the results of the numerical calculations for @xmath97 are well fitted by the function of the form in eq . ( [ eq : fitting ] ) , and that the pseudoscalar meson decay constant has the same critical behavior as the mass function has . calculated from the pagels - stokar formula for several values of @xmath54 ( indicated by @xmath83 ) . the dotted line shows the function of the form in eq . ( [ eq : fitting ] ) with the best fitted value @xmath101 . , height=264 ] in this subsection , we calculate the fermion - antifermion pair condensate @xmath102 in large @xmath0 qcd , and show that the system in the present analysis has the large anomalous dimension @xmath103 for the operator @xmath104 . we also show that the values of @xmath103 are not affected so much by the approximation for the running coupling used in the present analysis . the condensate @xmath102 is calculated from the following equation : @xmath105 where @xmath74 is the mass function obtained from the sd equation and @xmath106 represents uv cutoff introduced to regularize the uv divergence . in the improved ladder approximation , the high - energy behavior of the mass function is consistent with that derived using the operator product expansion ( ope ) . the chiral condensate calculated using the mass function was shown to obey the renormalization group equation derived with the ope ( see , e.g. , refs . then , as was adopted in refs.@xcite , we identify the condensate , which is calculated with uv cutoff @xmath106 , with that renormalized at the scale @xmath106 in qcd . ) , the integration in eq . ( [ eq : q - bar - q ] ) is effectively cut off at the scale of @xmath49 due to the truncation of the running coupling for any values of @xmath107 . ( see fig . [ fig : mass_function ] : @xmath108 for @xmath109 . ) ] we expect that infrared dynamics in large @xmath0 qcd is similar to that of strong coupling qed or walking gauge theories @xcite since the running coupling in large @xmath0 qcd is well approximated by the constant coupling ( see fig . [ fig : step ] ) @xcite . then , we also expect that the value of the anomalous dimension in large @xmath0 qcd becomes @xmath110 since the walking gauge theories have @xmath110 @xcite . when a considering system has the anomalous dimension @xmath103 , scaling properties of @xmath97 and @xmath111 with respect to @xmath54 near the critical point are expressed as follows @xcite : @xmath112 @xmath113 where @xmath87 represents the dynamical fermion mass . these equations mean that the relation between @xmath102 and @xmath97 can be written as @xmath114 where @xmath115 is a certain positive constant . from this equation , we can express the anomalous dimension as @xmath116 where @xmath117 here , we note that @xmath118 approaches @xmath103 for @xmath119 since @xmath97 becomes small , i.e. , @xmath120 , near the critical point ( see fig . [ fig : f_pi ] ) : @xmath121 in fig . [ fig : q - bar - q_over_fpi2 ] , we plot the values of @xmath118 for several values of @xmath54 as an estimation of the anomalous dimension . for several values of @xmath54 . @xmath83 and @xmath122 represent the values of @xmath118 calculated from eq . ( [ eq : gamma_m ] ) with and without approximation for the running coupling . ( we call them @xmath123 and @xmath124 respectively . ) , height=264 ] the data indicated by @xmath83 in fig . [ fig : q - bar - q_over_fpi2 ] is obtained with the approximated running coupling ( dashed line in fig . [ fig : step ] ) in the sd equation . ( we call this kind of data @xmath125 . ) on the other hand , the data indicated by @xmath122 is the result from the calculation with the two - loop running coupling given in eq . ( [ lan w ] ) . ( we call this kind of data @xmath126.)[multiblock footnote omitted ] from these results , we conclude that large @xmath0 qcd with two - loop running coupling as well as with approximated running coupling actually possesses obtained by fitting to the data of @xmath102 , we find @xmath127 for the approximated running coupling , and @xmath128 for the two - loop running coupling . ] moreover , fig . [ fig : q - bar - q_over_fpi2 ] shows that the data of @xmath123 is in good agreement with that of @xmath124 , which implies that the approximation for the running coupling used in the present analysis works well . we also expect that the approximation does not affect the results so much when we calculate the hbs equations for the bound states . as we will discuss in section [ sec : hbseq ] , we need the mass function for complex arguments when we solve the hbs equation for the massive bound state . in this section , we first introduce the sd equation for the complex argument following ref . @xcite ( see also ref . @xcite ) , and then solve it in the case of large @xmath0 qcd . the sd equation for the complex argument is expressed as @xcite @xmath130 \frac{\bar{g}^2(x+y)\ , \sigma ( y)}{y + { \sigma^2 ( y ) } } \ , \label{eq : complex_sd}\ ] ] where @xmath131 is the integral path on the complex plane . here , we took the same argument of the running coupling as that in eq . ( [ eq : approx_of_argument ] ) , and carried out the angle integration . note that the integral path @xmath131 must be taken so as to avoid the branch cut appearing in the integral . we first study the structure of the running coupling appearing in the sd equation ( [ eq : complex_sd ] ) to clarify the branch cut . in the improved ladder approximation it is essential to use the running coupling determined from the @xmath8-function in the high energy ( space - like ) region for consistency with perturbative qcd . in qcd with small @xmath0 , however , the running coupling obtained from the perturbative @xmath8-function diverges at some infrared scale , @xmath9 . in the ordinary sd equation in the space - like region , the infrared singularity is avoided by introducing infrared regularization such as the so - called higashijima - miransky approximation @xcite and its extension as in ref . however , since the running coupling in eq . ( [ eq : complex_sd ] ) is a complex function which has the complex argument , we need an extension with analyticity satisfied . several works such as in refs . @xcite proposed models of running coupling which are consistent with perturbative qcd in the high energy region . but they still have branch cuts on the complex plane , and it is a burdensome task to evade all the branch cuts by carefully selecting the integral path in eq . ( [ eq : complex_sd ] ) . one way to avoide such a complication might be to give up the consistency with perturbative qcd and use models of running coupling with analyticity such as the one used in ref . @xcite . here we point out that the situation dramatically changes in the large @xmath0 qcd . in the case of large @xmath0 qcd , as we explained in subsection [ sec : large_nf ] , the running coupling , as well as the two - loop @xmath8-function , is finite for any space - like momentum . this implies that we may be able to construct the running coupling by analytic continuation using the @xmath8-function . actually , an explicit solution of the two - loop renormalization group equation ( rge ) can be written in terms of the lambert @xmath11 function @xcite . when @xmath0 is close to @xmath6 , the solution of the rge has no singularity on the complex plane except for the time - like axis @xcite . as a result , for general complex @xmath62 except on the time - like axis ( @xmath132 ) , we can take the integral path @xmath131 in such a way that it just avoids the branch cut coming from the angle integration . in fig . [ fig : integralpath ] we show the branch cut together with a simple choice of the integral path @xcite . ) . here , the branch cut appears from the four - dimensional angle integration . , height=226 ] we stress again that the reason why we can take this simple integral path is that the running coupling has no singularity on the complex plane except for the time - like axis . for solving the sd equation on the complex plane , we here study the explicit form of the running coupling . in fig . [ fig : analytic_cont ] we show the real part of the running coupling on the complex plane which is obtained by performing the analytic continuation from the running coupling on the real axis determined from the two - loop @xmath8-function . on the complex plane obtained by the analytic continuation from the running coupling on the real axis ( we use the cauchy - rieman relation ) . the complex argument of @xmath133 is expressed as @xmath134 , where @xmath135 and @xmath136 are real . note that @xmath63 is in the space - like region for @xmath137 and in the time - like region for @xmath138 . , height=226 ] this figure shows that @xmath139 , i.e. , @xmath140 , in the range of @xmath141 , and that @xmath142 in the range of @xmath143 . thus we take the following approximation for the running coupling on the complex plane : @xmath144 which is smoothly connected to the approximation adopted in eq . ( [ eq : step ] ) for the running coupling on the space - like axis . now , let us solve the sd equation ( [ eq : complex_sd ] ) to obtain the mass function for complex variable @xmath62 . along the integral path shown in fig . [ fig : integralpath ] , the variables @xmath62 and @xmath63 are parametrized as @xmath145 where @xmath146 , @xmath135 , and @xmath136 are real . then the sd equation ( [ eq : complex_sd ] ) is rewritten as @xmath147 \frac{\bar{g}^2 ( x+y ) \ , \sigma ( y e^{i\theta})}{y + e^{-i\theta}\,\sigma^2 ( y e^{i\theta } ) } \ . \label{eq : complex_sd 2}\ ] ] from this we can easily see that the solution is expressed as @xmath148 where @xmath149 is real and satisfies the original sd equation on the real axis : @xmath150 \frac{\bar{g}^2 ( x+y ) \ , \bar\sigma ( y)}{y + { \bar\sigma ( y)}^2 } \ . \label{eq : real sd}\ ] ] note that the fermion propagator @xmath151 does not have any poles since the kinetic part @xmath62 and the mass part @xmath152 have the same phases as @xmath153 [ see eq . ( [ eq : sigma asum ] ) ] and the mass function in the space - like region satisfies @xmath154 . we should note that the above solution in eq . ( [ eq : sigma asum ] ) is a double - valued function on the complex plane : the variable @xmath155 can be parametrized as @xmath156 for which the solution takes @xmath157 . this corresponds to the fact that the sd equation has two solutions : when @xmath74 is a solution , @xmath158 also satisfies the equation . when we choose the range of @xmath136 as @xmath159 $ ] , the branch cut emerges on the time - like axis . this choice is natural because the appearance of the branch cut in the time - like region seems consistent with the analytic structure of the running coupling . we will see that this branch cut does not matter in calculating the bound state masses . in this section we briefly review the homogeneous bethe - salpeter ( hbs ) equation for the bound states of quark and antiquark and show how to solve it numerically . in this paper , we consider the scalar , pseudoscalar , vector , and axial - vector bound states of quark and antiquark , and we write these bound states as @xmath160 , @xmath161 , @xmath162 , and @xmath163 , respectively . here , @xmath164 represents the momentum of the bound states and @xmath165 represents the polarization vector satisfying @xmath166 and @xmath167 . now , we introduce the bethe - salpeter ( bs ) amplitudes @xmath168 for the bound states of quark and anti - quark as follows : @xmath169_{\alpha\beta},\ ] ] @xmath170_{\alpha\beta},\ ] ] @xmath171_{\alpha\beta},\ ] ] @xmath172_{\alpha \beta},\ ] ] where @xmath173 , @xmath174 is the generator of @xmath175 normalized as @xmath176 = 2 \delta_{ab}$ ] , and ( @xmath133 , @xmath8 ) , ( @xmath177 , @xmath178 ) , and ( @xmath179 , @xmath180 ) denote the spinor , flavor , and color indices , respectively . we can expand the bs amplitude @xmath168 in terms of the bispinor bases @xmath181 and the invariant amplitudes @xmath182 as follows : @xmath183_{\alpha \beta } = \ \sum_{i=1}^{4 } \left [ \gamma^i_{(s , p)}(p;q ) \right]_{\alpha \beta } \chi_{(s , p)}^{i } ( p;q ) , \label{eq : chi - sp}\ ] ] @xmath184_{\alpha \beta } = \ \sum_{i=1}^{8 } \left [ \gamma^i_{(v , a)}(p;q,\epsilon ) \right]_{\alpha \beta } \chi_{(v , a)}^{i } ( p;q ) . \label{eq : chi - va}\ ] ] the bispinor bases can be determined from spin , parity , and charge conjugation properties of the bound states . the explicit forms of @xmath185 , @xmath186 , @xmath187 , and @xmath188 are summarized in appendix [ app : bispinor - bases ] . we take the rest frame of the bound state as a frame of reference : @xmath189 where @xmath190 represents the bound state mass . after the wick rotation , we parametrize @xmath191 by the real variables @xmath192 and @xmath62 as @xmath193 consequently , the invariant amplitudes @xmath182 become functions in @xmath192 and @xmath62 : @xmath194 from the charge conjugation properties for the bs amplitude @xmath168 and the bispinor bases defined in appendix [ app : bispinor - bases ] , the invariant amplitudes @xmath195 are shown to satisfy the following relation : @xmath196 the hbs equation is the self - consistent equation for the bs amplitude ( see , for a review , ref . @xcite ) , and it is expressed as ( see fig . [ fig : hbseq ] ) @xmath197 the kinetic part @xmath198 is given by @xmath199 where @xmath151 is the full fermion propagator @xmath200 , and the bs kernel @xmath201 in the improved ladder approximation is expressed as @xmath202 in the above expressions we used the tensor product notation @xmath203 and the inner product notation @xmath204 it should be noticed that the fermion propagators included in @xmath198 in eq . ( [ def t ] ) have complex - valued arguments after the wick rotation . the arguments of the mass functions appearing in two legs of the bs amplitude are expressed as @xmath205 in general , it is difficult to obtain mass functions for complex arguments . however , as we have shown in section [ sec : complex_sd ] , it is easy to obtain them in the case of large @xmath0 qcd . we now modify eq . ( [ eq : hbseq ] ) so that we can solve it numerically . it is convenient to introduce the conjugate bispinor bases defined by @xmath206 multiplying these conjugate bispinor bases from the left , taking the trace of spinor indices , and summing over the polarizations , we rewrite eq . ( [ eq : hbseq ] ) into the following form : @xmath207 where the summation over the index @xmath180 is understood , and @xmath208 , \\ k_{ij}(u , x;v , y ) & = & \int_{-1}^{1 } d\cos\theta \ \sum_{\epsilon } \frac{1}{3 } \mbox{tr } \left [ \bar\gamma^i(p;q,\epsilon ) k(p , k ) \gamma^j(k;q,\epsilon ) \right ] \ , \end{aligned}\ ] ] with the real variables @xmath209 and @xmath63 introduced as @xmath210 and @xmath211 . here , we note that although the mass function @xmath74 has the branch cut on the time - like axis as mentioned in section [ sec : complex_sd ] , @xmath212 has no singularity and becomes a continuous function for all the range of @xmath192 and @xmath62 . as for @xmath213 , the branch cut of running coupling @xmath214 does not matter since its argument @xmath215 never takes a negative value . using the property of @xmath182 in eq . ( [ eq : even_chi ] ) , we restrict the integration range as @xmath216 : @xmath217 \chi^j(v , y).\ ] ] then , all the variables @xmath192 , @xmath62 , @xmath209 , and @xmath63 can be treated as positive values . to discretize the variables @xmath192 , @xmath62 , @xmath209 , and @xmath63 we introduce new variables @xmath218 , @xmath146 , @xmath219 , and @xmath135 as @xmath220 and set uv and ir cutoffs as @xmath221 , \ \ x , y \ \in \ [ \lambda_x,\lambda_x ] .\ ] ] we discretize the variables @xmath218 and @xmath219 into @xmath222 points evenly , and @xmath146 and @xmath135 into @xmath223 points . then , the original variables are labeled as @xmath224 } = \lambda \exp\left[\lambda_u/\lambda + d_u i_u \right ] , \ \ x_{[i_x ] } = \lambda \exp\left[\lambda_x/\lambda + d_x i_x \right ] , \nonumber\\ & & v_{[i_v ] } = \lambda \exp\left[\lambda_u/\lambda + d_u i_v \right ] , \ \ y_{[i_y ] } = \lambda \exp\left[\lambda_x/\lambda + d_x i_y \right ] , \nonumber\end{aligned}\ ] ] where @xmath225 and @xmath226 . the measures @xmath227 and @xmath228 are defined as @xmath229 as a result , the integration is converted into the summation : @xmath230 in order to avoid integrable singularities in the kernel @xmath231 at @xmath232 , we adopt the following four - splitting prescription @xcite : @xmath233 , \end{aligned}\ ] ] where @xmath234 , \ \ y_\pm = \exp\left[y \pm \frac{d_x}{4}\right ] .\ ] ] now that all the variables have become discrete and the original integral equation ( [ eq : hbseq ] ) turned into a linear algebraic one , we are able to deal it numerically . however , it is difficult to find the bound state mass @xmath190 and the corresponding bs amplitude @xmath168 directly since the hbs equation depends on @xmath190 nonlinearly . a way which enables us to solve the nonlinear eigenvalue problem is the _ fictitious eigenvalue method _ @xcite . in this method we introduce a fictitious eigenvalue @xmath235 and interpret the hbs equation ( [ eq : hbseq ] ) as a linear eigenvalue equation for a given bound state mass @xmath190 : @xmath236 consequently , the hbs equation turns into an ordinary eigenvalue problem which we can solve by standard algebraic techniques . by adjusting an input mass @xmath190 such that an eigenvalue @xmath235 equals unity , we obtain the bound state mass and the corresponding bs amplitude as a solution of the original hbs equation ( [ eq : hbseq ] ) . in appendix [ app : positronium ] , to show the validity of this method , we calculate the mass of the positronium using this method . in this section we show the results of our numerical analysis . as discussed in subsection [ sec : laddersd ] , the approximation to the argument of the running coupling in eq . ( [ eq : approx_of_argument ] ) breaks the chiral symmetry explicitly @xcite . so , before solving the hbs equation for the massive bound states , we solve that for the pseudoscalar bound state and see how much the chiral symmetry is explicitly broken by this approximation . the mass of the lowest - lying pseudoscalar bound state should become zero because it appears as a nambu - goldstone boson when the chiral symmetry is spontaneously broken . so , we substitute zero for the bound state mass and check whether the fictitious eigenvalue @xmath235 becomes unity . we use the following parameters for the calculations : @xmath237 = [ \ -18.0,\ 0\ ] , \ \ [ \ \lambda_x , \ \lambda_x \ ] = [ \ -8.5,\ 0\ ] , \ ] ] @xmath238 we calculate the fictitious eigenvalues for several values of @xmath54 and show them in table [ tab : pseudoscalar ] . . fictitious eigenvalues obtained by solving eq . ( [ eq : fictitious hbs ] ) for the pseudoscalar bound state with zero mass used as an input . [ cols="^,^,^,^",options="header " , ] we can see that @xmath239 is satisfied within @xmath240 uncertainty . this implies that our calculations actually reproduce the massless nambu - goldstone boson within the numerical error , and that the effect of explicit chiral symmetry breaking caused by the approximation for the running coupling is negligible . in this subsection we show the results of the numerical calculations for the masses of the vector , axial - vector , and scalar bound states . for the uv and ir cutoffs we adjust the values of them in such a way that the dominant supports of the integrands of the decay constant in eq . ( [ eq : f_v ] ) and the normalization condition in eq . ( [ eq : normalization ] ) lie in the energy region between the uv and ir cutoffs . as an example , we show the integrands of the decay constant and the normalization condition for the vector bound states in appendix [ app : uncertainties ] . from these figures , the dominant supports lie in the lower energy region for smaller value of @xmath54 . then , we use the following @xmath54-dependent uv and ir cutoffs for the vector and the axial - vector bound states : @xmath241 & = & [ \ -12.0 + 22.0 \times ( \alpha_\ast - 1.0 ) \ , \ -1.0 + 35.0 \times ( \alpha_\ast - 1.0 ) \ ] , \label{eq : ir cutoffs } \\ \left[\ \lambda_x , \ \lambda_x \ \right ] & = & [ \ -5.0 + 22.0 \times ( \alpha_\ast - 1.0 ) \ , \ -2.0 + 20.0 \times ( \alpha_\ast - 1.0 ) \ ] . \label{eq : uv cutoffs}\end{aligned}\ ] ] for the scalar bound state , on the other hand , we use the following fixed uv and ir cutoffs : @xmath237 = [ \ -18.0,\ 0\ ] , \ \ [ \ \lambda_x , \ \lambda_x \ ] = [ \ -10.0,\ 0\ ] .\ ] ] although the integrands of the normalization conditions are shown in appendix [ app : uncertainties ] only for the vector bound states , we have checked that the dominant supports always lie within the energy region between uv and ir cutoffs for all kinds of bound states and for all values of @xmath54 . as for the numbers of the discretization , we use @xmath242 for the vector and the axial - vector bound states , and @xmath243 for the scalar bound states . in appendix [ app : uncertainties ] , we show that these numbers of discretization are large enough for the present analysis . we should stress that we actually found a solution for eq . ( [ eq : fictitious hbs ] ) reproducing @xmath244 for all the types of the bound states in the range of @xmath92 $ ] . this means that there do exist the vector , axial - vector , and scalar bound states near the phase transition point in the broken phase . ( or , equivalently , @xmath245 ) . this fact seems consistent with the property of the conformal phase transition which has no bound states in the symmetric phase @xcite . ] now , let us show the critical behavior of the masses of the existing bound states . in fig . [ fig : all_masses ] , we plot all the bound state masses calculated for several values of @xmath54 together with the pseudoscalar meson masses obtained in the previous subsection . . , height=302 ] this figure shows that the masses of the vector , axial - vector , and scalar bound states go to zero simultaneously as the coupling approaches its critical value ( or , equivalently , @xmath246 ) : @xmath247 next , to study the critical behavior of @xmath248 , @xmath249 , and @xmath250 we use the function of the form in eq . ( [ eq : fitting ] ) and fit the value of @xmath90 by minimizing @xmath251 the resultant best fitted values of @xmath90 for the scalar , vector , and axial - vector bound states are @xmath252 respectively . we also plot ( the square of ) the ratio of the bound state masses to @xmath97 for several values of @xmath54 in fig . [ fig : m_over_fpi ] . , @xmath253 , @xmath254 and @xmath255 for several values of @xmath54 ( indicated by @xmath83 , @xmath122 , @xmath256 , and @xmath257 , respectively ) . dotted lines represent the values of @xmath258 , @xmath259 , and @xmath260 obtained from eqs . ( [ eq : d_mv ] ) and ( [ eq : d_fpi ] ) . , height=302 ] the dotted lines plotted together with the data in this figure represent the values of the following ratios obtained from eqs . ( [ eq : d_mv ] ) and ( [ eq : d_fpi ] ) : @xmath261 this figure clearly shows that all the masses of the scalar , vector , and axial - vector bound states have the same scaling property as that of @xmath97 : @xmath262 we can also say that these masses have the same scaling property as that of @xmath94 since @xmath97 and @xmath94 have the same scaling property . their ratios are summarized as follows : @xmath263 one might think that the vector and axial - vector boundstates decay into a fermion and an antifermion since @xmath264 and @xmath265 . however , this does not happen : as we noticed above eq . ( [ eq : sigma_m ] ) , @xmath82 is not the pole mass but the dynamical mass defined in the space - like region . furthermore , as we have shown in section [ sec : complex_sd ] , the fermion propagators do not have any poles in the entire complex plane including the time - like axis where the pole mass of fermion should be defined . thus the vector and axial - vector boundstates do not decay into a fermion and an antifermion . in this paper we first pointed out that , when we solve the schwinger - dyson ( sd ) equation in large @xmath0 qcd , we do not need to introduce any infrared regularizations for the running coupling since it takes a finite value for all the range of the energy region due to the existence of the ir fixed point . in the case of small @xmath0 , we have to regularize the infrared divergence of the running coupling , and different schemes of regularizations would give different results . furthermore , it is difficult to find the regularization which makes the analytic structure of the running coupling simple enough . on the contrary , the solution of the two - loop rge in large @xmath0 qcd is explicitly written in terms of the lambert @xmath11 function , and the running coupling does not have any singularities on the complex plane except for the time - like axis when @xmath0 is close to @xmath6 . this significant feature of the running coupling in large @xmath0 qcd enabled us to solve the sd equation on the complex plane . then , we solved the homogeneous bethe - salpeter ( hbs ) equations for the scalar , pseudoscalar , vector , and axial - vector bound states of quark and anti - quark in large @xmath0 qcd with the improved ladder approximation in the landau gauge . in the quark propagator included in the hbs equation , we used the quark mass function obtained from the sd equation with the same approximation , which is needed for the consistency with the chiral symmetry . we first showed that the hbs equation provides the massless pseudoscalar bound state in the broken phase which is identified with the nambu - goldstone boson associated with the spontaneous breaking of the chiral symmetry . next , we showed that there actually exist vector , axial - vector , and scalar bound states even near the phase transition point in the broken phase , and that their masses decreases as the number of massless quarks @xmath0 increases . at the critical point all the masses go to zero , showing the same scaling property as that of the pseudoscalar meson decay constant @xmath97 consistently with the picture expected from the conformal phase transition @xcite . let us discuss the pattern of the chiral symmetry restoration by considering the representation of chiral @xmath266 of the low - lying mesons extending the analyses done in refs . @xcite . for @xmath267 the pseudoscalar meson denoted by @xmath268 and the longitudinal axial - vector meson denoted by @xmath269 are an admixture of @xmath270 and @xmath271 , since the chiral symmetry is spontaneously broken @xcite @xmath272 where @xmath235 denotes the helicity in the collinear frame , and the experimental value of the mixing angle @xmath273 is given by approximately @xmath274 @xcite . on the other hand , the longitudinal vector meson denoted by @xmath275 belongs to pure @xmath276 and the scalar meson denoted by @xmath277 to pure @xmath278 : @xmath279 when the chiral symmetry is restored at the phase transition point , it is natural that the chiral representations coincide with the mass eigenstates : the representation mixing is dissolved . from eq . ( [ mix pi a ] ) one can easily see that there are two ways to express the representations in the wigner phase of the chiral symmetry : the conventional manifestation la the linear sigma model ( called the gl manifestation in ref . @xcite ) corresponds to the limit @xmath280 in which @xmath268 is in the representation of pure @xmath281 of @xmath282 together with the scalar meson , both being the chiral partners : @xmath283 on the other hand , the vector manifestation ( vm ) @xcite corresponds to the limit @xmath284 in which the @xmath269 goes to a pure @xmath285 , now degenerate with the scalar meson in the same representation , but not with @xmath275 in @xmath286 : @xmath287 namely , the degenerate massless @xmath268 and ( longitudinal ) @xmath275 at the phase transition point are the chiral partners in the representation of @xmath286 . now , what does our result say on the chiral representation of low - lying mesons ? as can be seen from fig . [ fig : m_over_fpi ] , the resultant values of the masses obtained from the hbs equation roughly satisfies the following relation @xcite : @xmath288 for all values of @xmath54 . this relation holds independently of the mixing angle @xmath273 given in eq . ( [ mix pi a ] ) when the low - lying mesons saturate the chiral algebra shown in ref . @xcite . then , it is reasonable to discuss the chiral representation without worrying about the influence of the excited states of the bound states . by using the relation @xmath289 @xcite and the values of @xmath249 and @xmath248 obtained from the hbs equation in subsection [ ssec : vb ] , the value of the mixing angle @xmath273 is roughly determined as @xmath290 this implies that @xmath268 and the longitudinal @xmath269 are still admixtures of the pure chiral representation even at the chiral restoration point : @xmath291 this may suggest the existence of a new type of manifestation of chiral symmetry restoration in large @xmath0 qcd which is neither of the gl manifestation nor the simple version of the vector manifestation ( vm ) . several comments are in order in appendix [ app : decay_const ] , we show the calculations of the coupling constants @xmath12 , @xmath13 , and @xmath14 of the vector , axial - vector , and scalar bound states to the vector current , axial - vector current , and scalar density . the results shows that they also have the same scaling properties as @xmath97 . these results indicate that all the dimension - full quantities determined by the infrared dynamics have the same scaling properties , as far as the ( improved ) ladder approximation is concerned . although the masses obtained from the hbs equation satisfy the condition ( [ sat cond ] ) needed for the saturation of the chiral algebra , the couplings @xmath97 , @xmath12 , and @xmath13 do not seem to satisfy the first weinberg s sum rule @xcite : @xmath292 . we have not fully understood what this means for the pattern of chiral symmetry restoration . apparently , reducing the numerical uncertainty will help us to reach the final understanding . in the present analysis we did not include the effect from the four - fermion interaction which is induced in the case of @xmath110 as was conjectured in strong coupling qed @xcite and was demonstrated in walking gauge theories @xcite . it is not clear at this moment whether or not the qualitative results in the present analysis will be changed when we include such an effect . in the present analysis we stressed that the running coupling in large @xmath0 qcd determined from a two - loop @xmath8-function is expressed as the lambert @xmath11 function which enables us to solve the hbs and sd equations with mutual consistency near the critical point . apparently , this lambert @xmath11 function can be used as an infrared regularization to solve the hbs and sd equations with mutual consistency in the case of qcd with small @xmath0 . it will be very interesting to study meson masses near the chiral phase transition in hot and/or dense qcd by using such an infrared regularization . we think that it is important to clarify which effective field theory ( eft ) describes the new pattern of the chiral symmetry restoration expressed in eq . ( [ new pattern ] ) . especially , it is very interesting to see how the matching between the eft and the underlying qcd with large @xmath0 can be done to determine bare parameters of the lagrangian of the eft . this work was supported in part by the jsps grant - in - aid for the scientific research ( b)(2 ) 14340072 . the work of m.h . was supported in part by the brain pool program ( # 012 - 1 - 44 ) provided by the korean federation of science and technology societies and usdoe grant # de - fg02 - 88er40388 . m.h . would like to thank professor mannque rho , professor gerry brown , and professor dong - pil min for their hospitality during his stays at kias , at suny at stony brook , and at seoul national university where part of this work was done . * appendices * in this appendix , to show the validity of the fictitious eigenvalue method for solving the hbs equation explained in subsection [ ssec : fem ] , we calculate the mass of the ortho - positronium which is the vector bound state of the electron and the positron . the same analysis was done in ref . @xcite , and here we follow the analysis . in the weak coupling limit the hbs equation for the ortho - positronium can be solved analytically , and the energy spectrum takes the following form @xcite : @xmath293 where @xmath294 is the mass of the electron and the positron , and @xmath295 is the coupling constant of qed . we use following parameters in our calculation : @xmath296 = [ \ -18.5,\ -2.9\ ] \ , \ \ [ \ \lambda_{x } , \ \lambda_{x } \ ] = [ \ -10.8,\ 2.2\ ] \ , \ ] ] @xmath297 ( we used the energy scale which satisfies the relation @xmath298 following ref . @xcite . ) in fig . [ fig : positronium_bssv ] , we show the resultant values of the fictitious eigenvalue @xmath235 for several values of the input parameter @xmath249 . finding the point where the smallest @xmath235 becomes unity , we determine the value of the mass of the ground state @xmath299 as the solution of the original hbs equation : @xmath300 from this the binding energy is calculated as @xmath301 these values are in good agreement with the values @xmath302 and @xmath303 derived from eq . ( [ eq : positronium_mass ] ) . this shows that our numerical method works well to obtain the mass of the ground state . in this appendix we show the explicit forms of the bispinor bases for the scalar , pseudoscalar , vector , and axial - vector bound states . here , we use the notation @xmath304 with @xmath190 being the mass of the bound states , and @xmath305 \equiv a[b , c ] + b[c , a ] + c[a , b]$ ] . bispinor base for the scalar bound state ( @xmath306 ) is given by @xmath307 , \end{aligned}\ ] ] and that for the pseudoscalar bound state ( @xmath308 ) is given by @xmath309\ \gamma_5 \ .\end{aligned}\ ] ] furthermore , for the vector bound state ( @xmath310 ) we use @xmath311 ( p \cdot \hat q ) , \ \ \gamma_{(v)}^3 = \frac{1}{2 } [ \fsl{\epsilon},\fsl{\hat q } ] , \ \ \gamma_{(v)}^4 = \frac{1}{3!}[\fsl{\epsilon},\fsl{p},\fsl{\hat q } ] , \\ & & \gamma_{(v)}^5 = ( \epsilon \cdot p ) , \ \ \gamma_{(v)}^6 = \fsl{p } ( \epsilon \cdot p ) , \ \ \gamma_{(v)}^7 = \fsl{\hat q}(p \cdot \hat q ) ( \epsilon \cdot p ) , \ \ \gamma_{(v)}^8 = \frac{1}{2 } [ \fsl{p},\fsl{\hat q}](\epsilon \cdot p ) , \nonumber \label{v bases}\end{aligned}\ ] ] and for the axial - vector bound state ( @xmath312 ) @xmath313 \gamma_5 , \ \ \ \gamma_{(a)}^3 = \frac{1}{2 } [ \fsl{\epsilon},\fsl{\hat q}]\ ( p \cdot \hat q ) \ \gamma_5 , \nonumber\\ \gamma_{(a)}^4 & = & \frac{1}{3!}[\fsl{\epsilon},\fsl{p},\fsl{\hat q } ] \ \gamma_5 , \ \ \ \gamma_{(a)}^5 = ( \epsilon \cdot p)\ ( p \cdot \hat q)\ \gamma_5 , \ \ \ \gamma_{(a)}^6 = \fsl{p } ( \epsilon \cdot p)\ \gamma_5 , \nonumber\\ \gamma_{(a)}^7 & = & \fsl{\hat q}\ ( \epsilon \cdot p)\ ( p \cdot \hat q ) \ \gamma_5 , \ \ \ \gamma_{(a)}^8 = \frac{1}{2 } [ \fsl{p},\fsl{\hat q}](\epsilon \cdot p)\ ( p \cdot \hat q)\ \gamma_5 .\end{aligned}\ ] ] in this subsection we calculate coupling constants @xmath12 , @xmath13 , and @xmath14 of the vector , axial - vector , and scalar bound states to the vector current , axial - vector current , and scalar density . they are defined by @xmath314 where @xmath174 is the flavor matrix normalized as @xmath315 = 2\delta_{a b}$ ] . by using the bs amplitudes for the vector bound state , @xmath12 is expressed as @xmath316 \ . \label{eq : f_v}\ ] ] in the above expression , the normalization of the bs amplitudes @xmath182 are determined by the following normalization condition @xcite : @xmath317 \ . \label{eq : normalization}\end{aligned}\ ] ] here , we notice again that @xmath318 has no singularity although the fermion propagator @xmath151 has a branch cut in the time - like region . so the integral in eq . ( [ eq : normalization ] ) is well - defined . once we have obtained @xmath249 and the corresponding bs amplitudes by solving the hbs equation , we can calculate @xmath12 from eq . ( [ eq : f_v ] ) . we can also calculate @xmath13 and @xmath14 in a similar way . in fig . [ fig : decay_const ] ( a ) we show the values of @xmath12 , @xmath13 , and @xmath14 together with @xmath97 for several values of @xmath54 . , @xmath13 , @xmath14 , and @xmath97 and ( b ) @xmath319 , @xmath320 , and @xmath321 for several values of @xmath54 . , title="fig:",height=188 ] , @xmath13 , @xmath14 , and @xmath97 and ( b ) @xmath319 , @xmath320 , and @xmath321 for several values of @xmath54 . , title="fig:",height=188 ] + ( a ) ( b ) to see the scaling properties , we plot the ratio of @xmath12 , @xmath13 , and @xmath14 to @xmath97 in fig . [ fig : decay_const ] ( b ) . this figure shows that @xmath12 , @xmath13 , and @xmath14 have the same scaling properties as that of @xmath97 . to solve the hbs equation for the bound states numerically , we introduced the uv and ir cutoffs and converted the hbs equation into a linear eigenvalue equation by discretizing the integral . as we have discussed in subsection [ ssec : vb ] , we adjust the values of the uv and ir cutoffs in such a way that the dominant supports of the integrands of the decay constant in eq . ( [ eq : f_v ] ) and the normalization condition in eq . ( [ eq : normalization ] ) lie in the energy region between the uv and ir cutoffs . in figs . [ fig : sft0885 ] and [ fig : sft1000 ] we show those integrands for @xmath322 and @xmath323 in the case of the vector bound state . ) and ( b ) the normalization condition in eq . ( [ eq : normalization ] ) for @xmath324 . , title="fig:",width=226 ] ) and ( b ) the normalization condition in eq . ( [ eq : normalization ] ) for @xmath324 . , title="fig:",width=226 ] + + ( a ) ( b ) ) and ( b ) the normalization condition in eq . ( [ eq : normalization ] ) for @xmath325 . , title="fig:",width=226 ] ) and ( b ) the normalization condition in eq . ( [ eq : normalization ] ) for @xmath325 . , title="fig:",width=226 ] + + ( a ) ( b ) these figures show that the dominant supports lie in the lower energy region for smaller value of @xmath54 , and that the present choices in eqs . ( [ eq : ir cutoffs ] ) and ( [ eq : uv cutoffs ] ) covers the supports . for other values of @xmath54 used in the present analysis we have checked that the dominant supports always lie within the energy region between the uv and ir cutoffs chosen as in eqs . ( [ eq : ir cutoffs ] ) and ( [ eq : uv cutoffs ] ) . as for the numbers of the discretization , due to the limitation of the computer resources we used @xmath326 and @xmath327 as shown in eq . ( [ nbs num ] ) . here we study the dependences of the mass and the decay constant of the vector bound state on the size of discretization . we show the typical values of the mass in fig . [ fig : nbs_dependence ] ( a ) and those of the decay constant in fig . [ fig : nbs_dependence ] ( b ) for five choices of the size of the discretization , @xmath328 , @xmath329 , @xmath330 , @xmath331 , and @xmath332 . and ( b ) @xmath333 for five choices of the size of discretization , @xmath328 , @xmath329 , @xmath330 , @xmath331 , and @xmath332 . , title="fig:",height=188 ] and ( b ) @xmath333 for five choices of the size of discretization , @xmath328 , @xmath329 , @xmath330 , @xmath331 , and @xmath332 . , title="fig:",height=188 ] + ( a ) ( b ) figure [ fig : nbs_dependence ] ( a ) clearly shows that the choice @xmath334 is large enough to obtain the mass of the vector bound state . on the other hand , fig . [ fig : nbs_dependence ] ( b ) shows that there are still uncertainties in the value of the decay constant which come from the size of the discretization . apparently , this uncertainty from the size of discretization is the dominant part of the numerical uncertainties in the present analysis . t. banks and a. zaks , nucl . b * 196 * , 189 ( 1982 ) . t. appelquist , j. terning and l. c. wijewardhana , phys . lett . * 77 * , 1214 ( 1996 ) [ arxiv : hep - ph/9602385 ] . v. a. miransky and k. yamawaki , phys . d * 55 * , 5051 ( 1997 ) [ erratum - ibid . d * 56 * , 3768 ( 1997 ) ] [ arxiv : hep - th/9611142 ] . t. appelquist , a. ratnaweera , j. terning and l. c. wijewardhana , phys . d * 58 * , 105017 ( 1998 ) [ arxiv : hep - ph/9806472 ] . j. b. kogut and d. k. sinclair , nucl . b * 295 * , 465 ( 1988 ) ; f. r. brown , h. chen , n. h. christ , z. dong , r. d. mawhinney , w. schaffer and a. vaccarino , phys . d * 46 * , 5655 ( 1992 ) ; y. iwasaki , k. kanaya , s. sakai and t. yoshie , phys . lett . * 69 * , 21 ( 1992 ) ; y. iwasaki , k. kanaya , s. kaya , s. sakai and t. yoshie , nucl . suppl . * 53 * , 449 ( 1997 ) ; prog . theor . phys . suppl . * 131 * , 415 ( 1998 ) . r. oehme and w. zimmermann , phys . d * 21 * , 471 ( 1980 ) . m. velkovsky and e. shuryak , phys . b * 437 * , 398 ( 1998 ) . t. hatsuda and t. kunihiro , phys . rept . * 247 * , 221 ( 1994 ) ; r. d. pisarski , hep - ph/9503330 ; g.e . brown and m. rho , phys . * 269 * , 333 ( 1996 ) ; f. wilczek , hep - ph/0003183 ; g. e. brown and m. rho , phys . * 363 * , 85 ( 2002 ) [ arxiv : hep - ph/0103102 ] . m. harada and k. yamawaki , phys . lett . * 83 * , 3374 ( 1999 ) [ arxiv : hep - ph/9906445 ] . m. harada and k. yamawaki , phys . lett . * 86 * , 757 ( 2001 ) [ arxiv : hep - ph/0010207 ] . m. harada and k. yamawaki , to appear in phys . [ arxiv : hep - ph/0302103 ] . m. bando , t. kugo , s. uehara , k. yamawaki and t. yanagida , phys . lett . * 54 * , 1215 ( 1985 ) ; m. bando , t. kugo and k. yamawaki , phys . rept . * 164 * , 217 ( 1988 ) . r. s. chivukula , phys . d * 55 * , 5238 ( 1997 ) [ arxiv : hep - ph/9612267 ] . n. nakanishi , prog . * 43 * , 1 ( 1969 ) . t. kugo , in _ proc . of 1991 nagoya spring school on dynamical symmetry breaking , nakatsugawa , japan , 1991 _ , ed . k. yamawaki ( world scientific , singapore , 1992 ) . t. maskawa and h. nakajima , prog . theor . phys . * 52 * , 1326 ( 1974 ) ; t. maskawa and h. nakajima , prog . theor . phys . * 54 * , 860 ( 1975 ) ; m. bando , m. harada and t. kugo , prog . theor . phys . * 91 * , 927 ( 1994 ) [ arxiv : hep - ph/9312343 ] . t. kugo and m. g. mitchard , phys . b * 282 * , 162 ( 1992 ) . t. kugo and m. g. mitchard , phys . b * 286 * , 355 ( 1992 ) . m. bando , m. harada and t. kugo , prog . phys . * 91 * , 927 ( 1994 ) [ arxiv : hep - ph/9312343 ] . k. higashijima , phys . d * 29 * , 1228 ( 1984 ) . v. a. miransky , sov . j. nucl . * 38 * , 280 ( 1983 ) [ yad . fiz . * 38 * , 468 ( 1983 ) ] . k. i. aoki , m. bando , t. kugo , m. g. mitchard and h. nakatani , prog . phys . * 84 * , 683 ( 1990 ) . k. i. aoki , m. bando , t. kugo and m. g. mitchard , prog . * 85 * , 355 ( 1991 ) . c. d. roberts and s. m. schmidt , prog . * 45 * , s1 ( 2000 ) [ arxiv : nucl - th/0005064 ] . t. kugo , m. g. mitchard and y. yoshida , prog . * 91 * , 521 ( 1994 ) [ arxiv : hep - ph/9312267 ] . p. maris , c. d. roberts and p. c. tandy , phys . b * 420 * , 267 ( 1998 ) [ arxiv : nucl - th/9707003 ] . k. i. aoki , t. kugo and m. g. mitchard , phys . b * 266 * , 467 ( 1991 ) . k. naito , k. yoshida , y. nemoto , m. oka and m. takizawa , phys . c * 59 * , 1095 ( 1999 ) [ arxiv : hep - ph/9805243 ] ; k. naito and m. oka , phys . c * 59 * , 542 ( 1999 ) [ arxiv : hep - ph/9805258 ] . r. alkofer , p. watson and h. weigel , phys . d * 65 * , 094026 ( 2002 ) [ arxiv : hep - ph/0202053 ] . e. gardi and m. karliner , nucl . b * 529 * , 383 ( 1998 ) [ arxiv : hep - ph/9802218 ] ; e. gardi , g. grunberg and m. karliner , jhep * 9807 * , 007 ( 1998 ) [ arxiv : hep - ph/9806462 ] . c. n. leung , s. t. love and w. a. bardeen , nucl . b * 273 * , 649 ( 1986 ) ; nucl . b * 323 * , 493 ( 1989 ) . t. appelquist , k. d. lane and u. mahanta , phys . lett . * 61 * , 1553 ( 1988 ) ; h. georgi , e. h. simmons and a. g. cohen , phys . b * 236 * , 183 ( 1990 ) . r. m. corless , g. h. gonnet , d. g. e. hare , d. j. jeffrey and d. e. knuth , adv . math . * 5 * , 329 ( 1996 ) . h. pagels and s. stokar , phys . d * 20 * , 2947 ( 1979 ) . b. holdom , phys . b * 150 * , 301 ( 1985 ) ; k. yamawaki , m. bando and k. matumoto , phys . lett . * 56 * , 1335 ( 1986 ) ; t. akiba and t. yanagida , phys . b * 169 * , 432 ( 1986 ) ; t.w . appelquist , d. karabali and l.c.r . wijewardhana , phys . lett . * 57 * , 957 ( 1986 ) . t. kugo and y. yoshida , soryushiron kenkyu * 91 * , b26 ( 1995 ) ; y. yoshida , ph.d thesis , kyoto university ( 1995 ) m. harada and y. yoshida , phys . rev . d * 53 * , 1482 ( 1996 ) [ arxiv : hep - ph/9505206 ] . f. j. gilman and h. harari , phys . rev . * 165 * , 1803 ( 1968 ) . s. weinberg , phys . rev . * 177 * , 2604 ( 1969 ) . s. weinberg , phys . lett . * 18 * , 507 ( 1967 ) . kondo , s. shuto and k. yamawaki , mod . a * 6 * , 3385 ( 1991 ) . k. i. aoki , k. morikawa , j. i. sumi , h. terao and m. tomoyose , prog . * 102 * , 1151 ( 1999 ) [ arxiv : hep - th/9908042 ] . see , for example , c. itzykson and j. b. zuber , _ quantum field theory _ ( mcgraw - hill , new york , 1980 ) .

axis is antero - posterior : anterior ( head ) towards the left , posterior ( abdomen ) towards the right . the @xmath0 axis is medio - lateral . crosses : approximate positions of severing , along the midline in the scutellum ( blue ) and off - axis in the scutum ( black ) . + ( b ) epithelial cell apical junctions marked by e - cadherin : gfp just before the severing , @xmath1 , here in an old pupa ( see text for classification ) . blue circles : two concentric circles define the annular severed region ; the distance between circles corresponds to about one cell size . yellow arrows : macrochaete used as spatial references to position the severed region . + ( c ) first image after severing , @xmath2 s. yellow : fitted ellipse @xcite ( see methods ) . ( d ) time @xmath3 s after the severing , showing a larger opening along @xmath0 than @xmath4 . bars @xmath5 mm ( a ) , @xmath6 m ( b - d ) . [ fig : metamorphosis ] [ fig : fit_ellipse ] ] an epithelial tissue is a sheet of cells that acts as a barrier , separating for instance the outside and the inside of a multicellular organism . its biological function relies in part on the formation of a network of adherens junction belts , connected to the acto - myosin cytoskeleton , where cells adhere to each other @xcite , and which transmits mechanical information over several cell diameters @xcite . a key issue is to understand and model the role of tissue mechanics ( forces , displacements , and their time evolution ) in the coordinated changes of cell shape and position which determine morphogenetic flows at the tissue level @xcite . several models describe tissues using _ continuum mechanics _ @xcite . one precondition is the existence of a mesoscopic scale defining a domain over which averages of cell properties are well - defined @xcite . this description further relies on the assumption that the tissue mechanical state can be quantified , at the same mesoscopic scale , by two variables @xcite : the stress characterizing in which directions , and to what extent , the domain is under tension or under compression ; and the strain characterizing how far the domain geometry is from that of a relaxed state . such description has the advantage of accounting for individual cell properties using only a few spatially averaged parameters , which determine for instance how fast and through which succession of states the domain reacts to external solicitations . beyond simple observation , _ in vivo _ mechanical measurement techniques include elastography @xcite , photoelasticity @xcite , magnetic micromanipulation @xcite , tonometry @xcite or nanoindentation @xcite . a large literature ( for review , see _ e.g. _ @xcite ) has established laser ablation of individual cell junctions as a tool to measure the tensions within an epithelium , in particular during _ drosophila _ dorsal closure @xcite . this technique has allowed to measure the material relaxation time @xmath7 and contributed to a better characterization of morphogenetic processes in _ @xcite . the analysis of single cell junction ablation is usually based on reasonable assumptions @xcite : that the tissue is at mechanical equilibrium just before severing ; that the ablation is effective in removing at least part of the cell junction tension ; and that during relaxation the velocity remains small ( reynolds number much smaller than 1 ) . within these assumptions , the initial retraction velocity yields the value of the cell junction tension removed by the ablation , up to an unknown prefactor which depends on the dissipation @xcite . the tensor @xmath8 denotes the component of the stress removed by ablation . it arises as an average over several individual cell junctions within a region of the tissue . by ablating straight lines in either of two perpendicular directions @xcite or by performing statistics on single cell junction ablations in several directions @xcite , it is possible to measure step by step the anisotropy of the ratio @xmath9 ( where @xmath10 denotes the tissue viscosity ) . the stress to viscosity ratio @xmath9 is the initial strain rate after the severing : the rate of displacement that the epithelium would spontaneously undergo if it were free . the continuous description applies to other cellular materials such as foams @xcite . in the present work , we investigate experimentally whether it also applies to epithelial tissues . more specifically , we ask whether the epithelium strain , stress and material properties can be directly measured on a mesoscopic scale . we address these questions in a model system , the dorsal thorax epithelium of _ drosophila _ pupa . the pupa is the life stage during which the larva starves , feeds upon its reserves and metamorphoses into an adult @xcite . the dorsal thorax is a single layer of cells whose apical surfaces , surrounded by the adherens junction belts , face the cuticle , which protects the pupa ; their basal surfaces face the hemolymph , which acts as a nutrient transporter . it is composed of a large anterior region , the scutum , and a posterior tip , the scutellum . the thorax , and especially the scutellum , undergoes extensive morphogenetic changes during the metamorphosis that shapes the adult fly ( figure [ fig : metamorphosis ] ( a ) ) , making it a useful model to decipher the mechanisms that control tissue morphogenesis during development @xcite . it has a bilateral symmetry axis : the midline ( axis @xmath4 , dashed line in figure [ fig : fit_ellipse ] ( a - b ) ) . it displays large cells , the macrochaete ( precursor cells of adult sensory hairs , figure [ fig : metamorphosis ] ( b ) ) , whose positions are precisely reproducible @xcite . we introduce an original type of severing experiment . we sever by short pulse laser the adherens junctions in an annular region around a @xmath1130 @xmath12 m radius circular tissue domain ( see figure [ fig : fit_ellipse ] ( b - d ) , methods and movies 1 - 3 ) . the inner and outer wound margins display a comparable speed , displacement and anisotropy . note that the same anisotropy in speed and displacement results in the outer ellipse boundary having minor and major axes orthogonal to the minor and major axes of the inner one . after severing , the wound heals within tens of minutes and the pupa develops to adult . experiments are done in the scutellum ( blue cross in figure [ fig : metamorphosis ] ( a ) ) . we classify them into three groups , according to the developmental age at which the severing is performed : earlier than 15 - 16 h , approximately 18 - 20 h , and later than 24 - 26 h after pupa formation . for brevity we call them young " ( @xmath13 ) , middle - aged " ( @xmath14 ) and old " ( @xmath15 ) pupae , respectively . similar experiments can be performed in other locations . as a proof of principle , we include three experiments performed in the scutum ( black cross in figure [ fig : metamorphosis ] ( a ) ) of middle - aged ( @xmath16 ) and old ( @xmath17 ) pupae . we analyse the retraction of the inner tissue domain and its margin , within the set of assumptions relevant for the retraction of a single severed cell - junction . the boundary of the retracting domain is fitted at each time point by an ellipse @xcite ( see methods and figure [ fig : fit_ellipse ] ( c - d ) ) , yielding values of the @xmath4 and @xmath0 axes as a function of time ( figure [ fig : tau_x_vs_v ] ) . our protocol offers several advantages . ( i ) the experiment directly yields measurements averaged over the severed tissue domain , comprising typically a hundred of cells . as discussed below , the domain size is thus an adequate mesoscopic scale , crossing - over from detailed cell - level to tissue - level continuous descriptions @xcite . ( ii ) the comparison between the initial and final states yields a direct measurement of the strain which existed before severing . ( iii ) several cell - cell junctions in various directions are severed at once ( rather than step by step ) and in the same pupa . hence one experiment yields a direct measurement of all stress to viscosity tensor components . it takes only a few tens of seconds , the typical value of the relaxation time . ( iv ) last but not least , by isolating in time and space an epithelium domain from its neighboring cells , we obtain a complete set of well - defined initial , boundary and final conditions for a spatio - temporal model of the retracting domain . we thus introduce a continuous model , in the spirit of mayer _ et al . _ @xcite and taking into account a coupling between space and time dependences . as explained below , fitting such a model to the displacements in the domain bulk tests the continuous description , suggests an interpretation of the relaxation time , and probes the competition between internal viscosity and external friction . model - independent measurement of strain @xmath18 , stress to viscosity ratio @xmath9 , and relaxation time @xmath7 . the wound margin position ( ellipse semi - axis ) is plotted _ versus _ time after severing ; data from figure [ fig : fit_ellipse ] ( b - d ) , esm movie 1 , along the @xmath0 axis . the difference between the initial ( @xmath19 ) and final ( @xmath20 ) positions ( blue dashed lines ) directly yields the value of @xmath21 . the initial velocity @xmath22 ( slope of the orange dashed line ) yields an estimate of @xmath23 . inset : velocity , estimated by finite differences of successive positions , _ versus _ the position during the first @xmath24 s. an arrow indicates the direction of increasing time @xmath25 . the slope of a linear fit ( purple line ) yields the inverse of the relaxation time , @xmath26 ( see equation [ eq : measure_tau ] ) . ] severing the epithelium reveals the displacement that the epithelium would spontaneously undergo if it were free . we measure the strain using a logarithmic definition @xcite , hencky s true strain " @xcite : @xmath27 where @xmath19 is the initial radius , and @xmath28 is the ellipse semi - axis at the end of the relaxation . similarly @xmath29 . the advantages of this definition are that @xcite : it is valid at all amplitudes ; it is adapted for tensorial measurements ; and finally it increases the range of validity of the linear elasticity approximation which applies here even to the highest strain value we measure , @xmath30 . each severing is followed by a relaxation to a final domain strictly smaller than the initial disk ( figure [ fig : fit_ellipse ] ( d ) ) , indicating that before the severing the tissue had a positive strain in all directions . since the midline is a symmetry axis , we expect that the strain axes are parallel and perpendicular to it . we check that this is the case for the fitted ellipse axes , and that accordingly the shear strain @xmath31 is indistinguishable from 0 . we thus plot @xmath32 and @xmath33 ( figure [ fig : strain_anisotropy ] ( a ) ) , obtained with absolute precision better than @xmath34 . the values are clustered according to the three pupa age groups : @xmath18 is low and isotropic at young age ( green cluster ) , moderate and isotropic at middle age ( red ) , high and anisotropic ( @xmath35 ) at old age ( blue ) . similarly , in each experiment , the initial retraction of the wound margins ( figure [ fig : fit_ellipse ] ( c ) ) indicates the sign of the stress : before the severing the tissue was under tensile stress in all directions . this is reminiscent of positive cell junction tensions observed in the wing @xcite and in the notum @xcite . the initial retraction velocity divided by @xmath19 yields the stress to viscosity ratio @xmath9 ( figure [ fig : tau_x_vs_v ] ) , in a way which is likely to be independent of any tissue rheological model , as suggested by the following argument . when ablating a single cell - cell junction , force balance shows that the force @xmath36 exerted by the ablated junction on the neighbouring vertex is proportional to the initial recoil velocity @xmath37 : @xmath38 , where @xmath39 is a friction coefficient @xcite . coarse - graining this relation , and assuming that the main source of dissipation is identical in our experiments which sever many junctions at once , yields @xmath40 , where the factor @xmath19 is introduced by dimensional analysis . results for the stress to viscosity ratio @xmath9 ( figure [ fig : sigma_anisotropy ] ( b ) ) are similar to those for @xmath18 : its amplitude and anisotropy are initially small and increase with age ; the values are clustered according to the three age groups ; no shear stress is detected . values of @xmath9 span almost two decades , reflecting the sensitivity and precision of the method . the relaxation time , @xmath7 , is obtained from the time evolution of the ellipse axes . instead of fitting an exponential to the data , we plot the velocity _ versus _ position ( inset of figure [ fig : tau_x_vs_v ] ) . this method is robust and independent of any assumption . a linear regression of the velocity as a function of position yields a slope of @xmath41 , as seen , _ e.g. _ , in the equation : @xmath42 we find that the relaxation time is approximately isotropic ( esm figure 3 ( a ) ) and use @xmath43 as an estimate of @xmath7 . it is close to 10 s and only slightly varies with the pupa age group ( figure [ tau_vs_logxi ] ( c ) , vertical axis ) . [ fig : sigma_anisotropy ] [ tau_vs_logxi ] mechanical state and material properties . color code according to pupa development ages : green , young ; red , middle - aged ; blue , old . the edges of the rectangular regions represent the mean values @xmath44 standard deviations for each group . the experiments in the scutum are in yellow . ( a ) strain anisotropy : @xmath33 _ versus _ @xmath32 . note the difference in horizontal and vertical scales ; the solid line is the first bisectrix @xmath45 , indicating the reference for isotropy . ( b ) same for the severed stress to viscosity ratio @xmath46 . ( c ) relaxation time @xmath7 and dimensionless friction to viscosity ratio @xmath47 ; values are the averages of the measurements along the @xmath4 and @xmath0 axes ( esm figure 3 ) . note the semi - log scale . the blue rectangle takes into account two very small values of @xmath48 , of order @xmath49 and @xmath50 ( below the plotted range ) . ] to describe the bulk of the retracting domain , we track the displacements of around a hundred features in a band of tissue ( figure [ fig : klt ] and methods ) . spatially averaging their positions by binning them into 8 equal - size bins improves the signal to noise ratio . we obtain the initial velocity profile _ versus _ position ( figure [ fig : fit : xi ] ) . for old pupae , it is spatially linear ; for middle - aged pupae it is spatially nonlinear ; in the young pupae , immediately after severing only the boundaries move significantly . to account for these observations , we formulate a spatio - temporal , viscoelastic , kelvin - voigt model in one dimension of space , where the coordinate @xmath51 denotes either @xmath4 or @xmath0 , and @xmath25 denotes time after severing ( see esm ) . thanks to the definition ( [ eq : def : hencky ] ) , the strain @xmath52 is related without approximation to the velocity @xmath53 through : @xmath54 we model the effect of friction of the epithelium against the hemolymph and the cuticle as an external fluid friction with coefficient @xmath55 . we find that when internal viscosity dominates , the strain remains uniform and all parts of the tissue relax exponentially with the same viscoelastic relaxation time @xmath56 where @xmath57 is the young modulus ( esm figure 4 ( a ) ) . when external friction dominates , strain diffuses from the boundaries with a diffusion coefficient @xmath58 and remains thus inhomogeneous at times @xmath25 smaller than @xmath59 ( esm figure 4 ( b ) ) . in the general case , the dynamical equation for the strain field @xmath60 reads : @xmath61 with the conditions @xmath62 initially , @xmath63 at boundaries , and @xmath64 at the end . the relative importance of external friction _ versus _ internal viscosity is quantified by the dimensionless parameter : @xmath65 solving equation [ eq : general : d ] at short time and integrating over space equation [ eq : velocity_strain ] at @xmath66 yields the initial velocity profile : @xmath67 whose curvature is controlled by @xmath48 . we thus estimate @xmath48 by fitting a hyperbolic sine to experimental initial velocity profiles ( figure [ fig : fit : xi ] ) . as for the relaxation time , we check that the values @xmath68 and @xmath69 determined in the directions @xmath4 and @xmath0 are roughly consistent ( esm figure 3 ( b ) ) : we thus use @xmath70 as an estimate of @xmath48 . we find that @xmath48 varies more than @xmath7 ( figure [ tau_vs_logxi ] ( c ) , horizontal axis ) . the middle aged group is characterized by intermediate values of @xmath48 ( figure [ fig : fit : xi ] , red ) , of order of 10 . according to equation [ eq : def : xibar ] , this indicates a possible effect of the external friction on the relaxation for scales larger than @xmath71 m . for both other age groups , the values of @xmath48 should be considered as bounds on the order of magnitude . for old pupae we find @xmath72 smaller than a few times unity ( figure [ fig : fit : xi ] , blue ) . for young pupae , we find @xmath72 larger than a few tens or hundred ( figure [ fig : fit : xi ] , green ) . this model depends on three parameters measured independently of each other : the initial strain @xmath73 , the viscoelastic time @xmath7 , and the friction to viscosity ratio @xmath48 . to validate the model , we use it to numerically compute the whole space - time map of strains and displacements we simulate each experiment using the values of @xmath7 and @xmath48 measured as described above . figure [ boxtraj ] shows that we find a good agreement with experimental data . moreover , we plot the square deviation between experimental and numerical data as a function of the parameter values @xmath74 ( see methods ) . we check that the error landscapes we obtain are broadly consistent with our experimental estimates of @xmath48 and @xmath7 ( esm figure 5 ) . model - dependent measurement of the friction to viscosity ratio @xmath48 . initial velocity profiles immediately after severing are plotted _ versus _ initial position prior to severing , for three typical pupae : young ( green ) , middle - aged ( red ) and old ( blue ) . open squares : features , from figure [ fig : klt ] ( b ) . closed triangles : same , spatially averaged in 8 bins . lines : fit by a sinh function ( equation [ eq : vinit ] ) , yielding for @xmath48 a value either above , in , or below the measurable range : @xmath75 ( green ) , @xmath765 ( red ) , @xmath77 ( blue ) , respectively . ] validation of the model . blue triangles : positions of features from figure [ fig : klt ] ( b ) , spatially averaged in bins _ versus _ time after severing ; bars : standard deviation . red lines : numerical resolution of equation 3 , using the measured values @xmath78 , @xmath79 . ] we report relaxation times @xmath7 in the range 5 to 20 s , and initial velocities in the range 0.1 to 0.6 @xmath12m.s@xmath80 . this agrees with values provided by single cell junction ablation : 10 - 100 s for the relaxation times , 0.1 to 1 @xmath12m.s@xmath80 for the initial velocities in other _ tissues @xcite . the model allows to interpret @xmath7 as a viscoelastic time . its order of magnitude is typical of the viscoelastic times corresponding to cell _ internal _ degrees of freedom @xcite , also measured in cell aggregates @xcite . conversely _ external _ degrees of freedom ( such as cell - cell rearrangements or cell divisions ) would have much larger viscoelastic times : in cell aggregates they are typically of order of several hours @xcite . the time scales of morphogenesis in this tissue are also of order of hours @xcite , as observed in other tissues as well , such as the _ drosophila _ pupal wing @xcite . indeed continuous models of morphogenetic processes presuppose the existence of such a separation of time scales . the present experimental data are well described assuming that a single viscoelastic time is relevant . other experiments display a wide distribution of viscoelastic times @xcite : the model can easily be modified so as to take this feature into account . note that in the presence of friction the model already predicts that different parts of the severed domain relax at different rates ( esm figure 4 ( b ) ) . measurements of @xmath18 , @xmath9 , @xmath7 and @xmath48 are performed as averages over the severed disk . the requirements on the averaging scale are that it should be ( i ) large enough to obtain a sufficient signal to noise ratio yielding significant results , and ( ii ) small enough to be able to focus simultaneously on the whole severed ring despite the epithelium surface curvature which is observed in old pupae . the @xmath81 symmetry with respect to the midline observed in figures [ fig : fit_ellipse ] ( b - d ) , [ fig : fit : xi ] , and [ boxtraj ] acts as a control of the signal to noise ratio . the experimentally available range of scales corresponds to 60 - 100 cell areas , in other words the semi - axis length @xmath19 corresponds to 4 - 5 cell diameters . for old pupae , scales smaller than @xmath19 are also probed , corresponding to the scale of bins ( at least @xmath82 or @xmath83 , see figs . [ fig : fit : xi ] and [ boxtraj ] ) . we experimentally determine the order of magnitude of the dimensionless friction to viscosity ratio , @xmath48 ( figure [ tau_vs_logxi ] ( c ) , horizontal axis ) . obtaining more precise values of @xmath48 would depend on model ingredients , such as depth of ablation along the apico - basal axis , detailed cell geometry , tissue compressibility , or type of friction . for simplicity we have modeled by a fluid friction @xcite the dissipation on either surfaces of the epithelium , namely the hemolymph at basal side and the cuticle at apical side . we do not observe any threshold effect due to solid friction , and the agreement between model and data justifies _ a posteriori _ our assumption . for young pupae , the initial velocity field is not significantly higher than the noise level ( figure [ fig : fit : xi ] , green ) . in this age group , it can not be used to validate the model , and the estimate of @xmath48 strongly depends on the motion of domain parts located close to the boundary ( equation [ eq : vinit ] ) . in the dorsal thorax , the severed domain fulfills the theoretical requirements for a continuous description : it contains a number of cells much larger than one ( large enough to allow for averaging ) , while being much smaller than the total number of cells in the whole tissue . one could ask whether averages at scale @xmath19 are meaningful , due to variation of cell properties from cell to cell . we speculate that if heterogeneities within the domain dominated the average , the domain boundary could in principle adopt any arbitrary shape compatible with the @xmath81 symmetry . however , in practice , we observe that the domain boundary at the beginning , as well as at the end , of the relaxation also displays a @xmath84 symmetry ( figure [ fig : fit_ellipse ] ( b - d ) and movies 1 - 3 ) and can be fitted by an ellipse : this is consistent with the measurements of @xmath9 and @xmath18 as tensors averaged over scale @xmath19 . limits of our model lie in its simplifications : one dimensional treatment , linear elasticity and viscosity , isotropy and homogeneity of material parameters including friction , identical properties of all cells contained in the severed region . however , the agreement between model and experiment is already good , and validates these approximations _ a posteriori_. we emphasize that one major goal of this model is to provide an order of magnitude of the friction to viscosity ratio @xmath48 . we do not claim that this model is unique , nor that it is general . in the future we hope to further test the model ingredients and relevance on this epithelium and other quasi - planar ones . since its ingredients are general , we expect our model to be of broad relevance when studying the relaxation of quasi - planar epithelia over similar time and length scales . however , the simplifications involved deserve further comments . in fact , completely different hypotheses could also be invoked to explain the spatial non - linearity in the velocity profile observed for young and middle - aged pupae . for instance , the viscous component of biological tissues could be a non - linear function of time , strain , and strain rate @xcite . a strain - dependent viscosity coefficient may lead to a non - linear velocity profile without the need to invoke any external friction . alternatively , a spatially heterogeneous friction might also explain the observed velocity profiles . since a linear velocity profile is observed in older pupae , the amplitude of these additional , more complex ingredients would have to decrease strongly during development . we believe that our model has the merit of simplicity . it successfully accounts for the description of the velocity profiles at any time . each of its hypotheses could be tested experimentally and , if experimental results require it , hypotheses could be relaxed to lead to more detailed explorations . we expect that additional ingredients could strongly affect the value of @xmath48 , but that other measurements presented here ( namely @xmath18 , @xmath9 and @xmath7 ) should be robust . the model thus provides a flexible and robust route to a mechanical description of quasi - planar epithelia . building upon classical single - junction laser ablations , our experiments enable to measure the time evolution of the mechanical state ( strain , and stress to viscosity ratio ) and material properties ( relaxation time , and friction to viscosity ratio ) during pupa metamorphosis , as well as their anisotropy . this provides relevant ingredients for the modeling of a tissue as a continuous , linear , viscoelastic material . moreover , our approach provides a method to explore the biological causality of the observed properties , by analysing mutant phenotypes or by micro - injecting drugs . we conjecture that @xmath7 would be affected by low doses of drugs known to modify cytoskeletal rheology such as cytochalasin d @xcite , ck-666 @xcite , y-27632 @xcite which respectively inhibit actin polymerization , arp2/3 complex activity or rho - kinase activity . we anticipate that modulating cell - cell adhesion by knock - down of e - cadherin or catenins @xcite could affect the viscosity . although we can not modulate the cell - cuticle adhesion , we could tune cell - matrix adhesion by knock - down of integrin or its linker to the cytoskeleton @xcite to affect the friction and thereby modify @xmath48 . in principle , by establishing a complete map of @xmath8 at different positions , one could determine the length scale at which @xmath8 varies within the tissue , and check _ a posteriori _ whether it is larger than the scale @xmath19 used to measure @xmath8 . however , the present experiments measure the stress up to a dissipative prefactor , the tissue viscosity . similarly , single cell - junction ablation experiments measure force up to a prefactor ( a friction coefficient ) , rarely discussed or measured in the literature . to separate the stress and viscosity variations , alternative methods to directly measure forces and viscosities in live epithelia are thus called for . these include the measurement of cell junction tensions from movies observations @xcite , where the unknown prefactor , namely the average tension over the whole image , is by definition the same for all junctions . external mechanical manipulation @xcite yields direct determination of out - of - plane elastic and/or viscous moduli , but the in - plane moduli are only indirectly determined . another possible strategy could rely on microrheology @xcite : either passive , by tracking the brownian diffusion of beads within the cells ; or active , by moving ( magnetically or optically ) a bead back and forth within a cell at a specified frequency . such measurements could yield a direct access to the value of @xmath10 , but at the intracellular level . measuring @xmath10 at the scale of several cells either directly , or indirectly by measuring @xmath57 and @xmath7 and using the relation @xmath85 , could probably be possible , but we are not aware yet of a published method . since @xmath18 is a dimensionless geometrical quantity , its value is insensitive to cell size : the observed @xmath117-fold increase in @xmath18 can not be related to changes of cell size over time . it should involve other causes , that remain to be determined . part of the measured increase in @xmath9 values may be due to cell divisions which decrease the average cell area by a factor of @xmath86 between young and old pupae . the observed increase is a factor of @xmath87 , and thus should involve other causes as well . one candidate is a decrease of the tissue viscosity @xmath10 . however the near constancy of the viscoelastic time would then imply a corresponding increase in tissue elasticity , a rather unlikely coincidence . we therefore expect that stronger tensile forces are at work in the tissue in older pupae . from 18 hours apf onwards , the lateral part of the scutellum undergoes apical cell contractions that shape the lateral domain of the tissue @xcite . this observation led us to divide the ablation experiments into the three age groups . it may also explain the changes in mechanical properties measured in the tissue for middle - aged and old pupae . to precisely analyse the causes of the increases in @xmath18 and @xmath9 would therefore require to identify genes that specifically affect the lateral contraction of the tissue . our experimental method should be applicable to other tissues . performing experiments in an epithelium which is flat enough to significantly increase @xmath19 , and with a sufficient signal to noise ratio on the estimate of @xmath48 , would enable to check the scaling of @xmath48 _ versus _ @xmath19 given by equation [ eq : def : xibar ] . _ drosophila melanogaster _ larvae were collected at the beginning of metamorphosis ( pupa formation ) @xcite . the pupae were kept at 25@xmath88c , then dissected and mounted as described in @xcite ( esm figure 1 ) . using the adherens junction protein e - cadherin fused to green fluorescent protein @xcite we imaged the apical cell junctions by fluorescent microscopy @xcite . the time - lapse laser - scanning microscope lsm710 nlo ( carl zeiss microimaging ) collected 512@xmath89512 pixel images in mono - photon mode at 488 nm excitation ( pixel size in the range 0.24 to 0.29 @xmath12 m ) through a plan - apochromat 63@xmath891.40 oil objective . before and after severing , the time interval between two consecutive frames was either 393 ms or 970 ms according to the scanning mode ( bi- or mono - directional ) , without any apparent effect on the results presented here . we focused on the adherens junctions to be severed ; due to the epithelial surface curvature especially in old pupae , adherens junctions of innermost and outermost regions were slightly out of focus ( figure [ fig : fit_ellipse ] and movies 1 - 3 ) . we defined a region of interest as an annular region between two concentric circles . we severed it at the 10@xmath90 frame ( figure [ fig : fit_ellipse ] ) and defined @xmath19 as the radius of its inner circle . laser severing was performed using ti : sapphire laser ( mai tai deepsee , spectra physics ) in two - photon mode at 890 nm , @xmath91 100 fs pulses , 80 mhz repetition rate , @xmath11 0.2 w at the back focal plane , used at slightly less than full power to avoid cavitation ( esm figure 2 ) @xcite . severing itself had a duration ranging from 217 to 1300 ms ( according to the size of the region of interest and the scanning mode ) during which no image was acquired . we could check that the severing had been effective to remove their adherens junctions , rather than simply bleach them ( esm figure 2 ) : after retraction the cells at the wound margin moved apart more than a cell diameter ; moreover , there was no fluorescence recovery ( figure [ fig : fit_ellipse ] ( d ) ) . at the tissue scale , for each frame after the severing the inner severed tissue boundary was fitted with an ellipse ( figure [ fig : fit_ellipse ] ( c - d ) ) using the ovuscule @xcite imagej plugin @xcite . we have modified it so that each fitting procedure started from the ellipse fitted on the preceding are linearly fitted over a 5 frames sliding window in order to improve the signal to noise of position and velocity determinations . independently , at the cell scale , images were denoised with safir software @xcite . we took the logarithm of the grey levels , to obtain comparable intensity gradients in differently contrasted parts of the image ( figure [ fig : klt ] ( a ) ) . this allowed us to use kanade - lucas - tomasi ( klt ) @xcite tracking algorithm @xcite by selecting @xmath92 features " of interest , corresponding to most cell vertices ( figure [ fig : klt ] ( a ) ) . features were tracked from frame to frame : they followed the cell vertex movement towards the center , and we have observed no cell neighbour swapping . we checked that the features center of mass had only a small displacement , which we subtracted from individual feature displacements without loss of generality . to implement a quasi-1d analysis , we collected the features in a band along the @xmath0 axis and similarly along the @xmath4 axis ( figure [ fig : klt ] ( b - c ) ) , and both axes were analysed separately . the band width ( here @xmath93 of the initial circle diameter ) was chosen to be large enough to perform large statistics on features , and small enough to treat the features displacement as one - dimensional . we numerically solved adimensionalized equations ( see esm ) using the matlab solver ` pdepde ` . for a given set of parameter values @xmath94 , we calculated @xmath73 from the strain of the feature of largest initial amplitude , and rescaled the solution of esm equation 21 so as to satisfy esm equation 10 . using the experimental initial positions of features @xmath95 and the calculated strain field @xmath52 , we simulated the trajectories @xmath96 , @xmath97 , where @xmath98 was the number of features . the mismatch @xmath99 with our computation was evaluated by comparison with the experimental trajectories , @xmath100 : @xmath101 where @xmath102 denoted the euclidean norm , and @xmath103 the number of movie images . we gratefully thank f. molino for modifying the ovuscule plugin , p. thvenaz and s. birchfield for advices regarding their softwares , h. oda for reagents , j .- allain for suggesting to use klt algorithm , lab members for discussions , and the pict - ibisa@bdd ( umr3215/u934 ) imaging facility of the institut curie . this work was supported by grants to y.b . from arc ( 4830 ) , anr ( blan07 - 3 - 207540 ) , erc starting grant ( cepodro 209718 ) , cnrs , inserm , and institut curie ; and by postdoc grants to i.b . by the frm ( spf20080512397 ) , to f.b . by the nwo ( 825.08.033 ) . 90 lecuit t , lenne pf 2007 cell surface mechanics and the control of cell shape , tissue patterns and morphogenesis . nat rev mol cell biol _ 8:633644 . chen cs , tan j , tien j 2004 mechanotransduction at cell - matrix and cell - cell contacts . _ annu rev biomed eng _ 6:275302 . ma x , lynch he , scully pc , hutson ms 2009 probing embryonic tissue mechanics with laser hole drilling . _ phys biol _ 6:036004 . lecuit t , le goff l 2007 orchestrating size and shape during morphogenesis . _ nature _ 450:189192 . mammoto t , ingber de 2010 mechanical control of tissue and organ development . _ development _ 137:14071420 . fung yc 1993 _ biomechanics : mechanical properties of living tissues _ ( springer ) . hufnagel l , teleman aa , rouault h , cohen sm , shraiman bi 2007 on the mechanism of wing size determination in fly development . _ proc natl acad sci usa _ 104:3835 - 3840 . bittig t , wartlick o , kicheva a , gonzalez - gaitan m , jlicher f 2008 dynamics of anisotropic tissue growth . _ new j phys _ 10:063001 . blanchard g , _ et al . _ 2009 tissue tectonics : morphogenetic strain rates , cell shape change and intercalation . _ nat meth _ 6:458 - 464 . butler l , _ _ 2009 cell shape changes indicate a role for extrinsic tensile forces in drosophila germ - band extension . _ nat cell biol _ 11:859 - 864 . ranft j , _ _ 2010 fluidization of tissues by cell division and apoptosis . _ proc natl acad sci usa _ 107:2086320868 . aigouy b , _ _ 2010 cell flow reorients the axis of planar polarity in the wing epithelium of drosophila . _ cell _ 142:773786 . batchelor gk 2000 _ an introduction to fluid dynamics _ ( cambridge university press ) ophir j , cespedes i , garra b , ponnekanti h , huang y , maklad n , 1996 elastography : ultrasonic imaging of tissue strain and elastic modulus in vivo _ eur j ultrasound _ 3:49 - 70 . nienhaus u , aegerter - wilmsen t , aegerter cm 2009 determination of mechanical stress distribution in drosophila wing discs using photoelasticity . _ mech dev _ 126:942949 . desprat n , supatto w , pouille p , beaurepaire e , farge e 2008 tissue deformation modulates twist expression to determine anterior midgut differentiation in drosophila embryos . _ dev cell _ 15:470477 . fleury v , _ et al . _ 2010 introducing the scanning air puff tonometer for biological studies . _ phys rev e _ 81:021920 . peaucelle a , braybrook s , le guillou l , bron e , kuhlemeier c , hfte h , 2011 pectin - induced changes in cell wall mechanics underlie organ initiation in _ arabidopsis_. _ curr biol _ 21:1720 - 1726 . rauzi m , lenne pf 2011 cortical forces in cell shape changes and tissue morphogenesis . _ curr top dev biol _ 95:93144 . hutson ms , _ _ 2003 forces for morphogenesis investigated with laser microsurgery and quantitative modeling . _ science _ 300:145149 . hutson ms , ma x 2007 plasma and cavitation dynamics during pulsed laser microsurgery in vivo . _ phys rev lett _ 99:158104 . hutson ms , _ et al . _ 2009 combining laser microsurgery and finite element modeling to assess cell - level epithelial mechanics . _ biophys j _ 97:30753085 . kiehart dp , galbraith cg , edwards ka , rickoll wl , montague ra 2000 multiple forces contribute to cell sheet morphogenesis for dorsal closure in drosophila . _ j cell biol _ 149:471490 . peralta xg , _ et al . _ 2007 upregulation of forces and morphogenic asymmetries in dorsal closure during drosophila development . _ biophys j _ 92:25832596 . farhadifar r , rper j , aigouy b , eaton s , jlicher f 2007 the influence of cell mechanics , cell - cell interactions , and proliferation on epithelial packing . _ curr biol _ 17:20952104 . rauzi m , verant p , lecuit t , lenne pf 2008 nature and anisotropy of cortical forces orienting drosophila tissue morphogenesis . _ nat cell biol _ 10:14011410 . landsberg kp , _ _ 2009 increased cell bond tension governs cell sorting at the drosophila anteroposterior compartment boundary . _ curr biol _ 19:19501955 . r , simoes sdm , rper j , eaton s , zallen ja 2009 myosin ii dynamics are regulated by tension in intercalating cells . _ dev cell _ 17:736743 . bosveld f , bonnet i , guirao b _ et al . _ 2012 mechanical control of morphogenesis by fat / dachsous / four - jointed planar cell polarity pathway . _ science _ , in press . mayer m , depken m , bois js , jlicher f , grill sw 2010 anisotropies in cortical tension reveal the physical basis of polarizing cortical flows . _ nature _ 467:617621 . martin ac , gelbart m , fernandez - gonzalez r , kaschube m , wieschaus ef 2010 integration of contractile forces during tissue invagination . _ j cell biol _ 188:735749 . janiaud e , graner f , 2005 foam in a two - dimensional couette shear : a local measurement of bubble deformation . _ j fluid mech _ 532:243267 . graner f , dollet b , raufaste c , marmottant p 2008 discrete rearranging disordered patterns , part i : robust statistical tools in two or three dimensions . _ eur phys j e _ 25:349369 . bainbridge sp , bownes m 1981 staging the metamorphosis of _ drosophila melanogaster_. _ j embryol exp morph _ 66:57 - 80 langevin j , _ et al . _ 2005 drosophila exocyst components sec5 , sec6 , and sec15 regulate de - cadherin trafficking from recycling endosomes to the plasma membrane . _ dev cell _ 9:365376 . thvenaz p , delgado - gonzalo r , unser m 2011 the ovuscule . _ ieee trans pattern anal mach intell _ 33:382393 . hoger a , 1987 the stress conjugate to logarithmic strain . _ int j solids struct _ 23:1645 - 1656 . farahani k , naghdabadi r , 2000 conjugate stresses of the seth - hill strain tensors . _ int j solids struct _ 37:5247 - 5255 . tanner r , tanner e 2003 heinrich hencky : a rheological pioneer . _ rheologica acta _ 42:93101 . wottawah f , _ _ 2005 optical rheology of biological cells . _ phys rev lett _ 94:098103 . marmottant p , _ et al . _ 2009 the role of fluctuations and stress on the effective viscosity of cell aggregates . _ proc natl acad sci usa _ 106:1727117275 . guevorkian k , gonzalez - rodriguez d , carlier c , dufour s , brochard - wyart f 2011 mechanosensitive shivering of model tissues under controlled aspiration . _ proc natl acad sci usa _ 108:1338713392 . guevorkian k , colbert m , durth m , dufour s , brochard - wyart f 2010 aspiration of biological viscoelastic drops . _ phys rev lett _ 104:218101 . flanagan md , lin s 1980 cytochalasins block actin filament elongation by binding to high affinity sites associated with f - actin . _ j biol chem _ 255:835 - 838 . nolen bj , tomasevic n , russell a , pierce dw , jia z , mccormick cd , hartman j , sakowicz r , pollard td 2009 characterization of two classes of small molecule inhibitors of arp2/3 complex . _ nature _ 460:1031 - 1034 . uehata m , ishizaki t , satoh h , ono t , kawahara t , morishita t , tamakawa h , yamagami k , inui j , maekawa m , narumiya s. 1997 calcium sensitization of smooth muscle mediated by a rho - associated protein kinase in hypertension . _ nature _ 389:990 - 994 . papusheva e , heisenberg cp 2010 spatial organization of adhesion : force - dependent regulation and function in tissue morphogenesis . _ embo j. _ 29:2753 - 2768 . brodland gw , _ et al . _ 2010 video force microscopy reveals the mechanics of ventral furrow invagination in drosophila . _ proc natl acad sci usa _ 107:2211122116 . mackintosh f , schmidt c 1999 microrheology . _ curr opinion coll interface science _ 4:300 - 307 . panorchan p , lee jsh , daniels br , kole tp , tseng y , wirtz d , 2007 probing cellular mechanical responses to stimuli using ballistic intracellular nanorheology . _ methods cell biol _ 83:115 - 140 crocker jc , hoffman bd 2007 multiple - particle tracking and two - point microrheology in cells . _ methods cell biol _ 83:141178 . roberts db 1998 _ drosophila : a practical approach _ ( oxford university press , usa ) , 2 edition . sgalen m , _ et al . the fz - dsh planar cell polarity pathway induces oriented cell division via mud / numa in drosophila and zebrafish . _ dev cell _ 19:740752 . oda h , tsukita s 2001 real - time imaging of cell - cell adherens junctions reveals that drosophila mesoderm invagination begins with two phases of apical constriction of cells . _ j cell sci _ 114:493501 . / thevenaz / ovuscule/. kervrann c , boulanger j 2006 optimal spatial adaptation for patch - based image denoising . _ ieee trans image process _ 15:28662878 . boulanger j , _ et al . _ 2010 patch - based nonlocal functional for denoising fluorescence microscopy image sequences . _ ieee trans med imaging _ 29:442454 . lucas b , kanade t 1981 an iterative image registration technique with an application to stereo vision . _ international joint conference on artificial intelligence _ pp 674679 . tomasi c , kanade t 1991 detection and tracking of point features . _ carnegie mellon university technical report _ cmu - cs-91 - 132 . feature tracker . http://www.ces.clemson.edu/~stb / klt/. electronic supplementary material ( esm ) the external dissipation is not accessible to single cell junction ablation experiments and its estimate requires modelling @xcite . our model is analogous to models of viscoelastic relaxation proposed for local severing experiments of acto - myosin bundles in living cells @xcite . we use an eulerian description . since the severed domain area visibly decreases with time , we do not enforce any in - plane incompressibility condition . note also that in - plane incompressibility would fix the velocity profile , which is incompatible with the experimental observation of different profiles ( figure 5 ) . we rather assume for simplicity that the out - of - plane thickness is a passive variable : hence the 3d poisson ratio is @xmath104 , but the in - plane effective 2d poisson ratio vanishes . we test this hypothesis by checking _ a posteriori _ that the model predictions are compatible with experimental observations . within this simplification , the antero - posterior @xmath4 and medio - lateral @xmath0 axes ( figure 1 ) then decouple . finally , we define @xmath8 as the part of stress which has been removed by ablation . we thus consider only one spatial dimension , with the variable @xmath51 denoting position along either @xmath4 or @xmath0 . we model the retracting slab of tissue as a linear viscoelastic kelvin - voigt material ( that is , with a steady response at long time ) with a young modulus @xmath57 and a viscosity @xmath10 . we neglect the line tension around the tissue domain : first , because we observe that the ellipse does not remain round ( this suggests the line tension is weak , without being a proof , since we could also imagine that it has the same anisotropy as the strain ) ; second , because even if the line tension exists , it is small enough that the ellipse contour remains visibly rough , see movies 1 - 3 . two terms contribute explicitly to @xmath8 : an elastic one ( which implicitly includes the external pressure ) and a viscous one : @xmath105 taking into account a fluid external friction , with a friction coefficient @xmath55 of dimension pa.s.m@xmath106 , the force balance equation reads : @xmath107 combining equations 3 , [ eq : stress ] and [ eq : newton ] yields equation 4 . the initial length of the retracting slab is @xmath108 : @xmath109 $ ] . assuming that the initial state respects mechanical equilibrium ( over the relevant time scale of 10 to 100 s ) , the initial strain is uniform , and we have for all @xmath51 : @xmath110 at the boundary of the severed region , we have for all @xmath25 : @xmath111 since the final state is at mechanical equilibrium , we have for all @xmath51 : @xmath112 injecting equation 3 into equation [ eq : stress ] , using conditions given by equations [ eq : ic]-[eq : fin : sigma ] yields by integration over time : @xmath113 where the relaxation time is viscoelastic : @xmath114 . equations 3 , [ eq : ic ] , [ eq : bc : e ] yield an approximate analytical prediction for the initial velocity profile at time @xmath115 , as follows . an expansion of @xmath18 at first order in @xmath116 : @xmath117 defines the initial strain rate as : @xmath118 substituting equations [ eq : e1:expand]-[eq : e1:def ] into equation 4 , we obtain an ordinary differential equation for @xmath119 : @xmath120 with a diffusion coefficient @xmath121 . with the boundary condition @xmath122 , the solution of equation [ eq : e1:eqdiff ] reads : @xmath123 where @xmath124 , or @xmath125 since @xmath126 . spatial integration of equation [ eq : e1:sol ] using equations 3 and [ eq : e1:def ] yields equation 6 . esm figure [ sfig : twolimits ] shows two limit regimes of equation 4 . @xmath127 when internal viscosity dominates : external friction may be neglected , and equation [ eq : newton ] reduces to @xmath128 so that @xmath8 is uniform and its value is fixed at the boundaries : @xmath129 for all @xmath130 . integration yields : @xmath131 the strain is uniform at all times ( esm figure [ sfig : twolimits ] ( a ) ) . since the center of the slab does not move , @xmath132 for all @xmath25 , the velocity profile is linear : @xmath133 as expected , in the limit of small friction @xmath134 , equation 6 reduces to equation [ eq : initv_modelfree ] when @xmath135 . in this regime , all parts of the tissue relax exponentially with the same time scale @xmath7 . @xmath127 when external friction dominates : the internal viscosity of the tissue may be neglected . combining equations 3 , [ eq : stress ] and [ eq : newton ] yields a diffusion equation for the strain field : @xmath136 with conditions given by equations [ eq : ic]-[eq : fin : sigma ] . strain diffuses from the boundaries on a time scale @xmath137 , and the strain field remains inhomogeneous until @xmath138 ( esm figure [ sfig : twolimits ] ( b ) ) . one may note that the strain diffusion equation ( equation [ eq : diff : e ] ) is identical to that obtained in the context of gel dynamics @xcite . taking into account interstitial flow ( _ i.e. _ the transport of water across the epithelium ) would only modify the diffusion constant : @xmath139 , where @xmath140 is the permeability coefficient . in the general case , to solve equation 4 numerically we normalize variables by defining : @xmath141 , @xmath142 and @xmath143 . equation 4 then becomes : @xmath144 with the initial condition : @xmath145 the boundary conditions : @xmath146 and the final conditions : @xmath147 since @xmath148 , the limit regimes are recovered when @xmath149 ( internal viscosity dominates , @xmath150 ) , and @xmath151 ( external friction dominates , @xmath152 ) . 90 mayer m , depken m , bois js , jlicher f , grill sw 2010 anisotropies in cortical tension reveal the physical basis of polarizing cortical flows . _ nature _ 467:617621 . colombelli j , et al . 2009 mechanosensing in actin stress fibers revealed by a close correlation between force and protein localization . _ j cell sci _ 122:16651679 . charras gt , mitchison tj , mahadevan l 2009 animal cell hydraulics . _ j cell sci _ 122:32333241 . doi m 2009 gel dynamics . _ j phys soc jpn _ 78:052001 . ( esm ) set - up for live imaging . ( a ) the pupal case ( light brown ) is partially removed , while the cuticle ( transparent ) is kept intact . blue cross : approximate position of severing , along the symmetry axis of the scutellum . ( b ) the pupa ( white ) is mounted between slide and coverslip with its back facing the objective ( dark grey ) . antero - posterior axis is horizontal ( anterior towards the left , posterior towards the right).,scaledwidth=70.0% ] ( esm ) range of acceptable severing powers . images shown are just before the severing ( top ) and 24.3 s after the severing ( bottom ) as a function of the proportion ( indicated at top left ) of the full laser power , here at 890 nm . below 20% , no severing occurs . from 30 to 50% , severing is partial . from 60 to 80% , severing is complete and cell - cell junctions move significantly ; experiments presented in this work are performed in this range . above 90% , cavitation occurs . these ranges vary slightly from one experiment to the other , so that the quality of severing must be carefully assessed _ a posteriori_.,scaledwidth=70.0% ] [ sfig : tau_anisotropy ] ( esm ) check of @xmath7 and @xmath48 isotropy . ( a ) relaxation time @xmath7 : for each experiment , represented by one point , we plot the value obtained along the @xmath0 direction ( estimated as in the inset of figure 2 ) _ versus _ the value obtained along the @xmath4 direction . age is color - coded as in figure 3 : green , young ; red , middle - aged ; blue , old . ( b ) same plot for the friction / viscosity ratio @xmath48 , fitted by equation 6 as in figure 5 . note the logarithmic scale . two values of @xmath48 of order @xmath49 and @xmath50 are below the plotted range . , scaledwidth=50.0% ] , @xmath153 s , @xmath154 m . ( a ) uniform strain relaxation , when internal viscosity dominates ( low @xmath48 , initial velocity profile similar to the blue curve in figure 5 ) : @xmath155 , @xmath156 . ( b ) strain diffusion , when external friction dominates ( high @xmath48 , initial velocity profile similar to the green curve in figure 5 ) : @xmath157 , @xmath158 . strain value in each grid square is color - coded from 0 ( dark blue ) to 1 ( dark red ) . [ sfig : twolimits],scaledwidth=50.0% ] ( esm ) error maps in the @xmath159 plane , semi - log axes . experiments are those of figure 5 and movies 1 - 3 ; error is estimated according to equation 7 . the minimal region of the error landscape ( its deepest valley ) is too wide to allow a direct measurement of the parameters . however , depending on the orientation of the minimal region of the error landscape , the computed error maps fall broadly into three categories , which agree with the three age groups . there is also a good agreement with hatched regions , which are the results of the independent measurements of the parameters @xmath7 and @xmath48 ( figure 3 ( c ) and esm figure [ sfig : xi_anisotropy ] ) . ( a ) small @xmath48 , old pupa ( movie 3 ) . minimal valley parallel to the @xmath48 axis : this suggests that only an upper bound , of the order of @xmath160 ( white bar ) , can be given for the value of @xmath48 . friction is negligible . ( b ) finite @xmath48 , middle - aged pupa ( movie 2 ) . minimal valley intermediate , parallel to the @xmath7 axis : friction can not be neglected , and @xmath48 admits a finite estimate . ( c ) large @xmath48 , young pupa ( movie 1 ) . minimal valley parallel to the @xmath7 axis , at very large @xmath48 : this suggests that only a lower bound can be given for the values of @xmath7 and @xmath48 . ]

type ia supernovae ( sne ) have recently been used extensively for the measurement of relative distances to galaxies ( , hamuy 1996 ; filippenko & riess 1999 ; branch 1998 ; and riess , press & kirshner 1996 ) . with peak @xmath3-band absolute magnitudes ranging ( typically ) from @xmath4 to @xmath5 , type ia sne have been discovered in ever - increasing numbers at redshifts ( @xmath6 ) well out in the hubble flow the 29 type ia sne of the survey ( hamuy 1996 ) , supplemented by the seven new sne from the center for astrophysics ( cfa ) survey ( riess 1998 ) , are testament to this fact . the dispersion in the raw survey @xmath7-band hubble diagram alone is only @xmath8 , @xmath90.15mag tighter than the @xmath7-band relationship ( sakai 1999 ) . in recent years , it has become clear that type ia sne are not standard candles of constant luminosity . significant dispersion in their peak luminosities , particularly in the @xmath3 and @xmath7 bands , has now been unequivocally demonstrated ( , phillips 1993 ; hamuy 1996 ; riess 1996 , 1998 ; phillips 1999 ) ; fortuitously , though , these same groups emphasize that there is a strong correlation between sne light - curve shape and peak luminosity . this is perhaps best appreciated through figure 8 of phillips ( 1999 ) , in which the correlation between peak luminosity and decline rate @xmath10 is shown . @xmath10 corresponds to the decline in @xmath3-band magnitude from maximum to @xmath11 days after @xmath3 maximum . correcting the observed peak magnitudes for this decline rate effect results in a @xmath7-band hubble diagram dispersion of 0.14mag ( hamuy 1996 ) . while the hubble diagrams , whether they be raw or corrected for decline - rate effects , can provide highly accurate relative distances , the derivation of an actual hubble constant , h@xmath1 , requires that a locally - calibrated zero point be provided ; with the launch of @xmath12 , this zero point can now be provided via independent , cepheid - based , distance determinations for nearby type ia sne - host galaxies . providing such a zero point is the goal of the sandage / saha _ type ia sne hst calibration program _ ( saha 1997 , and references therein ) . to date , six galaxies have been targeted by the sandage / saha program ( ngc 3627 , 4496a , 4536 , 4639 , 5253 , and ic 4182 ) ; each galaxy s associated type ia sn , metallicity , and _ hst _ proposal i d number is listed in table [ tbl : galaxies ] . this list of six is supplemented by two others ngc 3368 ( host to sn 1998bu ) , from tanvir ( 1995 ) and ngc 4414 ( host to sn 1974 g ) , from turner ( 1998 ) . three further local type ia - host galaxies will be observed by sandage / saha during _ hst _ cycles 79 ( ngc 1316 , 4527 , and 3982 ) and , for completeness , are also listed in table [ tbl : galaxies ] . as part of the _ hst key project on the extragalactic distance scale , cepheid - based distance determinations have been made for 18 galaxies , including one local , calibrating , type ia sne - host ( ngc 4414 turner 1998 ) . this ensemble is now being employed as calibration for a variety of secondary distance indicators probing further into the hubble flow , including the relationship ( sakai 1999 ) , surface brightness fluctuations ( ferrarese 1999 ) , fundamental plane ( kelson 1999 ) , and , of course , type ia sne ( this work ) . _ this paper differs somewhat from these other calibrations of secondary methods in that we now focus on the recalibration of extant cepheid data . it presents results from a search for cepheids in all galaxies observed with _ hst that are host to type ia supernovae . the goal of the present analysis is to process the remaining non-_key project , _ hst - observed , type ia sne - host galaxies in a manner consistent with that employed throughout the earlier papers in our series . while parallel double - blind reductions were not employed for the present recalibration work ( as was done for papers i - xxiii ) , the philosophy adopted for the seven galaxies described herein _ is consistent with that used for our earlier _ key project galaxies . this ensures that these galaxies are on the same footing as our original sample of 18 , and provides a useful cross - check of the sandage / saha and tanvir independent reductions . a benefit of our re - analysis is that not only do we have a better understanding of the subtleties of accurately calibrating wf / pc and wfpc2 data , but we are in a position to take advantage of additional archive data for one of the galaxies ( ngc 5253 ) , that was not available to the original authors . _ _ _ _ _ in section [ previous ] , results from previous type ia supernova studies are briefly summarized . in section [ data ] , we briefly remind the reader of this standard _ hst key project _ reduction and calibration pipeline , then discuss the new photometric reductions , cepheid period - luminosity ( pl ) relations , and distances to each cepheid calibrator galaxy , and then compare them to previous results . the cepheid photometry for our reanalysis is provided in an appendix . section [ implications ] is devoted to the implications for h@xmath1 work implied by our uniform re - analysis of the local type ia sne calibrators ; these implications hold for all the empirical studies of table [ tbl : h0_ia ] ; for brevity , only two analyses ( those due to suntzeff 1999 and saha 1997 ) are followed in detail . the role of a possible metallicity correction to the cepheid - derived distances is discussed in section [ metallicity ] , while the local flow field and its impact upon h@xmath1 is described in section [ flow ] . we re - visit the type ia sne - predicted distance to the fornax cluster in section [ fornax ] . finally , our results are summarized in section [ summary ] . using their five cepheid - based distances to local type ia sne - host galaxies , supplemented with an indirectly inferred distance to ngc 3627 , saha ( 1997 ) derived zero points to their uncorrected @xmath3- and @xmath7-band hubble diagrams ( themselves , based equally upon the and asiago ( patat 1995 ) surveys ) . ignoring external systematic effects , their resulting hubble constant was @xmath13 . using only the uncorrected sample , but now including the @xmath14-band data , and ignoring those calibrators with poor photometry or without a direct cepheid - based distance determination , both hamuy ( 1996 ) and riess ( 1996 ) found @xmath15 , in agreement with saha et al . correcting for the light - curve shape - peak luminosity relationship , in both the high-@xmath16 _ and local calibrator sample ) , hamuy ( 1996 ) , riess ( 1996,1998 ) , phillips ( 1999 ) , and suntzeff ( 1999 ) , each found @xmath17 , despite the different approaches in the treatment of foreground and host galaxy extinction , correcting the peak magnitudes for light - curve shape results in an approximate @xmath18% increase in . _ a variety of other type ia sne - related h@xmath1 determinations populate the literature . the subtle differences in each study are tied to the details of either the high-@xmath16 or local calibrator sample selection . for example , a preliminary analysis from our team ( madore 1999 ) led to a value for the hubble constant of h@xmath1=@xmath19 . for that study though , only three of the six type ia sne employed had direct cepheid - calibrated distances ; indirect distances to two fornax cluster sne were inferred via the cepheid distance to cluster member ngc 1365 ( silbermann 1999 ; see also section [ fornax ] ) . in the analysis which follows we have chosen to restrict ourselves only to those eight sne which have _ direct cepheid - calibrated distances ; as madore rightly note , the inclusion of the two fornax sne has negligible impact upon the derived h@xmath1 . _ an alternative method , independent of the local , empirical calibration of the hubble constant , is the derivation of h@xmath1 based upon purely physical models of type ia sne explosions . this method will not be discussed in any detail in what follows , except to note that it is encouraging that the weighted mean of three recent _ theoretical _ type ia sne - h@xmath1 analyses ( hflich & khokhlov 1996 ; ruiz - lapuente 1996 ; iwamoto & nomoto 1998 ) is @xmath20 , in close agreement with the recent empirical values , to be discussed in section [ hamuy ] . table [ tbl : h0_ia ] provides a useful summary of these recent type ia sne - based h@xmath1 determinations , including both physical and empirical methodologies . where available ( and where relevant ) , an attempt has been made to list the number of sne included in the adopted hubble diagrams , as well as the number of local calibrators employed in deriving the associated zero points . the associated mean peak @xmath3-band magnitude for each sample is likewise provided ; the notes to table [ tbl : h0_ia ] clarify whether said mean applies to the high-@xmath16 sample or the local calibrators . seven type ia sne - host galaxies , six from the sandage / saha team ( ngc 4496a , 4536 , 4639 , 5253 , 3627 and ic 4182 ) and one from tanvir ( ngc 3368 ) , were extracted from the _ hst archive and processed with the _ key project s allframe ( photometry stetson 1994 ) and trial ( variable finding stetson 1996 ) pipeline , as documented thoroughly by stetson ( 1998 ) and hill ( 1998 ) . for papers i - xxiii of the _ key project , we carried out a ( real - time ) parallel reduction , based on both dophot and allframe , using at least two different methods for finding the cepheids . we included in the final analysis only those cepheids which were in common to both analyses , but for consistency , generally quoted only the allframe results . in the present case , the dophot results were already previously published by the sandage / saha team , we did not undertake a parallel dophot reduction and , thus , the methodology employed is not _ identical to that used in papers i - xxiii of this series . _ _ _ _ for the allframe@xmath21trial analysis described here , photometry , cepheid selection , and pl analyses were handled identical to each of our previous galaxies , no special treatment was afforded any of the cepheid samples herein , and only those cepheids with good quality @xmath7- _ and @xmath14-band data were considered . as described in sections [ n4639]-[n4496a ] , for several of the galaxies analyzed by saha et al . , these authors either ( a ) used a special pl - fitting scheme ( , ngc 4639 and 3627 ) , or ( b ) treated the @xmath14-band data specially in the derivation of the true modulus ( , ngc 4496a and ic 4182 ) . for those galaxies with peculiarities not encountered in earlier _ key project papers , details are provided in the appropriate subsections below . we discuss the reasons for the differences between previously published results and those in this paper , so that the reasons for the difference in the overall value of h@xmath22 derived in this paper are clear . as it turns out , there is not a single explanation for the differences we see compared to our standard pipeline processing procedure , but a variety of reasons . the full set of cepheids for each of the seven galaxies , is listed in appendix a , along with positional and photometric information . as alluded to previously , we have only retained those cepheids which have high quality @xmath7- _ and @xmath14-band photometry ; we made a conscious decision to not retain the lower quality candidates . because of this , the cepheid numbers in tables a1-a7 are typically smaller than the number listed in the corresponding saha dophot analysis , since saha provided a complete listing of both bona fide _ and lower - quality suspected cepheid candidates . epoch - by - epoch photometry , light - curves , and finding charts , for all the cepheids discovered in this work are available through the _ hst key project s web page . _ _ _ _ _ the _ key project s data were calibrated without the adoption of an explicit , analytic , correction for charge transfer effects ( cte ) , although the hill ( 1998 ) calibration employed did correct for it in the mean . on the other hand , the current data _ were analyzed with a developmental calibration which does treat cte ( stetson 1998 ) , although our quoted moduli , and the subsequent h@xmath1 analysis which follows , adopt the mould ( 1999b ) transformation from the stetson ( 1998 ; s98 ) scale to that of hill ( 1998 ; h98 ) ; this amounts to mean shifts of @xmath210.02mag ( @xmath7 ) and @xmath210.04mag ( @xmath14 ) , in the sense of h98 being greater than s98 . this corresponds to a @xmath210.07mag shift ( s98@xmath23h98 ) in the true modulus , and retains consistency with previous ( and current ) papers in this series . the two systems do agree at the @xmath91@xmath24 level , but unresolved discrepancies between the two have led us to conservatively adopt the hill ( 1998 ) scale , in what follows . _ _ ngc 4639 , host to sn 1990n , is one of three local calibrators situated in the virgo cluster . of the 38 galaxies ( to date ) with cepheid - based distances ( ferrarese 1999 ; table 3 ) , ngc 4639 is the current record - holder ; saha ( 1997 ) derive a true modulus of @xmath25 , based upon 17 cepheid candidates . our allframe@xmath21trial analysis uncovered 17 high - quality cepheid candidates , seven of which were in common with the saha ( 1997 ; table 4 ) sample . the majority of our sample ( , 14/17 ) reside on chips 2 and 3 . positional and photometric information for the full sample is given in table [ tbl : n4639_ceph ] . the first entry of table [ tbl : ceph_comp ] shows that the agreement in assigned mean @xmath7-band magnitudes , for the seven cepheids in common between table [ tbl : n4639_ceph ] and saha ( table 4 ) , is excellent . however , as is the case for many of the galaxies discussed in this paper , the agreement in the @xmath14-band is somewhat inferior ( although the agreement does hold at the @xmath26 level ) . our @xmath7- and @xmath14-band pl relations are shown in figure [ fig : n4639 ] . standard pl - fitting to the full sample results in @xmath27 , @xmath28 , e@xmath29 , and @xmath30 ; the quoted uncertainties simply reflect the rms scatter of the fits , which includes the effects of the finite width of the instability strip as well as of random observational noise . a full accounting of the error budget ( detailed in table [ tbl : error ] ) leads to our final adopted value of @xmath31 . in comparison , recall that saha ( 1997 ) found @xmath25 , our result places ngc 4639 0.2mag nearer than that favored by saha et al ( although still in agreement at the @xmath32 level ) . saha ( 1997 ) interpret their dophot pl relations as showing evidence for significant scatter beyond that encountered in previous galaxies ( indicative of a substantial component of differential reddening ) , and thereafter employ a weighting scheme in averaging the sample of de - reddened cepheid moduli , in order to derive the galaxy s true modulus . we would suggest that upon closer inspection the evidence presented therein for differential reddening is perhaps not as suggestive as might at first seem . first , of the 15 cepheids employed by saha ( neglecting the same three which they discard in deriving the true modulus , but which are still plotted in their apparent @xmath7- and @xmath14-band moduli pl relations - , their c1-v3 , c3-v8 , and c3-v9 ) , only two lie outside the lmc @xmath7-band @xmath33 instability strip width ( @xmath34mag ) . our analysis ( upper panel of figure [ fig : n4639 ] ) found one candidate lying outside this region . formally , the random uncertainties associated with the dereddened pl fit to the respective samples ( , entry r3 of table [ tbl : error ] ) are @xmath35mag ( allframe ) and @xmath36mag ( dophot ) , so while the scatter in the saha dereddened moduli distribution is marginally larger than that found in our allframe@xmath21trial analysis , in neither case is the scatter anomalously large in comparison with the entire set of _ hst - observed galaxies . a simple visual inspection of our allframe pl relations ( and the listed @xmath7- and @xmath14-band scatter in each panel ) for ngc 4639 ( figure [ fig : n4639 ] ) , in comparison with those for the other galaxies in this study ( figures [ fig : n4536]-[fig : n4496a ] ) , do not show a significantly larger scatter . one can also do a formal fit to the distribution of cepheids in the differential reddening diagnostic plane ( , @xmath37 versus @xmath38 - figure 7 of saha 1997 ) for both the allframe@xmath21trial and dophot datasets , the formal slope of the least squares fit is @xmath39 . a slope of @xmath40 is expected in the case of photometric scatter alone , while a slope of @xmath41 is expected in the case where differential reddening is substantial - , statistically , we feel there is no compelling reason to invoke any substantial differential reddening component , and for these reasons , we have treated ngc 4639 in the same way as all of the other galaxies we have analyzed , taking the unweighted mean as the representative final galaxy modulus . the differing weighting schemes employed account for half of the discrepancy between the two analyses , while the remainder is due to the aforementioned ( unresolved ) @xmath14-band disagreement . _ ngc 4536 was host to sn 1981b , for which the light curve and photometry are of the highest quality . saha ( 1996a ) derived moduli of @xmath42 , @xmath43 , and @xmath44 , and a total line - of - sight reddening of e@xmath45 , based upon the 27 highest - quality cepheids ( , @xmath46 ) in their table 3 . our allframe@xmath21trial analysis resulted in the sample of 39 candidates listed in table [ tbl : n4536_ceph ] . the @xmath7- and @xmath14-band pl relations are shown in figure [ fig : n4536 ] ; because of incompleteness below periods of 20 days , the formal pl fits were restricted to the 27 cepheids for which @xmath47d , and are represented by the solid lines ( , @xmath48 and @xmath49 ) . our final true modulus is @xmath50 . our predicted true distance modulus appears 0.15mag discrepant with that of saha ( 1996a ) ( @xmath44 ) . examining the saha data on a chip - by - chip basis , demonstrates though that this discrepancy results ( partially ) from what appears to be a discrepant chip within their dataset . if we apply the same color ( @xmath51 ) , period ( p@xmath5220d ) , and quality - index ( qi@xmath53 ) cuts adopted by saha ( 1996a ) , and then perform the standard pl fitting to ( a ) the sum of chips @xmath54 ( 17 cepheids ) , and ( b ) chip 2 alone ( 12 cepheids ) , we find that ( a ) @xmath55(random ) and ( b ) @xmath56(random ) . the upper panel of figure [ fig : n4536_mod_comp ] illustrates this 0.66mag chip 2 discrepancy present within the saha dataset , where the solid symbols represent the 17 chip 1 + 3 + 4 cepheids and the open symbols the 12 chip 2 cepheids ; the solid and dotted horizontal lines correspond to the unweighted mean of each respective sample . the lower panel of figure [ fig : n4536_mod_comp ] shows that such a discrepancy does not exist in our allframe@xmath21trial dataset . because the saha chips 1 , 3 , and 4 , are self - consistent , we feel that the true modulus derived from the mean of the cepheids therein @xmath57 represents a better option than the originally published @xmath44 , as reflected by its entry under @xmath58 in table [ tbl : distances ] . the fact that our allframe@xmath21trial distance to ngc 4536 is 0.19mag greater than that derived from chips 1 + 3 + 4 of the saha sample is consistent with that expected , based upon a comparison of mean magnitudes for the cepheids in common in the two datasets ( see relevant entry in table [ tbl : ceph_comp ] ) . host to the highly - reddened , but well - photometered , sn 1989b ( wells 1994 ) , ngc 3627 is the most recently observed galaxy by the sandage / saha type ia sne calibration team . the wfpc2 field is replete with dust lanes , and suffers from an extremely bright background . based upon a sub - sample of 25 cepheids , drawn from a master list of 68 , saha ( 1999 ) derive an true modulus of @xmath59 . the allframe@xmath21trial analysis for ngc 3627 proved to be more challenging than that encountered for any previous galaxy . the problem stems , in part , from a distinct lack of bright local standards upon which to base the calibration . there are many bright objects in the field , but they are decidedly non - stellar , forcing the use of those comparatively faint calibrators which successfully avoid the dust lanes . in addition , even after eliminating all likely star clusters and background galaxies , the remaining stellar images are still @xmath950% broader than in the rest of the _ hst - observed galaxies . in practice , what this means , is that the median aperture correction ( in going from 2 to 5 pixel radius ) , for ngc 3627 , is @xmath600.362mag , while the median for all the other galaxies is @xmath600.192mag , there is a 0.17mag difference in the aperture corrections for ngc 3627 . _ the source of this discrepancy is not obvious . perhaps all of the ngc 3627 observations were out of focus ; however , contemporaneous _ key project wfpc2 ngc 3319 observations do not deviate from the median aperture correction noted previously . second , perhaps ngc 3627 is extremely rich in star clusters which are incorrectly being identified as stars . a third possibility is that the extremely high surface brightness of ngc 3627 is problematic ; however , the form of the stellar images in the low surface brightness regions of our frames do not appear appreciably different from those in the high surface brightness parts of the same images . _ in what follows , we have chosen the conservative approach , and adopted the second option - , using the full sample s median aperture corrections and making use of all of the data available to us . instead of the standard 0.05mag random photometry uncertainty used for other galaxies in this study ( , r1 in table [ tbl : error ] ) , for ngc 3627 we have assumed the more generous value of 0.15mag , to reflect the greater uncertainty associated with the adopted aperture corrections . our results should be reliable at this level . in total , 36 high - quality cepheid candidates were discovered , the properties for which are listed in table [ tbl : n3627_ceph ] . eighteen of our cepheids are in common with the saha ( 1999 ; table 4 ) sample , and as can be seen by the relevant entry to table [ tbl : ceph_comp ] , the agreement in the mean @xmath7-band magnitudes is excellent , although the @xmath14-band agreement is somewhat inferior ( agreement at the 2@xmath24 level only ) . based upon this common sample , we _ a priori expect that our allframe true modulus should be 0.17mag _ greater than that of saha ( 1999 ) . _ _ the pl relations for ngc 3627 are shown in figure [ fig : n3627 ] , and possibly provide evidence for the presence of differential reddening , by the excess scatter beyond that expected from the finite width of the lmc instability strip . larger errors as a result of the difficulty in working against the high background may also contribute to the higher scatter . incompleteness , for p@xmath6125d , is apparent , so our preferred allframe@xmath21trial distance has been based upon the 17 cepheids with periods in excess of 25d . standard pl fitting , incorporating the full error budget , yields @xmath62 . had we included cepheids with periods less than 25days , our derived distance to ngc 3627 would decrease . in contrast , saha ( 1999 ) derive the significantly larger modulus of @xmath59 . this 0.16mag _ greater modulus is at odds with that which we had predicted on the basis of the 18 cepheids in common ( recall , we expected the dophot modulus to be 0.17mag _ smaller than our allframe modulus ) . the source of the discrepancy appears to be due to the sample of cepheids used by both groups which are _ not part of the aforementioned `` common '' subset . restricting ourselves to only those cepheids used by each group in their respective pl fitting ( , the 25 used by saha and the 17 used in our re - analysis ) , we show in figure [ fig : n3627_mod_comp ] the distribution of individually dereddened true moduli for each sample . _ _ _ the upper panel of figure [ fig : n3627_mod_comp ] shows the 25 saha ( 1999 ) cepheids , while the lower panel shows our 17 ; in both panels , the filled symbols represent the five cepheids which occur in both samples . the horizontal lines represent the unweighted mean of the dereddened moduli , hence the @xmath63 shown in the upper panel , which differs from the aforementioned saha value of @xmath64 ( which represents their favored weighted mean ) . several points should be made regarding figure [ fig : n3627_mod_comp ] first , the mean magnitudes of the ( small ) overlap sample are in good agreement , as was found for the larger sample described earlier . indeed , just using these five cepheids in common leads to @xmath65(random ) , _ regardless of whose photometry is adopted . of the 0.23mag discrepancy noted in figure [ fig : n3627_mod_comp ] , 0.14mag is being driven by the four saha cepheids with @xmath66 ; removal of these four outliers leads to @xmath67(random ) , in better agreement with our favored @xmath68(random ) . further support for their excision comes from the fact that their inferred reddenings are all negative , as witnessed by their populating the lower - right quadrant of saha et al.s figure 11 . the residual 0.09mag offset can be traced primarily to those @xmath69d cepheids in the saha sample which were not recovered in our analysis ; those saha cepheids _ not included in our analysis ( , the 20 open squares in the upper panel of figure [ fig : n3627_mod_comp ] ) , appear to lie , in the mean , 0.14mag beyond their five which overlap our sample . conversely , again in the mean , our cepheids _ not included in the saha analysis ( , the 12 open circles in the lower panel of figure [ fig : n3627_mod_comp ] ) _ are self - consistent with the overlap sample . _ _ _ _ a clearer way in which to express the above is to simply note that the median _ and mean @xmath70 of our sample of 17 cepheids are equivalent ( @xmath71 versus 30.06 , respectively ) , while for the saha ( 1999 ) sample of 25 cepheids , the median and mean differ by 0.19mag ( @xmath72 versus 30.29 , respectively ) . we would argue , in this case , that the saha _ median @xmath70 is a more robust measure of the true dophot distance to ngc 3627 , in agreement with that favored by our allframe@xmath21trial analysis . _ _ ngc 3368 was not part of the original sandage / saha sne ia program , simply because at the time of proposing , the galaxy had not played host to a suitable sn ; this fortuitously changed on 1998 may 9 , with the eruption of sn 1998bu . while the sn itself is highly extinguished , its light - curve and photometry are of the highest quality , and thus ideally suited for the h@xmath1 analysis which follows . tanvir ( 1995 ) , as part of their leo i group study , had earlier derived a cepheid - based distance to ngc 3368 of @xmath73 ( after addition of the 0.05@xmath74mag `` long - vs - short exposure '' correction - casertano & mutchler 1998 ) . unlike the case for the _ key project sample , the roll angle was not fixed from epoch - to - epoch for ngc 3368 . hence , a master coordinate system was defined relative to the first ( chronologically - speaking ) epoch chip 2 image , and the allframe reduction performed on all four wfpc2 chips simultaneously . _ eleven candidates were identified , their location in the resulting pl - plane denoted by the filled circles in figure [ fig : n3368 ] . an incompleteness bias , for periods less than @xmath75 days , not apparent in the published data of tanvir ( 1995 ; figure 3 ) , is clearly present . because tanvir did not include positional information for their seven cepheid candidates , we identified their subset , within our larger dataset , based upon their proximity to one another in the pl - plane . their sample corresponds to cepheids c03 , c05 , c06 , c07 , c08 , c10 , and c11 , of table [ tbl : n3368_ceph ] . for these seven cepheids in common between the two studies , our allframe mean @xmath7- and @xmath14-band magnitudes are @xmath76 and @xmath77mag brighter , respectively , than those of tanvir et al ( see table [ tbl : ceph_comp ] ) . the substantial differences in these mean magnitudes remains unresolved at present , but can be revisited once tanvir publish the full photometry for their cepheids and local standards . we note in passing that identifying such standards for the allframe analysis proved challenging . bearing in mind the aforementioned incompleteness bias below p=20d , we adopt a final modulus based upon the 7 candidates in table [ tbl : n3368_ceph ] with periods in excess of this cutoff , @xmath78 . ngc 5253 represents one of the two wf / pc galaxies ( supplemented with some newer wfpc2 data , in the case of ngc 5253 ) analyzed for this work ; the second , ic 4182 will be discussed in section [ i4182 ] . the reduction of the earlier wf / pc photometry proved challenging . the original dophot analysis led to @xmath79 and e@xmath80 ( saha 1995 ) . the large uncertainty on the latter allows for consistency with the measured galactic foreground reddening e@xmath81 ( schlegel 1998 ) . ngc 5253 is unique in that it played host to two potential type ia sne local calibrators , sn 1895b and 1972e . the quality of the photometry associated with sn 1895b is of low quality , and therefore we do not consider it further . the lightcurve and photometry for sn 1972e ( hamuy 1996 ) are of a decidedly higher quality ( although not of the same level as those discussed in sections [ n4639 ] to [ n3368 ] , primarily because observations did not begin until 5d past the peak , and so one must extrapolate back to the peak using template lightcurves ) and will be incorporated into the h@xmath1 analysis of section [ implications ] . sn 1972e is situated well in the outskirts of ngc 5253 , @xmath82kpc from the optical center see figure 1 of saha ( 1995 ) . the intrinsic reddening local to sn 1972e is measured by suntzeff ( 1999 ) to be e@xmath83 ; if this is taken as a lower limit on the mean intrinsic reddening , then in combination with the aforementioned galactic foreground reddening , the implied total line - of - sight reddening is e@xmath84 . because both wf / pc and wfpc2 data exist for ngc 5253 , slightly different reduction and calibration steps were employed . first , for the wfpc2 data , the stetson ( 1998 ) @xmath14-band calibration and cte corrections , based upon @xmath85 cen and ngc 2419 , were used . second , the @xmath7-band images were not taken with the usual f555w filter , but with f547 m ; the f547 m calibration is derived from seven observations of @xmath85 cen taken with the gain=14 electronics , at odds with the gain=7 electronics employed in the ngc 5253 observations . the @xmath85 cen zeropoints were therefore corrected with a ( gain=7)@xmath60(gain=14 ) zeropoint correction derived from f555w data on @xmath85 cen ; a separate correction was applied to each chip . the separately calibrated wf / pc and wfpc2 data for ngc 5253 differ by @xmath86mag in v , in the sense of wfpc2 being brighter , and @xmath87mag in i. the weighted average was then constructed and compared with the ground - based photometry from john tonry ( as reported in table 2 of saha 1995 ) . this average is heavily weighted in favor of the wf / pc observations , simply because of the far greater number of observations . we note also that object 3a in saha et al.s ( 1995 ) table 2 turned out to be a clump of @xmath88 stars of comparable brightness in the higher - quality wfpc2 data , and was eliminated from the comparison with tonry s ground - based data . the @xmath7-band magnitudes differed by @xmath89 , in the sense of the ground - based photometry being brighter , and the @xmath14-band magnitudes differed by @xmath90 ( in the same sense ) . in comparison , saha ( 1995 ) applied a 0.14mag f555w zeropoint contamination correction , and found agreement with tonry s ground - based photometry at the level of @xmath91 ( @xmath7-band , in the sense of tonry being fainter ) and @xmath92 ( @xmath14-band , in the sense of tonry being brighter ) . because our re - analysis is anchored to the higher - quality wfpc2 data , an _ a posteriori contamination correction is not needed . in summary , based upon the tonry ground - based photometry , the saha ( 1995 ) reference star photometry , after application of their contamination correction , is @xmath93mag ( v ) and @xmath94mag ( i ) brighter than our allframe photometry , which would correspond to an _ a priori predicted allframe true modulus 0.04mag nearer than that of saha ( 1995 ) . _ _ seven cepheid candidates were identified in our allframe analysis ( see table [ tbl : n5253_ceph ] for the relevant properties ) , five of which were in common with those of saha ( 1995 ) . the remaining cepheid candidates listed in table 4 of saha ( 1995 ) were examined , but were either not recovered or ( more usually ) the light curve quality was not very high or the cepheids were located too close to the edge of the chip . the pl fit , based upon the full allframe sample of seven cepheids , is shown in figure [ fig : n5253 ] . the inferred apparent moduli @xmath95 and @xmath96 are significantly smaller ( , 0.15mag and 0.27mag in @xmath7 and @xmath14 , respectively ) than those favored by saha et al , and at odds with the conclusions of the previous paragraph , based upon the bright calibrator stars in the field . we discuss the reasons for this difference below . there are five cepheids in common between saha ( 1995 ) and table [ tbl : n5253_ceph ] from our census , c02-c06 correspond to saha et al.s c4-v2 , c3-v2 , c4-v4 c3-v3 , and c1-v2 , respectively . table [ tbl : ceph_comp ] shows that the mean magnitudes for these five cepheids agree extremely well . identical moduli ( both apparent and true ) and reddenings to those favored by our allframe analysis are found when using the saha et al . dophot data , provided the same cepheid sample is used . the reason for the difference with saha et al.s results is that of the five cepheids included in their @xmath14-band analysis their c2-v3 , c3-v6 , c4-v3 , c3-v2 , and c4-v2 only the latter two were included in our final sample . our photometry has benefited from the wfpc2 supplement not available in 1994 . in some cases , this has allowed us to better eliminate / include marginal cepheid candidates from the original saha analysis . for example , c1-v2 and c3-v3 were discounted in the @xmath14-band analysis by saha because of difficulty in assigning their mean magnitudes , while we found no reason to eliminate them based on their photometry ; both are clearly consistent with the pl fit of figure [ fig : n5253 ] , unlike the situation which faced saha ( their figure 5 ; lower panel ) . on the other hand , ( i ) we did not recover their c2-v3 ; ( ii ) c4-v3 sits on a bad column ; and ( iii ) the lightcurve of c3-v6 is clearly non - cepheid - like , even in their own data ( their figure 4 ) . we have treated the seven cepheids in our final sample in a manner consistent with all of the other galaxies in the _ key project , and see no reason to reject any of them , based upon our allframe@xmath21trial analysis . _ we note also that our inferred mean reddening ( e@xmath97 ) , is larger than that found by saha ( 1995 ) ( e@xmath80 ) . our higher value of reddening is in better agreement with other estimates of the reddening ( recall , the galactic foreground reddening is e@xmath98 schlegel 1998 ) . in addition , over the inner @xmath99 ( half of our cepheids lie within @xmath100 of this region ) , the total line - of - sight reddening ( from @xmath101h@xmath102 calzetti 1997 ) is e@xmath103 . in summary , our adopted distance modulus for ngc 5253 , incorporating the full error budget , is @xmath104 . the main difference from that of the published saha ( 1995 ) analysis ( , @xmath79 ) is due to the difference in adopted cepheid samples . the small scatter in our allframe pl relation and individual light curves , combined with the larger inferred reddening value , plus larger time baseline of the data analyzed here ( in a manner consistent with all previous _ key project galaxies ) , all give us confidence in the cepheid sample adopted in this paper . _ an independent check of our ( and saha et al.s ) conclusions is provided by the sakai ( 1999 ) distance determination based upon the magnitude of the tip of the red giant branch ( trgb ) ; a preliminary analysis , using only ngc 5253 halo stars , places the tip at @xmath14=23.99 . for a tip absolute magnitude m@xmath105 ( ferrarese 1999 ) , and a galactic foreground extinction a@xmath106mag ( schlegel 1998 ) , the inferred trgb distance to ngc 5253 would be @xmath107 . this value for @xmath108 lies between saha ( @xmath109 ) and our current result ( @xmath110 ) , in ( dis)agreement with both of the cepheid - based distance determinations at the @xmath111 level . , a distance modulus of @xmath112 can be derived for sn 1972e which , like the sakai ( 1999 ) t - rgb distance to the host galaxy , lies half - way between our new cepheid distance ( @xmath110 ) and the original saha ( 1995 ) result ( @xmath109 ) . ] an ultimate resolution to the discrepancy between each of these approaches to the distance of ngc 5253 is still forthcoming . we should stress that the larger random uncertainty associated with our allframe@xmath21trial distance ensures that under the weighting scheme employed in section [ implications ] , ngc 5253 , by itself , does not overly influence our final result . in passing , we note that the formal saha solution of @xmath109 was not adopted in the subsequent sandage ( 1996 ) and saha ( 1997 ) h@xmath1 analyses ; instead , a true modulus of @xmath113 appears to have been adopted . ic 4182 is the host galaxy for the sne ia calibrator sn 1937c . while the photographic calibration of the sn 1937c lightcurve is of lower quality ( in comparison with the sne of sections [ n4639 ] to [ n3368 ] ) , the quality of the lightcurve itself is high , and for that reason is included in the h@xmath1 analysis of section [ implications ] . the original dophot reduction for this galaxy uncovered 27 high - quality cepheids ( saha 1994 ) ; the published moduli were @xmath114 and @xmath115 . the inferred reddening e()=@xmath116 is clearly unphysical , which led saha to assume @xmath117 . for the allframe analysis , the calibration adopted parallels that of section [ n5253 ] and stetson ( 1998 ) , although in this case there were no supplementary wfpc2 data . twenty - eight cepheid candidates were identified ( properties as noted in table [ tbl : i4182_ceph ] ) , 27 of which were in common with those in table 3 of saha ( 1994 ) . as can be seen from inspection of table [ tbl : ceph_comp ] , there is a difference in the photometry for these cepheids in common the stetson ( 1998 ) allframe mean magnitudes are 0.038mag ( @xmath7 ) and 0.119mag ( @xmath14 ) brighter than the saha dophot magnitudes . the pl relations , and associated fits , for the 28 cepheids of table [ tbl : i4182_ceph ] are shown in figure [ fig : i4182 ] . our adopted allframe@xmath21trial true modulus , with the full error accounting , of @xmath118 ( table [ tbl : distances ] ) is identical to the saha ( 1994 ) result of @xmath119 , although this excellent agreement in true modulus is somewhat misleading . specifically , our re - analysis led to a formal value of the reddening of e()=@xmath120 , while saha found e()=@xmath121 . to avoid any bias , our quoted true modulus includes the correction for the small , but negative , reddening implied by our cepheid data ; saha applied no such correction . we note in passing that the galactic foreground reddening along the line - of - sight to ic 4182 is e()=@xmath122 ( schlegel 1998 ) , consistent with our cepheid - derived mean reddening . preliminary reduction of ground - based photometry of ic 4182 , provided by m. pierce , confirms the validity of our adopted zeropoints at the level of @xmath123mag , within the accuracy of the ground - based data . further confirmation of the ground - based tie - in is underway . in the case of sn 1960f , the light - curve shape and photographic calibration are of low quality ( see schaefer 1996 ) . indeed , suntzeff ( 1999 ) and earlier papers in their series , do not include this sn in their h@xmath1 analysis because of its questionable quality . for completeness , however , we have re - analyzed the data for its host galaxy , ngc 4496a , and provide solutions for h@xmath1 with and without its inclusion in the hubble diagram zero points , in section [ implications ] . based upon 44 cepheids with p@xmath5218d , color @xmath124 , and quality index q@xmath53 , saha ( 1996b ) found @xmath125 and e()=0.04@xmath1260.06 . our allframe@xmath21trial re - analysis uncovered 94 cepheids ( table [ tbl : n4496a_ceph ] ) , the pl relations for which are shown in figure [ fig : n4496a ] . restricting ourselves to the 51 cepheids with p@xmath5225d , to minimize the effects of incompleteness bias ( apparent in figure [ fig : n4496a ] ) , our distance modulus for ngc 4496a is @xmath127 . our modulus is consistent with the published saha ( 1996b ) value . in their h@xmath1 analyses , sandage ( 1996 ) and saha ( 1997 ) do not adopt their published modulus ( , @xmath125 ) , and instead adopt the apparent @xmath7-band modulus ( , @xmath128 ) in its place . for consistency with standard practice in the _ key project , we employ the true modulus @xmath70 , and not the apparent modulus @xmath38 , in section [ implications ] . _ table [ tbl : distances ] summarizes the analysis and discussion of sections [ n4639]-[n4496a ] . column ( 2 ) lists the new allframe@xmath21trial distances derived in the current study , while those values originally published in the tanvir ( 1995 ) and saha ( 1994,1995,1996ab,1997,1999 ) papers are provided in column ( 4 ) . the unweighted mean offset between the two datasets is @xmath129mag , while the weighted mean is @xmath130mag , a discrepancy at the @xmath92@xmath24 level . neglecting ic 4182 , since the saha ( 1994 ) uncertainty of @xmath131mag appears to be too small ,- band problems which led saha ( 1994 ) to infer a highly negative reddening for the ic 4182 cepheids , their quoted true modulus uncertainty ( @xmath131mag ) is significantly smaller than that associated with any of their other six galaxies ( see column ( 4 ) of table [ tbl : distances ] ) , and is also smaller than the uncertainty associated with the true modulus of the lmc , upon which the relative distance is based . ] yields @xmath132mag and @xmath133mag . further , neglecting ngc 5253 , due to its extreme outlier status , results in @xmath134mag and @xmath135mag , similar to what was found using the full sample of seven galaxies . using the result based upon this full sample , @xmath91/2 of the 0.12mag discrepancy can be traced to unresolved photometric calibration issues ( see table [ tbl : ceph_comp ] ) , with the remainder ascribed to the different cepheid samples and pl - fitting methodology employed . again , as there exist serious concerns regarding either the dophot cepheid sample or photometry , for each of the seven galaxies analyzed , applying an ad hoc _ a posteriori `` averaging '' of the dophot and allframe distance moduli is _ not recommended . consistent with the general _ key project practice of adopting the allframe results , we do so here as well . _ _ _ column ( 3 ) of table [ tbl : distances ] presents the true moduli favored by the standard _ key project pl - fitting methodology employed in the current study , but using the tabulated saha and tanvir photometry , while column ( 5 ) shows the true moduli used by sandage ( 1996 ) and saha ( 1997 ) in their analyses ( c.f . column 4 ) . table [ tbl : supernovae ] lists the adopted photometry and sne light curve characteristics for each of these nearby calibrators ; the peak luminosities of the sne ( columns 7 - 9 ) were calculated using the @xmath136 values of table [ tbl : distances ] . how the adoption of these subsets of true moduli ( table [ tbl : distances ] ) and sne photometry ( table [ tbl : supernovae ] ) impacts upon the derivation of h@xmath1 is explored in section [ implications ] . _ given the distance moduli derived from our allframe@xmath21trial re - analysis of the local type ia sne - host galaxies of table [ tbl : distances ] ( column 2 ) , we are now in a position to calibrate the distant sample of sne comprising the clan - tololo - based hubble diagrams . two approaches are considered here one based upon the adoption of an absolute magnitude - peak luminosity correlation correction ( section [ hamuy ] , paralleling suntzeff 1999 ) , and one with no correction ( section [ sandage ] , paralleling saha 1997 ) . this is not meant to be an exhaustive replication of all previous type ia sne - based h@xmath1 determinations , nor is one warranted ; for convenience , we have chosen the two extremes in the treatment of sne lightcurves . while a variety of different approaches have been taken ( , phillips 1993 ; tammann & sandage 1995 ; hamuy 1996 ; saha 1997 ; riess 1998 ; phillips 1999 ; jha 1999 ) , here we are simply interested in exploring the effects of our new distance determinations upon the zero points of the two extrema . because all empirically - based studies listed in table [ tbl : h0_ia ] rest squarely upon the zero point defined by the same local calibrators , the differential effect is essentially identical , regardless of the details in the treatment of the distant sample . we employ the reddening - corrected m@xmath137 hubble relations described by suntzeff ( 1999 ) and phillips ( 1999 ) . following suntzeff et al . , we adopt the zero points , coefficient uncertainties , and hubble diagram fit dispersions derived from the full sample of 35 clan - tololo / cfa type ia sne ; coefficient uncertainties and dispersions were kindly provided by mark phillips . the fits to the decline rate versus peak luminosity relation from phillips ( 1999 ) were derived from the low - extinction ( , e()@xmath610.05 ) subset of 18 type ia sne in the clan - tololo / cfa sample . while the suntzeff ( 1999 ) fits included corrections for both galactic and host galaxy reddening , the phillips fits include only foreground galactic reddening , although it should be stressed that the zero points for these two sets agree to @xmath138 . our adopted relations are represented by the following expressions for h@xmath1 in @xmath3 , @xmath7 , and @xmath14 . @xmath139- \nonumber \\ & & 1.010(\pm0.934)\big[\delta m_{15}(b)_t-1.1\big]^2 + 28.653(\pm0.042)\big\ } , \label{eq : hamuy_b } \\ \log{\rm h}_\circ(v ) & = & 0.2\big\{{\rm m^{max}_v}-0.672(\pm0.396)\big[\delta m_{15}(b)_t-1.1\big]- \nonumber \\ & & 0.633(\pm0.742)\big[\delta m_{15}(b)_t-1.1\big]^2 + 28.590(\pm0.037)\big\ } , \label{eq : hamuy_v } \\ \log{\rm h}_\circ(i ) & = & 0.2\big\{{\rm m^{max}_i}-0.853(\pm0.214)\big[\delta m_{15}(b)_t-1.1\big]+28.219(\pm0.034)\big\ } , \label{eq : hamuy_i}\end{aligned}\ ] ] where @xmath140 e@xmath141 represents the sum of e@xmath142 ( , foreground galactic reddening ) and e@xmath143 ( , intrinsic sn reddening ) from table 5 of suntzeff ( 1999 ) . for sn 1989b , we assign e@xmath144 ( wells 1994 ) , for sn 1960f , e@xmath145 ( schaefer 1996 ) , and for sn 1974 g , e@xmath146 ( schaefer 1998 ) . the dispersions for the fits of equations [ eq : hamuy_b]-[eq : hamuy_i ] are @xmath147 , @xmath148 , and @xmath149 ( phillips 1999 , priv comm ) . suntzeff ( 1999 ) provide a zeropoint to equations [ eq : hamuy_b]-[eq : hamuy_i ] using the following absolute distances to the host galaxies of five type ia sne ( 1 ) @xmath25 for sn 1990n in ngc 4639 ; ( 2 ) @xmath44 for sn 1981b in ngc 4536 ; ( 3 ) @xmath150 for sn 1998bu in ngc 3368 ; ( 4 ) @xmath151 for sn 1937c in ic 4182 ; and ( 5 ) @xmath152 for sn 1972e in ngc 5253 . the adopted moduli were then paired with the corresponding sn s apparent peak magnitude ( columns 3 - 5 of table 4 ) and intrinsic reddening ( sum of columns 10 and 11 of table 5 in suntzeff 1999 ) , in order to derive m@xmath153 . these peak luminosities and associated decline rates @xmath154 were then used in conjunction with equations [ eq : hamuy_b]-[eq : hamuy_i ] to derive a color- and sn - dependent h@xmath1 . the weighted average of the resulting set of hubble constants was h@xmath155kms@xmath156mpc@xmath156 . before demonstrating how our new distances to the galaxies in question ( column 2 of table [ tbl : distances ] ) impact upon the suntzeff ( 1999 ) value of h@xmath155 , we note the following : first , suntzeff use a ratio of total - to - selective absorption of r@xmath157 , while all distances quoted by the _ key project , including those of table [ tbl : distances ] , assume r@xmath158 . in the analysis which follows , whenever the suntzeff ( 1999 ) hubble relations ( , equations [ eq : hamuy_b]-[eq : red ] ) are employed ( sections [ hamuy ] , [ metallicity ] , and [ flow ] ) , we adjust the true moduli ( and associated cepheid reddenings ) of table [ tbl : distances ] , to ensure consistency with suntzeff et al.s use of r@xmath157 . we stress that the effect upon h@xmath1 is @xmath1591% , and entirely accounted for in our error budget ( entry r2 of table [ tbl : error ] ) . second , suntzeff correctly adopt @xmath150 for ngc 3368 , and not the published @xmath160 ( tanvir 1995 ) ; the additional 0.05mag corresponds to the `` long - versus short '' effect , not included by tanvir et al . third , suntzeff use ( slightly ) modified absolute moduli for ic 4182 and ngc 5253 , from those published by saha ( 1994 ) and saha ( 1995 ) , respectively . while the published values were @xmath119 ( ic 4182 ) and @xmath79 ( ngc 5253 ) , suntzeff used @xmath161 ( ic 4182 ) and @xmath152 ( ngc 5253 ) . in the case of ngc 5253 , a long - versus - short correction of 0.05mag was applied to an apparent @xmath7-band modulus of @xmath162 , followed by the subtraction of a total line - of - sight extinction of a@xmath163mag , resulting in a @xmath70 which is 0.16mag nearer than that favored by saha ( 1995 ) . this differs ( mildly ) from the saha ( 1995 ) published @xmath7-band modulus of @xmath164 . since there is no extant evidence for a long - versus - short effect in the older wf / pc data ( , for our purposes , ic 4182 and ngc 5253 ) , such a correction is not necessary . _ our re - analysis of the saha and tanvir galaxies allows us to circumvent some of the problems encountered by suntzeff ( in particular , the assumptions that had to be made in order to derive a suitable @xmath70 for ic 4182 and ngc 5253 ) . applying the self - consistent distances listed in column 2 of table [ tbl : distances ] , in lieu of those at the disposal of suntzeff et al . , along with the sne photometry of table [ tbl : supernovae ] , to the hubble relations of equations [ eq : hamuy_b]-[eq : hamuy_i ] , leads to the set of sn- and color - dependent hubble constants shown in table [ tbl : hamuy ] . a sample error budget for sn 1981b in ngc 4536 , is provided in table [ tbl : error ] , similar to that of table 4 of hamuy ( 1996 ) . using the full sample of eight calibrators in table [ tbl : hamuy ] leads to a weighted , color - independent , hubble constant of h@xmath165 . because of the previously documented shortcomings of sn 1960f and 1974 g , we favor a hubble constant tied to the remaining six calibrators ( , the five employed by suntzeff 1999 , supplemented by sn 1989b in ngc 3627 ) , although we should stress that numerically the results are identical . therefore , our favored result is @xmath166 comparing equation [ eq : ho_bkg ] with the suntzeff ( 1999 ) result of h@xmath155 demonstrates that our new allframe@xmath21trial - based distances result in a 6.1% increase in the predicted hubble constant . in contrast with the suntzeff ( 1999 ) methodology of section [ hamuy ] , saha ( 1997 ) , sandage ( 1996 ) do not apply a peak luminosity - decline rate correction aimed at reducing the scatter of the type ia sne hubble diagram . their adopted fits , which are based upon a combination of the and asiago ( patat 1995 ) surveys , are @xmath167 and are the direct analog of equations [ eq : hamuy_b ] and [ eq : hamuy_v ] . if we now take the peak absolute magnitudes for the six calibrators used in section [ hamuy ] , from table 6 of saha et al . , we find the weighted means to be m@xmath168 and m@xmath169 . while saha ( 1997 ) use sn 1895b in their h@xmath1 analysis , we have not , due to the larger errors in its photographic photometry . the good agreement between these mean magnitudes and the ones we derived in section [ hamuy ] ( see the final two entries of table [ tbl : supernovae ] ) , despite the @xmath0mag systematic offset in the absolute distances ( recall table [ tbl : distances ] ) is largely a fortuitous one . it comes about due to the fact that saha ( 1997 ) assume the observed sne peak magnitudes reflect the unextinguished values ( except in the cases of sn 1981b and sn 1989b ) , whereas suntzeff ( 1999 ) solve for both the intrinsic and galactic foreground reddening . coupled with equations [ eq : saha_b ] and [ eq : saha_v ] , these lead to h@xmath170 kms@xmath156mpc@xmath156 and h@xmath171 kms@xmath156mpc@xmath156 , respectively . the weighted mean of these @xmath3- and @xmath7-band hubble constants is h@xmath172 . saha ( 1997 ) use slightly different distances to the majority of their calibrators , in comparison with those published in their individual galaxy papers ; adopting these published distances to the galaxies in question leads to h@xmath173 , a 1.2% increase over that above . in comparison , adopting our best estimate of the distances to the six sandage / saha calibrators , based upon their original published dophot photometry ( column 3 of table [ tbl : distances ] ) , would yield h@xmath174 ( entry 7 of table [ tbl : h0_bkg ] ) . if we were to now replicate the analysis of saha ( 1997 ) , again using their sne photometry and reddening parameters ( table 6 of saha et al . ) , but now adopting the new allframe@xmath21trial distance moduli of table [ tbl : distances ] ( column 2 ) , the predicted mean peak magnitudes would be m@xmath175m@xmath176 , the mean @xmath3- and @xmath7-band peak magnitudes both decrease by @xmath90.21 mag . in conjunction with equations [ eq : saha_b ] and [ eq : saha_v ] , these revised mean magnitudes lead to hubble constants of h@xmath177=@xmath178 and h@xmath179=@xmath180 . the weighted mean of these values is h@xmath181 ( entry 5 of table [ tbl : h0_bkg ] ) . this value is directly comparable to the h@xmath182 alluded to earlier , in that the adopted hubble diagram ( equations [ eq : saha_b ] and [ eq : saha_v ] ) and sn lightcurve photometry ( table 6 of saha 1997 ) were chosen to be identical the only difference here is the adopted absolute distances to the host galaxies in question . the adoption of the new distance moduli leads to an upward revision of the saha et al . ( 1997 ) h@xmath1 of 8.9% . saha ( 1997 ) have noted that when correcting for the decline rate adopted by suntzeff ( 1999 ) , the calibrators become @xmath90.1 mag more luminous than the distant sample . suntzeff ( 1999 , and earlier papers ) suggest that such an anomaly may be real , reflecting the morphological type ( and stellar population ) mismatch between nearby and distant host galaxies . on balance , as summarized earlier , the case for a decline rate correction seems strong and hence our adoption of such an approach . saha use a different type ia sne sample for deriving their hubble diagram . their derived decline rate - peak luminosity relationships ( in both @xmath3 and @xmath7 ) are @xmath910% shallower than those derived by suntzeff ( 1999 ) and phillips ( 1999 ) . using these saha fits ( their equations 23 and 24 , with the associated zero points ) with the new allframe@xmath21trial distances to the same six calibrators employed above , leads to h@xmath1=@xmath183 , which is 4.5% greater than the value derived in the absence of any m@xmath184 correction ( , entries 5 and 6 of table [ tbl : h0_bkg ] ) . in comparison , the steeper slope favored by suntzeff ( 1999 ) results in an 8.3% increase in h@xmath1 ( , entries 1 and 5 of table [ tbl : h0_bkg ] ) . the results of sections [ hamuy ] and [ sandage ] were assuming no metallicity dependence of the cepheid pl relation . the dependence of the cepheid pl relation on metallicity has not yet been well established ( , musella 1999 ; storm , carney & fry 1999 ; nevalainen & roos 1998 ; saio & gautschy 1998 ; webb 1998 ; kennicutt 1998 ; sasselov 1997 ; gould 1994 ; madore & freedman 1990 ) , with perhaps the best empirical determination of the magnitude of such an effect to date being that due to kennicutt ( 1998 ) . kennicutt performed a differential analysis of the pl relations in two fields of m101 , spanning a range of @xmath185dex in oxygen abundance , and found @xmath186=-0.24\pm 0.16$]magdex@xmath156 . in other words , for any galaxy whose metallicity exceeds that of the reference galaxy - , the lmc , with @xmath187 by an order of magnitude , simply determining the absolute distance modulus via offsetting the apparent pl relations from those of the lmc , could lead to @xmath70 being underestimated by 0.24mag . for each of the six high- and intermediate - quality calibrating sne of table [ tbl : galaxies ] kennicutt ( 1999 ) have now shown that , in the mean , the metallicity of the cepheid fields , as measured by abundances , is 0.31dex greater than that of the lmc . recalling the kennicutt ( 1998 ) result of @xmath186=-0.24\pm 0.16$]magdex@xmath156 , this @xmath210.31dex metallicity offset from the lmc implies a typical correction factor of @xmath1880.07mag needs to be applied to the ensemble of calibrators . using the abundances from table [ tbl : galaxies ] , we applied the appropriate kennicutt ( 1998 ) correction to each of the individual calibrators , and then followed the identical decline rate - peak luminosity analysis described in section [ hamuy ] . this leads to a predicted hubble constant of h@xmath189 . in other words , applying a correction to the absolute distance moduli , based on kennicutt ( 1998 ) relationship , results in a reduction in our adopted h@xmath1 of 4.4% ( entries 1 and 3 of table [ tbl : h0_bkg ] ) . these two results agree to within @xmath91@xmath24 . the type ia supernova sample is sufficiently distant that use of the cosmic microwave background ( cmb ) reference frame is the standard to which the observed type ia sne host galaxy heliocentric velocities are transformed . here we test the assumption that correcting velocities for the local flow field does not introduce any significant systematic offset in the derived h@xmath1 . the cmb frame should , of course , be appropriate , as all but 5 of the 29 sne have recessional velocities in excess of 6000kms@xmath156 ( , a regime where peculiar velocities are @xmath190% the recessional velocity ) ; for the nearby sne , though , the peculiar velocities can be significant , so an effort has been made to estimate them . to that end , we have employed the three - attractor ( virgo@xmath21great attractor@xmath21shapley ) model described by mould ( 1999b ) , in order to correct the observed heliocentric velocities for the resultant flow field solution . we assume local group infall velocities of 200kms@xmath156 , 400kms@xmath156 , and 85kms@xmath156 , for virgo , the great attractor , and shapley , respectively . within this framework , the adopted infall velocities reproduce the amplitude of the cmb dipole ( @xmath9630kms@xmath156 ) , but with only these three attractors , the direction of local group motion is @xmath920@xmath191 away from the cmb direction . as we are simply interested in assessing the magnitude of the effect , at this stage , fine - tuning the model to improve this vector offset was deemed unnecessary . starting from 29 sne ( hamuy 1996 ; table 1 ) , we adopted the hamuy ( 1996 ) color cut to eliminate the three reddest sne ( sn 1990y , 1992k , and 1993h ) , and also eliminated the three which are either missing ( sn 1990 t and 1991s ) from zcat ( huchra 1999 ) , or have zcat velocities which are discrepant with those tabulated by hamuy et al ( sn 1993ag ) . to this final sample of 23 type ia sne , we performed a weighted least squares fit of the form @xmath192 + c_2 \label{eq : fit}\ ] ] including errors in m@xmath193 and @xmath154 , but neglecting those in recessional velocity @xmath194 . restricting ourselves to the peak @xmath7-band magnitude @xmath195 and the cmb reference frame ( , @xmath196 in equation [ eq : fit ] ) , for the adopted sample of 23 sne , we find @xmath197 and @xmath198 , with a dispersion @xmath199 . following the error analysis outlined in table [ tbl : error ] , and using the same six calibrators employed in section [ hamuy ] , this yields a @xmath7-band hubble constant of @xmath200 . conversely , if we now take the output of our three - attractor flow field model and replace @xmath201 , with the derived flow field velocities @xmath202 , the coefficients for equation [ eq : fit ] become @xmath203 and @xmath204 , with a dispersion @xmath205 . the resultant @xmath7-band hubble constant is then @xmath206 . comparison of the above results demonstrates that h@xmath179 increases by only 2.1% in going from the cmb to flow field reference frame . the same conclusion holds for h@xmath177 and h@xmath207 , not surprisingly . it is re - assuring to confirm that the two reference frames lead to hubble constants which are consistent at the @xmath138 level , as we had assumed , _ a priori . _ under the assumption that the fornax distance was defined by the silbermann ( 1999 ) cepheid distance to ngc 1365 - , @xmath208 - the suggestion had been made ( , figure 4 of suntzeff 1999 ) that the ( two ) fornax cluster type ia sne appeared to lie @xmath90.35mag behind ngc 1365 ( a discrepancy , albeit only at the @xmath91.5@xmath24 level ) . indeed , the same apparent discrepancy with this earlier cepheid distance had been noticed in surface brightness fluctuation ( sbf , ferrarese 1999 ) and fundamental plane ( fp , kelson 1999 ) fornax studies . with the addition of new cepheid distances to ngc 1326a ( prosser 1999 ) and 1425 ( mould 1999a ) , though , the _ key project s favored fornax distance is now @xmath209 ( ferrarese 1999 ) , reducing the cepheid - sne distance discrepancy from the earlier @xmath91.5@xmath24 , to @xmath611@xmath24 . further , direct , fornax distance determinations , including sandage / saha s upcoming attempt at uncovering cepheids in ngc 1316 , will be an important aid in determining whether the earlier cepheid versus sbf / fp distance `` dichotomy '' has truly been resolved . regardless , as far as a type ia sne - based determination of h@xmath1 is concerned , the inclusion of these two additional fornax sne has little effect upon the derived result , save for the expected reduction in the final random uncertainty ( madore 1999 ) . _ table [ tbl : h0_bkg ] summarizes the results of sections [ hamuy]-[flow ] , illustrating the dependence of the derived hubble constant to the important constituents of the analysis , including the adopted peak luminosity - light curve shape correction , number of local calibrators , cepheid pl - metallicity , and reference frame . it is important to stress that all of the scenarios based upon our uniform allframe@xmath21trial analysis ( all but the final entry of table [ tbl : h0_bkg ] ) are consistent , at the @xmath210 level , with our final , adopted , result of h@xmath211 ( , the initial entry of table [ tbl : h0_bkg ] ) . it is reassuring that this result is so robust , considering , for example , the different approaches adopted in the treatment ( or lack thereof ) of the sne light curve shapes ( contrast the first and fifth entries of table [ tbl : h0_bkg ] ) . we have re - derived distances to seven type ia sne - host galaxies , using archival _ hst data , collected by two different , non-_key project programs . the methodology employed parallels that used in reducing and analyzing the 18 galaxies observed as part of the _ hst key project on the extragalactic distance scale . this re - analysis has also benefited from the inclusion of additional archival data not available at the time of original reduction , as well as the experience and knowledge gained over the past five years in calibrating wf / pc and wfpc2 data . likewise , such cross - checking has uncovered several minor discrepancies in some of the earlier published results . _ _ _ the major result of our re - analysis is that , in the mean , the absolute distance modulus for these seven local type ia sne calibrators is @xmath0mag nearer ( weighted ) , on average , than those published originally by tanvir ( 1995 ) and saha ( 1994,1995,1996ab,1997,1999 ) . there are several reasons for the systematic differences , including differences in the photometry , cepheid samples , weighting of the data , and treatment of reddening . using these new distances , we have re - calculated the relevant zero points to the hubble diagrams used by suntzeff ( 1999 ) and saha ( 1997 ) , in order to quantify their effect upon the predicted value of the hubble constant . properly weighting each galaxy and sn by its associated uncertainties shows that any previous empirical calibration tied to these local type ia sne calibrators will have underestimated h@xmath1 by @xmath212% . formally , our favored , based upon type ia sne , is h@xmath213 . this result assumes the decline rate - peak luminosity relationship of suntzeff ( 1999 ) and no metallicity dependence of the underlying cepheid pl relationship . if the kennicutt ( 1998 ) empirical metallicity dependence is assumed , the predicted h@xmath1 is reduced to h@xmath214 . our type ia sne - based h@xmath1 determination agrees , to within the uncertainties , with the ( sakai 1999 ) and surface brightness fluctuation ( ferrarese 1999 ) results . the work presented in this paper is based on observations with the nasa / esa hubble space telescope , obtained by the space telescope science institute , which is operated by aura , inc . under nasa contract 5 - 26555 . the continued assistance of the nasa and stsci support staff , and in particular our program coordinator , doug van orsow , is gratefully acknowledged . support for this work was provided by nasa through grant go-2227 - 87a from stsci . smgh and pbs are grateful to nato for travel support via a collaborative research grant ( 960178 ) . we wish to thank nick suntzeff and brian schmidt for their input , and say a special word of thanks to the referee , mark phillips , for a most thorough and helpful analysis of the paper . tables [ tbl : n4639_ceph ] to [ tbl : n4496a_ceph ] include basic positional information , along with the cepheid mean magnitudes and periods , derived as part of our allframe@xmath21trial re - analysis of the type ia sne - host galaxy _ hst archive data . stetson s ( 1998 ) photometric calibration was employed , although transformation to the hill ( 1998 ) scale can be accomplished ( in the mean ) by adding @xmath210.02mag and @xmath210.04mag , respectively , to the cepheid mean @xmath7- and @xmath14-band magnitudes . epoch - by - epoch photometry , light curves , and finding charts , for all the cepheids contained herein , are available through the _ hst key project s web page http://www.ipac.caltech.edu/h0kp/.__ lcll ngc 3982 & n / a & sn 1998aq & 8100 ngc 4639 & 9.00@xmath1260.20 & sn 1990n & 5981 ngc 4536 & 8.85@xmath1260.20 & sn 1981b & 5427 ngc 3627 & 9.25@xmath1260.20 & sn 1989b & 6549 ngc 3368 & 9.20@xmath1260.20 & sn 1998bu & 5415 ngc 4527 & n / a & sn 1991 t & 7504 ngc 5253 & 8.15@xmath1260.15 & sn 1972e & 4277 ngc 1316 & n / a & sn 1980n & 7504,8100 ngc 1316 & n / a & sn 1981d & 7504,8100 ic 4182 & 8.40@xmath1260.20 & sn 1937c & 2547 ngc 4496a & 8.77@xmath1260.20 & sn 1960f & 5427 ngc 4414 & 9.20@xmath1260.20 & sn 1974 g & 5397 ngc 5253 & 8.15@xmath1260.15 & sn 1895b & 4277 lrrrrll 68@xmath1262@xmath1265 & 1@xmath215 & 34@xmath216 & 0@xmath217 & 6@xmath218 & -19.54@xmath1260.06 & this work 64@xmath1265@xmath1265 & 0@xmath215 & 40@xmath216 & 2@xmath217 & 4@xmath218 & @xmath219 ? & jha ( 1999 ) 60@xmath1262 & 0@xmath215 & 40@xmath216 & 13@xmath217 & 8@xmath218 & -19.49@xmath1260.07 & saha ( 1999 ) 63@xmath1262@xmath1264 & 1@xmath215 & 39@xmath216 & 0@xmath217 & 6@xmath218 & -19.47@xmath1260.03 & phillips ( 1999 ) 64@xmath1262@xmath1264 & 1@xmath215 & 34@xmath216 & 0@xmath217 & 5@xmath218 & -19.42@xmath1260.04 & suntzeff ( 1999 ) 65@xmath1261@xmath1266 & 11@xmath215 & 25@xmath216 & 1@xmath217 & 3@xmath218 & -19.45@xmath1260.01 & riess ( 1998 ) 60@xmath1266 & 1@xmath215 & 28@xmath216 & 0@xmath217 & 7@xmath218 & -19.48@xmath1260.07 & tripp ( 1998 ) 50@xmath1263@xmath1265 & 0@xmath215 & 20@xmath216 & 37@xmath217 & 4@xmath218 & -19.65@xmath1260.09 & lanoix ( 1998 ) 58@xmath1263@xmath1268 & 0@xmath215 & 40@xmath216 & 15@xmath217 & 7@xmath218 & -19.52@xmath1260.07 & saha ( 1997 ) 60@xmath1265 & 7@xmath215 & 25@xmath216 & 0@xmath217 & 7@xmath218 & -19.48@xmath1260.07 & tripp ( 1997 ) 63@xmath1263@xmath1263 & 1@xmath215 & 28@xmath216 & 0@xmath217 & 4@xmath218 & -19.14@xmath1260.03 & hamuy ( 1996 ) 64@xmath1263@xmath1263 & 1@xmath215 & 16@xmath216 & 3@xmath217 & 3@xmath218 & -19.22@xmath1260.09 & riess ( 1996 ) 66@xmath12612 & ? @xmath215 & ? @xmath216 & ? @xmath217 & -@xmath218 & @xmath219 ? & iwamoto & nomoto ( 1998 ) 68@xmath1266@xmath1267 & 0@xmath215 & 0@xmath216 & 6@xmath217 & -@xmath218 & -19.28@xmath1260.12 & ruiz - lapuente ( 1996 ) 67@xmath1269 & 1@xmath215 & 9@xmath216 & 16@xmath217 & -@xmath218 & -19.24@xmath1260.09 & hflich & khokhlov ( 1996 ) rcccc 7@xmath220 & @xmath600.015 & 0.030 & @xmath600.064 & 0.060 16@xmath220 & @xmath600.113 & 0.018 & @xmath600.006 & 0.034 18@xmath220 & @xmath600.011 & 0.036 & @xmath210.062 & 0.032 7@xmath220 & @xmath600.076 & 0.054 & @xmath600.122 & 0.036 5@xmath220 & @xmath210.028 & 0.093 & @xmath210.022 & 0.147 27@xmath220 & @xmath600.038 & 0.021 & @xmath600.119 & 0.039 45@xmath220 & @xmath600.039 & 0.014 & @xmath600.033 & 0.028 lllll ngc 4639 & @xmath221 & @xmath222 & @xmath223 & @xmath223 ngc 4536 & @xmath224 & @xmath225 & @xmath226 & @xmath227 ngc 3627 & @xmath228 & @xmath229 & @xmath230 & @xmath231 ngc 3368 & @xmath232 & @xmath233 & @xmath234 & n / a ngc 5253 & @xmath235 & @xmath236 & @xmath237 & @xmath238 ic 4182 & @xmath239 & @xmath240 & @xmath241 & @xmath242 ngc 4496a & @xmath243 & @xmath244 & @xmath245 & @xmath128 ngc 4414 & @xmath246 & n / a & @xmath247 & n / a llccccccc 1990n & ngc 4639 & @xmath248 & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 1981b & ngc 4536 & @xmath255 & @xmath256 & n / a & @xmath257 & @xmath258 & @xmath259 & n / a 1989b & ngc 3627 & @xmath260 & @xmath261 & n / a & @xmath262 & @xmath263 & @xmath264 & n / a 1998bu & ngc 3368 & @xmath265 & @xmath266 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & @xmath271 1972e & ngc 5253 & @xmath272 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 1937c & ic 4182 & @xmath279 & @xmath280 & n / a & @xmath275 & @xmath281 & @xmath282 & n / a 1960f & ngc 4496a & @xmath283 & @xmath284 & n / a & @xmath285 & @xmath286 & @xmath287 & n / a 1974 g & ngc 4414 & @xmath288 & @xmath289 & n / a & @xmath290 & @xmath291 & @xmath292 & n / a & & & @xmath293 & @xmath293 & @xmath294 & & & @xmath295 & @xmath296 & @xmath294 lllll 1990n & ngc 4639 & @xmath297 & @xmath298 & @xmath299 1981b & ngc 4536 & @xmath300 & @xmath301 & @xmath302 n / a 1989b & ngc 3627 & @xmath303 & @xmath304 & @xmath302 n / a 1998bu & ngc 3368 & @xmath305 & @xmath306 & @xmath307 1972e & ngc 5253 & @xmath308 & @xmath309 & @xmath310 1937c & ic 4182 & @xmath311 & @xmath312 & @xmath302 n / a 1960f & ngc 4496a & @xmath313 & @xmath314 & @xmath302 n / a 1974 g & ngc 4414 & @xmath315 & @xmath316 & @xmath302 n / a & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & @xmath319 rll & * errors on the cepheid distance scale * & a & lmc true modulus & @xmath322 b & lmc pl zero point & @xmath323 s1 & lmc pl systematic uncertainty ( = a+b ) & @xmath322 c & hst @xmath7-band zero point & @xmath324 d & hst @xmath14-band zero point & @xmath324 s2 & photometry systematic uncertainty ( @xmath325 ) & @xmath326 r1 & photometry random uncertainty ( allframe vs dophot ) & @xmath327 r2 & extinction treatment ( uncertainty in r@xmath328 ) & @xmath323 r3 & dereddened pl fit & @xmath329 r@xmath330 & total random ( = r1+r2+r3 ) & @xmath331 s@xmath330 & total systematic ( = s1+s2 ) & @xmath332 * ii . * & * errors on the sn distance scale * & e & apparent @xmath7 magnitude & @xmath324 f & sn line - of - sight reddening ( @xmath333 ) & @xmath334 g & distance modulus ( = r@xmath330 ) & @xmath331 h & absolute @xmath7 magnitude ( = e+f+g ) & @xmath335 i & m@xmath336 vs @xmath337 slope ( @xmath338 $ ] ) & @xmath339 j & m@xmath336 vs @xmath340 curvature ( @xmath341 ^ 2 $ ] ) & @xmath342 k & @xmath340 ( @xmath343 ^ 2 $ ] ) & @xmath324 l & corrected absolute @xmath7 magnitude ( = h+i+j+k ) & @xmath344 m & hubble diagram zero point & @xmath329n & hubble diagram scatter & @xmath332 & total random ( = l+m+n ) & @xmath345 & & = 6.1 & total systematic ( = s@xmath330 ) & @xmath332 & & = 4.9 lcccccc @xmath346 & s99 & 6 & s99 & g99 & none & cmb @xmath347 & s99 & 8 & s99 & g99 & none & cmb @xmath348 & s99 & 6 & s99 & g99 & k98 & cmb @xmath349 & s99 & 6 & s99 & g99 & none & ff @xmath350 & none & 6 & s97 & g99 & none & cmb @xmath351 & s97 & 6 & s97 & g99 & none & cmb @xmath352 & none & 6 & s97 & s / t & none & cmb ccrrccc c01 & 1 & 597.4 & 647.0 & @xmath353 & @xmath354 & @xmath355c02 & 2 & 455.5 & 295.0 & @xmath356 & @xmath357 & @xmath358c03 & 2 & 472.2 & 324.0 & @xmath359 & @xmath360 & @xmath361c04 & 2 & 226.9 & 620.2 & @xmath362 & @xmath363 & @xmath364c05 & 2 & 329.0 & 321.1 & @xmath365 & @xmath366 & @xmath367c06 & 2 & 155.1 & 529.9 & @xmath368 & @xmath369 & @xmath370c07 & 2 & 437.8 & 167.9 & @xmath371 & @xmath372 & @xmath373c08 & 2 & 92.3 & 381.8 & @xmath374 & @xmath375 & @xmath376c09 & 3 & 518.2 & 548.0 & @xmath377 & @xmath378 & @xmath379c10 & 3 & 361.2 & 557.0 & @xmath380 & @xmath381 & @xmath382c11 & 3 & 551.4 & 575.6 & @xmath383 & @xmath384 & @xmath385c12 & 3 & 780.0 & 584.8 & @xmath386 & @xmath387 & @xmath388c13 & 3 & 488.6 & 590.1 & @xmath389 & @xmath390 & @xmath391c14 & 3 & 528.9 & 574.7 & @xmath392 & @xmath393 & @xmath394c15 & 3 & 256.0 & 614.9 & @xmath395 & @xmath396 & @xmath397c16 & 4 & 554.7 & 83.8 & @xmath398 & @xmath399 & @xmath400c17 & 4 & 559.5 & 76.6 & @xmath401 & @xmath402 & @xmath403 ccrrccc c01 & 1 & 210.3 & 155.6 & @xmath404 & @xmath363 & @xmath405c02 & 1 & 228.6 & 291.5 & @xmath406 & @xmath407 & @xmath408c03 & 1 & 106.4 & 375.6 & @xmath409 & @xmath410 & @xmath411c04 & 1 & 252.4 & 651.5 & @xmath412 & @xmath413 & @xmath414c05 & 1 & 82.4 & 661.7 & @xmath415 & @xmath416 & @xmath417c06 & 1 & 375.9 & 471.8 & @xmath418 & @xmath419 & @xmath420c07 & 1 & 713.2 & 639.1 & @xmath421 & @xmath422 & @xmath423c08 & 2 & 89.4 & 202.4 & @xmath353 & @xmath424 & @xmath425c09 & 2 & 686.7 & 157.7 & @xmath426 & @xmath427 & @xmath428c10 & 2 & 423.7 & 186.6 & @xmath429 & @xmath430 & @xmath431c11 & 2 & 291.8 & 255.8 & @xmath432 & @xmath433 & @xmath434c12 & 2 & 593.7 & 328.2 & @xmath435 & @xmath436 & @xmath437c13 & 2 & 399.3 & 405.5 & @xmath438 & @xmath439 & @xmath440c14 & 2 & 342.9 & 454.0 & @xmath441 & @xmath442 & @xmath443c15 & 2 & 402.4 & 474.4 & @xmath444 & @xmath445 & @xmath446c16 & 2 & 144.7 & 523.5 & @xmath447 & @xmath448 & @xmath449c17 & 2 & 284.3 & 572.6 & @xmath450 & @xmath451 & @xmath452c18 & 2 & 73.6 & 86.4 & @xmath453 & @xmath454 & @xmath455c19 & 2 & 726.0 & 133.8 & @xmath456 & @xmath457 & @xmath458c20 & 2 & 535.2 & 340.3 & @xmath459 & @xmath460 & @xmath461c21 & 2 & 195.5 & 164.9 & @xmath462 & @xmath463 & @xmath464c22 & 2 & 373.4 & 127.8 & @xmath465 & @xmath466 & @xmath467c23 & 2 & 372.6 & 344.9 & @xmath468 & @xmath466 & @xmath469c24 & 3 & 196.2 & 252.6 & @xmath470 & @xmath471 & @xmath472c25 & 3 & 91.9 & 489.5 & @xmath473 & @xmath474 & @xmath475c26 & 3 & 483.4 & 543.1 & @xmath476 & @xmath477 & @xmath478c27 & 3 & 311.5 & 574.3 & @xmath479 & @xmath480 & @xmath481c28 & 3 & 621.3 & 646.8 & @xmath482 & @xmath483 & @xmath484c29 & 3 & 222.7 & 677.0 & @xmath485 & @xmath486 & @xmath487c30 & 3 & 359.8 & 703.0 & @xmath488 & @xmath489 & @xmath490c31 & 3 & 266.8 & 474.4 & @xmath491 & @xmath492 & @xmath493c32 & 3 & 279.0 & 216.0 & @xmath494 & @xmath495 & @xmath496c33 & 4 & 660.9 & 137.5 & @xmath497 & @xmath498 & @xmath499c34 & 4 & 536.6 & 574.9 & @xmath500 & @xmath372 & @xmath501c35 & 4 & 268.5 & 766.8 & @xmath502 & @xmath503 & @xmath504c36 & 4 & 500.8 & 392.0 & @xmath505 & @xmath506 & @xmath507c37 & 4 & 781.0 & 440.9 & @xmath508 & @xmath509 & @xmath510c38 & 4 & 393.0 & 352.2 & @xmath511 & @xmath512 & @xmath513c39 & 4 & 422.2 & 98.5 & @xmath514 & @xmath515 & @xmath516 ccrrccc c01 & 1 & 304.0 & 343.0 & @xmath517 & @xmath518 & @xmath519c02 & 1 & 254.7 & 641.0 & @xmath520 & @xmath521 & @xmath522c03 & 2 & 145.4 & 171.0 & @xmath523 & @xmath524 & @xmath525c04 & 2 & 124.1 & 199.4 & @xmath526 & @xmath527 & @xmath528c05 & 2 & 389.8 & 199.8 & @xmath529 & @xmath530 & @xmath531c06 & 2 & 308.4 & 227.3 & @xmath532 & @xmath533 & @xmath534c07 & 2 & 416.6 & 267.6 & @xmath535 & @xmath492 & @xmath536c08 & 2 & 517.9 & 282.3 & @xmath537 & @xmath538 & @xmath539c09 & 2 & 287.4 & 307.4 & @xmath540 & @xmath541 & @xmath542c10 & 2 & 170.9 & 308.7 & @xmath543 & @xmath544 & @xmath545c11 & 2 & 516.4 & 343.5 & @xmath546 & @xmath547 & @xmath548c12 & 2 & 549.2 & 245.8 & @xmath549 & @xmath550 & @xmath551c13 & 2 & 183.6 & 347.3 & @xmath552 & @xmath553 & @xmath554c14 & 2 & 89.8 & 388.5 & @xmath555 & @xmath556 & @xmath557c15 & 2 & 751.4 & 465.2 & @xmath558 & @xmath559 & @xmath560c16 & 2 & 642.4 & 762.0 & @xmath561 & @xmath562 & @xmath563c17 & 2 & 469.8 & 291.2 & @xmath564 & @xmath565 & @xmath566c18 & 2 & 484.5 & 349.4 & @xmath567 & @xmath568 & @xmath569c19 & 2 & 307.5 & 392.3 & @xmath570 & @xmath571 & @xmath572c20 & 2 & 742.8 & 289.0 & @xmath564 & @xmath573 & @xmath574c21 & 2 & 648.2 & 311.4 & @xmath374 & @xmath575 & @xmath576c22 & 2 & 353.3 & 776.1 & @xmath577 & @xmath578 & @xmath579c23 & 2 & 582.0 & 261.3 & @xmath561 & @xmath580 & @xmath581c24 & 3 & 350.0 & 120.1 & @xmath582 & @xmath583 & @xmath584c25 & 3 & 255.2 & 209.7 & @xmath585 & @xmath586 & @xmath587c26 & 3 & 749.1 & 275.6 & @xmath588 & @xmath589 & @xmath590c27 & 3 & 328.6 & 281.2 & @xmath591 & @xmath592 & @xmath593c28 & 3 & 187.3 & 329.4 & @xmath594 & @xmath595 & @xmath596c29 & 3 & 243.2 & 394.0 & @xmath594 & @xmath597 & @xmath598c30 & 3 & 416.0 & 652.3 & @xmath599 & @xmath600 & @xmath601c31 & 3 & 520.5 & 662.1 & @xmath602 & @xmath603 & @xmath604c32 & 3 & 418.0 & 235.3 & @xmath605 & @xmath606 & @xmath607c33 & 4 & 601.9 & 163.1 & @xmath608 & @xmath609 & @xmath610c34 & 4 & 585.6 & 185.5 & @xmath611 & @xmath612 & @xmath613c35 & 4 & 535.6 & 471.7 & @xmath614 & @xmath615 & @xmath616c36 & 4 & 135.7 & 125.4 & @xmath617 & @xmath618 & @xmath619 crrccc c01 & 723.8 & 575.5 & @xmath620 & @xmath621 & @xmath622 c02 & 702.1 & 316.5 & @xmath623 & @xmath624 & @xmath625 c03 & -689.0 & 266.5 & @xmath626 & @xmath627 & @xmath628 c04 & 628.9 & 527.4 & @xmath629 & @xmath630 & @xmath631 c05 & -328.9 & -470.8 & @xmath632 & @xmath633 & @xmath634 c06 & -536.7 & 392.0 & @xmath635 & @xmath636 & @xmath637 c07 & 764.6 & 380.5 & @xmath638 & @xmath639 & @xmath640 c08 & -113.3 & -334.4 & @xmath641 & @xmath642 & @xmath643 c09 & -94.8 & 656.5 & @xmath644 & @xmath645 & @xmath646 c10 & -446.5 & 252.3 & @xmath647 & @xmath648 & @xmath649 c11 & -42.4 & 690.9 & @xmath650 & @xmath651 & @xmath652 crrccr c01 & 336.1 & 118.8 & @xmath653 & @xmath654 & @xmath655 c02 & 535.7 & 163.1 & @xmath656 & @xmath657 & @xmath658 c03 & 738.8 & 386.8 & @xmath659 & @xmath660 & @xmath661 c04 & 299.3 & 529.2 & @xmath662 & @xmath663 & @xmath664 c05 & 1124.0 & 549.7 & @xmath665 & @xmath666 & @xmath667 c06 & 393.8 & -562.0 & @xmath668 & @xmath669 & @xmath670 c07 & 964.1 & 433.6 & @xmath671 & @xmath672 & @xmath673 crrccr c01 & -200.5 & 151.8 & @xmath674 & @xmath675 & @xmath676 c02 & -479.8 & 188.3 & @xmath677 & @xmath678 & @xmath679 c03 & -503.5 & 305.1 & @xmath680 & @xmath681 & @xmath682 c04 & 96.2 & 643.0 & @xmath683 & @xmath684 & @xmath685 c05 & -481.3 & 338.5 & @xmath686 & @xmath687 & @xmath688 c06 & -216.8 & 291.4 & @xmath689 & @xmath690 & @xmath691 c07 & -372.7 & 29.9 & @xmath692 & @xmath693 & @xmath694 c08 & -331.9 & 60.4 & @xmath695 & @xmath696 & @xmath697 c09 & -259.2 & -141.8 & @xmath698 & @xmath699 & @xmath700 c10 & -257.8 & 33.3 & @xmath701 & @xmath702 & @xmath703 c11 & -653.5 & -56.8 & @xmath695 & @xmath704 & @xmath705 c12 & -521.5 & -150.1 & @xmath706 & @xmath707 & @xmath708 c13 & -455.6 & 63.2 & @xmath709 & @xmath710 & @xmath711 c14 & -526.5 & -460.8 & @xmath712 & @xmath713 & @xmath714 c15 & -240.0 & -145.1 & @xmath715 & @xmath716 & @xmath717 c16 & -326.3 & -570.0 & @xmath718 & @xmath719 & @xmath720 c17 & -645.1 & -449.4 & @xmath526 & @xmath721 & @xmath722 c18 & -96.6 & -59.9 & @xmath723 & @xmath724 & @xmath725 c19 & -79.3 & -550.7 & @xmath726 & @xmath727 & @xmath728 c20 & 115.3 & 7.1 & @xmath729 & @xmath730 & @xmath731 c21 & 156.6 & 101.4 & @xmath732 & @xmath733 & @xmath734 c22 & 151.5 & -160.4 & @xmath735 & @xmath736 & @xmath737 c23 & 139.1 & -175.8 & @xmath738 & @xmath739 & @xmath740 c24 & 561.9 & -235.3 & @xmath741 & @xmath742 & @xmath743 c25 & 594.5 & -129.0 & @xmath744 & @xmath745 & @xmath746 c26 & 531.8 & 262.3 & @xmath747 & @xmath748 & @xmath749 c27 & 360.3 & 191.7 & @xmath750 & @xmath751 & @xmath752 c28 & 579.0 & 174.9 & @xmath753 & @xmath754 & @xmath755 ccrrccc c01 & 1 & 289.2 & 110.5 & @xmath756 & @xmath757 & @xmath758c02 & 1 & 779.8 & 465.3 & @xmath759 & @xmath760 & @xmath761c03 & 1 & 134.7 & 696.6 & @xmath762 & @xmath763 & @xmath764c04 & 1 & 343.8 & 279.7 & @xmath765 & @xmath766 & @xmath767c05 & 1 & 266.3 & 379.2 & @xmath768 & @xmath769 & @xmath770c06 & 1 & 241.8 & 481.2 & @xmath771 & @xmath466 & @xmath772c07 & 1 & 663.2 & 524.5 & @xmath520 & @xmath773 & @xmath774c08 & 1 & 364.4 & 535.7 & @xmath775 & @xmath776 & @xmath777c09 & 1 & 680.4 & 546.0 & @xmath778 & @xmath372 & @xmath779c10 & 1 & 659.2 & 564.8 & @xmath780 & @xmath547 & @xmath781c11 & 1 & 765.4 & 318.6 & @xmath782 & @xmath783 & @xmath784c12 & 1 & 424.9 & 415.1 & @xmath785 & @xmath786 & @xmath787c13 & 1 & 693.9 & 288.7 & @xmath788 & @xmath789 & @xmath790c14 & 2 & 285.7 & 164.8 & @xmath791 & @xmath792 & @xmath793c15 & 2 & 554.7 & 178.0 & @xmath794 & @xmath795 & @xmath796c16 & 2 & 735.7 & 279.0 & @xmath797 & @xmath798 & @xmath799c17 & 2 & 286.6 & 280.3 & @xmath800 & @xmath801 & @xmath802c18 & 2 & 261.8 & 336.5 & @xmath803 & @xmath804 & @xmath805c19 & 2 & 435.6 & 356.9 & @xmath778 & @xmath806 & @xmath807c20 & 2 & 389.2 & 375.5 & @xmath808 & @xmath809 & @xmath810c21 & 2 & 514.6 & 377.4 & @xmath488 & @xmath811 & @xmath812c22 & 2 & 361.4 & 395.8 & @xmath371 & @xmath813 & @xmath814c23 & 2 & 171.1 & 426.1 & @xmath815 & @xmath558 & @xmath816c24 & 2 & 473.3 & 437.1 & @xmath497 & @xmath783 & @xmath817c25 & 2 & 102.0 & 468.0 & @xmath818 & @xmath819 & @xmath820c26 & 2 & 162.1 & 494.8 & @xmath821 & @xmath822 & @xmath823c27 & 2 & 365.1 & 497.0 & @xmath824 & @xmath825 & @xmath826c28 & 2 & 228.6 & 557.9 & @xmath827 & @xmath828 & @xmath829c29 & 2 & 521.8 & 571.5 & @xmath830 & @xmath831 & @xmath832c30 & 2 & 161.6 & 591.6 & @xmath833 & @xmath834 & @xmath835c31 & 2 & 391.7 & 676.5 & @xmath836 & @xmath837 & @xmath838c32 & 2 & 622.8 & 752.0 & @xmath839 & @xmath840 & @xmath841c33 & 2 & 422.5 & 219.6 & @xmath842 & @xmath843 & @xmath844c34 & 2 & 450.9 & 269.1 & @xmath845 & @xmath846 & @xmath847c35 & 2 & 444.9 & 297.6 & @xmath511 & @xmath848 & @xmath849c36 & 2 & 561.9 & 420.9 & @xmath850 & @xmath851 & @xmath852c37 & 2 & 511.9 & 454.3 & @xmath853 & @xmath854 & @xmath855c38 & 2 & 210.9 & 130.1 & @xmath856 & @xmath857 & @xmath858c39 & 2 & 213.3 & 96.5 & @xmath859 & @xmath860 & @xmath861c40 & 3 & 324.0 & 178.1 & @xmath862 & @xmath863 & @xmath864c41 & 3 & 196.2 & 181.9 & @xmath865 & @xmath866 & @xmath867c42 & 3 & 230.4 & 187.3 & @xmath868 & @xmath869 & @xmath870c43 & 3 & 562.4 & 190.4 & @xmath871 & @xmath872 & @xmath873c44 & 3 & 455.5 & 194.4 & @xmath874 & @xmath875 & @xmath876c45 & 3 & 257.3 & 213.7 & @xmath877 & @xmath878 & @xmath879c46 & 3 & 335.4 & 214.0 & @xmath880 & @xmath881 & @xmath882c47 & 3 & 479.3 & 228.3 & @xmath883 & @xmath884 & @xmath885c48 & 3 & 614.2 & 231.1 & @xmath567 & @xmath671 & @xmath886c49 & 3 & 381.5 & 280.7 & @xmath887 & @xmath888 & @xmath889c50 & 3 & 412.2 & 294.8 & @xmath890 & @xmath891 & @xmath892c51 & 3 & 281.8 & 301.0 & @xmath893 & @xmath894 & @xmath895c52 & 3 & 761.0 & 323.7 & @xmath896 & @xmath897 & @xmath898c53 & 3 & 233.2 & 347.9 & @xmath883 & @xmath899 & @xmath900c54 & 3 & 503.7 & 422.6 & @xmath901 & @xmath902 & @xmath903c55 & 3 & 408.4 & 492.1 & @xmath904 & @xmath905 & @xmath906c56 & 3 & 306.6 & 786.6 & @xmath907 & @xmath633 & @xmath908c57 & 3 & 402.2 & 790.9 & @xmath374 & @xmath642 & @xmath909c58 & 3 & 567.6 & 732.9 & @xmath910 & @xmath822 & @xmath911c59 & 3 & 454.1 & 238.2 & @xmath912 & @xmath913 & @xmath914c60 & 3 & 357.0 & 282.0 & @xmath915 & @xmath916 & @xmath917c61 & 3 & 292.1 & 619.2 & @xmath918 & @xmath919 & @xmath920c62 & 3 & 120.9 & 428.7 & @xmath921 & @xmath922 & @xmath923c63 & 3 & 101.1 & 438.2 & @xmath924 & @xmath925 & @xmath926c64 & 3 & 724.1 & 450.1 & @xmath927 & @xmath928 & @xmath929c65 & 3 & 110.2 & 195.1 & @xmath930 & @xmath575 & @xmath931c66 & 3 & 547.8 & 661.5 & @xmath932 & @xmath933 & @xmath934c67 & 3 & 452.3 & 242.2 & @xmath441 & @xmath360 & @xmath935c68 & 4 & 309.2 & 137.0 & @xmath797 & @xmath936 & @xmath937c69 & 4 & 196.6 & 148.0 & @xmath938 & @xmath939 & @xmath940c70 & 4 & 172.6 & 275.4 & @xmath941 & @xmath942 & @xmath943c71 & 4 & 278.9 & 351.3 & @xmath944 & @xmath945 & @xmath946c72 & 4 & 169.8 & 450.0 & @xmath947 & @xmath948 & @xmath949c73 & 4 & 710.3 & 105.4 & @xmath950 & @xmath477 & @xmath951c74 & 4 & 246.2 & 121.9 & @xmath952 & @xmath953 & @xmath954c75 & 4 & 637.2 & 150.7 & @xmath494 & @xmath955 & @xmath956c76 & 4 & 145.2 & 163.0 & @xmath901 & @xmath957 & @xmath958c77 & 4 & 337.7 & 236.1 & @xmath959 & @xmath960 & @xmath961c78 & 4 & 637.4 & 252.9 & @xmath962 & @xmath963 & @xmath964c79 & 4 & 211.9 & 334.9 & @xmath371 & @xmath965 & @xmath966c80 & 4 & 159.4 & 358.9 & @xmath485 & @xmath967 & @xmath968c81 & 4 & 300.2 & 369.1 & @xmath969 & @xmath422 & @xmath970c82 & 4 & 717.8 & 408.7 & @xmath971 & @xmath972 & @xmath973c83 & 4 & 618.4 & 413.1 & @xmath859 & @xmath974 & @xmath975c84 & 4 & 492.4 & 679.0 & @xmath406 & @xmath976 & @xmath977c85 & 4 & 360.2 & 658.5 & @xmath883 & @xmath976 & @xmath978c86 & 4 & 391.5 & 645.4 & @xmath979 & @xmath980 & @xmath981c87 & 4 & 101.9 & 644.0 & @xmath982 & @xmath983 & @xmath984c88 & 4 & 427.0 & 636.2 & @xmath836 & @xmath763 & @xmath985c89 & 4 & 401.2 & 613.1 & @xmath986 & @xmath987 & @xmath988c90 & 4 & 375.2 & 619.5 & @xmath989 & @xmath990 & @xmath991c91 & 4 & 597.0 & 462.4 & @xmath992 & @xmath463 & @xmath993c92 & 4 & 138.7 & 463.2 & @xmath429 & @xmath994 & @xmath995c93 & 4 & 138.6 & 180.7 & @xmath432 & @xmath996 & @xmath997c94 & 4 & 643.6 & 333.3 & @xmath950 & @xmath998 & @xmath999

the morphology of galaxies is of great interest because it reflects not only the structural properties of galaxies but also their star formation histories . a morphological study of galaxies has been one of the main topics of extra - galactic studies since the pioneering work by @xcite . despite the enormous efforts to determine the morphological types of galaxies , the morphological types of the majority of galaxies from recent surveys , such as the sloan digital sky survey ( sdss ) , still need to be determined . the morphology of bright galaxies in the local universe has been well studied and summarized in catalogs , such as a revised shaply - ames catalog of bright galaxies ( rsa , @xcite ) and the third reference catalog of bright galaxies ( rc3 , @xcite ) . however , none of these catalogs are suitable for dwarf galaxy studies because most dwarf galaxies which are much fainter than the bright end of dwarf galaxies are omitted due to their selection criteria . the recent study of @xcite , who visually classified 14,000 sdss galaxies from data release 4 ( dr4 ) , focussed on the redshift range @xmath10 , and as a consequence included few dwarf galaxies . the galaxy zoo project @xcite has provided morphological information for a much larger number of objects , but only in broad categories with little information on dwarf galaxies . dwarf galaxies , however , are predicted to be the most dominant type of galaxy based on cold dark matter ( cdm ) cosmology @xcite . the importance of dwarf galaxies in galactic astronomy has been well demonstrated by the dominance of dwarf galaxies in the local group ( lg ) , in which the number of dwarf galaxies tripled recently , mostly thanks to sdss @xcite . they also dominate the nearby satellite systems such as those hosted by m81 @xcite and m106 @xcite . galaxies with distances less than @xmath11mpc and velocities relative to the lg ( @xmath12 ) less than 600kms@xmath13 , which were compiled in the updated nearby galaxy catalog @xcite , are also dominated by dwarf galaxies . @xcite presented an all - sky catalog of nearby galaxy groups that included 10,914 galaxies in the local universe ( @xmath14 ) with various astrophysical parameters including the morphological types in numerical code @xmath15 . on the other hand , they did not report the detailed morphological types of dwarf galaxies . @xcite provided a better treatment of the morphology of dwarf galaxies , but they confined the dwarfs to within @xmath11mpc . because galaxies in the local universe will provide a unique sample to explore the structural properties of dwarf galaxies , it is important to determine their detailed morphological types . sdss opens a new era of galaxy morphology by providing color images which are useful for obtaining information about the underlying stellar populations . because stellar populations are closely related to the star formation history of galaxies , the morphological types that consider the stellar populations using color information and the light distributions can be considered to be the best proxy for characterizing the integral property of galaxies . the blue elliptical galaxy @xcite , is an example of a new type of galaxy that has emerged in the new era of the galaxy morphology . moreover , it is deep enough to cover dwarf galaxies as faint as @xmath16 in the local universe . another major difference between the morphological properties observed in single - band images and color images were found in the dwarf elliptical ( de ) family . although de galaxies and dwarf spheroidal galaxies ( dsph ) are distinguished in the lg , they were used interchangeably for galaxies outside the lg in previous studies . in particular , @xcite used de and de , n for both de galaxies and dsph galaxies in a classification of virgo cluster dwarfs , whereas @xcite used sph for both galaxies . on the other hand , there appears to be some difference between de and dsph @xcite . for example , de galaxies are generally brighter and redder than dsph galaxies . although the environmental dependence of giant galaxies is well understood thanks to recent surveys , such as sdss @xcite , the dependence of dwarf galaxy morphology on environment has not been well studied due to the lack of dwarf galaxies in previous analyses . therefore , it is important to produce a homogeneous sample of dwarf galaxies with detailed morphological types . because the limiting magnitude of spectroscopic observations of the sdss is @xmath17 , it is possible to make a volume - limited sample brighter than @xmath18 if galaxies with redshift less than @xmath19 are used . this provides an unprecedented opportunity to classify the morphological types of local galaxies using the color images of the sdss dr7 with an emphasis on dwarf galaxies , whose morphological properties are still not well understood . particular attention was paid to the morphologies of de - like galaxies because they are the building blocks of larger structures in cdm cosmology . in particular , dsph are believed to be mostly dark matter dominated dwarfs ( e.g. , gilmore et al . 2007 ) . because de galaxies show different properties from dsph , such as the presence of gas and young stellar populations , at least for the lg dwarfs @xcite , it is important to determine whether they can be distinguished by morphology alone . the aim of this paper is to perform a census of the morphology distribution of local galaxies brighter than @xmath20 . to accomplish this , the morphological types of all sdss galaxies that have redshifts less than @xmath19 are classified by visual inspection of the color images of sdss dr7 . the results are presented in the form of a catalog that includes not only our classificiations , but also numerical t types , adopted distances , absolute @xmath21-band magnitudes , @xmath22 colors , and isophotal angular diameters and axis ratios . an additional aim is to better understand the environment of local galaxies , a factor which can play an important role in morphology . section 2 introduces the observational data used in this study and the procedure for selecting the sample galaxies . section 3 describes the methodology of morphology classification and section 4 provides a catalog of the morphological types of the local galaxies . section 5 presents the physical and morphological properties of the local galaxies and section 6 reports the environmental dependence of the morphology . the final section provides a brief summary and discussion . the basic sample of the local galaxies in this paper was derived from the korea institute for advanced study value - added galaxy catalog ( kias - vagc choi et al . 2010 ) , but was supplemented by galaxies with @xmath23 in the nasa extragalactic data base ( ned ) and the galaxies listed in @xcite that are not overlapped with the galaxies from the kias - vagc and ned . the kias - vagc is a value - added catalog of galaxies based on the new york university value - added catalog ( nyu - vagc ) dr7 which was derived from the sdss dr7 @xcite ) . the kias - vagc provides valuable astrophysical parameters such as absolute magnitude and colors corrected for the galactic extinction as well as distances and morphological types determined by automated classifier @xcite . it also lists the redshifts of 10,497 galaxies that are not included in nyu - vagc . they gathered redshift data from various catalogs ( see choi et al . ( 2010 ) for a detailed description ) . more detailed descriptions of sdss data were reported by @xcite and @xcite . all of the galaxies with redshifts less than 0.01 were selected from the kias - vagc . the color images of the individual galaxies listed in the kias - vagc were examined carefully to determine whether they are parts of a galaxy or multiply listed . a few tens of galaxies were removed from the selected galaxies because of their multiple entry or invalid redshift . the number of cleaned galaxies from kias - vagc is 4807 . in addition , 897 galaxies from ned , whose images and photometric data are available from sdss dr7 , were added . we present the number distribution of the sample galaxies as a function of redshift in figure 1 . the total number of sample galaxies is 5836 . figure 2 shows the distribution of sample galaxies in a color - magnitude diagram . the petrosian @xmath21 magnitude and @xmath22 color are used as the representative magnitude and color of a galaxy because the petrosian @xmath21 magnitude was used to select the spectroscopic target galaxies and the @xmath22 color is one of the best proxies of the galaxy morphology because of its close dependence on the star formation rate @xcite . most galaxies brighter than the limiting magnitude of the sdss spectroscopic target galaxies ( @xmath17 ) are located in a relatively narrow range of @xmath22 colors , whereas galaxies fainter than @xmath17 have a wide range of @xmath22 colors . the color range of galaxies brighter than @xmath17 is similar to that of the volume - limited sample of galaxies brighter than @xmath24 @xcite . the redshifts of galaxies fainter than @xmath17 have been observed by other surveys , such as 2dfgrs @xcite , which is deeper than sdss . figure 3 presents the distribution of @xmath21-band isophotal semi - major axis length ( @xmath25 ) and the axial ratio ( @xmath26 ) of sample galaxies whose isophotal semi - major and semi - minor axis lengths are available in the sdss dr7 . the isophotal axis lengths were measured at an isophote of 25 mag arcsec@xmath27 at each pass - band . most galaxies are larger than @xmath28 pixels . therefore , galaxies in the local universe ( @xmath23 ) are large enough for visual classification . approximately @xmath29 spiral galaxies are inclined more than @xmath30 . these edge - on galaxies comprise @xmath31 of the total galaxies and @xmath32 of spiral galaxies . galaxies less than @xmath33arcsec are mostly face - on galaxies . traditionally , morphological classification has been performed through inspection of photographic plates or prints sensitive to the blue region of the spectrum ( e.g. , de vaucouleurs 1959 ; sandage 1961 ; sandage and tammann 1981 ; sandage and bedke 1994 ) . more recently , galaxies have been classified using logarithmic , sky - subtracted digital @xmath34-band images in units of mag arcsec@xmath27 ( e.g. , buta et al . 2007 ) . therefore , in order to be `` true '' to the traditional classification systems , all of which are built on the original hubble ( 1926 , 1936 ) system , one should use sdss @xmath35-band images since these are close enough in wavelength to the @xmath34 band . in the present study , however , we use the color images provided by the sdss which are based on a combination of @xmath36 images @xcite . for some galaxy types , the use of such images could lead to systematic differences from blue light classifications because the inclusion of the @xmath37 band can enhance the prominence of galactic bulges , which play a role in the t type . our reasons for using color images over @xmath35-band images are twofold . first , color adds information to morphology by its ability to distinguish underlying stellar populations . second , our sample is dominated by galaxies that have very little or no bulge , making the possible impact of the @xmath37 band on t types less of an issue . @xcite show that color gradients and the colors of galaxies are effective for automatic classification . we therefore feel that the use of color images for visual classification is justified . the classification scheme of the morphological types of galaxies is based on the rc3 classification system @xcite , which is the de vaucouleurs revised hubble - sandage system , or vrhs @xcite . the description and sample color images given by @xcite are also consulted . the classification scheme is simplified by neglecting stages of elliptical and lenticular galaxies . that is , the three stages of elliptical galaxies , compact ellipticals ( ce ) , normal ellipticals ( e ) , and late - type ellipticals ( e@xmath38 ) and the three stages of lenticular galaxies , s0@xmath38 , s0@xmath39 and s0@xmath40 are not distinguished . the reason for this is that these e / s0 stages are based mostly on subtle aspects of structure and ignoring them has little impact on our analysis . on the other hand , all of the stages of spiral galaxies from 0/a to m are distinguished . de / dsph galaxies are separated from giant ellipticals because the surface brightness distribution of de galaxies is significantly different from that of the giant elliptical galaxies . elliptical galaxies are believed to have a simple structure . but , thanks to the color information that is sensitive to the underlying stellar populations , it is possible to detect peculiar morphologies induced by recent star formation in elliptical galaxies . the discovery of blue elliptical galaxies showing almost identical shapes to those of normal elliptical galaxies but with different colors @xcite is a good example of the usefulness of color images in extragalactic astronomy . earlier studies of blue elliptical galaxies revealed anomalous blue cores @xcite . for example . ngc 3156 and ngc 4742 have blue cores which are caused by the presence of young stellar populations in the nuclear regions @xcite . the distinction between elliptical galaxies and lenticular galaxies is made mainly by their surface brightness distribution . elliptical galaxies show steeper surface brightness gradients than lenticular galaxies , particularly in the outer parts . this is why the luminosity profiles of elliptical galaxies are best represented by the sersic function with @xmath41 whereas those of lenticular galaxies are best fit by the sersic function with @xmath42 . galaxies with characteristics intermediate between ellipticals and lenticulars are classified as e / s0 . lenticular galaxies are further divided into three bar families , sa0 , sab0 , and sb0 , according to their bar types . the sa0 types show no bar feature , whereas the sb0 types show a clear bar morphology crossing the central bulge . lenticular galaxies with intermediate bar types are assigned as sab0 . in the case of uncertain bar types , mostly due to high inclinations , they are classified as s0 galaxies . for spiral galaxies , the spiral arm morphology is differentiated according to the hubble stages adopted in rc3 @xcite . namely , the stages , 0/a , a , ab , b , bc , c , cd , d , dm , and m , of spiral galaxies are determined according to the relative size of the bulge and the openness of the spiral arms . on the other hand , the spiral varieties , ( r ) and ( s ) , are not distinguished because they are less fundamental than stage and family , and , as for the e / s0 stages , ignoring them has little or no impact on our analysis . we also exclude recognition of outer rings and pseudorings for the same reason . therefore , for example , a strongly barred spiral galaxy with an intermediate hubble stage is denoted as sbc , whereas a non - barred spiral galaxy with an intermediate hubble stage is denoted sac . for edge - on spirals , the bar families are not distinguished in this study and the hubble stage is denoted as sa , sab , sb , etc . irregular galaxies are divided into three types : i m , dwarf irregular ( di ) , and bcd . the first type , i m , is a natural extension of the late - type spirals and some i m galaxies are believed to have a bar component . in that case , a morphological type , ibm , is assigned . the second type , di , represents small dwarf galaxies with an amorphous shape . their physical properties , such as size , luminosity , and colors are similar to those of dsph and de@xmath2 galaxies , which are described below . the last type , bcd , stands for the blue compact dwarf and is characterized by star burst regions . they have amorphous shapes with high central surface brightness , and are likely to show an outer irregular envelope . however , there is no unique definition of bcd in the literature @xcite but the morphological properties of bcds in the present sample are similar to those of the bcds analyzed by @xcite . the shapes of bcd galaxies without an outer irregular envelope , i.e. , hii region - like bcds , are similar to the small blue dwarf elliptical galaxy ( de@xmath2 ) described below . bcds show an extremely blue color due to bursts of star formation . a significant fraction of irregular galaxies have small star burst regions . these galaxies are distinguished from i m and di as well as bcd . we designate these galaxies as ir / bcd . in addition to di galaxies , there are five types of dwarf galaxies : dwarf lenticular galaxies ( ds0 ) , de galaxies , blue - cored dwarf elliptical galaxies ( de@xmath43 ) , dsph galaxies , and de@xmath2 galaxies . we designate these five dwarf galaxies as de - like galaxies . the dwarf lenticular galaxies were introduced by @xcite for the morphological classification of dwarf galaxies in the virgo cluster . they are characterized by lens - like features that have a shallow brightness gradient interior to a sharp edge . ( see buta ( 2013 ) for further discussion of lenses in disk galaxies . ) in this respect , ds0 galaxies are different from their giant cousins , s0 galaxies , which are characterized by a bulge and a disk . a majority of ds0 galaxies show nucleation . the presence and absence of nucleation are distinguished by the subscripts @xmath44 and @xmath45 as ds0@xmath46 and ds0@xmath47 for a nucleated one and an un - nucleated one , respectively . some dwarf galaxies exhibit similar characteristics to ds0 galaxies but the outer parts of these galaxies show disturbed disks , mostly resembling disrupted spiral arms . these galaxies are classified as ds0@xmath48 . we do not distinguish ds0@xmath48 galaxies from ds0 galaxies , except for the description of a detailed morphology . de galaxies can be distinguished from giant elliptical galaxies by their surface brightness distribution , which is better represented by the sersic function with @xmath49 . most previous studies do not distinguish between de galaixes and dsph galaxies ( eg . , binggeli et al . 1985 ; kormendy & bender 2012 ) . in particular , @xcite use sph to represent spheroidal galaxies that can be placed in a position parallel to the i m galaxies in their revised parallel - sequence galaxy classification , which extends the van den bergh parallel - sequence galaxy classification @xcite . they assumed that the sph galaxies are stripped scd - im galaxies . however , our study considers dsph galaxies as a distinct population of dwarf galaxies because photometric properties , such as surface brightness and colors of dsph galaxies , seem to be different from those of de galaxies . moreover , if we consider that dsph galaxies are mostly dispersion - supported systems @xcite while a considerable fraction of de galaxies are supported by rotation @xcite , their origins might be different . the de and dsph galaxies are distinguished by their surface brightness while de@xmath2 galaxies are distinguished from others by their blue color . the de@xmath43 galaxies are characterized by the presence of blue cores which are supposed to be regions of active star formation . the de and dsph galaxies are further distinguished by their nuclear morphology : de@xmath47 , de@xmath46 , dsph@xmath47 and dsph@xmath46 . the distinction of nucleated dwarfs from un - nulceated dwarfs is somewhat arbitrary because the degree of nucleation varies continuously @xcite . blue - cored dwarfs , however , are easily distinguished from nucleated dwarfs because the cores are bluer and larger than the nuclear region of nucleated dwarfs . blue dwarf ellipticals ( de@xmath2 ) have similar color to irregular galaxies . their size is similar to or smaller than hii region - like bcds . small de@xmath2 and di galaxies have similar features except for the axial symmetry . on the other hand , there is no clear distinction in the degree of axial symmetry for de@xmath2 . therefore , it seems plausible that de@xmath2 galaxies are di galaxies with an extreme axial symmetry . for practical purposes , the numerical type @xmath50 borrowed from rc3 is used with some simplification for the ellipticals ( @xmath50=-5 ) and lenticulars ( @xmath50=-3 ) . however , we keep all of the hubble stages from t=0 to t=9 for spirals ( from @xmath51 to @xmath52 ) and @xmath50=10 ( i m ) and @xmath50=11 ( di ) for irregulars . bcd is assigned as @xmath50=13 , while irregulars containing small bcd - like components as @xmath50=12 . also , we assign new codes from -6 to -11 for dwarf elliptical - like galaxies : @xmath50=-6 ( de ) , @xmath50=-7 ( de@xmath43 ) , @xmath50=-8 ( dsph ) , @xmath50=-9 ( de@xmath2 ) , @xmath50=-10 ( ds0 , ds0@xmath48 ) , and @xmath50=-11 for transition type dwarf ( des / di di / des ) , respectively . in the present study , the color images of the sub - sample of galaxies were examined using the image display tool provided by the sdss for training before the classification of the full sample of galaxies . the morphological types given in the ned were consulted during the training period . because the number of subtypes to be classified is more than 20 , morphological types were classified in two steps . galaxies were divided into four broad types ( elliptical / lenticulars , spirals , irregulars , and de - like galaxies ) , and their fine subtypes were then classified after sorting the galaxies according to their broad types . the entire sample of galaxies was classified by each of the three authors . we followed two steps to finalize the morphological types of galaxies . in the first step , one of the authors ( seo ) determined the morphological types of the entire sample by consulting the results of other authors . the final morphological type of each galaxy was determined by the senior author after reviewing seo s classification . there is not much difference between the final morphological types and those of seo for giant galaxies ( e , s0 , and sp ) and irregular galaxies . however , about @xmath53 of seo s classification of de - like galaxies are revised , mostly within the category of de - like galaxies . + figure 4 shows representative images of elliptical , lenticular , spiral , and irregular galaxies classified in the present study . the elliptical and lenticular galaxies are displayed in the first and second rows , respectively . spiral galaxies are presented in the third , fourth , and fifth rows , and are grouped according to the bar types ( sa , sab , and sb ) . the spiral galaxies are arranged according to the hubble stage ( a , b , c , d , and m ) with the earliest types ( saa , saba , and sba ) in the leftmost column and the latest types ( sam , sabm , and sbm ) in the rightmost column . the irregular galaxies are presented in the sixth row . the first three columns present i m , di , and ir / bcd galaxies , respectively , whereas the fourth column and the fifth column display a bcd embedded in an irregular envelope and a bcd resembling hii regions , respectively . different scales are used for each stamp image in figure 4 for a better demonstration of their morphology . figure 5 shows images of de - like galaxies . the first and second rows display images of dwarf lenticular galaxies , with normal dwarf lenticulars ( ds0 ) in the first row and dwarf lenticular galaxies with peculiar disk morphology ( ds0@xmath48 ) in the second row . the morphology of ds0 galaxies is characterized by the presence of a lens - like feature in addition to the disk component . lens - like features are normally redder than the disk but some ds0 galaxies have blue lens - like features due to young stellar populations near the nuclei . the galaxies in the third rows are de galaxies . we arranged de@xmath47 galaxies in the first and second columns and de@xmath46 galaxies from third to fifth columns . in the fourth row , we display de@xmath43 galaxies that show intense star formation in the nuclei of the galaxies . the nuclear star formation is thought to be caused by recent gas accretion @xcite . the fifth and sixth rows show dsph and de@xmath2 , respectively the dsph galaxies are generally fainter and bluer than the de galaxies . some de@xmath2 galaxies have large blue cores , the radius of which is approximately half the galaxy size . in addition to these types of galaxies , there are compact elliptical galaxies that are occasionally observed near the centers of clusters . they are smaller than the giant elliptical galaxies but their central surface brightness is similar to that of giant elliptical galaxies . due to their high surface brightness , they are classified as compact elliptical galaxies rather than de galaxies . a prototype of compact elliptical galaxies is m32 in the lg . this section presents a catalog of the visually classified morphological types of 5836 galaxies whose redshifts are less than @xmath19 . the basic statistics of the morphological types of the local galaxies are reported along with a brief comparison with other studies . in addition , the new morphological types revealed from color images are briefly described . the morphological types of 5836 galaxies are presented along with some basic data obtained from the sdss dr7 . table 1 lists a part of the morphological catalog of local galaxies whose full content is available in the electronic version of this paper . the sdss object identification number is used as the name of a galaxy but the primary ned name is presented in the last column for reference . this paper uses the distance of a galaxy derived from the heliocentric redshift corrected for the motion of the lg . we used the prescription of @xcite to derive the velocity relative to the centroid of the lg ( @xmath12 ) . on the other hand , for those galaxies lying inside a @xmath54-cone around m87 with a heliocentric redshift less than z=0.007 @xcite , the distance of the virgo cluster is used . a hubble constant of h=75kms@xmath13 was used . the catalog contains 12 columns with the following information . + + + + + + + corrected for the galactic extinction @xcite . the sdss model magnitude is a magnitude derived by best - fitting model ( exponential profile or de vaucouleurs profile ) to the observed images . + + + + + ccrrrrrrrrcr 587731514215760014 & 17.283075 & 1.121029 & 0.003855 & di & 10 & 17.5 & -14.79 & 0.73 & 44.7 & 0.5 & shoc053 + 588015510351118339 & 18.232364 & 0.981506 & 0.003843 & sabdm & 8 & 17.43 & -18.79 & 1.86 & -9 & -9 & ngc0428 + 588015510351184046 & 18.413968 & 0.875452 & 0.003873 & sm / im & 10 & 17.55 & -13.73 & 0.69 & 26.2 & 0.71 & ugc00772 + 588015510351249607 & 18.584742 & 0.916687 & 0.003717 & di & 10 & 16.92 & -14.77 & 1.05 & 41.4 & 0.5 & sdssj011420.41 + 005503.2 + 588015507666960497 & 18.719523 & -1.095353 & 0.006761 & ds0n & -10 & 29.05 & -16.02 & 1.45 & 40.4 & 0.62 & sdssj011452.68 - 010543.2 + 587743981974650974 & 18.84512 & 8.099593 & 0.007812 & sacd & 6 & 33.24 & -17.09 & 2.39 & 61.4 & 0.74 & ugc00803 + 587731511532060697 & 18.876839 & -0.86098 & 0.005874 & sabcd & 6 & 25.51 & -18.95 & 1.54 & 149.1 & 0.77 & ngc0450 + 587731511532191838 & 19.087214 & -0.911365 & 0.005576 & di & 10 & 24.32 & -15.12 & 1.07 & 26.9 & 0.85 & sdssj011620.92 - 005440.8 + 758877280840910682 & 19.106185 & 33.431996 & -0.000627 & dsphun & -8 & 0.64 & -10.52 & 2.62 & 4.7 & 0.74 & andii + 758877278149148728 & 19.18239 & 16.400064 & 0.006615 & saa & 1 & 28.46 & -18.74 & 1.74 & 88.2 & 0.76 & cgcg459 - 021 + 587727226772258858 & 19.377137 & -9.296768 & 0.006479 & sam & 9 & 27.91 & -16.47 & 1.11 & 73.9 & 0.37 & sdssj011730.51 - 091748.3 + 587727179531747451 & 19.809026 & -9.597428 & 0.006379 & im / bcd & 12 & 27.5 & -14.81 & 1.16 & 29.5 & 0.43 & shoc061 + 587731512606326893 & 20.028347 & -0.205448 & 0.00581 & sdm & 8 & 25.23 & -17.18 & 1.52 & 106.6 & 0.29 & ugc00866 + 588015510351970406 & 20.26232 & 0.851495 & 0.008134 & di & 10 & 34.48 & -15.83 & 1.02 & 24.4 & 0.69 & sdssj012102.95 + 005105.3 + 587744045327319127 & 20.364876 & 7.018498 & 0.007495 & sa & 1 & 31.94 & -19.3 & 1.65 & 111 & 0.39 & ngc0485 + 588015509278359598 & 20.522909 & 0.08518 & 0.007569 & sadm & 8 & 32.23 & -15.84 & 1.58 & 77.8 & 0.14 & fgc0156 + 588015510352101409 & 20.537435 & 0.945465 & 0.007799 & scd & 6 & 33.14 & -19.91 & 1.92 & 310.6 & 0.19 & ngc0493 + 588015508204748937 & 20.810885 & -0.700637 & 0.006625 & sd & 7 & 28.46 & -16.83 & 1.38 & 88.3 & 0.23 & ugc00931 + 587743980364955756 & 20.93191 & 6.691629 & 0.007789 & sabcd & 6 & 33.09 & -17.14 & 1.74 & 50.7 & 0.96 & ugc00942 + rrrrrr number & 86 & 1092 & 286 & 1878 & 2498 + fraction & 0.015 & 0.187 & 0.049 & 0.321 & 0.428 + figure 6 presents the frequency distribution of the broad morphological types of 5836 galaxies . irregular galaxies are most abundant ( @xmath55 ) , and spiral galaxies are second - most abundant ( @xmath56 ) . these two types comprise @xmath57 of the local galaxies . the fractions of elliptical galaxies and lenticular galaxies are @xmath58 and @xmath59 , respectively . the fraction of de - like galaxies is @xmath60 . if we exclude the ds0 galaxies , it becomes @xmath61 . therefore , they are 10 times more frequent than their giant cousins , elliptical galaxies . the local universe is dominated by dwarf galaxies because the majority of irregular galaxies are dwarf galaxies . table 2 lists the basic statistics of the morphological types of galaxies presented in this catalog ( table 1 ) . figure 7 shows the frequency of the hubble stages ( top panel ) and that of the bar family ( bottom panel ) . the hubble stages are grouped into three stages of early ( e ) , intermediate ( m ) , and late ( l ) . the edge - on galaxies are included in the frequency distribution of the hubble stages but they are not included in the frequency distributions of the bar family . the number of spirals increases from the early types to the late types . the fraction of barred galaxies ( sab and sb ) is @xmath62 which is similar to the bar fraction in rc3 @xcite and that derived from the mid - ir images of nearby galaxies @xcite but somewhat higher than those derived from the sdss galaxies at @xmath63 @xcite . the bar fraction from the ellipse fitting technique ( e.g. , barazza et al . 2009 ) is somewhat lower than the visual classifications but that from fourier analysis @xcite is somewhat higher than that in the present study . the difference in the bar fractions is due mostly to the classification method and the criteria for bar selection . the lowest bar fraction from the sdss data is that derived by @xcite who reported a bar fraction of @xmath64 . figure 8 shows the frequency distribution of the subtypes of de - like galaxies ; de , de@xmath43 , dsph , de@xmath2 , and ds0 . as shown in figure 8 , the dominant population of the de - like galaxies is de@xmath2 galaxies , which is distinguished from the others by its blue color . the second largest population is dsph galaxies , but if de@xmath43 galaxies are considered to be a special type of de , then the number of de is similar to that for dsph . the ds0 galaxies include ds0@xmath48 galaxies which show signatures of spiral arm structures in the outer parts . figure 9 shows the frequency distributions of @xmath22 colors for the subtypes of de - like galaxies with those of di for comparisons . the de galaxies show the reddest color with narrow color ranges , whereas the de@xmath2 galaxies show the bluest color similar to that of di galaxies . the dsph galaxies show redder colors than de@xmath43 and de@xmath2 . their average color is similar to that of de galaxies but they have a u - r color distribution extended to blue colors up to @xmath65 . this classification is compared with the automated classification @xcite given in kias - vagc . because they presented broad morphological types ( early and late types ) only , the types in this study were also grouped into early and late types . the early type includes ellipticals ( e ) and lenticulars ( s0 ) with their dwarf cousins ( de , de@xmath43 , dsph , de@xmath2 , and ds0 ) and the late type includes spirals ( sp ) and irregulars ( ir ) . spiral galaxies show the best match ( @xmath66 ) and de - like galaxies show the worst match ( @xmath67 ) between the present classification and that reported by @xcite . ellipticals ( e ) and irregulars ( ir ) also show good matches of @xmath68 , whereas lenticulars show fair matches ( @xmath69 ) . the low match between the present classification and @xcite for de - like galaxies is due mostly to the extremely high mismatch of de ( de@xmath47 and de@xmath46 ) and dsph ( dsph@xmath47 and dsph@xmath46 ) , which show only @xmath70 matches . for both de and dsph , the un - nucleated ones show better matches . the worst match was found for nucleated dwarf spheroidals ( dsph@xmath46 ) which showed only a @xmath71 match . on the other hand , de@xmath43 and de@xmath2 show the @xmath72 matches . the reason for the worse classification of de and dsph galaxies appears to be the small number of dwarf ellipticals in the training sample of @xcite . in the case of de@xmath43 and de@xmath2 galaxies , they included a sufficient number in their training sample . figure 10 compares the morphological types determined in this study with those from ned . we use the numerical type @xmath50 adopted in the present catalog with three additional codes : @xmath50=12 , @xmath50=13 , and @xmath50=14 . the first and second ones are used for the ned i0 galaxies and bcd galaxies , respectively . the last one is assigned to the spiral galaxies with no hubble stage . they are usually given as s ? or sb ? , which suggests an uncertain morphology . in addition to these types , @xmath50=-7 is assigned to de - like galaxies . the circle sizes in figure 10 are proportional to the fractions of the types classified in the present study . as shown in figure 10 , there is good agreement between the present types and those from the ned . on the other hand , there are two types that show a considerable difference . one type is elliptical galaxies ( @xmath50=-5 ) and the other is i0 galaxies ( @xmath50=12 ) . some bias in the classification of s0/a galaxies can be seen . the ned s0/a galaxies ( @xmath50=0 ) are classified mainly as one of the three types , s0 , s0/a , and sa . if the s0/a is highest , as in figure 10 , then the sa fraction would be slightly higher than the others . for elliptical galaxies , approximately half of the ned ellipticals are classified as lenticulars or de - like galaxies in the present study . this is expected because there is frequent confusion between elliptical and lenticular galaxies in the literature . note that most ned lenticular galaxies are classified as lenticular galaxies , whereas ned elliptical galaxies are classified frequently as lenticular galaxies . the confusion between ellipticals and de - like galaxies is also conceivable because they differ only in the surface brightness and its gradient . as shown in figure 10 , ellipticals are more likely to be misclassified as dwarf ellipticals / spheroidals than lenticulars . i0 is a special type of irregular galaxy . it is not so irregular because it is not as disorganized , which is the main characteristics of irregular galaxies . i0 galaxies show a smooth structure similar to lenticular galaxies but have central irregular zones of active star formation @xcite . this type is not included as a separate class because they do not have properties that can be easily distinguishable from other types of galaxies . recent observations show that many elliptical / lenticular galaxies show similar features in their central regions . as expected , most i0 galaxies are classified in the present study as either lenticular galaxies ( @xmath50=-3 ) or irregular galaxies ( @xmath50=10 ) . * in the mid - ir , i0s like ngc 2968 , 3077 , and 5195 are much more regular and the class is no longer applicable @xcite . * a considerable difference between the present types and ned types can also be observed in bcd galaxies ( @xmath50=13 ) . most of our bcds are also classified as bcds in ned . however , the majority of ned bcds are classfied as irregulars ( @xmath50=10 ) . this is due to those ir / bcd galaxies which are considered as irregular galaxies in our classification but most of which are classified as bcds in ned . confusion with irregular galaxies is natural because many irregular galaxies have star burst regions , which is a basic characteristic of bcds . such irregular galaxies are considered to be bcds in the present study only if the star burst regions are large enough to dominate the luminosities of galaxies . the introduction of de@xmath2 galaxies is another cause of the large mismatch between the ned bcds and our types for these galaxies . similar fractions of ned bcds are classified as bcds and de@xmath2 galaxies in our classification . some de@xmath2 galaxies are small enough , i.e. , comparable to hii region - like bcds with similarly blue color with the bcd color . the circles at @xmath50=-7 represent early - type dwarf galaxies . they are not divided into subtypes of de , de@xmath43 , de@xmath2 , dsph , and ds0 because the ned types were not finely defined . most ned early - type dwarfs are classified by us as early - type dwarfs but a small fraction of ned early type dwarfs are classified as irregular galaxies . a better comparison of early - type dwarfs with others is given below ( figure 13 ) . finally , the small circles at @xmath50=14 represent spiral galaxies with no hubble stages in ned . most spiral galaxies we classify as spirals but some are classified as lenticulars , early - type dwarfs , irregulars , and bcds . figure 11 compares the present types with those of rc3 @xcite . we used the same convention of @xmath50 type . as shown in figure 10 and figure 11 , the two comparisons are very similar . this similarity is due to the fact that most of the ned types come from rc3 . for the galaxies in the present catalog , @xmath73 of ned types come from rc3 . the absence of galaxies at @xmath50 ( rc3)=13 is due to the lack of bcd types in rc3 . there are 18 galaxies whose morphological types are classified as peculiar galaxies ( p ) in rc3 . the majority of these galaxies are classified as irregulars ( 7 galaxies ) and bcds ( 5 galaxies ) . the remaining 6 galaxies consist of four spirals , one lenticular galaxy , and one de galaxy . figure 12 shows the comparison of the morphological types with those in the virgo cluster catalog ( binggeli et al . 1985 , hereafter vcc ) . since in vcc , ds0 galaxies are distiguished from other de - like galaxies , we assign @xmath50 ( vcc)=-6 to des and @xmath50 ( vcc)=-10 to ds0 in figure 12 . however , we keep our subtypes of de ( -6 ) , de@xmath43 ( -7 ) , dsph ( -8 ) , de@xmath2 ( -9 ) , ds0 ( -10 ) , and des / di ( -11 ) in figure 12 to see how vcc des distribute in our de - like galaxies . for the types from @xmath50=-5 to @xmath50=14 , we used the same conventions as those in figure 11 . as shown in figure 12 , there is a fairly good agreement between the morphological types determined in this study and those in vcc . however , the degree of agreement is worse than rc3 , especially for the hubble stages of spiral galaxies later than @xmath50 ( vcc)=3 . the best agreement between this study and vcc is observed for lenticular galaxies . elliptical galaxies also give quite good agreement . on the other hand , the greatest discrepancy between this study and vcc is observed for vcc bcds . the majority of vcc bcds are classified as irregulars in this study . there are two dwarf spiral galaxies in vcc . one is ic 3094 , which is given a type of @xmath74ds?@xmath75 . the ned type of this galaxy is @xmath74s ; bcd@xmath75 and we classified it as @xmath74saa@xmath75 . the other is ngc 4192 which is calssified as @xmath74d : sc@xmath75 in vcc . we classified it as @xmath74sacd@xmath75 while ned gives @xmath74sab(s)ab@xmath75 . figure 13 compares the types in this study with those reported by @xcite . these studies ( kn hereafter ) include the updated nearby galaxy catalog . we plot the fractional distributions of the morphological types of galaxies in the kn that are listed in our catalog . our morphological types are grouped into five broad types e , des , s0 , sp , and ir , and we display the fractional frequency distributions of @xmath50 in kn at each panel , which are labeled according to morphological type with the number of galaxies in parentheses . the @xmath50 codes adopted by @xcite , which are slightly different from the @xmath50 codes used in the present catalog , are used for convenience . the elliptical galaxies are presented in the top panel . there are only 16 shared galaxies , 9 of which are classified as ellipticals in kn while all of the other galaxies except one ( mcg-01 - 33 - 007 ) are classified as des or lenticulars . the type @xmath50=-2 and @xmath50=-3 were assigned for de and dsph , respectively , in kn . the second panel shows the fractional distribution of the morphological types of galaxies classified as de - like galaxies in our catalog . our dwarf galaxies designated as des include de , de@xmath43 , dsph , de@xmath2 , and ds0s . here , we present here 131 des that are shared with kn . as shown in the second panel of figure 13 , there is a significant disagreement in the morphological types between ours and theose of kn . among 130 des , 18 galaxies are classified as irregular galaxies and 42 galaxies as spirals , which are mostly late - types of stages dm and m ( @xmath50=8 ) . the third panel from the top in figure 13 shows 48 lenticulars listed in the present catalog , which show a wide range of @xmath50 in @xcite . about @xmath76 of our lenticular galaxies are classified as @xmath50=-1 , which is assumed to designate lenticular galaxies in kn . among the 22 galaxies that were not classified as lenticulars in kn , 4 galaxies are ellipticals and 5 are des . in particular , 10 galaxies that correspond to @xmath77 of the s0 galaxies are classified as early - type spirals of stage @xmath51 and @xmath25 . the fourth panel from the top shows the fractional distribution of the kn morphological types for our spiral galaxies . there is a good agreement between the current classification and those in kn . the fraction of spirals with @xmath78 is @xmath79 , which is in good agreement with that of the present catalog ( @xmath80 ) . as shown in the bottom panel of figure 13 , most of irregular galaxies ( @xmath81 ) are classified as @xmath50=8 , @xmath50=9 , and @xmath50=10 in the kn where late - type spiral galaxies ( dm and m ) , irregulars including bcds , and transition types are classified as @xmath50=8 , @xmath50=9 , and @xmath50=10 , respectively . the absolute @xmath21-band magnitude and @xmath22 color are basic properties of a galaxy . the former is related directly to the mass of a galaxy , which is the most important parameter in the formation and evolution of the galaxy , whereas the latter is closely related to the star formation history . because most of the sample galaxies do not have redshift - independent distances , we use redshift - dependent distances for all of them to calculate @xmath82 . we corrected the motion of the lg following @xcite but the k - correction and evolution correction are not applied in the derivation of @xmath82 because they are negligibly small for local galaxies . figure 14 shows the @xmath22 color versus @xmath82 diagram of the 5836 galaxies listed in the present morphological catalog . some features should be noted . first , galaxies with a different morphology occupy different parts of the color - magnitude diagram . the so called red sequence is well defined from @xmath83 to @xmath84 within which the luminosity - color relationship is well established . the upper part of the red sequence consists of elliptical galaxies and s0 galaxies ( red color ) , whereas the lower part consists of de - like galaxies ( cyan color ) . they overlap around @xmath85 , which suggests that @xmath86 is the bright limit of dwarf galaxies . second , irregular galaxies ( blue color ) display the blue cloud . they are clustered around the mean values of @xmath87 and @xmath88 , respectively . some de - like galaxies are located in the blue cloud . they are mostly de@xmath2 galaxies . third , spiral galaxies are located mostly in the green valley . the majority of spiral galaxies are brighter than the dwarf limit of @xmath86 but a non - negligible fraction of spiral galaxies are fainter than @xmath86 . this is due mainly to the late - type spiral galaxies of the hubble stages d , dm , and m. finally , there are a number of galaxies whose photometry needs to be refined . for example , galaxies located in the far red envelope of the red sequence are probably due to poor photometry , even though some are due to heavy extinction . in addition , @xmath82 of galaxies fainter than @xmath89 are also due mostly to poor photometry . figure 15 shows the fractional frequency distributions indicating the luminosity distribution of local galaxies . the galaxies are grouped according to their broad morphological types : e , des , s0 , sp , and ir . although hii region - like bcd galaxies are supposed to be dwarf galaxies similar to de@xmath2 , we grouped all the bcds as a subtype of the ir class because the majority of bcd components are imbeded in the irregular galaxies . luminosity distributions of galaxies with different morphological types differ significantly . elliptical galaxies show a highly asymmetric distribution that peaks at m@xmath90 , whereas des and irregular galaxies show symmetrical distributions with similar peak magnitudes of m@xmath91 . lenticular galaxies show a narrower distribution than elliptical galaxies with a peak at m@xmath92 . among the three giant galaxies , e , s0 , and sp , spiral galaxies show the most symmetrical distribution with a peak luminosity at m@xmath93 . therefore , spiral galaxies are likely to be faintest among the three types of giant galaxies . it is due to the dominance of late - type spirals which becomes fainter toward the later hubble stages @xcite . figure 16 presents the @xmath22 color distributions of the local galaxies which are sorted according to the broad morphological types . the color distributions depend more strongly on the morphological types than the luminosity distribution . the dependence of the color distribution on the morphological type is expected because the color of a galaxy is a widely used proxy of the galaxy morphology . as shown in figure 16 , elliptical galaxies show the narrowest color distribution centered on @xmath94 . approximately @xmath95 of elliptical galaxies have @xmath22 colors between 2.2 and 3.0 . the blue elliptical galaxies , which comprise @xmath96 of all elliptical galaxies , have a blue tail . lenticular galaxies are also likely to have red colors similar to elliptical galaxies . approximately @xmath69 of lenticulars have @xmath22 colors redder than 2.2 . on the other hand , spiral galaxies are likely to have blue colors . approximately @xmath97 of spiral galaxies have @xmath22 colors bluer than @xmath98 . this makes their color distribution significantly different from that of lenticular galaxies . this shows that the use of the colors of a galaxy as a proxy for the galaxy morphology will make a significant contamination inevitable . the color distribution of de - like galaxies shows a double peak , one at @xmath99 and the other at @xmath100 . the blue peak is caused by the blue colors of de@xmath2 galaxies and the red peak is due to de and dsph galaxies . the irregular galaxies show a completely different color distribution from the others . their peak @xmath22 color is 1.2 , which is 0.4 mag bluer than that of spiral galaxies . only @xmath31 of irregular galaxies are redder than @xmath98 . figure 17 shows the luminosity function of the local galaxies , which extends to @xmath20 , together with the best - fitted schechter luminosity function @xcite of the form @xmath101^{\alpha+1}exp[-10^{-0.4(m - m^{\star})}]dm,\ ] ] where @xmath102 is the characteristic luminosity of a galaxy and @xmath103 is the number density in units of mpc@xmath104 . we determined the three parameters , @xmath105 , @xmath106 , and @xmath107 by minimizing @xmath108 . the drop at @xmath109 is attributed to the observation limit of the sdss spectroscopic observation , which is set to @xmath1 . this confines the volume - limited sample to galaxies brighter than @xmath18 . the rapid drop at @xmath110 is somewhat different from the luminosity functions of galaxies derived from the different volume - limited samples of different limiting magnitudes . @xcite derived the luminosity functions for six volume - limited samples from sdss dr4plus . their slopes are much shallower than the slope of the present luminosity function . the reason for the steeper gradient , i.e. , the smaller @xmath111 in the schechter functions , for the present sample is that the local volume is less dense than the regions outside the local universe . a similar trend for the steeper gradient in the luminosity functions of nearer samples was also observed among the volume - limited samples @xcite . the size of a galaxy can be defined in a variety of ways . we adopted the isophotal semi - major axis length as a representative radius of the local galaxies . isophotal diameters can be affected by galactic extinction and inclination . we suppose that galactic extinction is negligible for the present sample due to the sdss - covered area , however , inclination might affect the isophotal diameter . there have been a number of studies on the effect of inclination on the isophotal diameter @xcite . while @xcite assumed that the change in diameter due to the inclination of a galaxy is negligible , @xcite thought that the inclination effects are need to be corrected . however , as shown by @xcite and @xcite , it depends on the optical nature of galaxies . in the case of spiral galaxies , their disks are believed to be optically thick @xcite , whereas the disks of lenticular galaxies are thought to be optically thin @xcite . the isophotal diameters of optically thin disks increase with inclination . the isophotal radius is calculated from the isophotal semi - major axis of the @xmath21-band image obtained from sdss dr7 . the isophotal semi - major and semi - minor axes are measured at a surface brightness ( @xmath112 ) of 25 mag arcsec@xmath27 at five pass - bands ( @xmath113 ) in the sdss , and are given in units of pixels . the isophotal axis lengths in the pixels are converted to kiloparsecs using a pixel size of 0.396 arcsec and the distances used to calculate the absolute magnitude . this isophotal radius is close to the holmberg radius which is measured at @xmath114 mag arcsec@xmath27 . figure 18 shows the frequency distribution of the isophotal radii of local galaxies sorted according to their broad morphological types : e , des , s0 , sp , and ir . as shown in figure 18 , the distributions of the isophotal radii of galaxies depend strongly on the morphological types of the galaxies . as expected , de - like galaxies show the narrowest distribution , which peaked at @xmath115kpc , and are limited to @xmath116kpc . the ir class of galaxies shows a similar narrow distribution with the same peak but it extends to @xmath117kpc . this is because most ir galaxies are dwarf galaxies with sizes similar to des . some galaxies larger than @xmath118kpc are late - type spiral - like galaxies or peculiar galaxies , some of which are merging galaxies . the ranges of isophotal radii of the ellipticals and lenticulars are similar , even though there are large fluctuations in the ellipticals due to the small number of galaxies . interestingly , the most probable sizes of the lenticulars and spirals are similar to each other ( @xmath119kpc ) whereas the ranges of the distributions are significantly different . the sizes of the largest spirals are about two times larger than those of the lenticulars . the same is true for those elliptical galaxies which show a similar size distribution to that of the lenticular galaxies . the reason for the double size of the spiral galaxies is thought to be the late accretion of the intergalactic material @xcite , which makes the inside - out growth of galaxies similar to that assumed for the formation of the milky way . the basic properties of a galaxy such as mass , luminosity , radius , and morphology are believed to be determined by the galaxy formation process . these parameters are supposed to be correlated with each other because they have the same origin . the radius - luminosity relation @xcite is an example of such a correlation . because the galaxy size depends on the morphology , as shown in figure 18 , the interplay between the size and luminosity is supposed to be more complex . there have been many studies to investigate the morphology dependence of the radius - luminosity relation of galaxies since the pioneering work by @xcite . most of the previous studies used proxies of galaxy morphology rather than morphology itself @xcite . for example , @xcite used the concentration index as a proxy of the morphology . as is well known from previous studies @xcite , proxies are at best a crude approximation of galaxy morphological types . in this regard , @xcite distinguished early - type galaxies and late - type galaxies by visual inspection of the images for an analysis of the structure of central galaxies in groups and clusters . therefore , the morphological types presented in this study will be useful to explore the interplay between the size and luminosity of galaxies . we are planning to do this in a series of forthcoimng papers . we are particularly interested in the radius - luminosity relation of dwarf galaxies . the axial ratio of a galaxy ( @xmath26 ) is closely related to the intrinsic shape of the galaxy , which is determined by its dynamical structure . figure 19 shows the fractional frequency distributions of the axis ratios of 5836 galaxies sorted according to their morphological types . the isophotal semi - major and semi - minor axes in @xmath21-band , which are measured at a surface brightness of 25 mag arcsec@xmath27 , are used . figure 19 plots the three morphological types of giant galaxies ( e , s0 , and sp ) compared to their dwarf cousins ( de and ds0 ) for e and s0 galaxies and irregular galaxies for spiral galaxies . the axis ratio distributions of the ellipticals and de ( left panel ) are similar , whereas a significant difference was observed in the lenticular and dwarf lenticular galaxies ( middle panel ) as well as in spiral galaxies ( right panel ) . the solid lines represent the giant galaxies and the dotted lines indicate the dwarf counterparts . although the general shapes of the axis ratio distributions of the giant elliptical galaxies and de galaxies are similar , there are some differences between the two . dwarf elliptical galaxies have a higher fraction of galaxies with @xmath120 than giant elliptical galaxies , even though the smallest @xmath26 is observed in a giant elliptical galaxy . on the other hand , the morphological type of this galaxy ( ngc 2768 ) is controversial . it is classified as e6 in rc3 , whereas the morphological type adopted by the ned is s0 , which was taken from the carnegie atlas of galaxies @xcite . the more flattened structure of de galaxies is consistent with those observations showing fast rotation in a considerable fraction of de galaxies @xcite . on the contrary , the general similarity of the axis ratio distributions of the giant and dwarf elliptical galaxies suggests that a large fraction of dwarf elliptical galaxies are dispersion - supported systems like giant elliptical galaxies . the dissimilarity of the axis ratio distributions of giant and dwarf lenticular galaxies is quite interesting because ds0s are considered to be low - luminosity versions of s0s @xcite . the axis ratio distribution for ds0s in figure 19 seems to show well that the ds0s should not be thought of as small versions of s0s . rather , they are close to de galaxies because their axis ratio distribution is similar to that of de galaxies . the lack of ds0s with an isophotal axis ratio ( @xmath26 ) less than 0.3 suggests that they are intrinsically thicker than s0s . the lack of small axis ratios is also observed in de and irregulars , suggesting less flattened shapes for these galaxies . the thick intrinsic shape of irregular galaxies is consistent with earlier studies @xcite . the dissimilar distributions of the axis ratios of spiral galaxies and irregular galaxies shown in the right panel of figure 19 are believed to be caused by the real difference in their dynamical structures because the statistical noises are small enough to be ignored due to the large number of spiral and irregular galaxies . in the case of lenticular galaxies , the axis ratio distribution of which is supposed to be similar to that of the spiral galaxies . we can see a large fluctuation in the histogram due to the small number of lenticular galaxies , which is about one - seventh that of the spiral galaxies . the larger fraction of small axis ratios in spiral galaxies is due to the rotating disks of spiral galaxies , the orientations of which are random , whereas the lack of small axis ratios in the irregular galaxies is believed to be due to the triaxial nature of irregular galaxies @xcite . figure 20 shows the fractional distribution of the axis ratios of the dwarf elliptical - like galaxies ; de@xmath43 galaxies in the left panel , dsph galaxies in the middle panel , and de@xmath2 galaxies in the right panel . the axis ratio distribution of de galaxies ( dotted lines ) is also plotted for comparison . due to the small numbers , there are large fluctuations in the histograms but there seem to be some differences among the subtypes of de - like galaxies . the most pronounced difference is observed between de and de@xmath2 galaxies . it is due to the narrow range of axis ratio distribution of de@xmath2 galaxies with a lower limit of @xmath121 . approximately 80@xmath122 of de@xmath2 galaxies have axis ratios between @xmath123 and @xmath124 . this siggests that they are mostly dispersion - supported systems but are not as round as dsph galaxies and dispersion - supported de galaxies . the lack of a systematic difference between de and de@xmath43 is consistent with the morphological similarity between the two types of galaxies , except for the presence of blue cores in de@xmath43 galaxies . the axis ratio distribution of dsph galaxies is similar to that of de galaxies but the lower fraction of galaxies with small @xmath26 appears to be real because they are systematically lower than that of de galaxies for a small axis ratio ( @xmath7 ) . the lack of galaxies with a small @xmath26 is consistent with the kinematic observations of dsph galaxies @xcite . the morphology of dwarf elliptical galaxies is characterized by their round appearance and red colors , similar to that of giant elliptical galaxies . on the other hand , there are some differences in the photometric properties between de and giant ellipticals . for example , de galaxies exhibit lower surface brightness and a shallower luminosity gradient than giant elliptical galaxies . the mean luminosity of de galaxies is @xmath125 , which is approximately 3.5 mag fainter than the mean luminosity ( @xmath126 ) of elliptical galaxies listed in the present catalog . @xcite , @xcite , and @xcite argued that these differences result from the structural difference between de and giant ellipticals . therefore , de galaxies can not simply be a small version of giant elliptical galaxies ( kormendy & bender ( 2012 , and references therein ) . the real dwarf counterparts of giant elliptical galaxies are compact elliptical galaxies , exemplified by m32 @xcite . they have sizes comparable to dwarf elliptical galaxies but have a high surface brightness . on the other hand , @xcite and @xcite agrued that the scaling relations of the dwarf elliptical galaxies are continuations of those of giant elliptical galaxies because the nonlinear relations between the absolute magnitude and the effective surface brightness or effective radius are consequences of the two linear relations between the stellar concentration , quantified by the sersic index @xmath44 , and stellar mass or central surface brightness . their structural characteristics are not distinguished from those of giant elliptical galaxies ( graham 2013 , and references therein ) . whether or not de galaxies are distinct populations , they are thought to be a mixture of kinematically heterogeneous populations @xcite , the origins of which are thought to be different @xcite . the three subtypes of de - like galaxies , de@xmath47 , de@xmath46 and de@xmath43 , have similar structure except for the nuclear morphology . normally , the blue cores of de@xmath43 galaxies are larger than the star - like nuclei in de@xmath46 galaxies but in some cases , they appear small . however , we classify them as de@xmath43 galaxy if their colors are blue enough to assume the presence of young stellar populations . the number of de@xmath43 galaxies that have a small core is negligibly small . the global colors of de@xmath43 galaxies are similar to those of de@xmath47 and de@xmath46 galaxies . however , in some cases , their colors appear quite blue close to de@xmath2 galaxies . the well - known examples of de@xmath43 galaxies are ngc 185 @xcite and ngc 205 @xcite , both of which are satellite galaxies of m 31 . the blue cores of de@xmath43 galaxies are thought to be due to recent star formation . different scenarios have been proposed for the origin of the gas in the center of de@xmath43 galaxies . the gas left in the spiral and irregular galaxies which are transformed into de by ram pressure stripping @xcite or galaxy harassment @xcite is widely considered for the origin of the central star formation of de@xmath43 galaxies in cluster environment @xcite . on the other hand , gas accretion from the cosmic web @xcite can be resposible for the isolated de@xmath43 galaxies such as ic 125 @xcite . in addition , it is possible that small cores such as star - like blue nuclei are made from recycled gas from evolved stars @xcite . the morphology of dsph galaxies is similar to that of dwarf elliptical galaxies . they share a round appearance and red color , lower surface brightness , and shallower gradient in the surface brightness than elliptical galaxies . this is why many studies did not distinguish them @xcite . on the other hand , there are some differences in their morphological properties . on average , they are fainter than de galaxies . they have a mean luminosity of @xmath127 , which is @xmath128 mag fainter than the de galaxies . in addition , they differ in their surface brightness distribution . the surface brightness of dsph galaxies is , on average , fainter than that of de galaxies . also , dsph galaxies show a gentler gradient in their surface brightness distribution than de galaxies . the low luminosity and low surface brightness of dsph galaxies are consistent with the relation between the luminosity and surface brightness of a galaxy @xcite . in addition to the difference in surface brightness between dsph galaxies and de galaxies , there are some properties that are quite different . the presence of a blue core is barely observed in the dsph galaxies , whereas a significant fraction of de galaxies have blue cores . some difference in the axis ratio distributions is present , showing smaller fractions of highly flattened dsph galaxies . because axis ratios , i.e. , ellipticities are closely related to the degree of rotational support @xcite , dsph galaxies are believed to be mostly a dispersion - supported system , whereas de galaxies are considered to be partially supported by rotation . this picture of kinematic properties of dsph galaxies is frequently addressed in previous observations @xcite . on the other hand , the distinction of ellipticity distributions between dsph and de is not as clear as that in the fast and slow rotating elliptical galaxies observed in the atlas @xcite due to the fact that the fractions of rotationally supported des and dispersion - supported des are similar @xcite . however , the differences in photometry as well as in kinematics between the dsph and de galaxies suggest that they have different origins . it is quite likely that dsph galaxies are primodial objects formed at @xmath129 @xcite , while de galaxies , at least fast rotating de galaxies , are stripped late type galaxies . the morphological distinction of de@xmath2 galaxies from other dwarf galaxies is one of the most successful uses of color images in morphology classification . while the blue colors of blue elliptical galaxies @xcite , which distinguish them from normal elliptical galaxies , are believed to be due to recent star formation in the core regions , the colors of the de@xmath2 galaxies are due to global star formation that makes their colors much bluer than the dwarf elliptical galaxies . these galaxies have similar colors to those of irregular galaxies . their shapes , however , are quite similar to those of de galaxies although their axis ratio distribution is somewhat different . de@xmath43 are considered to be much different from the de@xmath2 galaxies because de@xmath43 galaxies have red colors outside the blue core . in some aspects , the de@xmath2 galaxies are similar to hii region - like bcds or di . the former shows somewhat bluer colors than de@xmath2 but with a similar shape , and the latter shows a lack of axial symmetry compared to de@xmath2 . the morphology of dwarf lenticular galaxies ( ds0 ) is similar to their giant cousin , lenticular galaxies ( s0 ) , in the sense that they show subtle structure , such as the bulge and disks in s0s . on the other hand , they are significantly different from s0s in many aspects . the morphology of the central excess of light in dwarf lenticular galaxies appears to be lens - like @xcite and much less massive than a bulge . in some cases , the lens component is quite large in size but their luminosity seems to be low . they are likely to have star - like nuclei , which causes them to be classified as ds0@xmath46 , similar to the nucleated dwarf elliptical galaxies , de@xmath46 . their colors are also similar to dwarf ellipticals . in addition , it is quite rare to have bars inside and there appears to be no inner and outer rings in the dwarf lenticular galaxies . the absence of inner and outer rings in dwarf lenticular galaxies is expected if they are stripped , former late - type galaxies . rings are a phenomenon mainly of massive early - type galaxies . they are virtually never seen among extreme late - type spirals or magellanic irregulars @xcite . on the other hand , they are likely to show disrupted features in the outer parts of the galaxies . these features are probably remnants of spiral arms . if this is the case , dwarf spirals can be formed in disks as small as those of dwarf lenticular galaxies but can not survive long due to the lack of sufficient mass to protect them from tidal or hydrodynamic disturbances . compact galaxies are distinguished from others by their high surface brightnesses which are brighter than 20 mag arcsec@xmath27 @xcite . a bcd galaxy has very blue color indicative of a likely starburst . bcds show a range of morphologies in the present catalog . one extreme case is a bcd whose morphology is characterized by a hii region - like appearance , and the other extreme case is characterized by localized compact regions of intense bursts of star formation within a larger irregular galaxy . examples of the two extreme cases are presented in the two rightmost columns of the bottom row in figure 5 . the hii region - like bcds similar to those observed by @xcite who chose their samples mostly from the earlier objective prism surveys . the shape of these galaxies is similar to the small - sized de@xmath2 galaxies but they have deep bluer colors . because there is no clear color distinction between these two types of galaxies , they can be classified as genuine bcds or isolated extragalactic hii regions @xcite . on the other hand , the selection of small - sized de@xmath2 galaxies from the de@xmath2 sample of galaxies is also subjective . the other types of bcds are basically irregular galaxies with localized compact regions of intense star - bursts . the size of the compact regions of extreme star - burst varies from the size of hii regions to the galaxy size . therefore , it is also subjective to set a size - limit when selecting bcds . in the present study , a galaxy is classified as bcd if the size of the compact regions of extreme star burst is larger than @xmath32 of the galaxy size . in order to distinguish the irregular galaxies with small star burst regions , we assign im / bcd as the morphological type of these galaxies . they are amount to @xmath32 of irregular galaxies . the majority of galaxies in the local universe are dwarf galaxies if we consider irregular galaxies fainter than m@xmath130 to be dwarf galaxies , even though we classify them as i m galaxies because of morphologies resembling the late - type spirals . di galaxies show a range of morphologies from very simple shapes to a high degree of complexity . di with a simple morphology resemble de@xmath2 or bcds . the only difference is the degree of roundness . di galaxies with complex morphologies are caused mainly by sites of intense star formation . they are too small to drive the global density waves . star formation in di galaxies is believed to be caused by self - propagating star formation @xcite . interactions with neighboring galaxies makes their morphology rounder and rounder if the interactions occur frequently . the morphology of some di galaxies show a transition type from di to de / dsph or vice versa . the morphology of a galaxy is dependent on its environment . this dependence has been represented by the morphology - density relation since the pioneering work of @xcite . there are a variety of ways to measure the local background density of a galaxy @xcite . recently , @xcite explored the dependence of disk morphology on the local background density derived from the @xmath44th nearest neighbor method with @xmath131 , @xmath132 , which is defined as @xmath133 where @xmath134 is the projected distance to the fifth nearest galaxy brighter than @xmath18 with @xmath135kms@xmath13 . we normalized @xmath132 by the mean local background density ( @xmath136 ) . figure 21 shows the fractional density distribution of local galaxies sorted by the broad morphological types . as shown in figure 21 , the distributions of the local background density of galaxies with different morphology differ greatly . elliptical galaxies are most frequent in the high - density regions , forming a narrow peak at the high - density regions with an extended tail toward the low - density regions , whereas spiral and irregular galaxies show approximately symmetrical distributions with a peak density at log ( @xmath137 . lenticular and de - like galaxies show similar density distributions to that of elliptical galaxies with a slightly larger fraction of galaxies in the low - density regions than elliptical galaxies . it is worth noting that de - like galaxies do not obey the luminosity - density relation @xcite because it dictates that dwarf galaxies which are basically much less luminous than elliptical and lenticular galaxies are located preferentially in the low - density regions . this means that the morphology of a galaxy , particularly a dwarf galaxy , can be affected greatly by other environmental properties . the interactions with large galaxies to which most of dwarf galaxies are bound as satellite galaxies are considered to be the main cause of the secondary parameter affecting the dwarf galaxy morphology . the distance to the nearest neighbor ( @xmath138 ) appears to be a good measure of the galaxy environment because the effect of interactions depends on the separation between the interacting galaxies . to see the dependence of the galaxy morphology on the local background density , the fraction of galaxy morphological types is plotted as a function of the local background density ( figure 22 ) . in general , they follow the morphology - density relation , i.e. , the fractions of early type galaxies ( e , s0 , and des ) increase with increasing local background density , whereas those of late type galaxies ( sp and ir ) decrease with increasing density . although the general pattern of increasing fractions with increasing local background density is the same for the three classes of early type galaxies , there are some differences between them . fractions of the two early type giant galaxies ( e and s0 ) show an almost flat distributions at log ( @xmath139 and gradual increase thereafter , whereas that of the early type dwarf galaxies ( des ) shows a nearly monotonic increase with increasing density with a sudden change of the slope at log ( @xmath140 . irregular galaxies show a nearly opposite trend from that of des at log ( @xmath141 and a somewhat flat distriobution at log ( @xmath142 with a local minimum around log ( @xmath143 . on the other hand , spiral galaxies demonstrate behavior opposite of the general trend for early - type galaxies , that is , an almost monotonically decreasing spiral fraction with increasing local background density . also , there is not much variation in the fractions of hubble stages and bar types of spiral galaxies along the local background density except for an increasing early spiral fraction and a decreasing late spiral fraction with increasing local background density in high - density regions , log ( @xmath144 . to observe the dwarf morphology dependence on the local background density , we plot the distribution of the morphology fractions of dwarf galaxies as a function of the local background density in figure 23 . a clear difference in the morphology distribution can be seen between the two groups of dwarf galaxies . one group of dwarf galaxies ( de , dsph , and ds0 ) shows similar fractions with an increasing trend for the density ranges of log ( @xmath145 and become different fractions at the higher density regions . the other group ( de@xmath43 and de@xmath2 ) shows dominant fractions at local background densities less than the mean local background density . they comprise more than @xmath95 of the des galaxies at log ( @xmath146 , and decrease with increasing density to become the same fractions as those of the former group at log ( @xmath147 . the predominance of de@xmath2 galaxies in the low - density regions seems to reflect the environmental dependence of the star formation rate observed in the local universe , and higher star formation rate in the low - density regions @xcite , in accord with down - sizing @xcite . the high fraction of star forming dwarfs in the low - density regions is due to the dependence of collapse time ( @xmath148 ) of the pre - galactic cloud on the background density ( @xmath149 ) . in the low - density regions , a large amount of gas is left to make dwarf galaxies in the present epoch , including de@xmath2 galaxies . the preference for de@xmath43 galaxies in low - density regions can be understood if late gas accretion is the main cause of the central star formation in the isolaed de@xmath43 galaxies . although the local background density appears to be the primary parameter dictating the morphology of a galaxy , its role is different for giant galaxies and dwarf galaxies . for giant galaxies , it constrains the galaxy morphology through the star formation efficiency , which depends on the dynamical timescales ( @xmath149 ) at the initial collapse phase and the subsequent hierarchical merging phase of galaxy formation . in contrast , for dwarf galaxies , it constrains the interaction rates with giant galaxies . in cases of dwarf galaxies that are bound to giant galaxies as satellites , the interaction rate is inversely proportional to the orbital period that is determined by the dynamical timescales . therefore , de , dsph , and dwarf lenticulars , which are likely to be located in the high - density regions , are supposed to lose their gas quickly due to the frequent interactions with large galaxies . on the other hand , de@xmath2 galaxies have fewer interactions than other de - like galaxies to retain their gas longer . in the case of the de@xmath43 , the role of local background density seems to be complicated . since de@xmath43 are similar to de galaxies except for the blue core , their red color outside the nucleus suggests early removal of gas by frequent interactions with giant galaxies , but there is leftover gas in their environment to accrete the gas removed from des to make blue core . the morphologies of galaxies are affected by the neighboring galaxies . one of the measures of the influence of the neighbor galaxies is the projected distance , @xmath150 . this can be expressed in absolute units such as mpc , but @xmath151 , which is normalized by the neighbor s virial radius , is a good measure of the local environment because the morphology of a galaxy is affected significantly by the neighbor galaxy when it is located within the neighbor s virial radius @xcite . the virial radius of a galaxy is defined as the radius where the mean density inside the radius becomes the virialized density , which is set to 766 @xmath152 @xcite . the virial radius of a galaxy was calculated according to that described in @xcite . we derived @xmath150 of the 5836 local galaxies using the same neighbor search conditions as those applied to derive the local background density . figure 24 shows the frequency distribution of @xmath150 ( mpc ; upper panel ) and the scatter diagram of @xmath151 and @xmath150 ( mpc ) . the frequency distribution in the upper panel shows a very rapid decrease of number of galaxies with increasing @xmath150 . approximately @xmath153 of the local galaxies have a neighbor galaxy within @xmath154mpc and similar fractions of the local galaxies are located inside the virial radius of the nearest neighbor . owing to the rapid decrease of the frequency distribution of @xmath150 , the number of galaxies that have their nearest neighbors at @xmath155mpc is only 204 ( @xmath156 ) and it is reduced to 66 ( @xmath157 ) for @xmath158mpc . a good correlation is observed between @xmath151 and @xmath150 ( mpc ) with an upper envelope of an almost straight line . the scatter is caused by the wide range of virial radii of galaxies . the upper envelope is due to the smallest dwarf galaxies while the lower envelope is due to the largest giant galaxies . because of the small number of large giant galaxies , the lower envelope is not so straight as the upper envelope . figure 25 shows the fractional distributions of the morphological types as a function of @xmath151 . it is apparent that the majority of local galaxies are located within the virial radius of the nearest neighbor . among the morphological types of giant galaxies , the elliptical galaxies are most likely located within the virial radius of the nearest neighbor , whereas more than half of the spiral and irregular galaxies are located outside the neighbor s virial radius . this means that the effects of hydrodynamical interactions with neighboring galaxies are more important in elliptical galaxies because hydrodynamical interactions are likely to be effective when a galaxy is located within the neighbor s virial radius . lenticular galaxies show an intermediate distribution between elliptical galaxies and spiral galaxies . among the de - like galaxies , de , dsph , and ds0 galaxies show fractional distributions similar to that of lenticular galaxies , whereas de@xmath2 galaxies show a more extended tail toward large @xmath151 . more than @xmath159 of de , dsph , and ds0 galaxies are located at @xmath160 . the de@xmath43 galaxies show intermediate distribution between de and de@xmath2 . an examination of the morphology fractions of the 5 broad morphological types as a function of @xmath151 shows similar trends to those observed in figure 22 where the morphological fractions are plotted as a function of the local background density . that is , the three early types ( e , s0 , and des ) show increasing fractions with decreasing @xmath151 and the two late types ( sp and ir ) show the opposite trend . figure 26 shows the early type fractions of the target galaxies as a function of @xmath151 . the early - type fractions of the target galaxies increase with decreasing @xmath151 . however , the increasing rates depend on the neighbor s morphology . here , early type represents three broad types of galaxy , e , s0 , and des , whereas late type represents spiral and irregular galaxies . the early type fractions increase rapidly with decreasing @xmath151 when the neighbor galaxy is the early - type galaxy . whereas early - type fractions increase slowly with decreasing @xmath151 when the neighbor galaxy is a late type . this makes the early - type fraction of the target galaxies with the early - type neighbor larger than @xmath161 when they are located in the neighbor s virial radius , whereas that of the target galaxies with the late type neighbor becomes @xmath162 . this phenomenon was interpreted as morphological conformity between close neighbors @xcite . the gradual increase in the early - type fraction of the target galaxies with late - type neighbor when they come closer to the neighbor suggests that the effect of the tidal interactions dominates other interactions such as hydrodynamical interactions for gas transfer from the late - type neighbor to the target galaxy . in the case of early - type neighbors , hydrodynamical interactions enhance the effect of tidal interactions by removing gas to produce a rapid increase in the early type fraction when they become closer to the neighbor . this paper presents a catalog of the morphological types of 5836 galaxies whose redshifts are less than @xmath19 . the morphological types are determined by a visual inspection of the color images provided by sdss dr7 . the majority of galaxies in the present sample come from the kias - vagc which is based on the spectroscopic target galaxies of the sdss dr7 complemented by the bright galaxies with known redshifts from various catalogs . galaxies fainter than the limiting magnitude of the sdss spectroscopic target galaxies were also included if redshifts are available in the ned . the number of galaxies fainter than @xmath1 is 809 which is approximately @xmath163 of the present sample . therefore , the present sample is not flux - limited but a flux - limited sample can be defined if galaxies brighter than @xmath1 are used because galaxies brighter than the sdss bright limiting magnitude of @xmath164 are believed to be compiled almost completely in the ned . the galaxy morphology classification system adopted for bright galaxies in the present study is basically the de vaucouleurs revised hubble - sandage system of stages and families @xcite . inner and outer ring varieties , lenses and stages of e and s0 galaxies are not recognized in our catalog . for dwarf galaxies , we use notation similar to @xcite , modified to allow for features , such as blue cores or blue colors , that are recognizable in color images . that is , we distinguish subtypes ; de , de@xmath43 , dsph , de@xmath2 , and dwarf lenticulars ( ds0 ) . the de , dsph and ds0 galaxies are further divided into de@xmath47 , de@xmath46 , dsph@xmath47 , dsph@xmath46 , ds0@xmath47 , and ds0@xmath46 to denote the presence of nucleation . in the analysis of the frequency distributions , however , we do not distinguish them because physical properties , such as the color and luminosity are similar , and the numbers of de@xmath47 and dsph@xmath47 galaxies are much smaller than those of de@xmath46 and dsph@xmath46 galaxies . the present catalog contains 85 elliptical galaxies ( e : 1.5% ) , 1093 dwarf elliptical - like galaxies ( des : 18.7% ) , 286 lenticular galaxies ( s0 : 4.9% ) , 1874 spiral galaxies ( sp : 32.1% ) , and 2498 irregular galaxies ( ir : 42.8% ) . since most irregular galaxies are dwarf galaxies , the local universe is dominated by dwarf galaxies . among giant galaxies , spiral galaxies dominate the local universe . the late - type spirals are more frequent than the early - type spirals and @xmath165 of spiral galaxies have bars including weak bars . a significant number of de@xmath2 galaxies , which are generally smaller than the ordinary de galaxies , are included in the present catalog . some de@xmath2 galaxies have large cores that are somewhat bluer than the global colors . most de@xmath2 galaxies without cores are quite compact , similar to hii region - like bcds . the only difference between these two types is the degree of star formation . hii region - like bcds are bluer than de@xmath2 galaxies . a considerable number of di galaxies show somewhat round shapes with size , luminosity , and color similar to de@xmath2 galaxies . their environments , represneted by the local background density , are also similar . therefore , we suppose that de@xmath2 galaxies are the same populations as di galaxies both of which served as the building blocks of giant galaxies . some evidence exists for dichotomy in the morphological properties of de galaxies and dsph galaxies , which is in contrast to the conventional practice to treat them as the same class of galaxies @xcite . the dichotomy is not as clear as that between the giant elliptical galaxies and the de galaxies in terms of size and luminosity . however , they are different in the surface brightness and color . the colors of the de galaxies are similar to those of giant elliptical galaxies , whereas they differ appreciably from dsph galaxies . the surface brightness and surface brightness gradient of dwarf elliptical galaxies are intermediate between the giant elliptical galaxies and dwarf spheroidal galaxies . if we consider the kinematic difference between dsph and de galaxies @xcite , the origins of dsph and de galaxies thought to be different , at least for the disperion - supported ds0 galaxies and the fast rotating de galaxies , dispersion - supported dsph galaxies can be promodial objects @xcite while fast rotating de galaxies are stripped late type spirals @xcite . we explored the physical properties , such as luminosity , color , size , and axial ratios , of the local galaxies . in general , distributions of the physical parameters of galaxies with different morphological types are quite different in terms of the most probable values and the shapes of their distributions . elliptical galaxies show the highest luminosity but their maximum size is smaller than that of spiral galaxies . the larger size of spiral galaxies is supposed to be due to the accretion of the intergalactic material after the collapse phase of galaxy formation . the luminosity and size distributions of the two types of dwarf galaxies , des and dir , are very similar , implying similar masses . the dependence of galaxy size and luminosity on galaxy morphology seems to have originated from the morphology - density relation because the local background density determines the dynamical timescales ( @xmath149 ) , which is closely related to the star formation rates . elliptical and lenticular galaxies that are likely to be formed in dense environment are assumed to build up their bodies rapidly , due to short dynamical times , without a late accretion phase , whereas spiral galaxies build up their outer parts by accreting intergalactic material after the initial collapse phase . this picture of spiral galaxy formation is in good agreement with the inside - out growth picture of spiral galaxies . although the local background density appears to be the primary parameter dictating the morphology of a galaxy , its role is different for giant galaxies and dwarf galaxies . for giant galaxies , the local background density constrains the galaxy morphology through the star formation efficiency in the early phase of galaxy formation . in contrast , the local background density constrains the interaction rates between the host galaxies and their satellites , which are mostly dwarf galaxies because interaction rates are closely related to the orbital period that is determined by the dynamical timescales . therefore , we suppose that des , dsphs , and dwarf lenticulars , which are likely to be observed in the high - density regions , lose their gas quickly due to frequent interactions with large galaxies . on the other hand , de@xmath2 galaxies must have fewer interactions than other dwarf galaxies to keep their gas longer . in the case of the de@xmath43 , the role of local background density is twofold . it is high enough for frequent interactions to remove gas from the galaxy but it allows delayed accretion of gas to the nucleus of the de galaxy to make the blue core . most of the local galaxies have nearby companion galaxies . the mean projected distance to the nearest neighbor @xmath166 is @xmath8kpc , which is similar to the virial radius of a bright spiral galaxy . approximately @xmath167 of the local galaxies have the nearest neighbor within @xmath168 . on the other hand , a considerable number of galaxies are extremely isolated . the number of extremely isolated galaxies depends on the isolation criteria . if @xmath169mpc is taken as the minimum projected separation between a target galaxy and its nearest neighbor , the number of extremely isolated galaxies is 113 but it becomes 66 if we consider @xmath158mpc as the criterion for the extreme isolation . abazajian , k. n. , adelman - mccarthy , j. k. , agueros , m. a. , et al . 2009 , apjs , 182 , 543 abraham , r. g. , valdes , f. , yee , h. k. c. , & van den bergh , sidney , 1994 , , 432 , 75 abraham , r. g. , van den bergh , sidney , & nair , p. 2003 , , 588 , 218 aguerri , j. l. a. , iglesias - paramo , j. , vilchez , j. m. , munoz - tunon , c. , & sanchez - janssen , r. 2005 , , 130 , 475 ann , h. b. , park , c. , & choi , y .- y . 20008 , , 389 , 86 ann , h. b. 2014 , journal of the korean astronomical society , 47 , 1 . barazza , f. d. , jablonka , p. , desai , v. , et al . 2009 , astronomy and astrophysics , 497 , 713 bernardi , m. , sheth , r. k. , annis , j. , et al . 2003 , aj , 125 , 1849 bernardi , m. , meert , a. , vikram , v. , et al . 2014 , , 443 , 874 binggeli , b. , sandage , a. , & tammann , g. a. 1985 , , 90 , 1681 binney , j. 1978 , , 183 , 501 binney , j. , & de vaucouleurs , g. 1981 , , 194 , 679 boselli , a. , boissier , s. , cortese , l. , & gavazzi , g. 2008 , , 674 , 742 buta , r. j. , & combes , f. 1996 , fund . cosmic physics , 17 , 95 buta , r. j. , alpert , a. j. , cobb , m. l. , croker , d. a. , & purcell , g. b. 1998 , , 116 , 1142 buta , r. j. 2013 , planets , stars and stellar systems vol . oswalt , terry d. , & keel , william c. ( new york : springer science+business media dordrecht ) buta , r. j. , corwin , h. g. , odewahn , s. c. , & dejaiffe , r. 2007 , the de vaucouleurs atlas of galaxies ( cambridge : cambridge university press ) buta , r. j. , sheth , k. , athanassoula , e. , et al . 2015 , arxiv:150100454 cheng , j. y. , faber , s. m. , simard , l. , graves , g. j. , lopez , e. d. , yan , r. , & cooper , m. c. 2011 , , 412 , 727 chiboucas , k. , karachentsev , i. d. , & tully , r. b. 2009 , , 137 , 3009 cibinel , a. , carollo , c. m. , lilly , s. j. , et al . 2013 , apj , 776 , 72 choi , y .- y . , han , d .- h . , & kim , s. s. 2010 , journal of the korean astronomical society , , 43 , 191 choi , y .- y . , park , c. , & vogeley , m. s. 2007 , , 658 , 884 colless , m. et al . 2001 , , 328 , 1039 combes , f. 2014 , aspc , 480 , 211 conselice , c. j. 2003 , , 147 , 1 cowie , l. l. , songaila , a. , hu , e. , & cohen , j. g. 1996 , , 112 , 839 de vaucouleurs , g. 1959 , handbuch der physik , volume 53 , p.275 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , buta , r. j. , paturel , g. , & fouque , p. 1991 , third reference catalogue of bright galaxies ( new york : springer ) disney , m. , davies , j. , & phillipps , s. 1989 , , 239 , 939 dressler , a. 1980 , , 236 , 351 ferguson , h. c. , & binggeli , b. 1994 , astronomy and astrophysics review , 6 , 67 geha , m. , van der marel , r. p. , guhathakurta , p. , gilbert , k. w. , kalirai , j. , & kirby , e. n. 2010 , , 711 , 361 gerola , h. , & seiden , p. e. 1978 , , 223 , 129 gil de paz a. , madore b. f. , & pevunova , o. 2003 , , 147 , 29 gilmore , g. , wilkinson , m. i. , wyse , r. f. g. , kleyna , j. t. , koch , a. , evans , n. w. , & grebel , e. k. 2007 , , 663 , 948 gomez et al . 2003 , , 584 , 210 goto , t. , yamauchi , c. , fujita , y. , okamura , s. , sekiguchi , m. , smail , i. , bernardi , m. , & gomez , p. l. 2003 , , 346 , 601 graham , a. w. , & guzman r , . 2003 , , 125 , 2936 graham , a. w. 2013 , in planets , stars and stellar systems , vol . 6 , p91 - 140 , eds t. d. oswalt & w. c. keel ( springer : berlin ) ( arxiv:1108.0997 ) gu , q. , zhao , y. , shi , l. , peng , z. , & luo , x. 2006 , , 131 , 806 gunn , james e. , & gott , j. richard , iii , 1972 . , 176 , 1 guo , y. , mcintosh , d. h. , mo , h. j. , katz , n. , van den bosch , f. c. , weinberg , m. , weinmann , s. m. , pasquali , a. , yang , x. 2009 , , 398 , 1129 hallenbeck , g. , papastergis , e. , huang , s. , haynes , m. p. , giovanelli , r. , boselli , a. , boissier , s. , heinis , s. , cortese , l. , & fabello , s. 2012 , , 144 , 87 heidmann , j. , heidmann , n. , & de vaucouleurs , g. 1972 , memoirs of the royal astronomical society , 75 , 85 hodge , p. w. 1963 , , 68 , 691 hodge , p. w. 1973 , , 182 , 671 holmberg , e. 1958 , lund medd astron .. obs , ser . ii , 136 , 1 hubble , e. 1926 , , 64 , 321 hubble , e. 1936 , realm of the nebulae ( new haven : yale university press ) huertas - company , m. , rouan , d. , tasca , l. , soucail , g. , & le fevre , o. 2008 , , 478 , 971 jones , h. , davies , j. i. , & trewhella , m. 1996 , , 283 , 316 jerjen , h. , & binggeli , b. 1997 , in asp conf . 116 , the nature of elliptical galaxies ; 2nd stromlo symposium , ed . m. arnaboldi , g. s. da costa , and p. saha ( san francisco , ca : asp ) , 239 karachentsev , i. d. , nasonova , o. g. , & helene m. 2011 , , 743 , 123 karachentsev , i. d. , makarov , d. i. , & kaisina , e. i. 2013 , , 141 , 101 karachentsev , i. d. , & nasonova , o. g. 2013 , , 429 , 2677 karachentsev , i. d. , nasonova , o. g. , & courtois , h. m. 2013 , , 429 , 2264 karachentsev , i. d. , karachentseva , v. e. , & nasonova , o. g. 2014 , astro - ph1406.2467 kelly , brandon c. ; mckay , timothy a. 2004 , , 127 , 625 kim , e. , kim , m. , hwang , n. , lee , m. g. , chun , m .- y . , & ann , h. b. 2011 , , 411 , 1881 kormendy , j. 1985 , , 295 , 73 kormendy , j. , & freeman , k. c. 2014 , 2014arxiv1411.2170k ( arxiv : 1411 2170 ) kormendy , j. , & bender , r. 2012 , 198 , 2 kormendy , j. , fisher , d , b. , cornell , m. e. , & bender , r. 2009 , , 182 , 216 kraan - korteweg , r. c. 1986 , astronomy and astrophysics supplement series , 66 , 255 kraljic , k. , bournaud , f. , & martig , m. 2012 , , 757 , 60 kreckel , k. , peebles , p. j. e. , van gorkom , j. h. , van de weygaert , r. , & van der hulst , j. m. 2011 , , 141 , 204 kunth , d. & ostlin , g. 2000 , the astronomy and astrophysics review , 10 , 1 larson , r. b. 1976 , , 176 , 31 lee , g .- h . , woo , j .- h . , lee , m. g. , hwang , h. s. , lee , j. c. , sohn , j. , & lee , j. h. 2012 , , 750 , 141 lewis , i. et al . 2002 , , 334 , 673 lintott , c. j. , et al . , 2008 , , 389 , 1179 lisker t. , glatt k. , westera p. , & grebel e. k. 2006 , , 132 , 2432 lisker t. , grebel e. k. , binggeli b. , & glatt k. 2007 , , 660 , 1186 lisker , t. 2009 , astronomische nachrichten , 330 , 1043 lisker , t. , weinmann , s. m. , janz , j. , & meyer , h. t. 2013 , , 432 , 1162 lotz , j. m. , primack , j. , & madau , p. 2004 , , 128 , 163 lupton , r. , blanton , m. r. , fekete , g. , hogg , d. w. , omullane , w. , szalay , a. , & wherry , n. 2004 , , 116 , 133 makarov , d. , & karachentsev , i. 2011 , , 412 , 2498 matteucci f. , & francois p. 1989 , , 239 , 885 mcconnachie , a. w. 2012 , , 144 , 4 mcintosh , d. h. et al . 2005 , , 632 , 191 michard , r. , & simien , f. 1993 , 274 , 25 michard , r. 1999 , , 137 , 245 mcintosh , d. t. et al . , 2005 , , 632 , 191 moore b. , katz n. , lake g. , dressler a. , & oemler a. 1996 , nature , 379 , 613 moore b. , lake g. , & katz n. 1998 , , 495 , 139 mould , j. r. , huchra , j. p. , freedman , w. l. , et al . 2000 , apj , 529 , 786 muldrew , s. i. , croton , d. j. , skibba , r. a. , et al . 2012 , mnras , 419 , 2670 nair , p.b . , & abraham , r.g . , 2010 , , 186 , 427 naab , t. , & ostriker , j. p. 2006 , , 366 , 899 oh , s. , oh , k. , & yi , s. k. 2012 , , 198 , 40 pak , m. , rey , s .- c . , lisker , t. , lee , y. , kim , s. , sung , e .- c . , jerjen , h. , chung , j. 2014 , , 445 , 630 park , c. , & choi , y .- y . 2005 , , 635 , l29 park , c. , choi , y .- t . , vogeley , m. c. , gott , j. , r. , & blanton , m. 2007 , , 658 , 898 park , c. , gott , j. , r. , & choi , y .- y . 2008 , , 674 , 784 park , c. , & choi , y .- y . 2009 , , 691 , 1828 2002 , , 332 , 59 peppel , u. c. , drapatz , s. , ghosh , s. k. 1986 , mitteilungen der astronomischen gesellschaft , 67 , 392 roychowdhury , s. , chengalur , j. n. , begum , a. , karachentsev , i. d. 2010 , , 404 , 60 salucci , p. , wilkinson , m. i. , walker , m. g. , gilmore , g. f. , grebel , e. k. , koch , a. , frigerio miatins , c. , & wyse , r. f. g. 2012 , , 420 , 2034 sancisi , r. , fraternali , f. , oosterloo , t. , van der hulst , t. 2008 , the astronomy and astrophysics review , 15 , 189 sandage , a. r. 1961 , the hubble atlas of galaxies ( washington , dc : carnegie institution of washington ) sandage , a. , & bedke , j. 1994 , the carnegie atlas of galaxies ( washington , dc : carnegie institution of washington ) sandage , a. , & binggeli , b. 1984 , , 89 , 919 sandage , a. , & tammann , g. a. 1981 , a revised shapley - ames catalog of bright galaxies ( washington , dc : carnegie inst . ) sargent , w. l. w. , & searle , l. 1970 , , 162 , l155 scarlata , c. , carollo , c. m. , lilly , s. , et al . 2007 , apjs , 172 , 406 strom , s. e. 1980 , apj , 237 , 686 suh , h. , jeong , h. , oh , k. , yi , s. k. , ferreras , i. , & schawinski , k. 2010 , , 187 , 374 tanaka , m. , goto , t. , okamura , s. , shimasaku , k. , brinkmann , j. 2004 , , 128 , 2677 thuan , t. x. , & martin , g. e. 1981 , , 247 , 823 toloba , e. , boselli , a. , peletier , r. f. , falcon - barroso , j. , van de ven , g. , & gorgas , j. 2012 , , 548 , 78 van den bergh , s. 1960 , , 131 , 215 van den bergh , s. 1976 , , 206 , 883 van den bergh , s. 1988 , pasp , 100 , 344 van der hulst , t. , & sancisi , r. 2004 , in iau symp . recycling intergalactic and interstellar matter , ed . m. arnaboldi , g. s. da costa , o. saha ( san fransico , ca : asp ) , 122 walker , m. g .. mateo , m. , olszewski , e. w. , penarrubia , j. , wyn e. n. , gilmore , g. 2009 , , 704 , 1274 walker , m. g. 2013 , planets , stars and stellar systems vol . 5 , 1039 ( new york : springer ) weijmans , , de zeeuw , p. t. , emsellem , e. , et al . 2014 , mnras , 444 , 3340

different aspects of nonclassical properties of electromagnetic field have been studied since the advent of quantum optics . however , the interest on nonclassical states has been considerably escalated with the progress of interdisciplinary field of quantum computation and quantum communication in recent past as a large number of applications of nonclassical states have been reported in context of quantum computation and quantum communication @xcite . specifically , it is shown that squeezed states can be used for the implementation of continuous variable quantum cryptography @xcite and teleportation of coherent states @xcite , antibunched states can be used to build single photon sources @xcite , entangled states are essential for the implementation of a set of protocols of discrete @xcite and continuous variable quantum cryptography @xcite , quantum teleportation @xcite , dense - coding @xcite , etc . , states violating bell s inequality are useful for the implementation of protocols of device independent quantum key distribution @xcite . study of the possibility of generation of nonclassical states ( specially , entanglement and nonlocal characters of quantum states ) in different quantum systems have recently become extremely relevant and important for the researchers working in different aspects of quantum information theory and quantum optics . existence of entanglement and other nonclassical states in a large number of bosonic systems have already been reported ( see @xcite and references therein ) . however , it is still interesting to study the possibility of generation of nonclassical states in experimentally realizable simple systems . a specific system of this kind is a nonlinear optical coupler which can be easily realized using optical fibers or photonic crystals . optical couplers are interesting for several reasons . firstly , in a coupler the amount of nonclassicality can be controlled by controlling the interaction length and the coupling constants . further , optical couplers are of specific interest as recently matthews et al . have experimentally demonstrated manipulation of multiphoton entanglement in quantum circuits constructed using optical couplers @xcite , mandal and midda have shown that nand gate ( thus , in principle a classical computer ) can be build using nonlinear optical couplers @xcite . motivated by these facts we aim to systematically investigate the possibility of observing nonclassicality in nonlinear optical couplers . as a first effort we have reported lower order and higher order entanglement and other higher order nonclassical effects in an asymmetric codirectional nonlinear optical coupler which is prepared by combining a linear waveguide and a nonlinear ( quadratic ) waveguide operated by second harmonic generation @xcite . waveguides interact with each other through evanescent wave . extending the earlier investigation @xcite here we study a similar asymmetric nonlinear optical coupler for contradirectional propagation of fields . this type of contradirectional asymmetric optical coupler was studied earlier by some of us @xcite . however , in the earlier studies @xcite intermodal entanglement and some of the higher order nonclassical properties studied here were not studied . further , in those early studies second - harmonic mode ( @xmath7 was assumed to be pumped with a strong coherent beam . in other words , @xmath8 mode was assumed to be classical and thus it was beyond the scope of the previous studies to investigate single - mode and intermodal nonclassicalities involving this mode . interestingly , completely quantum mechanical treatment adopted in the present work is found to show intermodal squeezing in a compound mode involving @xmath8 mode . in addition , conventional short - length solutions of heisenberg s equations of motion were used in earlier studies @xcite , but recently it is established by some of us that using improved perturbative solutions obtained by sen - mandal approach @xcite we can observe several nonclassical characters that are not observed using short - length ( or short - time approach ) @xcite . keeping these facts in mind present paper aims to study nonclassical properties of this contradirectional asymmetric nonlinear optical coupler with specific attention to entanglement using perturbative solution obtained by sen - mandal technique . nonclassical properties of different optical couplers are extensively studied in past ( see @xcite for a review ) . specially , signatures of nonclassicality in terms of photon statistics , phase properties and squeezing were investigated in codirectional and contradirectional kerr nonlinear couplers having fixed and varying linear coupling constant @xcite , raman and brillouin coupler @xcite and parametric coupler @xcite , asymmetric @xcite and symmetric @xcite directional nonlinear coupler etc . here it would be apt to note that the specific coupler system that we wish to study in the present paper has already been investigated @xcite for contradirectional propagation of classical ( coherent ) input modes and it is shown that depth of nonclassicality may be controlled by varying the phase mismatching @xmath9 @xcite . existing studies are restricted to the investigation of lower order nonclassical effects ( e.g. , squeezing and antibunching ) under the conventional short - length approximation . only a few discrete efforts have recently been made to study higher order nonclassical effects and entanglement in optical couplers @xcite , but except a recent study on codirectional nonlinear optical coupler reported by us @xcite , all the other efforts were limited to kerr nonlinear optical coupler . for example , in 2004 , leonski and miranowicz reported entanglement in kerr nonlinear optical coupler @xcite and pumped kerr nonlinear optical coupler @xcite , subsequently entanglement sudden death @xcite and thermally induced entanglement @xcite were reported in the same system . amplitude squared ( higher order ) squeezing was also reported in kerr nonlinear optical coupler @xcite . however , no effort has yet been made to rigorously study the higher order nonclassical effects and entanglement in contradirectional nonlinear optical coupler . keeping these facts in mind in the present letter we aim to study nonclassical effects ( including higher order nonclassicality and entanglement ) in contradirectional nonlinear optical coupler . remaining part of the paper is organized as follows . in section [ sec : the - model - and ] , the model momentum operator that represents the asymmetric nonlinear optical coupler is described and perturbative solutions of equations of motion corresponding to different field modes present in the momentum operator are reported . in section [ sec : criteria - of - nonclassicalities ] , we briefly describe a set of criteria of nonclassicality . in section [ sec : nonclassicality - in - contradirectional ] the criteria described in the previous section are used to investigate the existence of different nonclassical characters ( e.g. , lower order and higher order squeezing , antibunching , and entanglement ) in various field modes present in the contradirectional asymmetric nonlinear optical coupler . finally , section [ sec : conclusions ] is dedicated for conclusions . a schematic diagram of a contradirectional asymmetric nonlinear optical coupler is shown in fig . [ fig : schematic diagram ] . from fig . [ fig : schematic diagram ] one can easily observe that a linear waveguide is combined with a nonlinear @xmath10 waveguide to constitute the asymmetric coupler of our interest . further , from fig . [ fig : schematic diagram ] we can observe that in the linear waveguide field propagates in a direction opposite to the propagation direction of the field in the nonlinear waveguide . electromagnetic field characterized by the bosonic field annihilation ( creation ) operator @xmath11 propagates through the linear waveguide . similarly , the field operators @xmath12 corresponds to the nonlinear medium . specifically , @xmath13 and @xmath14 denote annihilation operators ( wave vectors ) for fundamental and second harmonic modes , respectively . now the momentum operator for contradirectional optical coupler is @xcite@xmath15 where @xmath16 stands for the hermitian conjugate and @xmath17 represents the phase mismatch between the fundamental and second harmonic beams . the linear ( nonlinear ) coupling constant , proportional to susceptibility @xmath18 @xmath19 , is denoted by the parameter @xmath20 @xmath21 . it is reasonable to assume @xmath22 as in a real physical system we usually obtain @xmath23 . as a consequence , in absence of a highly strong pump @xmath24 . earlier this model of contradirectional optical coupler was investigated by some of the present authors ( @xcite and references therein ) . using ( [ eq:1 ] ) and the procedure described in @xcite we can obtain the coupled differential equations for three different modes as follows @xmath25 here it would be apt to mention that momentum operator for contradirectional asymmetric nonlinear coupler ( [ eq:1 ] ) is same as that of codirectional asymmetric nonlinear optical coupler @xcite . however , for the contradirectional couplers the sign of derivative in the heisenberg s equation of motion of the contra - propagating mode is changed ( i.e. , in the present case @xmath26 is replaced by @xmath27 as mode @xmath28 is considered here as the contra - propagating mode ) . the method used here to obtain ( [ eq : aaa ] ) is described in refs . @xcite . further , this particular description of contradirectional coupler is valid only for the situation when the forward propagating waves reach @xmath29 and the counter ( backward ) propagating wave reach @xmath30 . thus the coupled equations described by ( [ eq : aaa ] ) and their solution obtained below are not valid for @xmath31 @xcite . earlier these coupled equations ( [ eq : aaa ] ) were solved under short - length approximation . here we aim to obtain perturbative solution for these equations using sen - mandal method which is already shown to be useful in detecting nonclassical characters not identified by short - length solution @xcite . keeping this in mind , we plan to solve ( [ eq : aaa ] ) using sen - mandal approach . ) with a nonlinear ( @xmath32 ) waveguide operated by second harmonic generation . the fields involved are described by the corresponding annihilation operators , as shown ; @xmath33 is the interaction length . ] using sen - mandal approach we have obtained closed form perturbative analytic solutions of ( [ eq : aaa ] ) as@xmath34 with@xmath35\sinh|k|l\right\ } , \\ h_{4 } & = & \frac{c^{*}|k|k}{k^{*}}f_{1}\left\ { 2|k|^{2}f_{1}g_{-}^{*}+\delta k\sinh|k|l\left(g_{+}^{*}-1\right)\left[2i|k|+\delta k\tanh|k|l\right]\right\ } , \end{array}\label{eq : soln}\ ] ] where @xmath36 and @xmath37 . the solution obtained above is verified by escr ( equal space commutation relation ) which implies @xmath38=\left[b_{1}\left(l\right),b_{1}^{\dagger}\left(l\right)\right]=\left[b_{2}\left(l\right),b_{2}^{\dagger}\left(l\right)\right]=1 $ ] while all other equal space commutations are zero . further , we have verified that the solutions reported here satisfies constant of motion . to be precise , in ref . @xcite , it was shown that the constant of motion for the present system leads to @xmath39 here we have verified that the solution proposed here satisfies ( [ eq : const - of - motion ] ) . we have also verified that the solution of contradirectional coupler using short - length solution method reported in @xcite can be obtained as a special case of the present solution . specifically , to obtain the short - length solution we need to expand the trigonometric functions present in the above solution and neglect all the terms beyond quadratic powers of @xmath33 and consider phase mismatch @xmath40 . after doing so we obtain@xmath41 which coincides with the short - length solution reported earlier by some of the present authors @xcite . clearly the solution obtained here is valid and more general than the conventional short - length solution as the solution reported here is fully quantum solution and is valid for any length , restricting the coupling constant only . further , in case of codirectional optical coupler we have already seen that several nonclassical phenomena not identified by short - length solution are identified by the perturbative solution obtained by sen - mandal method @xcite . keeping this in mind , in what follows we investigate nonclassical characters of the fields that have propagated through a contradirectional asymmetric nonlinear optical coupler . as the criteria for obtaining signatures of different nonclassical phenomena are expressed in terms of expectation values of functions of annihilation and creation operators of various modes , we can safely state that eqs . ( [ 2 ] ) and ( [ eq : soln ] ) provide us sufficient resource for the investigation of the nonclassical phenomena . to illustrate this point we may note that the criteria for quadrature squeezing in single - mode ( @xmath42 and compound mode ( @xmath43 are @xcite @xmath44 and @xmath45 where @xmath46 and the quadrature operators are defined as @xmath47 and @xmath48 similarly , the existence of single- and multi - mode nonclassical ( sub - poissonian ) photon statistics can be obtained through the following inequalities@xmath49 and @xmath50 where ( [ eq : antib-1 ] ) provides us the condition for single - mode antibunching and ( [ eq : antib2 ] ) provides us the condition for intermodal antibunching . many of the early investigations on nonclassicality were limited to the study of squeezing and antibunching only , but with the recent development of quantum computing and quantum information it has become very relevant to study entanglement . interestingly , there exist a large number of inseparability criteria ( @xcite and references therein ) that can be expressed in terms of expectation values of moments of field operators . for example , hillery - zubairy criterion i and ii ( hz - i and hz - ii ) @xcite are described as @xmath51 and @xmath52 respectively . another , interesting criterion of inseparability that can be expressed in terms of moments of creation and annihilation operators is duan et al.s criterion which is described as follows @xcite : @xmath53 where @xmath54 clearly our analytic solution ( [ 2])-([eq : soln ] ) enables us to investigate intermodal entanglement using these criteria and a bunch of other criteria of nonclassicality that are described in ref . interestingly , all the inseparability criteria described above and in the remaining part of the present work can be viewed as special cases of shchukin - vogel entanglement criterion @xcite . in ref . @xcite , miranowicz et al . have explicitly established this point . so far we have described criteria of nonclassicality that are related to the lowest order nonclassicality . however , nonclassicality may be witnessed via higher order criterion of nonclassicality , too . investigations on higher order nonclassical properties of various optical systems have been performed since long . for example , in late seventies some of the present authors showed the existence of higher order nonclassical photon statistics in different optical systems using criterion based on higher order moments of number operators ( cf . @xcite and chapter 10 of @xcite and references therein ) . however , in those early works , higher order antibunching ( hoa ) was not specifically discussed , but existence of higher order nonclassical photon statistics was reported for degenerate and nondegenerate parametric processes in single and compound signal - idler modes , respectively and also for raman scattering in compound stokes - anti - stokes mode up to @xmath55 . criterion for hoa was categorically introduced by c. t. lee @xcite in 1990 . since then hoa is reported in several quantum optical systems ( @xcite and references therein ) and atomic systems @xcite . however , except a recent effort by us @xcite no effort has yet been made to study hoa in optical couplers . the existence of hoa can be witnessed through a set of equivalent but different criteria , all of which can be interpreted as modified lee criterion . in what follows we will investigate the existence of hoa in the contradirectional optical coupler of our interest using a simple criterion of @xmath56-th order single - mode antibunching introduced by some of us @xcite @xmath57 here @xmath58 and @xmath59 refer to the usual antibunching and the higher order antibunching , respectively . similarly , we can also investigate the existence of higher order squeezing , but definition of higher order squeezing is not unique . to be precise , it is usually investigated using two different criteria @xcite . the first criterion was introduced by hong and mandel in 1985 @xcite and a second criterion was subsequently introduced by hillery in 1987 @xcite . in hong and mandel criterion @xcite , the reduction of higher order moments of usual quadrature operators with respect to their coherent state counterparts are considered as higher order squeezing . however , in hillery s criterion , reduction of variance of an amplitude powered quadrature variable for a quantum state with respect to its coherent state counterpart is considered as higher order squeezing . in what follows we have restricted our study on higher order squeezing to hillery s criterion of amplitude powered squeezing . specifically , hillery introduced amplitude powered quadrature variables as @xmath60 and @xmath61 it is easy to check that @xmath62 and @xmath63 do not commute and consequently we can obtain an uncertainty relation and thus a criterion of @xmath64 order amplitude squeezing as@xmath65\right\rangle \right|}{\normalcolor < 0}}.}\ ] ] thus , for the specific case , @xmath58 , hillery s criterion for amplitude squared squeezing can be obtained as @xmath66 where @xmath67 similarly , we can obtain specific criteria of amplitude powered squeezing for other values of @xmath68 further , there exists a set of higher order inseparability criteria . to be precise , all the criteria that are used for witnessing the existence of multi - partite ( multi - mode ) entanglement are essentially higher order criteria @xcite as they always uncover some higher order correlation . interestingly , even in bipartite ( two - mode ) case one can introduce operational criterion for detection of higher order entanglement . for example , hillery - zubairy introduced following two criteria of higher order intermodal entanglement @xcite @xmath69 and@xmath70 here @xmath71 and @xmath56 are non - zero positive integers and the lowest possible values of @xmath71 and @xmath56 are @xmath72 which reduces ( [ hoe - criteria-1 ] ) and ( [ hoe - criteria-2 ] ) to usual hz - i criterion ( i.e. , ( [ hz1 ] ) ) and hz - ii criterion ( i.e. , ( [ hz2 ] ) ) , respectively . thus these two criteria may be viewed as generalized versions of the well known lower order criteria of hillery and zubairy ( i.e. , the criteria described in ( [ hz1 ] ) and ( [ hz2 ] ) ) . however , these generalized criterion can also be obtained as special cases of more general criterion of shchukin and vogel @xcite . for the convenience of the readers , we refer to ( [ hoe - criteria-1 ] ) and ( [ hoe - criteria-2 ] ) as hz - i criterion and hz - ii criterion respectively in analogy to the lowest order cases . in what follows , a quantum state will be called ( bipartite ) higher order entangled state if it is found to satisfy ( [ hoe - criteria-1 ] ) and/or ( [ hoe - criteria-2 ] ) for any choice of integers @xmath71 and @xmath56 satisfying @xmath73 existence of higher order entanglement can also be viewed through the criteria of multi - partite entanglement . for example , li et al . @xcite proved that a three - mode ( tripartite ) quantum state is not bi - separable in the form @xmath74 ( i.e. , compound mode @xmath75 is entangled with the mode @xmath8 ) if the following inequality holds for the three - mode system @xmath76 where @xmath77 are positive integers and annihilation operators @xmath78 correspond to the three modes . a quantum state that satisfy ( [ eq : tripartite ent1 ] ) is referred to as @xmath74 entangled state . the three - mode inseparability criterion mentioned above is not unique . there exist various alternative criteria of three - mode entanglement . for example , an alternative criterion for detection of @xmath74 entangled state is @xcite @xmath79 for @xmath80 this criterion coincides with miranowicz et al.s criterion @xcite . using ( [ eq : tripartite ent1 ] ) and ( [ eq : tripartite ent2 ] ) , we can easily obtain criteria for detection of @xmath81 and @xmath82 entangled states and use them to obtain a simple criterion for detection of fully entangled tripartite state . specifically , using ( [ eq : tripartite ent1 ] ) and ( [ eq : tripartite ent2 ] ) respectively we can conclude that the three modes of our interest are not bi - separable in any form if one of the following two sets of inequalities are satisfied simultaneously @xmath83 @xmath84 further , for a fully separable pure state we always have @xmath85^{\frac{1}{2}}.\label{eq : fully ent2}\ ] ] thus , a three - mode pure state that violates ( [ eq : fully ent2 ] ) ( i.e. , satisfies @xmath86 and simultaneously satisfies either ( [ eq : fully enta 0 ] ) or ( [ eq : fully enta 1 ] ) is a fully entangled state as it is neither fully separable nor bi - separable in any form . spatial evolution of different operators that are relevant for witnessing nonclassicality can be obtained using the perturbative solutions ( [ 2])-([eq : soln ] ) reported here . for example , using ( [ 2])-([eq : soln ] ) we can obtain the following closed form expressions for number operators of various field modes @xmath87,\end{array}\label{eq : na}\ ] ] @xmath88,\end{array}\label{eq : nb}\ ] ] @xmath89.\label{eq : nb2}\ ] ] it is now straight forward to compute the average values of the number of photons in different modes with respect to a given initial state . in the present work we consider that the initial state is product of three coherent states : @xmath90 where @xmath91 and @xmath92 are eigen kets of annihilation operators @xmath93 and @xmath8 , respectively . thus , @xmath94 and @xmath95 are the number of input photons in the field modes @xmath93 and @xmath8 , respectively . for a spontaneous process , @xmath96 and @xmath97 whereas , for a stimulated process , the complex amplitudes are not necessarily zero and it seems reasonable to consider @xmath98 using ( [ 2])-([eq : soln ] ) and ( [ eq : condition for squeezing])-([eq : two - mode quadrature ] ) we obtain analytic expressions for variance in single - mode and compound mode quadratures as @xmath99 & = & \frac{1}{4}\left[1\pm\left\ { \left(f_{1}f_{4}+f_{2}f_{3}\right)\gamma+{\rm c.c}.\right\ } \right],\\ \left[\begin{array}{c } \left(\delta x_{b_{1}}\right)^{2}\\ \left(\delta y_{b_{1}}\right)^{2}\end{array}\right ] & = & \frac{1}{4}\left[1\pm\left\ { \left(g_{1}g_{4}+g_{2}g_{3}\right)\gamma+{\rm c.c}.\right\ } \right],\\ \left[\begin{array}{c } \left(\delta x_{b_{2}}\right)^{2}\\ \left(\delta y_{b_{2}}\right)^{2}\end{array}\right ] & = & \frac{1}{4},\end{array}\label{eq : squeezing}\ ] ] and @xmath100 & = & \frac{1}{4}\left[1\pm\frac{1}{2}\left\ { \left(\left(f_{1}+g_{1}\right)\left(f_{4}+g_{4}\right)+\left(f_{2}+g_{2}\right)\left(f_{3}+g_{3}\right)\right)\gamma+{\rm c.c}.\right\ } \right],\\ \left[\begin{array}{c } \left(\delta x_{ab_{2}}\right)^{2}\\ \left(\delta y_{ab_{2}}\right)^{2}\end{array}\right ] & = & \frac{1}{4}\left[1\pm\frac{1}{2}\left\ { \left(f_{1}f_{4}+f_{2}f_{3}\right)\gamma+{\rm c.c}.\right\ } \right]=\frac{1}{2}\left[\begin{array}{c } \left(\delta x_{a}\right)^{2}\\ \left(\delta y_{a}\right)^{2}\end{array}\right]+\frac{1}{8},\\ \left[\begin{array}{c } \left(\delta x_{b_{1}b_{2}}\right)^{2}\\ \left(\delta y_{b_{1}b_{2}}\right)^{2}\end{array}\right ] & = & \frac{1}{4}\left[1\pm\frac{1}{2}\left\ { \left(g_{1}g_{4}+g_{2}g_{3}\right)\gamma+{\rm c.c}.\right\ } \right]=\frac{1}{2}\left[\begin{array}{c } \left(\delta x_{b_{1}}\right)^{2}\\ \left(\delta y_{b_{1}}\right)^{2}\end{array}\right]+\frac{1}{8},\end{array}\label{eq : intermodal squeezing}\ ] ] respectively . from eq . ( [ eq : squeezing ] ) it is clear that no squeezing is observed in @xmath8 mode . however , squeezing is possible in @xmath28 mode and @xmath101 mode as illustrated in fig . [ fig : squeezing - del k - g ] a - b , d - e . further , we observed intermodal squeezing in quadratures @xmath102 and @xmath103 by plotting right hand sides of eq . ( [ eq : intermodal squeezing ] ) in fig . [ fig : squeezing - del k - g ] c and fig . [ fig : squeezing - del k - g ] f. variation of amount of squeezing in different modes with phase mismatch @xmath9 and nonlinear coupling constant @xmath104 are shown in fig . [ fig : squeezing - del k - g ] a - c and fig . [ fig : squeezing - del k - g ] d - f , respectively . we have also studied the effect of linear coupling constant @xmath20 on the amount of squeezing , but its effect is negligible in all other quadratures except the quadratures of compound mode @xmath105 in compound mode @xmath75 , after a short distance depth of squeezing is observed to increase with the decrease in the linear coupling constant @xmath20 ( this is not illustrated through figure ) . intermodal squeezing in compound mode quadratures @xmath106 and @xmath107 can be visualized from the last two rows of eq.([eq : intermodal squeezing ] ) . specifically , we can see that variance in compound mode quadrature @xmath108 and @xmath109 have _ bijective _ ( both one - to - one and onto ) correspondence with the variance in @xmath110 and @xmath111 , respectively , where @xmath112 to be precise , quadrature squeezing in single - mode @xmath110(@xmath111 ) implies quadrature squeezing in @xmath113 and vice versa . for example , @xmath114 or , @xmath115 this is why we have not explicitly shown the variance of compound mode quadratures @xmath116 and @xmath117 with different parameters as we have done for the other cases . as we have shown squeezing in @xmath118 and @xmath119 through fig . [ fig : squeezing - del k - g ] a - b , and c - d this implies the existence of squeezing in quadrature @xmath120 and @xmath121 thus we have observed intermodal squeezing in compound modes involving @xmath8 . this nonclassical feature was not observed in earlier studies @xcite as in those studies @xmath8 mode was considered classical . the plots do not show quadrature squeezing in @xmath122 @xmath123 , and @xmath102 ( for some specific values of @xmath104 ) . however , a suitable choice of phase of the input coherent state would lead to squeezing in these quadratures . for example , if we replace @xmath124 by @xmath125 ( i.e. , if we chose @xmath126 instead of present choice of @xmath127 ) then we would observe squeezing in all these quadratures , but the squeezing that is observed now with the original choice of @xmath124 would vanish . this is so as all the expressions of variance of quadrature variables that are @xmath128 have a common functional form : @xmath129 ( c.f . , eqs . ( [ eq : squeezing ] ) and ( [ eq : intermodal squeezing ] ) ) . after establishing the existence of squeezing in single - modes and compound modes we now examine the possibility of higher order squeezing using eqs . ( [ 2])-([eq : soln ] ) , ( [ eq : na])-([eq : nb2 ] ) and ( [ eq : criterion - amplitude squared ] ) . in fact , we obtain @xmath130 & = & \pm\frac{n^{2}}{4}\left[\gamma\left(f_{1}f_{4}+f_{2}f_{3}\right)\left(f_{1}\alpha+f_{2}\beta\right)^{2n-2}+{\rm c.c.}\right],\end{array}\label{eq : asq-1}\ ] ] @xmath131 & = & \pm\frac{n^{2}}{4}\left[\gamma\left(g_{1}g_{4}+g_{2}g_{3}\right)\left(g_{1}\alpha+g_{2}\beta\right)^{2n-2}+{\rm c.c.}\right],\end{array}\label{eq : as - q-2}\ ] ] and @xmath132 & = & 0.\end{array}\label{eq : as - q-3}\ ] ] thus , we do not get any signature of amplitude powered squeezing in @xmath8 mode using the present solution . in contrary , mode @xmath28 @xmath133 is found to show amplitude powered squeezing in one of the quadrature variables for any value of interaction length as both @xmath134 and @xmath135 ( @xmath136 and @xmath137 ) can not be positive simultaneously . to study the possibilities of amplitude powered squeezing in further detail we have plotted the spatial variation of @xmath138 and @xmath139 in fig . [ fig : amplitude - squared - squeezing ] . negative regions of these two plots clearly illustrate the existence of amplitude powered squeezing in both @xmath28 and @xmath101 modes for @xmath58 and @xmath140 . extending our observations in context of single - mode squeezing and intermodal squeezing we can state that the appearance of amplitude powered squeezing in a particular quadrature can be controlled by suitable choice of phase of input coherent state @xmath124 as the expressions for amplitude powered squeezing reported in ( [ eq : asq-1 ] ) and ( [ eq : as - q-2 ] ) have a common functional form @xmath141 . the condition of hoa is already provided through the inequality ( [ hoa-1 ] ) . now using this inequality along with eqns . ( [ 2])-([eq : soln ] ) and ( [ eq : na])-([eq : nb2 ] ) we can obtain closed form analytic expressions for @xmath142 for various modes as follows @xmath143 @xmath144 @xmath145 further , using the condition of intermodal antibunching described in ( [ eq : antib2 ] ) and eqns . ( [ 2])-([eq : soln ] ) we obtain following closed form expressions of @xmath146 @xmath147 @xmath148 @xmath149 from the above expressions it is clear that neither the single - mode antibunching nor the intermodal antibunching is obtained involving @xmath8 mode . as the expressions obtained in the right hand sides of ( [ eq : dan ] ) , ( [ eq : dbn ] ) and ( [ eq : dab1 ] ) are not simple , we plot them to investigate the existence of single - mode and compound mode antibunching . the plots for usual antibunching and intermodal antibunching are shown in fig . [ fig : antibunching ] . existence of antibunching is obtained in single - mode @xmath28 for @xmath127 and the same is illustrated through fig . [ fig : antibunching ] a. however , in a effort to obtain antibunching in @xmath101 mode , we do not observe any antibunching in @xmath101 mode for @xmath127 . interestingly , from ( [ eq : dbn ] ) it is clear that if we replace @xmath127 by @xmath150 as before and keep @xmath151 unchanged , then we would observe antibunching for all values of rescaled interaction length @xmath152 this is true in general for all values of @xmath124 . to be precise , if we observe antibunching ( bunching ) in a mode for @xmath153 we will always observe bunching ( antibunching ) for @xmath154 if we keep @xmath151 unchanged . this fact is illustrated through fig . [ fig : antibunching ] b where we plot variation of @xmath155 with @xmath156 and have observed the existence of antibunching . in compound mode @xmath157 we can observe existence of antibunching for both @xmath127 and @xmath158 . however , in fig . [ fig : antibunching ] c we have illustrated the existence of intermodal antibunching in compound mode @xmath75 by plotting variation @xmath159 with rescaled interaction length @xmath156 for @xmath158 as region of nonclassicality is relatively larger ( compared to the case where @xmath127 and @xmath151 are same ) in this case . now we may extend the discussion to hoa and plot right hand sides of ( [ eq : dan ] ) and ( [ eq : dbn ] ) for various values of @xmath56 . the plots are shown in fig . [ fig : hoa ] which clearly illustrates the existence of hoa and also demonstrate that the depth of nonclassicality increases with @xmath68 this is consistent with earlier observations on hoa in other systems @xcite . we first examine the existence of intermodal entanglement in compound mode @xmath75 using hz - i criterion ( [ hz1 ] ) . to do so we use eqns . ( [ 2])-([eq : soln ] ) and ( [ eq : na])-([eq : nb2 ] ) and obtain@xmath160 similarly , using hz - ii criterion ( [ hz2 ] ) we obtain @xmath161.\end{array}\label{eq : hz2-ab1}\ ] ] it is easy to observe that eqns . ( [ eq : hz-1-ab1 ] ) and ( [ eq : hz2-ab1 ] ) provide us the following simple relation that is valid for the present case : @xmath162 , which implies that for any particular choice of rescaled interaction length @xmath156 either hz - i criterion or hz - ii criterion would show the existence of entanglement in contradirectional asymmetric nonlinear optical coupler as both of them can not be simultaneously positive . thus the compound mode @xmath75 is always entangled . the same is explicitly illustrated through fig . [ fig : ent - hz1-ii ] . similar investigations using hz - i and hz - ii criteria in the other two compound modes ( i.e. , @xmath163 and @xmath164 failed to obtain any signature of entanglement in these cases . further , signature of intermodal entanglement was not witnessed using duan et al . criterion as using the present solution and ( [ duan-1 ] ) we obtain @xmath165 however , it does not ensure separability of these modes as hz - i , hz - ii and duan et al . inseparability criteria are only sufficient and not necessary . ( color online ) hillery - zubairy criterion i ( solid line ) and criterion ii ( dashed line ) for entanglement are showing intermodal entanglement between modes @xmath28 and @xmath166 here @xmath167 ( solid line ) and @xmath168 ( dashed line ) are plotted with rescaled interaction length @xmath169 for mode @xmath75 with @xmath170 for @xmath171 ( thin blue lines ) and @xmath172 ( thick red lines ) . ] we may now study the possibilities of existence of higher order entanglement using eqns . ( [ hoe - criteria-1])-([eq : fully ent2 ] ) . to begin with , we use ( [ 2])-([eq : soln ] ) and ( [ hoe - criteria-1 ] ) to yield @xmath173 similarly , using ( [ 2])-([eq : soln ] ) and ( [ hoe - criteria-2 ] ) we can produce an analytic expression for @xmath174 and observe that @xmath175 from relation ( [ eq : emnprimeab1 ] ) , it is clear that the higher order entanglement between @xmath28 mode and @xmath101 mode would always exist for any choice of @xmath176 @xmath71 and @xmath56 . this is so because @xmath177 and @xmath174 can not be simultaneously positive . using ( [ eq : emnab1 ] ) and ( [ eq : emnprimeab1 ] ) , it is a straight forward exercise to obtain analytic expressions of @xmath177 and @xmath174 for specific values of @xmath71 and @xmath68 such analytic expressions are not reported here as the existence of higher order entanglement is clearly observed through ( [ eq : emnprimeab1 ] ) . however , in fig . [ fig : hoe ] we have illustrated the variation of @xmath178 and @xmath179 with the rescaled interaction length , @xmath156 . negative parts of the plots shown in fig . [ fig : hoe ] illustrate the existence of higher order intermodal entanglement in compound mode @xmath105 as expected from ( [ eq : emnprimeab1 ] ) , we observe that for any value of @xmath156 compound mode @xmath75 is higher order entangled . further , it is observed that hillery - zubairy s higher order entanglement criteria ( [ hoe - criteria-1])-([hoe - criteria-2 ] ) fail to detect any signature of higher order entanglement in compound modes @xmath163 and @xmath180 one can also investigate higher order entanglement using criterion of multi - partite ( multi - mode ) entanglement as all the multi - mode entangled states are essentially higher order entangled . here we have only three modes in the coupler and thus we can study higher order entanglement by investigating the existence of three - mode entanglement . a three - mode pure state that violates ( [ eq : fully ent2 ] ) ( i.e. , satisfies @xmath86 and simultaneously satisfies either ( [ eq : fully enta 0 ] ) or ( [ eq : fully enta 1 ] ) is a fully entangled state . using ( [ 2])-([eq : soln ] ) and ( [ eq : tripartite ent1])-([eq : fully ent2 ] ) we obtain following set of interesting relations for @xmath181 : @xmath182 @xmath183 and @xmath184 from ( [ eq : relation1 ] ) , it is easy to observe that three modes of the coupler are not bi - separable in the form @xmath81 and @xmath185 for any choice of rescaled interaction length @xmath186 further , eqn . ( [ eq : relation3 ] ) shows that the three modes of the coupler are not fully separable for @xmath187 ( c.f . positive regions of plot of @xmath167 shown in fig . [ fig : ent - hz1-ii ] ) . however , ( [ eq : relation2 ] ) illustrate that the present solution does not show entanglement between coupled mode @xmath75 and single - mode @xmath188 thus the three modes present here are not found to be fully entangled . specifically , three - mode ( higher order ) entanglement is observed here , but signature of fully entangled three - mode state is not observed . further , we have observed that in all the figures depth of nonclassicality increases with @xmath189 ( color online ) higher order entanglement is observed using hillery - zubairy criteria . solid lines show spatial variation of @xmath178 and dashed lines show spatial variation of @xmath179 with @xmath170 for @xmath171 ( thin blue lines ) and @xmath172 ( thick red lines ) . it is observed that depth of nonclassicality increases with increase in @xmath190 . ] in the present paper , various lower order and higher order nonclassical phenomena have been observed in contradirectional asymmetric nonlinear optical coupler . however , so far we have discussed only the stimulated cases as no nonclassical phenomenon is expected to be observed in spontaneous case . this is so because all the useful non - vanishing expressions for witnessing nonclassicality ( i.e. , eqs . ( [ eq : asq-1])-([eq : relation3 ] ) ) are proportional to @xmath191 thus all these expressions would vanish for @xmath192 . approximation @xcite & present work + 1 . & squeezing & @xmath8 & not investigated & not observed + 2 . & intermodal squeezing & @xmath193 & not observed & observed + 3 . & amplitude squared squeezing & @xmath93 & not investigated & observed + 4 . & amplitude squared squeezing & @xmath8 & not investigated & not observed + 5 . & lower order and higher order intermodal entanglement & @xmath75 & not investigated & observed + 6 . & lower order and higher order intermodal entanglement & @xmath193 & not investigated & not observed + 7 . & three - mode ( higher order ) bi - separable entanglement & @xmath81 , @xmath185 & not investigated & observed + 8 . & three - mode ( higher order ) bi - separable entanglement & @xmath81 & not investigated & not observed + in the present study we report lower order and higher order nonclassicalities in a contradirectional asymmetric nonlinear optical coupler using a set of criteria of entanglement , single - mode squeezing , intermodal squeezing , antibunching , intermodal antibunching etc . variation of nonclassicality with various parameters , such as number of input photon in the linear mode , linear coupling constant , nonlinear coupling constant and phase mismatch is also studied and it is observed that amount of nonclassicality can be controlled by controlling these parameters . the contradirectional asymmetric nonlinear optical coupler studied in the present work was studied earlier using a short - length solution and considering @xmath8 mode as classical @xcite . in contrast , a completely quantum mechanical solution of the equations of motion is obtained here using sen - mandal approach which is not restricted by length . the use of better solution and completely quantum mechanical treatment led to the identification of several nonclassical characters of a contradirectional asymmetric nonlinear optical coupler that were not reported in earlier studies . all such nonclassical phenomena that are observed here and were not observed in earlier studies are listed in table [ tab : nonclassicalities - observed - in ] . further , there exist a large number of nonclassicality criteria that are not studied here and are based on expectation values of moments of annihilation and creation operators ( c.f . table i and ii of ref . as we already have compact expressions for the field operators it is a straight forward exercise to extend the present work to investigate other signatures of nonclassicality , such as , photon hyperbunching @xcite , sum and difference squeezing of an - tinh and hillery , inseparability criterion of manicini et al . @xcite , simon @xcite and miranowicz et al . @xcite etc . further , the present work can be extended to investigate the nonclassical phenomena in other types of contradirectional optical couplers ( e.g. , contradirectional parametric coupler , contradirectional raman coupler , etc . ) that are either not studied till date or studied using short - length solution . recently , allevi et al . @xcite and avenhaus et al . @xcite have independently reported that they have experimentally measured @xmath194 which is sufficient to completely characterize bipartite multi - mode states . it is easy to observe that ability to experimentally measure @xmath194 ensures that we can experimentally detect signatures of nonclassicalities reported here . it is true that most of the nonclassicalities reported here can be observed in some other bosonic systems , too . however , the present system has some intrinsic advantages over most of the other systems as it can be used as a component in the integrated waveguide based structures in general and photonic circuits in particular [ @xcite and references therein ] . thus the nonclassicalities reported in this easily implementable waveguide based system is expected to be observed experimentally . further , the system studied here is expected to play important role as a source of nonclassical fields in the integrated waveguide based structures . * acknowledgment : * k. t. and a. p. thank the department of science and technology ( dst ) , india , for support provided through dst project no . sr / s2/lop-0012/2010 and a. p. also thanks the operational program education for competitiveness - european social fund project cz.1.07/2.3.00/20.0017 of the ministry of education , youth and sports of the czech republic . j. p. thanks the operational program research and development for innovations - european regional development fund project cz.1.05/2.1.00/03.0058 of the ministry of education , youth and sports of the czech republic . j. lugani , studies on guided wave devices and ultracold atoms in optical cavity for applications in quantum optics and quantum information , phd thesis , iit delhi ( 2013 ) , http://web.iitd.ac.in/~sankalpa/jasleen_corrected_thesis.pdf .

hii galaxies can be considered as the strong emission line subset of the blue compact dwarf galaxies ( bcd ) . their optical emission is dominated by strong and narrow emission lines produced by the interstellar gas ionized by a young and luminous cluster . their integrated optical spectra are indistinguishable from a normal hii region . hii galaxies are gas rich and metal poor . these facts had led to the early proposal that these systems may be very young perhaps undergoing their first burst of star formation ( sf ) @xcite . however , many authors have found clear evidences of the existence of a low surface brightness , old , non - ionizing stellar population in the majority of these galaxies , as explained by ( * ? ? ? * here in after mmdt ) and references therein . the existence of an old population plus the relative paucity of hii galaxies with extreme low metal content are consistent with a scenario in which hii galaxies are suffering at present a starburst , instead of being old systems with a slow evolution . therefore we use the term _ starburst _ to denote a violent episode of star formation where a large number of massive stars have been formed in a small volume of space and over a time scale of a few million years . galaxies called _ starbursts galaxies _ are a very heterogeneous category , including from bcds to ulirgs . we will handle here only hii galaxies , those dwarf galaxies with active star formation whose effects actually dominates their uv - optical emission . in these galaxies the starburst is considered a phase , a process of strong star formation in terms of intensity and duration . nevertheless , the fundamental details of how the star formation history proceeds in hii galaxies is still an unresolved problem . until now , three basic evolutionary scenarios have been postulated : * * bursting star formation * : the stars form in short but intense episodes separated by long quiescent periods of very low or null activity @xcite * * gasping star formation * : the star formation takes places as long episodes of sf of moderate intensity separated by short quiescent periods @xcite . * * continuous star formation * : the process of stellar formation is continuous and of low intensity during the galaxy life , with superimposed sporadic bursts @xcite . the low metallicities , lack of dust and the optical colors argue in favour of a bursting star formation with long quiescent periods ( e.g. * ? ? ? however , inactive star forming periods seem to rarely occur , and inter - burst stages characterized by low - levels of star formation are perhaps more plausible than its complete cessation @xcite . if this is the case , a continuous low level state of star formation would dominate during the vast majority of the time during which many or perhaps most of the stars in bcd galaxies would be formed . chemical evolution models with gasping or bursting star formation modes seem to be appropriate to describe local dwarf irregulars ( dirr ) @xcite , but the chemical evidence alone can not differentiate between gasping and bursting scenarios . the gasping scenario can be viewed as a transition between the continuous star formation and the bursting scenario @xcite . the most remarkable difference between gasping and bursting scenarios is that short episodes of star formation enrich the ism in a time scale of few tens of myr while long - lasting episodes enrich gradually the ism in a longer time scale , and any further episode of sf does not leave an appreciable imprint on the chemical evolution . the abundances , which increase rapidly in bursting models , change more slowly and during a longer time in gasping models , but with older ages in their stellar populations @xcite . this picture , however , does not entirely exclude the possibility that star formation histories appear to be composed by burst cycles when examined with a high time resolution , but looks constant when averaged over longer time scales . it does not exclude either the possibility that the sf propagates through the galaxy , taking place in independent luminous short lived hii regions : `` the short timescales associated to starbursts in dwarf galaxies may be understood as _ flickering _ events , small components of a larger starburst in the galaxy '' @xcite these observations of stellar populations suggest that it is most likely that the star formation in dwarf galaxies is sporadic , separated by millions to billions of years , even in isolated systems , as most of hii are . if the sf of hii galaxies is dominated by intermittent starburst episodes separated by long inactive time spans , quiescent blue dwarf galaxies without the dominant starburst and showing similar properties to those with such intense star formation events , should be relatively common . support to this framework comes from the fact that blue compact dwarfs ( bcd ) , with emission lines on average much weaker than these of hii galaxies , are more common than those ones . following this idea @xcite obtain a relationship between the duration time of the starburst phase and the quiescent periods , based on the number of objects of each type of their sample . they find that , if the duration of the burst phase is 10myr , the time in quiescence must be at least 0.27gyr , implying several mayor starburst episodes along the life of a bcd galaxy . therefore , the diversity in properties exhibited by dwarfs may be at least partially the consequence of observing them at different times in their star formation cycles , and the frequency distributions of galaxies in the various phases could correspond to the time spent in each of them . it would be interesting to see which is the most probable star formation scenario given by the _ ab initio _ cosmological hydrodynamical simulations . however this is difficult since the dwarf galaxies produced in the existing works are usually poor gas , dwarf elliptical ( de ) or dwarf spheroidals , ( dsp ) objects @xcite . other simulations , such as @xcite used fixed initial conditions for stellar and gas masses , and this way the mass to halo mass ratio is not a result of the simulation . in @xcite simulations , the ultraviolet ( uv ) background is not included , which probably contributes to a high star formation efficiency and to large final stellar masses . @xcite follow the evolution of a galaxy with a halo mass of 10@xmath0m@xmath1 but only until z=5 . @xcite have calculated some complete hydrodynamical simulations , in which baryonic processes , as gas cooling , heating from the cosmic uv field , star formation and supernova - driven gas heating , are included with sufficient spatial resolution , with clumps as low as 10@xmath2m@xmath1 resolved . the created dwarf galaxies without bulges are similar to the observed dwarf irregular galaxies . however , besides the authors do not show the star formation histories , these galaxies show a maximum star formation rate ( sfr ) of 0.25m@xmath3 and a present sfr of 0.01m@xmath3 , much lower than observational values for hii galaxies , and not comparable to these objects that we will try to model here . only @xcite has recently presented some star formation histories , obtained from cosmological hydrodynamical simulations for dwarf galaxies , with enough spatial resolution for gravitational masses of 10@xmath0m@xmath1 . their star formation histories seem to be sporadic , and the stars continue forming sporadically even at late times . however , the simulations are still having some problems to reproduce the adequate number of galaxies and the author claims that more work is still necessary . as @xcite explains , all dwarf galaxies ( 10@xmath4m@xmath1 ) formed in the current hydrodynamical simulations are more than an order of magnitude more luminous than expected for these masses . in any case they are much more massive than bcds and hii galaxies that we want modeling . for what refers to other models existing in the literature , there are not models for the study of bcd or hii galaxies in the way we want to use them . a number of models have computed purely chemical evolution models for bcds or dwarf irregular galaxies ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and many others ) . most of them assume that the star formation occurs in bursts and include the effects of galactic winds and/or gas infall . however , they limit the study to the evolution of n and o abundances and/or to the luminosity - metallicity relation . @xcite and @xcite used the information coming from the chemical evolution models from @xcite to perform the next step and combine chemical and spectral evolution for irregular galaxies . these models , however , exclude the early stages of evolution , i.e. during the nebular phase when most massive stars dominate the energy output budget . @xcite or @xcite compute chemical and photometric evolution models in a way more or less consistent for starburst , spiral , early types or low surface brightness galaxies , respectively , any of them applied to hii galaxies . other works , in turn , are focused on the ionized gas properties and make models using single stellar populations ( ssps ) spectral energy distributions ( seds ) of a given metallicity for applying a photo - ionization code to obtain the emission lines and studying diagnostic diagrams and/or abundances obtained by empirical calibrations @xcite . most of them ignore the star formation history of the galaxy and not take into account the underlying stellar population , so these studies are only valid for the study of the current stellar generations of the galaxy . summarizing a code combining chemical , evolutionary synthesis and photo - ionization models has not been applied for the analysis of bcds and hii galaxies , with the exception of the ours ( mmdt ) , but only to elliptical massive galaxies @xcite . the aim of this work is to critically analyze the possibility of dwarf galaxies undergo recurrent phases of star formation by comparing the predictions of our self - consistent evolutionary models with the most evolution sensitive observed parameters , i.e. stellar continuum colors , emission line equivalent widths and chemical composition in a large sample of bcd galaxies . we have used an updated version of our models from mmdt , developed to predict the main characteristics of hii galaxies , such as the emission lines , continuum color , continuum plus the contribution of emission lines colors , equivalent widths and chemical abundances . our code uses the chemical evolution model results for the computation of the seds , which are , in turn , used as the ionizing sources for a photo - ionization code . in mmdt we studied the viability of this tool , showing its adequacy in the reproduction of the data defined by the stellar populations ( colors , equivalent width of absorption lines or spectral indices , spectral energy distributions .. ) , which define the time evolution , as well as the characteristics of the gas phase ( emission lines , equivalent widths , elemental abundances , gas densities ) , which defines the present time state of the galaxy . this is done in a self - consistent way , that is , using the same assumptions regarding stellar evolution , model stellar atmospheres and nucleosynthesis , and using a realistic age - metallicity relation . one of the most important results obtained from our previous work was that observational data are reproduced only if the mass involved in the last burst is much smaller than the mass of the old underlying stellar population . now , once checked the possibilities of our tool , we try to wide the number of models , changing the star formation scenarios to see if data are reproduced with more or less success for some of them . it is a widely assumed idea that starburst galaxies are strongly affected by gas infall and outflow and that chemical abundances and star formation can not modeled without these issues . however , the existence of outflows in dwarf starburst galaxies is by no means a settled issue ; while winds able to escape the galaxy have been found in two prototypical starburst galaxies like ngc1569 and ngc1705 , recent works about mass loss , for example @xcite , find that ionized and neutral hydrogen expansion velocities measured are , in all cases , too low to allow the gas to escape from the gravitational potential of their studied galaxies some of which , like ngc2363 or ngc4861 , can also be considered as _ prototypical_. in fact , according to @xcite:while the observational support for the presence of galactic winds in massive galaxies and gas - rich mergers is quite strong , the case for galactic winds in dwarf galaxies is much weaker . the need of winds is also related with the low oxygen abundance found for dwarf galaxies for their gas fractions , and , in this case , both infall and outflow can help . however , galaxies loosing a large fraction of their gas should show red colors , contrary to what is observed . in fact , a good fraction of the galaxies in the sample of @xcite can be reproduced with closed box models . furthermore , it is necessary to take into account the n / o vs o / h relationship for dwarf galaxies . selective winds models as @xcite have recognized problems in reproducing the observed n / o ratios , and the work by @xcite demonstrates that the observed n / o ratios , as well as their dispersion , can be reproduced with closed box models while selective winds can be ruled out . actually , @xcite showed that it is mainly the combination of the different time scales in nitrogen production by stars of different masses and the galaxy star formation histories as proceeding from different star formation efficiencies , what establishes the n / o ratio . therefore , since there is enough proof to conclude that infall and/or outflows are not necessary to reproduce the general trends of hii galaxies , galactic supernova - driven winds are not included in our models . this scenario is also in agreement with models from @xcite , who find that scaling relations of dwarf galaxies may be reproduced by simulated galaxies without supernova - driven outflows . for the present work , new and updated theoretical codes have been applied for the computation of the models . the main changes correspond to the evolutionary synthesis models , now from ( * ? ? ? * hereinafter mgvb09 ) instead @xcite . as explained before , to guarantee the self - consistency of the approach we have used the same assumptions in the stellar evolution , model stellar atmospheres and nucleosynthesis parts of the code . the use of these new models allows to be more consistent than in our previous work mmdt , since there the imf used in the chemical evolution models @xcite and in evolutionary synthesis models @xcite was not the same . the new models mgvb allow to us use the same imf for chemical and evolutionary synthesis ( photometrical ) calculations . moreover , with the previous models from @xcite we had also a problem for the lowest metallicities stellar populations since models for the youngest ages of low metallicities were not available . now we have models for the same number of ages for all metallicities . to include the lowest metallicities of the youngest ages change appreciably the colors of the stellar populations and allows to calculate emission lines for very low metallicity regions , as it is shown in @xcite . the contribution of the emission lines is quite different than expected from the simple extrapolation from other metallicities , also modifying the final results of colors . the data shown in the graphs have been extracted mainly from two main sources . first , the compilation from @xcite which provides emission line measurements , corrected for extinction , published for local hii galaxies . the sample comprises 450 objects and constitutes a large sample of local hii galaxies with good - quality spectroscopic data . the sample is rather inhomogeneous in nature , since the data proceed from different instrumental setups , observing conditions and reduction procedures , but have been analysed in a uniform way . data for these sample objects include the emission line intensities of : [ oii]@xmath5 3727,29 [ oiii]@xmath5 4959,5007 and [ nii]@xmath5 6548,84 all of them relative to h@xmath6 , and the equivalent widths of the [ oii ] and [ oiii ] emission lines ew([oii ] ) ew([oiii ] ) , and the h@xmath6 line , ew(h@xmath6 ) . as a second source , we have used the metal poor galaxy data from the data release 3 of sloan digital sky survey , taken from @xcite . the sloan digital sky survey @xcite constitutes a large data base of galaxies with well defined selection criteria and observed in a homogeneous way . the sdss dr3 @xcite provides spectra in the wavelength range from 3800 to 9300 for @xmath7 530000 galaxies , quasars and stars . @xcite extracted @xmath7 2700 spectra of non - active galaxies with the [ oiii]@xmath84363 emission detected above 1@xmath9 level . this initial sample was further restricted to the objects with an observed flux in the h@xmath6 emission line larger than 10@xmath10 erg s@xmath11 @xmath12 and for which accurate abundances could be derived . they have also excluded all galaxies with both [ oiii]@xmath84959/h@xmath6 @xmath13 0.7 and [ oii]@xmath83727/h@xmath6 @xmath14 1.0 . applying all these selection criteria , they obtain a sample of @xmath7 310 sdss objects . data for these sample objects include the emission line intensities of : [ oiii]@xmath84959,5007 and [ nii]@xmath86584 relative to h@xmath6 and the equivalent width of h@xmath6 . they also include the intensity of the [ oii ] @xmath5 3727,29 emission line for the lowest redshift objects . in the next section an explanation of the theoretical star - bursting models the section 3 presents the results , the meaning of each input parameter and the implication of their variation over the results of the models . we will discuss the influence of these parameters in the reproduction of the observable characteristics of star - forming galaxies , their impact over possible star - formation scenarios and the connection among different type of dwarf galaxies in section 4 . finally , a summary and the conclusions of this work are presented in section 5 . all tables of the models will be available in electronic format . the viability of our model technique to reproduce the observable characteristics of hii galaxies was discussed in mmdt : in a first stage the chemical evolution model is computed by using certain input parameters to obtain the star formation history and the gas and stars chemical abundances . secondly , the evolutionary population synthesis code is applied to obtain the seds corresponding to each time step of the chemical evolution . finally , the ionizing part of the sed and the resulting abundances are used as an input to the photo - ionization code to compute the time evolution of the emission lines of the ionized gas . the chemical evolution is computed with a simplified version of the classical _ multiphase chemical evolution model _ from @xcite . in our version of the code there is only one region ( without the two zones halo and disk ) with a given mass of gas which form stars ( in this case we do not consider the phase of formation of molecular clouds as in the classical multiphase models ) . we have run models considering the star formation as a set of successive bursts followed by quiescent periods in a region with a total mass of gas of @xmath15 . in each burst a given amount of gas is consumed to form stars , and this process comes defined by the star formation efficiency . the code solves the chemical evolution equations to obtain , in each time step , the abundances of 15 elements : h , d , @xmath16 , @xmath17 , c , @xmath18 , o , n , ne , mg , si , s , ca , fe , and _ nr _ ( where _ nr _ are the isotopes of the neutron rich elements , synthesized from @xmath19 , @xmath18 , @xmath20n and @xmath21 inside the co core ) . the stellar yields are those from @xcite for massive stars , @xmath22 , and those from @xcite for low and intermediate mass stars . the supernova ia yields used proceed from @xcite . the initial mass function ( imf ) is taken from @xcite with a range between 0.15 and 100 @xmath23 . more recent imfs such as @xcite or @xcite there exist but , for chemical evolution models , it is necessary to use a combination imf+stellar yields sets calibrated with milky way galaxy data , that is , able to reproduce the elemental abundances and gas and star densities as observed . this calibration is not still done for those imfs , although an update of models taking these imfs and some new stellar yield sets will be shown in a next future ( molla et al . in preparation ) . we have taken time steps of @xmath24t@xmath7 0.7myr from the initial time , @xmath25 , up to the final one , @xmath26gyr . at each time step , the star formation rate and the mass in each phase low mass , massive stars and remnants , total mass in stars created , and mass of gas are also obtained . the seds are taken from the popstar evolutionary synthesis models by mgvb09 . isochrones are an update from those from @xcite for 6 different metallicities : @xmath27 , 0.0004 , 0.004 , 0.008 , 0.02 and 0.05 . the very low metallicity model of @xmath28 had not been included before in similar works . the age coverage is from @xmath295.00 to 10.30 with a variable time resolution of @xmath30 in the youngest stellar ages . we have used the ferrini imf results @xcite with mass limits between 0.15 and 100m@xmath1 in order to avoid any inconsistency with the chemical evolution code , which also uses this imf . to calculate the sed integrated over the whole history of the galaxy , seds of ssps with the corresponding metallicity and age must be convolved with the star formation history ( sfh ) : @xmath31 where @xmath32 is the age of the stellar population created in a time @xmath33 and @xmath34 is the sed for each ssp of age @xmath35 and metallicity z reached in that time @xmath33 . a sed from the ssp library , @xmath34 , must be assigned to each time step according to its corresponding age and metallicity . taking into account the sfh , @xmath36 , and the age - metallicity relation , @xmath37 , obtained from the chemical evolution model , we know the age and metallicity assigned to each time step or stellar generation . for each one of them we have chosen the sed closest in age among those available in the grid of popstar . however , in our models the metallicity changes continuously while the available seds of the library are computed only for 6 possible values . therefore , we have interpolated logarithmically between the two ssp of the same age @xmath35 and closest in metallicities to @xmath37 to obtain the corresponding @xmath38 . the final result is the total luminosity at each wavelength @xmath8 , l@xmath39 ) , corresponding to the whole stellar population , including the ionizing continuum proceeding from the last formed stellar population . to compute the photo - ionization models , we have used cloudy ( version c06.02 * ? ? ? the gas is assumed to be spherically symmetric around a point source of radiation and the pressure or density in the gas is imposed by external conditions . this allows a plane - parallel geometry treatment in which the gas may be regarded as a thin shell . a closed geometry has been taken for the calculations . all the photons which escape from the illuminated face of the cloud towards the star , go on to strike the other side of the nebula , ensuring the case b of recombination and the approximation on the spot . the number of ionizing photons , q(h ) , striking the illuminated face of the cloud , have been calculated directly from the total resulting sed of the models . we derive the radius of the modelled region from the mechanical energy from massive stars with strong winds ( taken from mgvb09 ) , instead of using the radius required to maintain a constant density of stellar mass through the successive bursts , as we did in mmdt . @xcite demonstrated that an early - type star with a strong stellar wind can blow out a large cavity or _ bubble _ in the surrounding gas , if it is assumed to be compressed into a thin spherical shell . the wind - driven shell begins to evolve with an initial phase of free expansion followed by an adiabatic expansion phase , and then the material collapses into a thin , cold shell as a result of radiative cooling . at this stage the gas traps the ionization front and the radiative phase begins . in this phase the ionizing photons are absorbed and the region cools via emission in the balmer lines . in this process , the radius of the outer shock , r@xmath40 , evolves as : @xmath41 where @xmath42 is the total injected mechanical energy ( sn and stellar winds ) per unit time in units of 10@xmath43ergs s@xmath11 , @xmath44 is the interstellar medium density in units of @xmath45 , and @xmath46 is the age of the shell in units of 10@xmath47yr . since we use the value of the energy at each time - step , this radius represents the instantaneous size of the region obtained by adding the winds and sne from the previous age to this age . therefore l@xmath48 is this energy divided by the time step . then , the ionized gas is assumed to be located in a thin spherical shell at that distance r@xmath40 from the ionizing source . this approach has the advantage of eliminating the ionization parameter as a free variable in the models since now it is computed from the physical parameters of the evolving young cluster . the ionized gas abundances are assumed to be those reached at the end of the starburst previous to the current one . fifteen element abundances have been introduced in the photo - ionization code : he , c , n , o , ne , na , mg , al , si , s , ar , ca , fe and ni , obtained from the chemical evolution model , except for na , ar and ni which are not computed in the model and are scaled to the solar ratio @xcite . the models assume that the nebula is ionization bounded and no dust has been included in the chemical evolution models neither in the photo - ionization calculations . however dust grains mixed with the ionized gas have been partially taken into account , since we have included the depletion in refractory elements ( si , fe , ca , si , mg ) taken from @xcite . the grains can affect the absorption of the _ uv _ photons and decrease the electronic temperature . the density has been assumed constant for simplicity and equal to 100@xmath45 , which is appropriate for modeling hii galaxies @xcite and large circumnuclear hii regions @xcite , frequently found around the nuclei of starbursts and agns . although the constant density hypothesis is probably not realistic , it can be considered representative when the integrated spectrum of the nebula is analyzed . the shape of the ionizing continuum is defined by the pair of values ( @xmath49(ryd ) , log@xmath50 ) ( eq . [ sp ] ) obtained from the ionizing spectrum given directly by the evolutionary synthesis code . taking into account the simplified version used for our chemical evolution models , the free input parameters related with the infall time - scale and the molecular cloud formation disappear . the total mass of the initial gas is within the region from the initial time . the star formation rate is zero except during the bursts . we need to define the intensity of these bursts , and the time elapsed between them . each model is therefore characterized by three input parameters : * * 1 - the initial efficiency ( _ @xmath51 _ ) * : it is the amount of gas consumed to form stars in the first burst of star formation , that is , @xmath52 + we present here models computed with 2 values of @xmath51 : 1 . high efficiency models : the first burst of star formation involves 33 per cent of the total initial mass of gas ( 33@xmath5310@xmath54m@xmath1 ) , 2 . low efficiency models : the first burst of star formation involves 10 per cent of the total initial mass of gas ( 10@xmath5310@xmath54m@xmath1 ) . * * 2 - attenuation * : the star formation efficiency of the following bursts is attenuated in two different ways : 1 . by a factor which changes with the number of the burst , _ n _ , according to the expression : + @xmath55 + this attenuation type corresponds to a soft attenuation model ( hereinafter sam ) . 2 . by a constant factor , _ k@xmath56 _ , according to the expression : + @xmath57 + corresponding to a high attenuation model ( hereinafter ham ) . in this last expression _ k _ is the attenuation factor which , in this case , takes values from 0 to 1 . the lower the _ k _ value , the stronger the attenuation , the burst being less efficient each time ; the higher the _ k _ value , the weaker the attenuation , and then the stronger the bursts . we only show models with @xmath58 * * 3 - time between bursts ( * _ @xmath59 t _ * ) * : every burst takes place instantaneously and it is followed by quiet periods , whose duration can change . for this work we have taken @xmath59t= 1.3gyr for the inter - burst time , that is , one burst every 1.3gyr as the generic case , although we will also show some models with @xmath60gyr and @xmath61gyr in order to compare the different results . the models we show here are a selection of the ones calculated and described more widely in martn - manjn , moll & lpez - snchez , 2012 ( in preparation ) . in that work , the results of 20 models are compared with data for generic and particular low mass dwarf galaxies . we have selected the most relevant of them in order to show the effects of the input parameter variation on the resulting evolutionary history of each galaxy . the input parameters of these six models are shown in table [ inputs ] . .input parameters for the theoretical models . in column 1 it is defined the type of model according to the attenuation : soft ( sam ) or high attenuation ( ham ) ; the second column gives the parameter @xmath62 which defines the attenuation ; column 3 is the star formation initial efficiency @xmath51 , and column 4 shows the time between burst @xmath63 ingyr . [ cols="^,^,^,^,^ " , ] we will show here the main results obtained with the selected models . the * efficiency * determines the initial star formation rate and the initial metallicity of the gas . we have plotted in fig . [ fig1 ] the sfr for the 6 models . the first burst is strong in all panels , while the subsequent ones are less intense due to the decrease of the available gas to form stars and the attenuation . in a ) and b ) the last time computed is @xmath64gyr while in c ) and d ) the final time is around 1.3 and 0.6gyr respectively . in all panels we show the observational limits as dashed ( green ) lines . taking these limits into account , the dwarf galaxies may suffer 11 bursts for sam , or 8 - 9 for ham since later bursts show rates lower than observed . the oxygen abundances for these same models are shown in fig . [ fig2 ] . in this plot we can see that the two efficiencies chosen , 33 per cent and 10 per cent , give the upper and lower limits respectively for the observed oxygen abundance range in this type of galaxies . the models with the same initial efficiency and different attenuation type , are very similar , and if @xmath63 is shorter , the low oxygen abundance limit is reached very quickly . figs . [ fig3 ] and [ fig4 ] show diagnostic diagrams involving the ratios of intense emission lines . the number of the model ( see table 1 ) is given in each panel . the higher the efficiency , the more ionizing photons produced and the higher ionization parameter , leading to a higher excitation of the gas . the low efficiency models cover the region of the diagram occupied by the young and less metallic galaxies ( high [ oiii]@xmath55007,4959/h@xmath65 , low [ oii]@xmath83727/h@xmath65 and low [ nii]@xmath86584/h@xmath66 ) , as expected . the high initial efficiency models reproduce high excitation and high abundance galaxies , with high [ oiii]@xmath55007,4959/h@xmath65 and high [ nii]@xmath86584/h@xmath66 . the ionization degree is driven by the efficiency of the bursts , not by the attenuation mode or attenuation factor.thus , the models with different attenuation modes and the same initial efficiency do not show noteworthy differences , since it is the current burst of star formation which produces the observed emission lines , and the underlying population does not affect the ionization parameter . the attenuation sets the sfr of the successive bursts and determines the contribution of the underlying population . a higher attenuation implies a larger contribution of previous bursts to the total sed . therefore , it is important to see the attenuation factor effect on the the evolution of the broad - band continuum colors . in fig . [ fig5 ] we show the evolution of the modeled continuum colors for each burst of star formation . the inclusion of nebular emission continuum in the computation of the ssps reddens the colors of very young populations significantly , mainly at low metallicity as explained in mgvb09 . the ham cases , compared with sam , have a higher contribution from the non ionizing underlying continuum , which makes the colors redder for subsequent bursts and the young and blue population features disappear . meanwhile , sam maintains blue colors , characteristic of the current burst of star formation . in fig . [ fig6 ] the evolution of the colors is shown including the emission lines contribution to the wide band filters . we have done so taking into account the strongest emission lines that contribute to the color in each broad - band spectral interval at redshift zero . these are mainly [ oii]@xmath53727 in _ u _ , h@xmath65 in _ b _ , [ oiii]@xmath55007,4959 in _ v _ , h@xmath67 in _ r _ and [ siii]@xmath59069,9532 in _ i_. in that case the more intense the last burst , the larger the change in the colors , so the dotted lines are more similar to colors calculated only with the stellar populations than those shown by solid lines for higher efficiencies . on other hand , not all bands are equally modified at the same time . a fine tuning of the attenuation factor can be obtained from the evolution of the broad - band continuum colors versus the equivalent width of h@xmath65 , ew(h@xmath65 ) . these are important observations which give also an effective method to uncover the presence of old underlying stellar populations @xcite . the comparison of models and observations shows that in most star - forming dwarf galaxies , ssps are unable to reproduce the reddest colors shown by the data with the lowest ew(h@xmath65 ) values . in fact the observed h@xmath65 equivalent width values and colors require a large contribution from previous stellar generations @xcite . the evolution of ew(h@xmath65 ) _ vs. _ a pseudo - color of the continuum , the intensities of the adjacent continua of [ oii]@xmath83727 and [ oiii]@xmath85007 lines ( similar to u - v ) of our models compared with the data is shown in figure [ fig7 ] . the sam and ham cases for @xmath68gyr are plotted in the two first panels . it can be seen that to reproduce the trend of observational data , shifted to red colors at low values of ew(h@xmath65 ) with respect to the ssp predictions , the contribution to the total continuum of the non - ionizing red population must be much higher than the contribution of the current burst which creates the emission line spectrum . this trend can not be reproduced just with ssps neither by increasing their metallicity or age separately nor simultaneously ( mmdt ) . a strong attenuation in the sfr is needed , as can be seen in the center panel , to reproduce the whole range in ew(h@xmath65 ) and continuum colors simultaneously . the time between bursts ( @xmath59 t ) is a parameter which may also have an effect on the models similar to the attenuation . the reduction of the time between bursts offsets the effect of increasing the attenuation : the underlying population is less evolved and produces less reddening . in figure [ fig7 ] , right panel , ham with @xmath69gyr and @xmath70gyr are plotted . the ew(h@xmath65 ) decreases rapidly while resulting colors are not shifted to the red sufficiently to cover the range shown by the observational sample . the colors of the models with shorter inter - burst time are more similar to those with soft attenuation ( strong bursts ) , and require an extra reddening to reproduce the needed effects of the underlying non ionizing populations . however , the ew(h@xmath65 ) decreases more from burst to burst than in the case of a soft attenuation , where it maintains a high value at the beginning of every bursts . in order to reproduce the observed trend , the inter - burst time must be longer than 100myr . in brief , to reproduce the observable characteristics of star - forming galaxies we have to adjust the three input parameters : * the initial efficiency must be between 0.10 and 0.60 in order to produce a first burst which provides oxygen abundances within the observed values . in fact , efficiencies lower than 10% seems more probable to reproduce most of observations . * most ( perhaps all ) hii galaxies require the contribution of previous stellar generations to explain the observed trends shown by their continuum and emission line properties . the chemical and spectro - photometrical parameters ( equivalent widths and colors ) obtained by our models reproduce the observed relations if the contribution of the underlying population from previous bursts to the total continuum is higher than the contribution of the current burst of star formation which dominates the observed emission line spectrum . this implies a history of star formation higher in the past than at present , and , even in that case , attenuated burst along the time . the fine tuning of observations is obtained by the adjusting of the attenuation factor , once the initial efficiency is fixed . ( see martn - manjn et al . 2012 for details ) * the inter - burst time must be shorter than 1.3gyr and longer than 100myr . with shorter periods than 100myr the underlying continuum is too blue and its contribution to the color do not reproduce the trend shown by hii galaxies . three star formation scenario are usually discussed in relation with the evolution of dwarf galaxies , ( a ) burst : short star - formation episodes with large quiescent periods , ( b ) gasp : long moderate star - formation episodes with short quiescent periods and ( c ) continuous or almost continuous star formation with few over - imposed sporadic bursts . the three cases can be simulated by our models simply with a change of parameters : * the star - bursting scenario ( a ) is this one shown in the previous sections . the effects of an instantaneous burst can be seen in the emission lines during 10myr after the burst takes place . long quiescent periods or , in our case , null star formation periods , have been considered to last more than 1gyr , which would be the minimum age of the underlying population belonging to the previous burst . bursts occurred before the immediately previous one have very poor contribution to the current total continuum luminosity , and the use of this long time for the inter - burst period produces a similar result to assuming a two - burst model for some parameters , as ew(h@xmath65 ) . however , differences appear between both scenarios since the ism is enriched by every single burst occurring during the galaxy life time , stellar populations older than 1gyr should be present , and in consequence colors would be shifted to the red . * the gasping star formation scenario ( b ) can be simulated by substantially increasing the attenuation of the burst and reducing the inter - burst time . with the change in these two parameters we make a smoother and more moderate the star formation . since the subsequent bursts are closer in time , the quiescent periods are shorter . models with extreme attenuation and inter - burst times shorter than 100myr can be considered a good approximation to gasping models . this particular choice of parameters has important consequences when compared with the observations . increasing the attenuation a major contribution to the continuum by the previous bursts is obtained . however , if , in addition , the inter - burst time is reduced , the underlying population will become younger and bluer . this approach would predict the presence of intermediate age stellar populations , younger than 1gyr , even after several star bursts . with larger inter - burst times only the immediately previous stellar generation contributes substantially to the present continuum , but even older stellar generations would be noticeable if we further reduce the time between bursts . although complicated , it is also possible to simulate more extended star - burst phases by concatenation of several star - bursts . however , population synthesis modeling of bcd spectra does not favor extended periods of star formation @xcite . then , if the starburst phase is short , there should be many of these episodes during the galaxy lifetime . according to @xcite , there should be one bcd phase each 0.3gyr , which agrees with our models . * under a continuous star formation scenario ( c ) , the sfr must be very low and extended . if we reduce the inter - burst time to a minimum value and , simultaneously , we reduce the intensity of each burst , we obtain a very low star formation rate , not comparable to a starburst , but similar to a continuous star formation history . in fact , many of the observable features of star - forming galaxies can be modeled with a young stellar population 3 - 5myr old and most massive stars being essentially coeval . however , as @xcite demonstrated , it is also possible to obtain similar results with a continuous star formation , lasting at least 20myr since the beginning of the star formation . a low continuous star formation rate can not be neglected , especially in low metallicity galaxies . otherwise , if the sfr is very low , ( log ( sfr ) @xmath13 -3 approximately , according to * ? ? ? * models ) , there are not enough ionizing photons to produce emission lines . in this case we are not reproducing the so - called hii or bcd galaxies , but another type of galaxy or another stage of their evolution . actually , we may consider that , instead of a different scenario , the low and continuous star formation could be the phase occurring between bursts . in this way we could change the perspective of the problem and study the characteristics of the galaxies for the two phases : the burst and the following @xmath71myr , and the inter - burst , low activity periods , that is , after the first 10myr from the last star formation episode . in fact this is related with the two different time scales involved in the evolution , and related with different kind of data : on the one hand , a short time - scale in which the emission lines and othe effects of the ionizing stellar populations are observed . on the other hand there is long time - scale which defines the red color evolution mainly due to the the stellar populations older than 10myr . since a hii galaxy can be considered just a star - forming phase or a stage in the evolution of a gas rich dwarf galaxy , we can consider the inter - burst time in our models as a quiescent star - forming dwarf ( qbcd ) phase @xcite . then the hii galaxy may be in the phase 10@xmath72 yr after the star burst , in which its effects are still visible ( emission lines or changes in metallicity due to massive stars ejections ) and the qbcd phase would correspond to the inter - burst periods . the contribution of the young and old stellar populations in each galaxy may be obtained from the comparison of data coming from the two different time - scales , in particular , from observations as emission lines , proceeding from the ionizing stellar populations , with colors or abundances given by the older than 10myr stars . as we have already seen in our previous works , mmdt and @xcite , and we also show in fig . [ fig6 ] , when both types of stellar populations are contributing to the light , as occurs in bcd galaxies , the colors are contaminated by the emission lines , showing in most cases different trends than those standard sequences defined by the stellar populations , either young or old . taking into account this contamination , the colors observed are those corresponding to the ionizing continuum of a given galaxy , always during the bcd phase , that is , the colors of the young population which dominates the emitted light . in order to study the colors in the inter - burst periods and to know the properties of the underlying population , we should further eliminate the continuum arising from the starburst and isolate the properties of the underlying host galaxy . figures [ fig8 ] , [ fig9 ] and [ fig10 ] show the b - v _ vs. _ v - i , v - i _ vs. _ r - i and g - r _ vs. _ r - z color - color diagrams for our 6 models . black circles show the stellar continuum colors in all panels , while the colors computed including the contribution by the strongest emission lines are shown as green and red for models 1 and 2 respectively , brown and orange for models 3 and 4 , and blue and magenta for models 5 and 6 . in order to represent these older phases we have also represented , as colored open squares , the colors corresponding to ssps , as given in mgvb09 , for ages older than 10myr that is , without taking into account the youngest stellar populations phases , and metallicities from z=0.0001 to 0.02 . we have not computed exactly the colors of our inter - burst periods , but they must fall in this same region of the diagrams . in the b - v _ vs. _ v - i diagram ( fig . [ fig8 ] ) the inclusion of the emission lines contribution to the continuum colors shifts the position of the model points almost perpendicularly to the originally computed ones . the location of the points is mainly determined by the contribution of strong [ oiii]@xmath55007,4959 emission lines to the _ v _ band , and , logically , they are closer to the normal stellar populations locus when efficiency is low . this means that the excursion from locus defined by the main sequence for stellar populations is stronger when the star formation efficiency is higher , leading the points far . we also show in this figure the data by @xcite as cyan dots with error bars . this set of observations corresponds to readily observed colors , including both continuum and line emission , and no reddening correction has been applied . it can be seen that some of the data are impossible to be reproduced by the models which do not take into account the contribution by emission lines , even if some amount of reddening is invoked . on the other hand the data by @xcite of resolved locations in mrk 370 , shown in the figure as open green triangles are very well reproduced by our continuum colors . the data of mrk 370 are based on _ ubvri _ broadband and h@xmath66 narrow band observations . in this case , the authors subtracted the contribution of the underlying continuum from the old stellar population , removing the contribution from emission lines and correcting for extinction , measuring in this way true colors of the young star - forming knots . this way while open green triangles correspond to the observed mrk 370 colors , including the emission lines contribution and the underlying stellar population , the full green triangles are the stellar population colors after correction of extinction , emission lines and old stellar populations . in this case we see that some points fall out of the expected region of models which we interpret as probably due to a over - correction with ssps models . the underlying component has colors redder than the bcd ones , and they are also well reproduced by our models . in fig . [ fig9 ] we plot v - i _ vs. _ v - r , with the same data and code from @xcite and @xcite than in the previous figure . moreover , we have also included the data of bcds and its hosts from @xcite . this sample consists of 15 bcds for which they observed the total colors . the data corrected of extinction are the green stars . when they correct of the emission lines effect , the obtain the colors shown by magenta stars . these points corresponding to the actual young stellar populations are closer to the region where our models are represented by the same coloured dots as in fig . they subtracted the contribution of the underlying galaxy by assuming the mean surface brightness of the extensions , or hosts , to be constant and representative of the underlying galaxy within the starburst regions . summarizing , they give separately the colors of this underlying stellar continuum ( blue stars ) within the starburst , the contributions including the emission lines ( green stars ) and the colors of the bcd stellar continuum ( magenta stars ) , corrected from emission lines , which are close to black dots , corresponding to our continuum bcd phase . the colors of the isolated host galaxy ( blue stars ) show redder colors than the bcd component , in agreement with qbcds , mrk 370 host and model colors with ages older than 10myr . again we note some differences with our models that we assign to an over - correction of the emission lines . finally we represent in fig.[fig10 ] the g - r vs. r - z diagram with the same symbols that before . the selected qbcds @xcite are assumed to be like bcd host galaxies @xcite . the colors of the ssps with ages older than 10myr ( colored squares ) can reproduce these observations , indicating that they correspond to the colors of a more evolved population than the one corresponding to the current burst in bcds . however , some of the observed dots lie in the bcd zone , indicating that this data sample also contains a certain proportion of bcds , following a continuous evolutionary sequence . in this color - color diagram there is a larger difference than in the previous ones between models and data for the bluest points , showing these data higher g - r values than predicted for similar low r - z colors . probably this is due to the transformation used to calculate sdss colors with our johnson colors , which would be not valid for this type of objects with emission lines contaminating the wide band filter magnitudes . we see in the three figures that models start to separate off the main sequence of colors at different points following the type of attenuation . in the top panel , there are points with large contribution of lines starting even at the lowest and left corner , where the youngest stellar populations lie . the same points in medium panel begin to separate at redder colors , this implies that the underlying continuum corresponds to an older age . these models cover all observational range . however in the bottom panel the points with the contribution of the lines fall in a region much bluer where there are no observations . the underlying stellar populations are much younger than the populations of the data . in any case it seems that bcds and qbcds colors are well reproduced by the same basic models , during the first 10myr after the beginning of star formation and after 10myr up to more than 1gyr respectively , and therefore we can say that both types of galaxy colors overlap and could be considered as different phases in the evolution of the same object . historically , the most common scenario assumed for the star formation history of the star - forming galaxies is that stars form in a bursting mode . this scenario consists in successive instantaneous star formation bursts interposed between long periods of null or very low star formation activity . if these periods are shorter and the bursts are more extended and moderate , we are under a gasping star formation scenario . on the contrary , if the star formation is low and moderate during the whole life of the galaxy , and only some sporadic bursts take place , a continuous star formation scenario is appealed . we have analyzed the possibilities of these scenarios by means of theoretical models based on the successive bursts star formation hypothesis . each galaxy is modeled assuming an initial amount of unprocessed gas of 10@xmath73 m@xmath1 . the evolution is computed along a total duration of 13.2gyr during which successive star - bursts take place . since any of the possible scenarios of the star formation histories described show evidences of being able to explain the observed data , we have tried to check the hypotheses suggested . the grid of theoretical models are computed by the combination of three tools : a chemical evolution code , an evolutionary synthesis code and a photo - ionization code , all of them previously calibrated . they have been used in a self - consistent way , _ i. e. _ taking the same assumptions about stellar evolution and nucleosynthesis , and the resulting metallicity in every time step . these models have three free input parameters which can be changed to obtain different model results : ( a ) the initial star formation efficiency which determines the initial star formation rate ( sfr ) and the initial metallicity of the gas , ( b ) the attenuation of the successive bursts , which determines the evolution of the gas , keeping metallicity between the observational data limits , and ( c ) the inter - burst time , which sets the age of the non ionizing underlying population . we have shown that ( a ) also leads the behavior of the ionizing gas since emission lines are determined by the number of massive stars and the metallicity of the emitting gas . with this parameter we can control the number of old pristine stars that we are going to find in the next star formation burst . the strength of the bursts is controlled by ( b ) . the more attenuated , the less the star formation rate , which favors the previous stellar generations to have more weight on the total current spectrum , hence modifying the equivalent width and the colors of the continuum . finally , the contribution of the underlying population can be determined by ( c ) . these periods are the quiescent phases of the bcd . the shorter the inter - burst time , the younger the underlying stellar populations . these stars will be part of the host bcd galaxy , that is , all those generations of stars which have not ionizing stars , showing ages of more than 10@xmath72 yr . the models can reproduce simultaneously the data relative to the current ionizing population and the data which give us information about the evolutionary history of the star formation . in order to reproduce the observational data , the parameters are constrained : * the initial star formation efficiency must be lower than 50% and more probably is around 20% . * the subsequent star - forming burst must be attenuated , in order to have the adequate proportion of underlying to young / ionizing stellar populations ; and * the inter - burst time must be longer than 100myr but shorter than 1.3gyr . with our models we are able to compute the stellar continuum colors as well as the broad band colors corrected from the contribution of the more intense emission lines . this effect shifts the colors almost perpendicularly to the stellar continuum ones , specially during the first bursts of star formation , when the emission lines are more intense . the contamination of the continuum by the emission lines can make the observed galaxies appear younger than they really are , hiding evidence of underlying old populations . on the other hand , the different characteristics shown by the star formation histories of our models , imply that we may be observing the same objects in different evolutionary stages , that is , just during the burst , in the post - burst phase or in the inter - burst periods . if we consider that the different star formation scenarios correspond to different phases in the life of a single galaxy , our models are able to reproduce the characteristics of each one of these phases . considering the bcd phase as the first 10myr after a star formation burst , and the quiescent phase as the inter - burst periods , we can reproduce the galaxies both in a high activity and in a post - starburst ( or pre - starburst ) phase . the main difference between these phases are the existence of emission lines produced by the ionized gas . their intensity , and their contribution to the total continuum color , can be modified by the initial efficiency . with the attenuation factor we can produce a more or less reddened stellar continuum due to the contribution of the host galaxy , the age of these old stars being determined by the inter - burst time . this work has been partially supported by dgicyt grants aya200767965-c03 - 03 and aya201021887c0402 . also , by the comunidad de madrid under grant s-0505/esp/000237 ( astrocam ) and by the spanish micinn under the consolider - ingenio 2010 program grant csd2006 - 00070 : first science with the gtc ( http://www.iac.es/consolider-ingenio-gtc ) which are acknowledged . rt is grateful to the mexican research council ( conacyt ) for supporting this research under grants cb-2006 - 49847 , cb-2007 - 01 - 84746 and cb-2008 - 103365-f gallagher , j. s. , mould , j. r. , de feijter , e. , holtzman , j. , stappers , b. , watson , a. , trauger , j. , ballester , g. e. , burrows , c. j. , casertano , s. , clarke , j. t. , crisp , d. , griffiths , r. e. , hester , j. j. , hoessel , j. , krist , j. , matthews , l. d. , scowen , p. a. , stapelfeldt , k. r. and westphal , j. a. , 1996 apj , 466 , 732 g

the modulation - doped algaas / gaas heterostructure is a materials platform of great importance to the study of nanoscale electronic devices with quantum mechanical functionalities , and their development towards future technologies . @xcite studies of the two - dimensional electron gas ( 2deg ) formed in an algaas / gaas heterostructure at temperatures below @xmath0 k have met with great success , both in terms of the novel physics of 2d electrons , @xcite and as an underpinning technology for quantum wires , @xcite quantum dots , @xcite and other ballistic transport devices . @xcite devices based on two - dimensional hole gases ( 2dhgs ) have received less attention ; this is not from a lack of interesting physics . the higher effective mass of holes leads to stronger carrier interactions , making 2dhgs of interest for studies of the metal - insulator transition @xcite and bilayer quantum hall effect . @xcite additionally , the spin-@xmath1 nature of holes , arising from strong spin - orbit interactions , has driven interest in novel phenomena in 2dhgs , such as @xmath2-factor anisotropy @xcite and anomalous spin - polarization , @xcite as well as studies of the @xmath3 plateau in quantum point contacts ( qpcs ) , @xcite the quantum dot kondo effect @xcite and berry s phase in aharonov - bohm rings . @xcite the reduced hyperfine interaction for holes in gaas @xcite leads to reduced spin - decoherence time compared to electrons , @xcite making gaas hole quantum dots of interest for quantum computing . @xcite studies of low - dimensional hole devices are impeded by difficulties in making devices with high electronic stability and low noise / drift under electrostatic gating . telegraph noise , instability and gate hysteresis were particularly problematic in initial attempts to realize hole qpcs @xcite using si - doped ( 311)a - oriented 2dhgs . similar issues were reported for gated 2dhgs in c - doped ( 100 ) algaas / gaas heterostructures . @xcite numerous explanations have been offered including surface diffusion of ohmic contact metal producing a low mobility layer at or close to the surface , @xcite charge transfer to states either at the interfaces or in the semiconductor , @xcite carrier trapping in deep acceptor levels or in insulating parallel doping layers , @xcite or that metallic gates on p - gaas are ` inherently leaky ' because of a reduced schottky barrier relative to n - gaas . @xcite firmly establishing the origin of the gate instability / hysteresis will contribute towards the development of improved materials and devices for the study of low - dimensional hole systems . here we show that gate instability / hysteresis in p - type modulation - doped algaas / gaas heterostructures is caused by a complex interplay between surface - state trapping and gate - induced charge migration within the doping layer . we focus here on si - doped ( 311)a - oriented heterostructures , but similar physics may occur in c - doped ( 100)-oriented heterostructures also . we use three different experimental approaches : first , we rule out direct charge leakage between the gate and semiconductor by studying devices where an insulator layer is deposited underneath the gates ; two insulators , al@xmath4o@xmath5 and polyimide , were investigated . we find that insulating the gates does not eliminate the hysteresis , instead , it makes the hysteresis significantly worse . a recent study of a qpc featuring hfo@xmath4-insulated gates on a c - doped ( 100 ) algaas / gaas heterostructure @xcite also shows that hysteresis can remain despite insulating the gates . in our experiment , the semiconductor surface / gate interface is the only aspect that is different between the insulated and uninsulated gate devices ; they are on the same heterostructure with the same doping . the resulting large difference in hysteresis points to surface - states as an important contributing factor . hence our second approach was to investigate surface effects by studying matched algaas / gaas heterostructures , both with and without an al@xmath4o@xmath5 gate insulator layer , and schottky - gated p - type heterostructures with ( nh@xmath6)@xmath4s@xmath7 surface passivation treatment @xcite performed prior to gate deposition . in contrast to hole devices , the addition of the insulator does not induce hysteresis for electron devices but alters the pinch - off voltage . sulfur passivation does not bring consistent improvement in p - type devices , despite increasing the photoluminescence yield equivalently on ( 100 ) and ( 311 ) surfaces . @xcite in the one instance where passivation did bring significant improvement , hysteresis was still observed . this led us to suspect charge migration in the doping layer . hence our third approach was to investigate dopant effects through variable temperature studies . in particular , we compare the hysteresis in hole devices to electron devices at elevated temperatures @xmath8 k , where deep donors known as dx centers @xcite begin to detrap , @xcite allowing charge to migrate between si dopant sites . we observe gate hysteresis in electron devices at @xmath9 k very similar to that in hole devices at @xmath10 k. we also show that the hysteresis can be reduced in hole devices by reducing the thickness of the dopant layer from @xmath11 nm to less than @xmath12 nm ( @xmath13-doped ) . overall , our results point to competition between surface - state and dopant related processes with different time / energy scales as the cause of the hysteresis , and have wider implications given recent interest in surface charge as a source of scattering @xcite and dopant charge migration as a source of noise @xcite in algaas / gaas heterostructures and quantum devices . the paper is structured as follows . section ii briefly addresses materials and methods , with an extended discussion presented in appendix a. in section iii we focus on gate leakage and discuss hole devices featuring al@xmath4o@xmath5 and polyimide gate insulators . sections iv and v concentrate on surface - states and charge migration in the dopant layer , respectively . finally , in section vi , we draw conclusions based on the overall results and discuss the broader implications of the work for the study of low - dimensional devices in algaas / gaas heterostructures . appendices b - d contain additional supporting data . five separate algaas / gaas heterostructures denoted 1-e , 1-h , 2-e , 2-h and 3-h were made for this study ; the first four are ` matched ' electron ( -e ) and hole ( -h ) heterostructures produces by deposition onto side - by - side halves of 2 " diameter gaas substrate , one ( 100)-oriented and the other ( 311)-oriented . this relies on the orientation - dependent amphoteric nature of si dopants in algaas . @xcite the first four wafers feature an @xmath11 nm si - doped algaas modulation doping layer separated from the 2dhg/2deg by @xmath14 nm of undoped algaas , the fifth features a si - doping layer separated from the 2dhg/2deg by @xmath15 nm of undoped algaas . these heterostructures have typical carrier densities @xmath16 @xmath17 and mobilities @xmath18 @xmath19/vs at temperature @xmath20 mk . devices were produced by standard gaas device processing methods and involved definition of a hall bar structure by wet etching , deposition and annealing of ohmic contacts and deposition of ti / au gates and interconnects . insulated gate samples feature a @xmath21 nm al@xmath4o@xmath5 layer deposited using atomic layer deposition at @xmath22c after hall bar definition , with access for ohmic contacts provided using a buffered hf etch . the polyimide insulated gate sample was produced using photo - processable polyimide ( hd microsystems ) deposited after the ohmic contact anneal . sulfur passivation was performed immediately before gate deposition by @xmath23 min immersion in a @xmath24 dilution of stock ( nh@xmath6)@xmath4s@xmath7 solution prepared by adding @xmath25 g of elemental sulfur ( aldrich ) to @xmath26 ml of @xmath27 ( nh@xmath6)@xmath4s solution ( aldrich ) . the passivation is performed at @xmath28c and care is taken to minimize air exposure between passivation and gate / interconnect deposition . electrical measurements at @xmath29 k and @xmath30 k were obtained using a liquid helium dip - station and an oxford instruments heliox @xmath31he cryostat , respectively . standard two- and four - terminal ac lock - in techniques were used to measure the conducting channel s drain current @xmath32 , typically with a @xmath33v constant voltage excitation at @xmath34 hz applied to the source . the gate bias @xmath35 was applied using a keithley 2400 source - measure unit enabling continuous measurement of gate leakage current @xmath36 down to @xmath26 pa . more complete details of heterostructure growth , device fabrication and electrical measurement are given in appendix a. a common explanation for instability and gate hysteresis in schottky - gated p - type algaas / gaas heterostructures is charge leakage from the metal surface - gate into the heterostructure when the gate is positively biased to deplete the 2dhg . this often relies on the argument that schottky gates on p - type heterostructures are inherently more leaky than on n - type heterostructures . @xcite ref . @xcite suggests this occurs because : ( a ) gaas surface states cause the surface fermi level to be pinned slightly closer to the valence band than the conduction band , making the schottky barrier lower for p - type heterostructures , @xcite and ( b ) depletion of the hole gas requires the schottky gates to be forward biased . the latter argument is incorrect depletion is a reverse bias process whether the underlying structure is n - type or p - type ; the bias convention is reversed under an inversion in doping type . @xcite the former argument requires caution for two reasons . firstly , in most heterostructures the gaas cap is intrinsic , and the dopants in the heterostructure , whether n- or p - type , are in the algaas layers more than @xmath12 nm beneath the surface . hence the schottky barrier is the same for either doping type , suppressing any relative difference in gate leakage . secondly , schottky barrier measurements are usually performed at @xmath20 k , where thermionic emission dominates any leakage current , @xcite whereas most quantum devices @xcite are studied at @xmath37 k , where thermionic emission is quenched due to its exponential temperature dependence . this should further suppress the leakage current difference between schottky gates on n- and p - type heterostructures . on a log - axis vs gate bias @xmath35 for a schottky - gated modulation - doped p - type heterostructure . at positive @xmath35 , @xmath36 remains less than @xmath38 pa up to @xmath39 v , sufficient to achieve pinch - off for all uninsulated gate devices studied . in an n - type heterostructure , gate leakage would normally occur at @xmath40 v ( indicated by the arrow ) and is suppressed for negative @xmath35 . ( b ) channel current @xmath32 vs gate voltage @xmath35 at six different gate sweep rates for device a. the horizontal arrows indicate the direction of travel around the hysteresis loop.,width=302 ] thus metal gates on p- and n - type heterostructures should behave similarly at low temperature aside from a sign - reversal in bias convention ( forward / reverse ) . to confirm this , fig . 1(a ) shows the gate leakage current @xmath36 versus gate bias @xmath35 for a schottky - gated p - type heterostructure ( this data is presented on a linear @xmath36 axis for comparison in appendix b ) . the gate begins to pass significant current under forward bias conditions at @xmath41 v. under reverse bias , where 2dhg depletion occurs , the leakage current gradually rises as @xmath35 is made increasingly positive , with @xmath42 pa for the entire range @xmath43 v ; this is sufficient to achieve pinch - off in all of our uninsulated devices . in contrast , a negative @xmath35 , i.e. , forward bias , produces strong gate leakage at @xmath44 v. the misconception that schottky gates on p - type algaas / gaas heterostructures leak at positive @xmath35 may arise from the knowledge that schottky gates on n - type algaas / gaas heterostructures leak for positive @xmath45 v ( indicated by the arrow in fig . reverse bias leakage for n - type heterostructures usually does not occur until well beyond @xmath46 v , consistent with the reverse bias behavior for the p - type heterostructure in fig . despite this , schottky gates on n - type algaas / gaas heterostructures leak relatively tiny amounts of charge under normal operation . @xcite this was demonstrated by pioro - ladrire _ et al _ , @xcite using a device consisting of a small quantum dot , isolated from the adjacent 2deg , and coupled to a quantum point contact ( qpc ) charge sensor ( see fig . 7 of ref . the qpc detects charge leaking from the schottky gates forming the dot , with extremely small leakage currents @xmath47 a observed . these tiny currents are often associated with charge noise in n - type heterostructures , @xcite but are not commonly known to produce strong gate hysteresis in n - type heterostructures to the level seen in p - type devices ; the gates in electron devices are generally very stable ( see section iv - a ) . given this , it is not possible using fig . 1(a ) and the earlier discussion alone to definitively rule out gate leakage as a possible cause of gate hysteresis / instability in p - type heterostructures . the ultimate test , insulating the gates from the heterostructure , will be presented in section iii - c ; first we characterize the hysteresis we observe in schottky - gated 2dhgs . device a has ti / au gates deposited directly on the surface of 1-h . figure 1(b ) shows the measured drain current @xmath32 versus gate voltage @xmath35 from device a at six @xmath35 sweep rates between @xmath48 and @xmath49 mv / s . hysteresis occurs for all @xmath35 , but is most prominent for @xmath50 v. note that @xmath32 reflects the channel conductivity , which can vary due to changes in either carrier density or mobility . hall measurements with a small perpendicular magnetic field applied versus @xmath35 show a qualitatively identical hysteresis to that in fig . 1(b ) , confirming that changes in @xmath32 with @xmath35 are predominantly density related . the direction of travel around the hysteresis loop ( indicated by horizontal arrows in fig . 1(b ) ) provides important clues about the origin of the hysteresis . first , it allows the most obvious cause of apparent hysteresis : recording delay in the measurement apparatus ( sweep lag ) to be ruled out . any delay on the upsweep to positive @xmath35 causes a given depletion to occur at a higher apparent @xmath35 , on the downsweep to @xmath51 , reaccumulation of carriers occurs at a lower apparent @xmath35 . the net result is a clockwise hysteresis loop the direction of travel in fig . 1(b ) is clearly counterclockwise . another key characteristic of sweep lag is that the hysteresis is strongest where the derivative @xmath52 is greatest . in fig . 1(b ) the hysteresis is strongest where @xmath52 is the smallest ( see appendix c ) . these two observations confirm that the hysteresis is not an instrumental issue ; it instead originates within the device . the effect of gate leakage depends on where the charge leaks to . in an entirely dc measurement , charge leaking directly to the 2dhg would add / subtract from the channel current . this would modify the @xmath32-@xmath35 characteristics but should not produce hysteresis . here we measure @xmath32 using an ac lock - in technique ; any gate leakage direct to the 2dhg does not appear in the measured @xmath32 versus @xmath35 unless it is an ac current at the reference frequency . an alternative is that charge leaks into trap states between the gate and 2dhg ; this will produce hysteresis if the trap time is not very small . because a positive @xmath35 is applied to the gates to deplete the 2dhg , any charge leaking from the gate will be positive , and its closer proximity to the 2dhg will increase the depletion at a given @xmath35 on the downsweep . similarly , for a 2deg the leaked charge will be negative , and it will also increase depletion ; this is what produces the downward steps in the downsweep arm of the gate characteristics in fig . 7(a ) of ref . returning our attention to holes , on the upsweep , reaccumulation in the 2dhg will be delayed by the need for the trapped positive charge to dissipate . this produces a clockwise hysteresis loop for holes , opposite to that in fig . 1(b ) and consistent with our argument in section iii - c below that gate leakage does not cause the hysteresis . note that gate leakage corresponds to an counterclockwise hysteresis loop for electrons due to the hysteresis loop mirroring about @xmath51 for carrier sign inversion ; this is exactly the loop direction obtained in fig . 7 of ref . @xcite where gate leakage does cause the hysteresis . the steps in ref . @xcite arise from the high sensitivity of their measurement configuration , single / few trap resolution will not be observed in the large area gate devices studied here . note that the hysteresis loop for the hole qpc in ref . @xcite is counterclockwise , opposite the direction expected for hysteresis due to gate leakage . we will present a possible explanation for the counterclockwise loop direction in section iii - d , first , we continue towards ruling out gate leakage as the cause of the hysteresis by looking at insulated - gate devices . and ( b ) gate leakage current @xmath36 versus gate voltage @xmath35 at three different gate sweep rates for device b ( @xmath21 nm al@xmath4o@xmath5 layer on 1-h ) . horizontal arrows indicate sweep direction . in ( b ) the left / right axis and lower / upper data are for the up / down sweep , respectively . once @xmath36 reaches @xmath53a the gate voltage source implements current limiting , holding @xmath35 fixed . hence for @xmath54 v the data in ( a ) should be considered as @xmath55 versus time @xmath56 at fixed @xmath35 , with each @xmath57 v minor tick in @xmath35 corresponding to @xmath11 , @xmath58 and @xmath21 s for sweep rates of @xmath59 , @xmath12 and @xmath49 mv / s , respectively.,width=302 ] figure 2(a / b ) shows @xmath32 and @xmath36 versus @xmath35 for device b , with ti / au gates insulated by a @xmath21 nm al@xmath4o@xmath5 layer . we would expect hysteresis to be heavily suppressed in this device if it is caused by gate leakage . instead the hysteresis becomes stronger , and we need to drive @xmath35 to al@xmath4o@xmath5 insulator breakdown to even approach pinch - off . a higher pinch - off voltage is expected ; device b has increased gate-2dhg separation and an extra dielectric layer . we find a conversion factor @xmath60 relating the insulated and uninsulated gate biases @xmath61 and @xmath35 using a parallel - plate capacitor model , assuming gaas , algaas and al@xmath4o@xmath5 dielectric constants of @xmath62 , @xmath63 and @xmath64 , respectively . it is clear from figs . 1(b ) and 2(a ) that the shift in pinch - off bias from @xmath65 v to @xmath66 v , i.e. , by a factor of @xmath67 , far exceeds that expected from simply adding @xmath21 nm of al@xmath4o@xmath5 . to confirm that device b s large pinch - off voltage is not a fabrication problem , we measured a second device differing only in top - gate pattern / area ( see appendix d ) . this gave similar behavior , with a large @xmath68 v required to achieve even modest depletion ( @xmath69 reduction in @xmath32 ) . the features in figs . 1(b ) and 2(a ) are qualitatively identical , but we highlight two places where the quantitative differences are substantial . firstly , the current plateau at intermediate @xmath35 is much longer for device b , delaying pinch - off accordingly . this creates the illusion that the initial drop in @xmath32 in the upsweep is steeper ; the opposite is true with the initial drop in fig . 1(b ) complete by @xmath70 v whereas the same drop in @xmath32 takes @xmath71 v in fig . secondly , there is a distinct asymmetry in the sweep - rate dependence of upsweeps and downsweeps : the upsweep path depends heavily on sweep - rate while the downsweep path is largely independent of sweep - rate . this is also evident in fig . 1(b ) , but is heavily exacerbated by the al@xmath4o@xmath5 layer . the insulator - semiconductor interface can have a radical effect on performance in devices with a shallow conducting channel . @xcite oxide insulators are particularly troublesome due to high interface trap densities ; improvement is often obtained using polymeric insulators e.g. , polyimide . @xcite to explore this we studied device c , which features gates insulated with @xmath72 nm of polyimide . this gave qualitatively similar results to device b ( see appendix d ) . pinch - off can not be achieved in device c either , largely due to the much lower breakdown voltage @xmath73 v for the polyimide layer . at breakdown , @xmath55 has only fallen by @xmath74 of its @xmath51 value . a possible explanation for an counterclockwise hysteresis loop and its particular shape in fig . 1(b ) is the gradual population / depopulation of a layer of _ net _ negative charge between the gate and 2dhg ( n.b . , the charge on the gate is positive , this can not be gate leakage ) . on the upsweep , the positive @xmath35 places positive charge on the gate , and this ideally results in depletion of the 2dhg ( i.e. , reduced carrier density ) . but , if adding positive charge to the gate instead results in net negative charge accumulation between gate and 2dhg , then depletion stalls , producing a reduced transconductance @xmath75 and perhaps even an @xmath32 plateau . the link between a plateau in @xmath32 versus @xmath35 and stalling of 2dhg depletion was confirmed by a corresponding plateau in hall measurements of the 2dhg . if the capacity for net negative charge accumulation is finite , depletion eventually resumes , leading to pinch - off ( @xmath76 ) . on the downsweep , positive charge is gradually removed from the gate , leading to 2dhg repopulation and loss of the accumulated net negative charge . if the net negative charge is held in deep trapping sites , its loss may be very slow . repopulation of the 2dhg will occur first , producing a rapid rise in @xmath32 to a level close to its initial value at @xmath51 , followed by a long current plateau extending to @xmath51 as the net negative charge is lost . an immediately obvious mechanism is charge trapping by surface states ; it is well - known in both oxide - insulated iii - v fets @xcite and organic semiconductor fets . @xcite another possible mechanism is charge migration in the modulation - doping layer . here the use of ` net ' charge accumulation is deliberate and important . silicon donors in the al@xmath7ga@xmath77as modulation doping layer of an n - type heterostructure can take two different configurations : a shallow hydrogenic donor occupying a ga site , or for @xmath78 , a metastable donor where the si atom is displaced into an interstitial position along the @xmath79 direction by lattice distortion . @xcite the dx centers act as deep traps ; when the device is cooled below @xmath80 k , the dx centers capture free electrons in the doping layer to become dx@xmath81 , locking some fraction of the remaining hydrogenic donors in a positive charge state . this ` freezing ' dopant layer charge is vital to the stability and reproducibility of the electronic properties of devices based on si - doped n - type algaas / gaas heterostructures at low temperatures . @xcite comparatively little is known about the defect physics of si dopants in ( 311)-oriented p - type algaas / gaas heterostructures . in addition to acting as a simple substitutional acceptor by occupying an as site , si has also been suggested to form an acceptor complex / deep trap known as si - x . @xcite initial studies @xcite suggested that si - x consisted of a si@xmath82 acceptor adjacent to a ga site vacancy v@xmath83 ( subscript denotes site occupied ) . raman and ir spectroscopy regarding the existence of si - x in heavily si - doped ( 311)a gaas layers is controversial , with data suggesting that it does @xcite and does not @xcite exist . further work , suggested modified structures for si - x , first as a v@xmath83-si@xmath82-as@xmath83 complex , @xcite later ruled out in favor of a perturbed si@xmath83-v@xmath83 center . @xcite these studies were all for gaas ; the existence and properties of si - x - like complexes in al@xmath84ga@xmath85as is unknown . we comment further on possibilities for si acceptor complexes and deep traps in section vi . returning to hysteresis mechanisms , if the dopant layer traps are shallow or thermal energy is sufficient for a high detrapping rate , charge can migrate in response to the balance of charge between the gate and 2dhg . for example , when @xmath51 it is energetically more favorable for the 2dhg side of the doping layer to be net negative and the gate side net positive . at pinch - off , the positively charged gate and depleted 2dhg favour the opposite ( 2dhg side net positive / gate side net negative ) . this gate - induced ` tidal flow ' of doping layer charge is equivalent to negative charge accumulation between gate and 2dhg towards generating an counterclockwise hysteresis loop , as demonstrated in section v. to better understand the comparative behavior of devices a and b , we prepared devices d and e on 1-e : device d has no al@xmath4o@xmath5 layer , device e has a @xmath21 nm al@xmath4o@xmath5 layer grown simultaneously with that in device b. figures 3(a / b ) present the electrical characteristics for devices d and e. in each case the apparent hysteresis is minimal and the loop direction is consistent with sweep lag ( n.b . , reflection of gate characteristics about @xmath51 due to sign - inversion reverses loop direction for electrons sweep lag goes counterclockwise ) . on its own , the lack of device - induced hysteresis in devices d and e does not allow us to pinpoint the origin of the hysteresis solely to surface - states or charge migration , as the shift from 1-h to 1-e entails a change in both the surface orientation and the dopant physics . while the ( 100 ) gaas surface consists only of double - dangling bonds , the ( 311 ) surface contains equal densities of single- and double - dangling bonds . @xcite vs gate voltage @xmath35 for devices ( a ) without ( device d ) and ( b ) with a @xmath21 nm al@xmath4o@xmath5 gate insulator ( device e ) on 1-e . the inset to ( b ) shows a close up of the data in the main panel to highlight the hysteresis . the arrows indicate hysteresis loop direction . the purple dot - dashed lines are guides to the eye highlighting the low - bias non - linearity.,width=302 ] figure 3 has two notable features : the first is pinch - off bias , which increases from @xmath86 v to @xmath87 v upon addition of the al@xmath4o@xmath5 layer . this @xmath88 increase in pinch - off bias is less than the @xmath89 found for holes , but more than double the @xmath90 expected from a parallel - plate capacitor model . the second feature is the distinct non - linearity in @xmath32 versus @xmath35 at low bias in device e. although this low bias behavior in fig . 3(b ) looks different to that in fig . 2(a ) , in both cases it represents a reduced depletion rate for a given change in gate bias , and may have similar origin . a comparison of figs . 3(a ) and ( b ) suggests the low - bias non - linearity arises from the al@xmath4o@xmath5 layer ; we believe it is caused by competition between the filling of surface states and 2deg depletion . this hypothesis is further supported by extrapolating the linear trend at moderate @xmath35 to lower @xmath35 , as per the purple dot - dash lines in fig . 3(a / b ) . the vertical purple arrows indicate where the low - bias non - linearity ends ; @xmath91 v and @xmath92 v to the right of pinch - off in fig . 3(a ) and ( b ) , respectively . the latter is @xmath93 the former , very close to the @xmath90 expected with the al@xmath4o@xmath5 layer . the @xmath94 difference between the actual value of @xmath92 v for device e and the expected value of @xmath95 v is well explained by the @xmath94 difference in measured @xmath51 carrier density between devices d and e. the findings above suggest that most 2deg depletion occurs in the linear region in fig . 3(a / b ) , with surface - state filling dominant for @xmath96 v in device d and @xmath97 v in device e. another way to envision this is as threshold voltage shift induced by the surface - states ; as per oxide - insulated iii - v fets @xcite and organic semiconductor fets . @xcite a final point of note is that the data in fig . 3(a ) is not linear all the way to @xmath51 . this suggests surface - states have a measurable impact on schottky - gated devices also . this is not surprising ; one naturally expects a finite surface - state density for uninsulated gaas surfaces also . one approach to reducing the surface - state density is chemical passivation , the aim being to remove the native oxide and covalently satisfy all of the ga- and as- dangling bonds . this ideally shifts the surface - states out of the band - gap and into the valence or conduction bands . @xcite passivation is commonly achieved using aqueous and alcoholic chalcogenide solutions , particularly those containing sulfur e.g. , na@xmath4s @xcite or ( nh@xmath6)@xmath4s . @xcite a comprehensive review of chemical passivation of iii - v surfaces is provided by lebedev . @xcite vs gate voltage @xmath35 for ( a ) device f on 1-h and ( b ) device g on 1-h with sulfur passivation . the data in ( a ) is what we typically obtain with sulfur passivation , even with different treatments . @xcite the data in ( b ) is a repeatable measurement , but to date , a non - reproducible device . we are still working to establish reliable conditions that produce this outcome . the solid black and dashed blue traces were obtained at sweep rates of @xmath49 and @xmath59 mv / s . the arrows indicate hysteresis loop direction.,width=302 ] the best approach to sulfur passivation involves difficult decisions amongst competing benefits . for example , while na@xmath4s treatment produces a surface passivation that is more robust to light / oxygen than ( nh@xmath6)@xmath4s treatment , @xcite the latter produces surfaces with less o , more s and no traces of na . @xcite alcoholic solutions are more effective than aqueous solutions , @xcite but alcoholic solutions are incompatible with photolithography resist , making them difficult it implement with patterned gates . @xcite the passivation treatment used here results from an extended study of different approaches to sulfur passivation of patterned - gate hole devices that will be reported elsewhere . @xcite figure 4(a / b ) shows gate hysteresis data from passivated devices on 1-h ( device f / g ) . we usually obtain data like that in fig . 4(a ) on including sulfur passivation in the fabrication process ; pinch - off voltages between @xmath98 and @xmath99 v are typically obtained depending on the treatment formulation . @xcite the device g data in fig . 4(b ) represents an isolated instance where elimination of the current plateaus and a much lower pinch - off voltage @xmath100 v was observed . note that a small counterclockwise hysteresis loop remains , consistent with that obtained when we examine the hysteresis generated by charge reorganization in the dopant layer in section v. it is also consistent with the hysteresis we observe at @xmath101 mk and the data in ref . caution is needed with fig . 4(b ) because although the measurement itself is repeatable , i.e. , if we remeasure this device we get the same result , after several months work we are unable to produce another device showing the same behavior . the loss of the current plateaus in fig . 4(b ) bears further discussion . comparing with fig . 4(a ) and earlier data , device g pinch - off occurs before the current plateau normally begins . the initial depletion is very strong compared to the other devices ; by @xmath102 v the current has dropped to zero as opposed to the @xmath103 found in other devices without the @xmath51 conductivity being significantly less than normal . the very similar @xmath32 at @xmath104 for devices f and g shows that the radically different pinch - off voltages are not due to a correspondingly large difference in @xmath105 carrier density . although the rapid initial depletion in device g shows a strong reduction in the effect of the surface states on @xmath32 versus @xmath35 , it is unclear this requires a major change in the surface state spectrum . a reduction in surface state density in the tail of the distribution causing the current plateau may suffice to ensure that depletion is completed before current plateau onset , giving the radical difference in pinch - off voltage between devices f and g. putting device g aside momentarily , our work suggests that passivation does little to reduce the hysteresis . @xcite an apparent initial explanation is that the passivation solution is ineffective on ( 311 ) surfaces there are no prior studies on this gaas surface , development has focussed on the more commonly used ( 100 ) , ( 110 ) and ( 111 ) surfaces . @xcite we have performed comparative studies of the efficacy of sulfur passivation on ( 100 ) and ( 311 ) gaas surfaces using photoluminescence measurements , @xcite and find similar improvement in photoluminescence yield for both surfaces . @xcite thus , while it is evident that passivation significantly affects the surface states , as expected from earlier work , @xcite this does not translate into a substantial change in the observed hysteresis . a possible explanation is that the single - dangling bonds present on the ( 311 ) surface interrupt the surface chemistry , reducing passivation treatment effectiveness . the importance of dangling - bond presentation to surface chemistry is well known ; for example , it affects the incorporation probability of si into al / ga sites versus as sites during growth . @xcite xps studies show that ammonium sulfide treatment of ( 100 ) gaas surfaces leads to disulfide bridges between adjacent surface as atoms . @xcite it would be interesting to investigate whether this changes for the ( 311 ) surface ; the corresponding effect on surface - state spectrum could be established using deep level transient spectroscopy ( dlts ) . @xcite a focused surface chemistry study may ultimately reveal a passivation formulation that produces the improvement found for device g consistently . device g shows that passivation does not eliminate the hysteresis entirely and that surface - states are not the whole story . there is additional data supporting this , for example : the addition of the al@xmath4o@xmath5 layer to a ( 100)-oriented n - type heterostructure ( fig . 3 ) causes a threshold shift consistent with a large change in surface - state spectrum but does not introduce hysteresis ; yet hysteresis is reported for ( 100)-oriented p - type heterostructures @xcite with a loop direction ( counterclockwise ) consistent with our observations . these results suggest that dopants also play a role in the hysteresis ; we now explore this possibility . the key to the remarkable stability and performance of n - type algaas / gaas heterostructures is the dx center , a deep trap consisting of a lattice - distorted si@xmath83 site . @xcite if an n - type heterostructure is warmed above @xmath80 k , the dx centers begin releasing their electrons allowing the doping layer charge distribution to change . @xcite this provides an ideal system for studying whether charge motion in the doping layer generates hysteresis similar to that observed for p - type heterostructures . gate hysteresis data from device d at @xmath106 and @xmath9 k is shown in figure 5(a / b ) . the @xmath107 k data looks like an intermediate between that in fig . 3(a ) and fig . 3(b ) ; the pinch - off voltage is slightly higher ( @xmath108 v rather than @xmath109 v ) due to the increased temperature . while the apparent hysteresis in fig . 3 runs counterclockwise , indicative of sweep - lag , the hysteresis in fig . 5(a ) runs clockwise , the direction corresponding to that observed in p - type heterostructures . the hysteresis becomes more pronounced at @xmath9 k and its shape is interesting . as fig . 5(c ) illustrates , if one takes the characteristic shape obtained for holes , e.g. , fig . 1(b ) , removes the current plateaus ( green dotted segments ) , closes the gap and mirrors about @xmath51 to account for carrier charge inversion , then a hysteresis loop with the same shape as that in fig . 5(a ) is obtained . the loss of the current plateaus is consistent with the lack of hysteresis in fig . 3(a / b ) . this suggests the current plateaus are specific to holes , and likely a surface - state effect , consistent with figs . 2 and 4(b ) . a notable feature of fig . 5(a ) is that the slow cycle ( dashed blue trace ) has a much higher pinch - off voltage than the fast cycle ( solid black trace ) ; this also occurs for p - type heterostructures at @xmath110 k , as discussed below . vs gate voltage @xmath35 for device c on 1-e without al@xmath4o@xmath5 at @xmath111 ( a ) @xmath9 k and ( b ) @xmath107 k. ( c ) a schematic illustrating an evident relationship between the hysteresis loop shape in hole ( left ) and electron devices ( right ) , as discussed in the text . ( d - g ) @xmath55 vs @xmath35 for device d on 1-e with a @xmath21 nm al@xmath4o@xmath5 layer at @xmath112 ( d ) @xmath72 k , ( e ) @xmath9 k , ( f ) @xmath107 k , and ( g ) @xmath113 k. in ( a , b , d - g ) the solid black and dashed blue traces were obtained at sweep rates of @xmath49 and @xmath59 mv / s . the arrows indicate hysteresis loop direction . , width=302 ] unfortunately , the gates in device d begin to leak directly to the 2deg for @xmath114 k , preventing higher temperature measurements . to go higher in @xmath115 and further explore this behavior , we performed the same study using device e ( fig . 5(d - g ) ) . starting at @xmath116 k , no hysteresis appears ; the data resembles that in fig . 3(b ) from device d , albeit with a slightly higher pinch - off voltage . as @xmath115 is increased very similar hysteresis to that in device d emerges for @xmath117 k. the pinch - off voltage increases markedly with @xmath115 , this limited the measurements to @xmath118 k. beyond this , the pinch - off voltage exceeds al@xmath4o@xmath5 layer breakdown causing gate leakage . device e gives similar results to device d , both in terms of the overall hysteresis loop shape and direction of travel , and the trend for higher pinch - off voltage at lower sweep - rate . versus gate voltage @xmath35 for device f on 2-h without al@xmath4o@xmath5 at @xmath119 ( a ) @xmath120 k and ( b ) @xmath121 k , ( c ) @xmath122 k , ( d ) @xmath123 k , ( e ) @xmath124 mk , ( f ) @xmath125 mk , ( g ) @xmath126 mk and ( h ) @xmath127 k. ( i ) @xmath32 versus @xmath35 for device h on the @xmath13-doped heterostructure 3-h without al@xmath4o@xmath5/passivation at @xmath37 k. the solid black and dashed blue traces were obtained at sweep rates of @xmath49 and @xmath59 mv / s . the arrows indicate hysteresis loop direction.,width=302 ] in fig . 6(a - h ) we show the corresponding behavior for holes . as @xmath115 is reduced from @xmath128 k to @xmath129 mk , two obvious changes result : first , the current plateau shortens and drops to lower @xmath32 . the plateau shortening and lowering are linked if plateau onset shifts to higher @xmath35 then the low @xmath35 depletion proceeds further before the plateau appears . second , for @xmath130 k the pinch - off voltage for the slow trace significantly exceeds that for the fast trace . a careful inspection of figs . 6(c - e ) reveals that this caused by loss of the fast trace current plateau , while the slow trace current plateau remains until @xmath131 mk . for the fast trace hysteresis loop at @xmath130 k , the hysteresis remains despite the loss of the current plateau and strongly resembles that in fig . 5 and 4(b ) . this behavior clearly indicates that two processes are involved in generating the hysteresis surface - state trapping and charge redistribution amongst the dopants each with different time and energy scales . attributing the current plateau to surface state trapping , it appears that the surface - states ` freeze out ' at @xmath110 k leaving the dopant effects behind . this might explain the lack of hysteresis in the hfo@xmath4-insulated hole qpc studied by csontos _ @xcite compared to our al@xmath4o@xmath5 device ( fig . the data shown in fig . 3(a ) of ref . @xcite was obtained at @xmath132 mk ; here the surface - state component of the hysteresis is likely quenched leaving only the hysteresis due to charge migration in the doping layer . this is why the hysteresis occurs primarily closer to pinch - off , consistent with fig . 4(b ) where surface - state hysteresis also appears to be quenched . the hysteresis loop direction in ref . @xcite is consistent with this explanation ; data at higher @xmath115 for this device would be enlightening . to further demonstrate the influence of dopants in the hysteresis we discuss one final device . in figure 6(i ) we present hysteresis data from device h , made on the @xmath13-doped heterostructure 3-h with schottky - gates on an unpassivated surface . the key difference is that the dopants only have a vertical spread @xmath133 nm in device h , compared to @xmath134 nm for devices a - g . the 2dhg is closer to the surface ( @xmath135 nm for 3-h versus @xmath107 nm for 1-h ) and the @xmath12 nm gaas cap in 3-h is si doped unlike heterostructures 1 and 2 . doping of the cap is normally performed to improve ohmic contact formation ; however , the doping level is necessarily limited to prevent shorting of gates and ohmic contacts . comparing fig . 6(i ) with fig . 5(a / b ) there are two notable differences . first and foremost , the hysteresis loop s vertical extent is reduced from @xmath136 to @xmath137 . this reduction is likely due to the @xmath138-fold reduction in dopant layer spread for device h. second , the pinch - off voltage and width of the current plateau are increased . a cap doping of @xmath139 @xmath140 corresponds to replacing only roughly 1 in every @xmath141 surface atoms with si . considered alongside the sulfur passivation results , the cap doping is unlikely to be the dominant cause for the increase in current plateau width . it is more likely that the shallower 2dhg in 3-h exacerbates the surface - state contribution to the hysteresis . this would be consistent with scattering studies in shallow undoped n - type heterostructures where the surface - charge scattering contribution increased with reduced 2deg depth . we set out to identify the exact origin of the gate hysteresis in p - type algaas / gaas heterostructures several divergent explanations exist in the literature and a better understanding will enable development of low - dimensional hole devices with improved stability / performance . a commonly accepted explanation is that gates on p - type heterostructures are inherently leaky . we show that this not the case ; schottky gates on p - type heterostructures are not significantly more leaky than schottky gates on n - type heterostructures under reverse bias conditions . hysteresis due to gate leakage in p - type heterostructures should give a clockwise hysteresis loop and the hysteresis we observe runs counterclockwise . we also find that the hysteresis becomes drastically worse rather than much better if the gates are insulated . we note that hysteresis was also still observed in qpcs with hfo@xmath4-insulated gates on c - doped p - type heterostructures , @xcite despite these producing improved tunability compared to equivalent schottky - gated devices . the direction of the hysteresis loop in ref . @xcite is counterclockwise and is inconsistent with gate leakage . the measurements by csontos _ et al _ were obtained at very low temperature @xmath142 mk and based on our data , we propose that this hysteresis may be due to charge redistribution in the dopant layer . our work focussed on the investigation of devices with insulated / uninsulated gates on custom - grown , matched electron and hole heterostructures . this relies on si being an n - type dopant on ( 100 ) substrates and a p - type dopant on ( 311)a substrates . @xcite despite the strong hysteresis in p - type devices at @xmath37 k , we observe no hysteresis in n - type devices until the temperature exceeds @xmath106 k , where the si dx centers in the n - type heterostructure begin to detrap and migration of charge within the doping layer occurs . @xcite the hysteresis that emerges at @xmath143 k for electrons bears a strong resemblance to that in holes at lower temperatures , particularly at @xmath110 k , where the current plateau at intermediate @xmath35 drops to @xmath144 and is quenched . this correspondence , and the lack of hysteresis for electrons at @xmath145 k , strongly suggests that surface states are not the sole cause of the hysteresis ; migration of charge in the dopant layer is likely involved as well . we return to surface states following a discussion of dopants . comparatively little is known about the physics of si acceptors in ( 311)a heterostructures . although si clearly acts as a substitutional acceptor , the presence and properties of acceptor complexes such as si - x is debated . @xcite our data can not provide insight at the atomic level , but there are clearly no acceptor sites in a p - type heterostructure s doping layer that act like the deep - trapping dx centers in n - type heterostructures . regarding si - x specifically , we can draw two conclusions : if si - x exists in al@xmath84ga@xmath85as in ( 311)a heterostructures , then it must be a very shallow trap ( more than 100 times shallower than dx ) if present at high density , and only a deep - trap if it is present at such low density that it can not ` freeze ' the doping layer s charge configuration . while the lack of dx - like deep traps answers the obvious question of why p - type heterostructures are so unstable / hysteretic under gating , it is interesting to invert the thinking and consider instead why n - type heterostructures are often so impressively stable . the answer is clear the deep - trapping dx centers ` lock down ' the vast majority of free charge in the dopant layer but inspirational : one way to stabilize modulation - doped p - type heterostructures may be to deliberately poison the modulation doping layer with deep trapping sites , perhaps by engineering the growth conditions to obtain a high density of deep - trapping si acceptor complexes , or failing that , by co - doping with transition metal impurities , e.g. , cu , fe , ni or zn . @xcite cu is probably more optimal than fe or ni , which may bring magnetic side - effects , and zn , which is a rapid diffuser in gaas and may be incompatible with ohmic contact formation . the changes in hysteresis obtained by changing insulator composition or surface passivation are small compared to the differences arising between the presence / absence of a gate insulator . one possible explanation is that the metal / gaas interface quenches the surface - state density , either chemically by forming ti - o - ga or ti - o - as bonds ( or ti - s - ga or ti - s - as bonds ) , or physically by providing a nearby high electron density that partially screens the surface states . @xcite this is in the same spirit as the addition of a doping layer to compensate the surface - states in ultra - high mobility n - type algaas / gaas heterostructures . @xcite this idea could be applied here with the lightly - doped cap in 3-h replaced by adding a uniformly doped layer between the cap and modulation - doping for surface - state compensation . another alternative is to abandon schottky - gates and use a degenerately - doped cap as a semiconductor gate . @xcite this effectively places the gate underneath the semiconductor surface , enabling the gate to screen the 2dhg from the surface - states . indeed , this explains the high stability and lack of hysteresis in undoped p - type semiconductor - insulator - semiconductor field - effect transistor ( sisfet ) devices . @xcite there both hysteresis contributions are dealt with the modulation doping is removed and the gate screens the surface - states . given the success of undoped sisfets , one might ask : why bother making semiconductor - gated modulation - doped devices ? in undoped sisfets it is essential that the gate overlaps the ohmic contacts ; this makes fabrication more difficult and lowers yield . @xcite this overlap is unnecessary in modulation - doped devices . hence , if the doping layer charge migration issue described earlier can be successfully overcome , modulation - doped semiconductor - gated structures may provide a formidable platform for studying low - dimensional hole devices . this work was funded by australian research council grants dp0772946 , dp0877208 , ft0990285 and ft110100072 . dr and adw acknowledge support from dft spp1285 and bmbf quahl - rep 01bq1035 . this work was performed in part using the nsw and act nodes of the australian national fabrication facility ( anff ) . we thank l. eaves for helpful discussions on dx centers , and l.h . ho for the initial measurements that motivated our interest in this problem . @xmath146 : dw and dc contributed equally to the work . experiments were performed on five separate algaas / gaas heterostructures grown by molecular beam epitaxy ( mbe ) , denoted 1-e , 1-h , 2-e , 2-h and 3-h . the first four were produced in two separate growth runs , each performed onto side - by - side halves of 2 " diameter semi - insulating gaas substrate , one ( 100)-oriented and the other ( 311)-oriented , to give ` matched ' electron ( -e ) and hole ( -h ) heterostructures , respectively . production of electron and hole heterostructures in a single growth is enabled by the facet - specific amphoteric nature of si dopants in algaas ( n - type on ( 100 ) and p - type on ( 311 ) ) @xcite . both growth runs have a nominally identical epilayer structure , the active region consisting of @xmath147 nm undoped gaas , @xmath14 nm undoped al@xmath148ga@xmath149as , @xmath11 nm si - doped al@xmath148ga@xmath149as and a @xmath12 nm undoped gaas cap . the @xmath13-doped wafer 3-h begins with @xmath150 nm undoped gaas and @xmath15 nm undoped al@xmath84ga@xmath85as grown at @xmath151c . growth is then interrupted and the substrate cooled to @xmath152c . the si source is opened for @xmath153 s and then @xmath12 nm of undoped al@xmath84ga@xmath85as is grown . growth is interrupted again to return the substrate to @xmath151c before finishing the device with @xmath154 nm of undoped al@xmath84ga@xmath85as and a @xmath12 nm gaas cap . the devices have typical carrier densities @xmath155 @xmath17 and mobilities @xmath18 @xmath19/vs at temperature @xmath37 k ; see table i for specific values for each heterostructure at @xmath20 mk using four - terminal shubnikov - de haas and hall resistivity measurements . all measurements were performed in the dark . details for each of the eight devices studied are presented in table ii . [ cols="^,^,^,^,^",options="header " , ] hall bars with a @xmath72 nm high mesa were produced using photolithography and a @xmath156 buffered hf : h@xmath4o@xmath4:h@xmath4o wet etch . the buffered hf solution is @xmath157 nh@xmath6f : hf . the photolithographically defined ohmic contacts consist of @xmath158 nm aube alloy for p - type contacts and a stack containing @xmath12 nm ni , @xmath14 nm ge , @xmath159 nm au , @xmath160 nm ni and @xmath38 nm au for n - type contacts . contacts were annealed at @xmath161c for @xmath162 s ( p - type ) and @xmath163c for @xmath164 s ( n - type ) . the photolithographically defined gates consist of @xmath21 nm ti and @xmath26 nm au deposited after the ohmic contact anneal . figure 7(a ) shows an optical micrograph of a completed device . ohmic contacts ( three at each end and four near the middle ) and a single top - gate covering the midsection of the hall bar with interconnects at top and bottom . the remaining four metal gates were not used . ( b ) nomarski phase - contrast optical micrograph showing four etched penetrations ( off - white ) in the @xmath21 nm al@xmath4o@xmath5 layer ( pink ) obtained using a buffered hf etch . these correspond to the four central ohmic contacts in ( a ) . ( c ) scanning electron micrograph of the etch penetration ( dark ) in the al@xmath4o@xmath5 layer ( light ) for the ohmic contact at the far right of the hall bar in ( a ) . ( b ) and ( c ) were obtained after the buffered hf etch and before ohmic metallization.,width=226 ] al@xmath4o@xmath5 insulated gate samples were produced by adding the following steps between mesa etch and ohmic contact metallization . a @xmath21 nm al@xmath4o@xmath5 layer was deposited using ( ch@xmath5)@xmath5al and h@xmath4o gaseous precursors at @xmath22c in a cambridge nanotech savannah @xmath26 atomic layer deposition ( ald ) system . access to the heterostructure for the ohmic contacts was obtained by a @xmath165 s buffered hf etch ( @xmath157 nh@xmath6f : hf in h@xmath4o ) following photolithographic definition of the contact regions and prior to metallization . in the absence of h@xmath4o@xmath4 this etch self - terminates , stripping the native gaas surface oxide in the process . the openings in the al@xmath4o@xmath5 layer are visible optically ( fig . 7(b ) ) , but are clearer using scanning electron microscopy ( fig . although traces of residual oxide remain , they do not adversely affect ohmic contact formation . the polyimide insulated gate sample was produced with an added step between the ohmic contact anneal and gate deposition . a @xmath72 nm layer of patterned polyimide is obtained by spin - coating a diluted mixture ( 1:1.4 ) of photo - processable polyimide ( hd microsystems hd-4104 ) in thinner ( hd microsystems t-9039 ) , performing a @xmath166c soft - bake for @xmath162 s , exposing / developing aligned ohmic contact openings , and finishing with a @xmath167c hard - bake for @xmath165 min under @xmath168 atm n@xmath4 . the sulfur passivation treatment was performed immediately before gate deposition . a stock solution of ( nh@xmath6)@xmath4s@xmath7 was prepared by adding @xmath25 g of elemental sulfur ( aldrich ) to @xmath26 ml of @xmath27 ( nh@xmath6)@xmath4s solution ( aldrich ) and mixing until completely dissolved . the passivation solution was a @xmath24 dilution of the stock solution in deionized water ( millipore ) . passivation was performed by immersion for @xmath23 min in @xmath169 ml of passivation solution heated to @xmath28c in a water bath , followed by a deionized water rinse . the sample is stored under deionized water during transfer to the vacuum evaporator . the total time between passivation and sample at vacuum was @xmath170 min , with the sample exposed to air for no longer than a few minutes during evaporator loading and pump - down . electrical measurements at @xmath29 k were obtained using a liquid helium dip - station , with @xmath171 k achieved using the natural stratification of the he atmosphere inside the dewar . data at @xmath172 k was obtained using an oxford instruments heliox @xmath31he cryostat . standard two- and four - terminal ac lock - in techniques were used to measure the conducting channel s drain current @xmath32 , typically with a @xmath33v constant voltage excitation at @xmath34 hz applied to the source . pinch - off ( i.e. , @xmath76 ) is interpreted as @xmath173 pa ; this triggers the software to commence the downsweep to avoid pointlessly driving the device beyond pinch - off . the gate bias @xmath35 was applied using a keithley 2400 source - measure unit enabling continuous measurement of gate leakage current @xmath36 down to @xmath26 pa . this instrument has a built - in current limiter , with @xmath35 curtailed to keep @xmath36 at a specified limit even if a higher @xmath35 if requested . on a linear axis vs gate bias @xmath35 for a schottky - gated modulation - doped p - type heterostructure . the data matches that shown in fig . 1(a ) . at positive @xmath35 , @xmath36 remains less than @xmath38 pa to @xmath174 v , sufficient to achieve pinch - off for all uninsulated gate devices studied . in an n - type heterostructure , gate leakage would normally occur at @xmath175 v ( indicated by the arrow ) and is suppressed for negative @xmath35.,width=302 ] figure 8 shows the data in fig . 1(a ) plotted on a linear - linear scale for comparison . the gate leakage current @xmath36 is @xmath176 pa for the entire range @xmath177 v. and ( b ) the numerical transconductance @xmath178 for the upsweeps vs gate voltage @xmath35 . the traces in ( b ) are sequentially offset vertically by @xmath179 with increasing sweep rate for clarity.,width=302 ] figure 9 shows the vertical extent of the hysteresis loop and the slope of the upsweeps versus @xmath35 . the slope curves for the down sweeps are very similar to those in fig . 9 except there is less structure at low @xmath35 and the peak at higher @xmath35 is sharper and shifted to more positive @xmath35 . if the hysteresis is caused by sweep lag , one would expect the maximum vertical extent in the hysteresis to coincide in @xmath35 with the maximum slope @xmath178 . this is clearly not the case , supporting our assessment that the hysteresis is not caused by sweep lag . vs gate voltage @xmath35 for 22 hysteresis loop sweeps of device b2 . the first sweep is show in blue , the @xmath23nd to @xmath180nd sweeps evolve continuously in color from green to magenta . the sweep rate in each case is @xmath12 mv / s . the points of discontinuity at high @xmath35 are where gate voltage current limiting comes on / off for the up / downsweep.,width=302 ] figure 10 shows hysteresis loop data for device b2 , which is nominally identical to device b aside from a change in the gate design used . the gate in device b2 covers less area . in both cases the gate does not extend to the edges of the al@xmath4o@xmath5 layer ; hence in neither case is leakage by proximal direct shorting between gate and ohmic contact . the characteristics are very similar to those in device b , and the breakdown voltage is of a similar magnitude @xmath181 v. repeated sweeps appear to increase the breakdown voltage , enabling the device to slowly progress towards pinch - off . the exact cause for this is unclear . and ( b ) gate leakage current @xmath36 vs gate voltage @xmath35 at two different gate sweep rates for a device c featuring a @xmath72 nm polyimide layer on 2-h . the horizontal arrows indicate the direction of travel around the hysteresis loop . in ( b ) the left / right axis and lower / upper set of data are for the sweep to / from positive @xmath35 , respectively . note that once @xmath36 reaches @xmath38 na the gate voltage source implements current limiting by holding @xmath35 fixed . hence for @xmath182 v the data in ( a ) should be considered as @xmath55 versus time @xmath56 at fixed @xmath35 . to aid in converting the data at @xmath182 v into time , each @xmath57 v minor tick in the figure corresponds to @xmath58 and @xmath21 s for sweep rates of @xmath12 and @xmath49 mv / s , respectively.,width=302 ] the radical effect that the composition and properties of the insulator - semiconductor interface can have on transistor performance is well - known for organic fets . @xcite due to high interface - state densities , inorganic oxide insulators are particularly troublesome , causing significant hysteresis and shifts in threshold voltage . polymeric insulators such as polyimide often bring significant improvement . to explore whether the current plateau and difficulty in attaining pinch - off in device b is tied to insulator composition we studied device c containing gates insulated with a @xmath72 nm thick polyimide layer on 2-h . figures 11(a / b ) show @xmath55 and @xmath36 versus @xmath35 for this device . pinch - off can not be achieved in device c , largely due to the much lower breakdown field for the polyimide insulator . breakdown occurs for @xmath73 v , and once @xmath36 reaches @xmath38 na at @xmath183 v the voltage source holds the gate bias fixed , as in fig . 2 . thus the data at @xmath184 v in fig . 11(a ) should instead be considered as @xmath55 versus time @xmath56 at constant @xmath35 . here each minor sub - tick corresponds to a time of @xmath49 s for the @xmath49 mv / s trace and @xmath21 s for the @xmath12 mv / s trace . we show the data obtained beyond pinch - off in fig . 11 to better facilitate comparison with fig . 2 , and to highlight the equilibration behavior that occurs when a sweep is stopped . the @xmath55 versus @xmath35 characteristics for device c bear a striking resemblance to those of device b ( figs . 2 and 10 ) . on the upsweep there is an initial drop in current that plateaus for intermediate @xmath35 , followed by a recovery in @xmath55 when @xmath35 is held constant . this is consistent with slow accumulation of net negative charge between the gate and 2dhg , as per the explanation for hysteresis loop shape / direction in section iii - d . the downsweep gives an initial rapid rise in @xmath55 that plateaus as @xmath35 approaches zero again , and as in fig . 2 , there is a distinct asymmetry between the upsweeps and downsweeps regarding the dependence of the path taken on sweep rate . pinch - off occurs beyond @xmath183 v , thus the shift in pinch - off bias must exceed @xmath185 . the capacitive conversion factor for this device @xmath186 assuming @xmath72 nm of polyimide with dielectric constant @xmath187 , giving an anticipated pinch - off voltage @xmath188 v. b.c . scannell , i. pilgrim , a.m. see , r.d . montgomery , p.k . morse , m.s . fairbanks , c.a . marlow , h. linke , i. farrer , d.a . ritchie , a.r . hamilton , a.p . micolich , l. eaves and r.p . taylor , phys . b * 85 * , 195319 ( 2012 ) . a.m. see , i. pilgrim , b.c . scannell , r.d . montgomery , o. klochan , a.m. burke , m. aagesen , p.e . lindelof , i. farrer , d.a . ritchie , r.p . taylor , a.r . hamilton and a.p . micolich , phys . lett . * 108 * , 196807 ( 2012 ) .

diffusion is a fundamental transport mechanism in nature and industry , with applications ranging from physics to biology , chemistry , engineering , and economics . this process has attracted much attention during the last decades , particularly in statistical and condensed matter physics : diffusion - reaction processes ; transport in porous media and biological tissues ; trapping in heterogeneous systems ; kinetic and aggregation phenomena like dla , to name a few fields . from an intuitive point of view , brownian motion is often considered as a continuous limit of lattice random walks . however , a more rigorous background is needed to answer subtle questions . in mathematical textbooks , brownian motion is defined as an almost surely continuous process with independent normally distributed increments @xcite . the deceptive simplicity of this definition relies on the notion of `` almost surely '' that , in turn , requires a sophisticated formalism of wiener measures in the space of continuous functions , filtrations , sigma - algebra , etc . although this branch of mathematics is well developed , it is rather difficult for non - mathematicians , that is , the majority of scientists studying brownian motion in their every - day research . in this paper , we discuss a different , but still rigorous , approach to define and operate with brownian motion as suggested by p. lvy @xcite . we construct from scratch a simple and intuitively appealing representation of this process that gives a closed formula mapping of the unit interval onto the functional space of brownian paths . in this framework , sampling a brownian path is nothing else than picking up uniformly a point from the unit interval . figuratively speaking , brownian motion is constructed here by adding randomly wavelet - based geometrical features at multiple length scales . the explicit formula elucidates many classical results about brownian motion ( e.g. , non - differentiability of its path ) . the illustrative character of the wavelet representation , along with the simple structure of the underlying probability space , is different from the usual presentation of most classical textbooks . among various amazing properties , brownian motion is known to have a self - similar structure : when a fragment of its path is magnified , it `` looks '' like the whole path . in other words , any fragment obeys the same probability law as the whole path . as a consequence , brownian paths exhibit their features at ( infinitely ) broad range of length scales . as a matter of fact , multiscale geometrical structures are ubiquitous in nature and material sciences @xcite . for instance , respiratory and cardiovascular systems start from large conduits ( trachea and artery ) that are then split into thinner and thinner channels , up to the size of few hundred microns for the alveoli and several microns for the smallest capillaries @xcite . another example is a high - performance concrete which is made with grains of different sizes ( from centimeters to microns ) , smaller grains filling empty spaces between larger ones . the best adaptive description of such self - similar structures relies on intrinsicly multiscale functions to capture their mechanical or transport properties at different length scales . we illustrate this idea by constructing brownian motion using wavelets , a family of functions with compact support and well defined scaling @xcite . wavelets appear as the natural mathematical language to describe and analyze multiscale structures , from heterogeneous rocks to biological tissues @xcite . the wavelet construction of brownian motion naturally extends to fractional brownian motion and other gaussian processes and fields , allowing one to efficiently simulate , for instance , turbulent diffusion with high reynolds numbers or financial markets . in particular , we discuss the gaussian ( or massless ) free field and fractional gaussian fields which appear as basic models in different areas of physics , from astrophysics ( cosmic microwave background ) to critical phenomena , quantum physics , and turbulence @xcite . written in a spectral form in one dimension , brownian motion and the fractional gaussian field look very similar , one of them being the fractional derivative of the other . putting together these two processes reveals deep relations between them , and this correspondence carries over to higher dimensions . most importantly , the wavelet representation is a starting point for a number of highly efficient numerical methods to simulate various gaussian processes and fields . though wavelet representations and the related numerical methods are all known , they are not easily available in a single source . indeed the totality of these methods seems to be poorly understood , even amongst specialists . the purpose of this article is to present a unified and intuitive framework that is based on elementary mathematical structures like , for example , dyadic subdivision or wavelet tree . we also present some simple number - theoretic shortcuts and consequent numerical algorithms . finally , we discuss fast simulations of brownian motion in confining domains , where one of the difficulties is the ability to quickly access the local geometry near the boundary . these techniques can be applied for studying brownian motion and related processes or solving partial differential equations in complex multiscale media . we hope that this didactic article will provoke interesting discussions amongst the experts and will help for a better understanding of both theory and implementation of brownian motion and other gaussian processes and fields for a much broader community of their practical `` users '' , namely , physicists , biologists , chemists , engineers , and economists . in this section , we derive a wavelet representation of brownian motion in a simple explicit way , allowing one to gain an intuitive feeling of this constructive approach . in order to illustrate the basic idea of a wavelet representation , we revisit the first single particle tracking experiment by r. brown who looked through a microscope at stochastic trajectories ( now known as brownian paths ) of pollens of clarkia ( primrose family ) @xcite . first examples of such trajectories for mastic grains in water were reported by j. perrin @xcite . the essence of the wavelet representation can be recognized in his description of three trajectories ( fig . [ fig : perrin ] ) that were recorded at 30-second intervals @xcite : `` _ ils ne donnent quune ide trs affaiblie du prodigieux enchevtrement de la trajectoire relle . si , en effet , on faisait des points de seconde en seconde , chacun de ces segments rectilignes se trouverait remplac par un contour polygonal de 30 cts relativement aussi compliqu que le dessin ici reproduit , et ainsi de suite . _ '' following this idea , let us record the positions of a particle ( e.g. , pollen or grain ) at successive time moments with a selected time resolution @xmath0 . each particle submerged in water is permanently `` bombarded '' by water molecules . since the number of the surrounding water molecules is very large and their tiny actions are mostly uncorrelated , net microscopic displacements of the particle can not be considered deterministicly as in classical mechanics , but random . since the interaction of water molecules between them is very rapid as compared to the macroscopic resolution scale , there is no memory effect in their action on the heavy particle . as a consequence , the microscopic displacements of the particle are ( almost ) independent and have a finite variance @xmath1 . the macroscopic displacement during the resolution time @xmath0 is the sum of a large number @xmath2 of these displacements with zero mean ( no coherent flow ) . although the average displacement is also zero , the stochastic fluctuations around this value are of the order of @xmath3 . moreover , the central limit theorem gives us a precise probabilistic description of the fluctuations , resulting in a normal ( or gaussian ) distribution of macroscopic displacements of the particle @xcite . it is worth stressing that the gaussian character of the macroscopic displacements appears _ without any specific knowledge about the microscopic interactions_. the only important information at microscopic level was stationary , uncorrelated character of the interaction , and finite variance ( if one of these conditions is missing , the resulting macroscopic process may exhibit anomalous behavior , see @xcite and references therein ) . this is known as coarsening or upscaling : complex interactions and the specific features of the underlying microscopic dynamics are averaged out on macroscopic scales . this is the reason why brownian motion ( or diffusion in general ) is so ubiquitous in nature and science . once we know that the microscopic details are irrelevant ( under the conditions mentioned above ) , we can extend the gaussian behavior from macroscopic scales , where it has been established , to microscopic scales . this procedure can be called downscaling , when we explicitly and purposefully transpose the universal macroscopic behavior even onto smaller scales . the resulting _ model _ of the microscopic dynamics exhibits gaussian features at all scales . while the true dynamics and its gaussian model can be completely different at microscopic scales , they become identical at macroscopic scales . knowing that a particle moves continuously , we connect its successive positions separated on time @xmath0 by a continuous line . since the experimental setup is limited to the selected time scale @xmath0 , nothing can be said about the trajectory of the particle in between two records . in other words , the only condition for the trajectory to pass through the recorded points leaves us a variety of choices for the shape of the connecting continuous line . the common choice is connecting the successive positions by linear segments . as one will see , this choice fixes a particular wavelet representation , the haar wavelets . we shall show that other wavelets , corresponding to other choices of continuous connections , are as well useful . when the magnification and time resolution of the experimental setup are increased , smaller details of the particle s trajectory appear , allowing one to refine the above piecewise linear approximation . repeating this procedure , in theory up to infinity , one recovers all geometrical features and thus constructs the whole brownian path . in what follows , we put this schematic description into a more rigorous mathematical frame . we start with the position `` records '' at every unit time : @xmath4 ( e.g. , every second ) along one coordinate ( two- and higher dimensional brownian motion is then obtained by taking @xmath5 independent copies of the one - dimensional process ) . we focus on the time interval between @xmath6 and @xmath7 , the construction being applicable to any interval @xmath8 $ ] . for convenience , brownian motion is started at @xmath9 from the origin : @xmath10 . by definition ( or as a consequence of the central limit theorem if one relies on the above physical reasoning ) , the position at time @xmath11 , @xmath12 , is a random variable @xmath13 distributed according to the standard normal ( or gaussian ) law , @xmath14 , with mean zero and variance one @xmath15 in physics , the variance @xmath1 is related to the diffusion coefficient , @xmath16 , where @xmath17 is the one step duration ; here , @xmath18 yielding @xmath19 throughout the paper . a linear approximation at this time scale ( @xmath20 ) is simply @xmath21 that connects the positions @xmath10 and @xmath22 by a linear segment . if the time resolution is doubled , a new , intermediate position @xmath23 can now be seen ( fig . [ fig : tent ] ) . the random variable @xmath24 is conditioned by the fact that brownian motion passes through the points @xmath25 and @xmath26 , the value of @xmath13 being already known . it is distributed according to the normal law with mean value @xmath27 and variance @xmath28 ( see appendix [ sec : cond_law ] ) . in other words , one can write @xmath29 , where the new random variable @xmath30 ( i.e. , distributed according to the standard normal law ( [ eq : normal ] ) ) is independent of @xmath31 and @xmath22 . the linear approximation at the time scale @xmath32 connects three successive points @xmath25 , @xmath33 and @xmath26 by two linear segments : @xmath34 the shape of this approximation looks like a skewed tent ( fig . [ fig : tent ] ) that can be represented as a sum of a linear shift and a `` symmetric tent '' function : @xmath35 where the `` symmetric tent '' function @xmath36 is @xmath37 the decomposition ( [ eq : b1_decomp ] ) into the linear function @xmath38 and the tent function @xmath36 is unique . the new approximation @xmath39 is obtained from the previous one , @xmath40 , by adding the new term representing a smaller geometrical detail . by a segment at scale @xmath41 ; * ( b ) * linear approximation @xmath39 by two segments at scale @xmath42 . the latter `` skewed tent '' can be uniquely represented as the sum of a `` symmetric tent '' and a linear shift . , width=321 ] the same concept is applicable at every scale . assume that an approximation @xmath43 of brownian motion is already constructed at the time scale @xmath44 , i.e. , the positions @xmath45 are known at successive times @xmath46 , @xmath47 ranging from @xmath6 to @xmath48 . the approximate brownian path is a piecewise linear function passing successively through these points . at the next time scale @xmath49 , a new , intermediate position @xmath50 of brownian motion at time @xmath51 should be determined for each @xmath47 . as previously , the random variable @xmath24 is conditioned by the fact that brownian motion is _ known _ to pass through the points @xmath52 and @xmath53 . it is again distributed according to the normal law , with mean value @xmath54 and variance @xmath55 . in other words , one can write @xmath24 as @xmath56 where the new normal random variable @xmath57 is independent of the other positions . the linear segment between @xmath52 and @xmath53 is then replaced by two linear segments connecting the three successive points @xmath52 , @xmath58 , and @xmath53 . this is a new `` skewed tent '' function which can be uniquely decomposed as the sum of the previous linear segment ( drift ) and a symmetric tent function @xmath59 with the weight @xmath60 , where @xmath61 is a rescaled symmetric tent function on the interval @xmath62 ( see fig . [ fig : haar]a ) . we stress again that the new approximation is obtained from the previous one by simply adding the tent function @xmath59 , representing a smaller geometrical detail at the new scale @xmath49 , weighted by a normally distributed coefficient @xmath60 which is independent of the previously determined positions . * ( a ) * and the related haar function @xmath63 * ( b ) * , both having the support on the interval @xmath64 . , width=321 ] this construction is applicable to all linear segments ( all @xmath47 ) at the given scale @xmath65 , and it is valid for any scale . repeating this procedure from the scale @xmath66 up to infinity , one obtains the haar wavelet representation of brownian motion on the unit interval : @xmath67 where all weights @xmath68 and @xmath69 are independent @xmath14 random variables . here we introduced the superscript @xmath70 in order to stress that a sampling of gaussian weights @xmath60 yields a random realization of brownian motion . we will discuss in sect . [ sec : weights ] that all these gaussian weights can be constructed from a single random number @xmath70 from the unit interval that provides a natural parameterization of brownian paths . the dyadic structure of the intervals implies that for any @xmath71 , there exists only one interval @xmath72 of length @xmath73 containing the point @xmath38 : @xmath74 . the index @xmath47 is simply the integer part of @xmath75 : @xmath76 ( i.e. , the largest integer that does not exceed @xmath75 ) . as a consequence , the convergence in the above formula is very rapid . in fact , if one needs to obtain the value of function @xmath77 with a desired precision @xmath78 , it is sufficient to calculate the first @xmath79 terms , @xmath80 being the logarithm of @xmath81 on the base of @xmath82 . subtracting the linear term @xmath83 from eq . ( [ eq : bomega1 ] ) yields the haar wavelet representation of a brownian bridge on the unit interval , i.e. , brownian motion conditioned to return to @xmath6 at time @xmath11 . in the wavelet representation ( [ eq : bomega1 ] ) , the first sum is carried over all scales @xmath65 , while the second sum covers all @xmath48 subintervals @xmath72 at the scale @xmath65 . in some cases , it is convenient to enumerate all the dyadic subintervals @xmath72 by a single index @xmath84 as shown on fig . [ fig : intervals ] . since @xmath47 is ranging from @xmath6 to @xmath85 , the new index @xmath86 uniquely identifies the interval @xmath72 . in particular , one easily retrieves the pair @xmath87 from @xmath86 as @xmath88 using the notations @xmath89 we can write eq . ( [ eq : bomega1 ] ) in a more compact form @xmath90 as a result , brownian trajectory is decomposed into a sum of tent functions ( plus a linear drift ) with random independent identically distributed gaussian weights . as illustrated below , all these weights can be determined from a single uniformly distributed random variable . . , width=283 ] in this subsection , we illustrate how all random gaussian weights @xmath91 can be explicitly related to a single uniformly distributed random number @xmath70 . in other words , we show that the complicated abstract probability space of brownian paths can have a simple parameterization . however , this construction is not relevant from practical point of view , and the subsection can be skipped at first reading . we consider the binary expansion of a given real number @xmath70 from the unit interval @xmath92 where @xmath93 are equal to @xmath6 or @xmath7 ( note that @xmath94 is expanded as @xmath95 instead of its equivalent form @xmath96 ) . a uniform picking up of @xmath70 in @xmath97 $ ] is equivalent to _ independent _ random choice of its binary digits ( or bits ) @xmath93 . then we choose a prime number @xmath98 and construct another number @xmath99 using the bits of @xmath70 at positions @xmath98 , @xmath100 , @xmath101 , @xmath102 @xmath103 for example , @xmath104 , @xmath105 , etc . [ fig : omega ] ) . if @xmath98 and @xmath106 are two different prime numbers then @xmath99 and @xmath107 are independent as being constructed from separate sets of independent bits @xmath93 . moreover , if @xmath70 is chosen uniformly from the unit interval , then each @xmath99 is also uniformly distributed on the unit interval . consequently , a single random number @xmath70 gives rise to an infinite sequence @xmath108 of independent uniformly distributed random variables . in a more formal way , the real numbers @xmath99 can be written as @xmath109 where @xmath110 is the rademacher function . , @xmath111 , @xmath112 , @xmath102 using binary expansion of a single number @xmath70 from the unit interval . for instance , @xmath113 is constituted of the @xmath114 , @xmath115 , @xmath116 , @xmath102 bits of @xmath70 . , width=283 ] at last , we need to pass from uniformly distributed to normally distributed variables . for this purpose , we define the inverse @xmath117 of the error function : for @xmath118 $ ] , the value @xmath119 of the function @xmath117 satisfies @xmath120 although there is no simple analytic form for the function @xmath117 , many properties can be easily derived , and the whole function can be tabulated with any required precision . if @xmath121 denotes the @xmath122 prime ( e.g. , @xmath123 , @xmath124 , @xmath125 ) , then we set @xmath126 by construction , @xmath127 are independent normally distributed random variables . in other words , eqs . ( [ eq : omegap ] , [ eq : erf_inv ] ) map a uniformly distributed @xmath70 onto a sequence of gaussian weights @xmath128 . as a result , @xmath77 is constructed as a mapping from the unit interval , @xmath129 $ ] , onto the space of real - valued functions ( more precisely , the hlder space @xmath130 with any @xmath131 ) . in other words , any brownian path is explicitly encoded by the real number @xmath70 . picking up the real number @xmath70 from the unit interval ( with uniform measure ) is thus equivalent to choosing a brownian path ( with wiener measure ) . in this representation , the probability space for brownian motion is nothing else than the unit interval with uniform measure . it is intuitively much simpler than the classical construction of the probability space by means of wiener measure , filtrations , etc . at the same time , this mapping is evidently neither continuous , nor injective ( e.g. , two numbers @xmath70 and @xmath132 that differ at 6-th bit correspond to the same sequence of gaussian random numbers ) . the mapping remains a rather formal construction whose only purpose was to show the equivalence between two spaces . in eqs . ( [ eq : bomega1 ] ) or ( [ eq : bomega2 ] ) , brownian motion is decomposed into a sum of a linear function and tent functions . figure [ fig : haar]a illustrates that any tent function @xmath59 can actually be represented as the integral of a piecewise - constant function @xmath133 where @xmath63 is called the haar function and defined to be @xmath6 on the complement of @xmath72 and to take values @xmath134 and @xmath135 on its left and right subintervals , respectively ( fig . [ fig : haar]b ) . in fact , all haar functions are obtained by translations and dilations of a single `` mother '' function @xmath136 : @xmath137 as illustrated on fig . [ fig : haars ] , one has @xmath138 it is convenient to use previously introduced single index @xmath86 to denote different haar functions : @xmath139 eq . ( [ eq : bomega2 ] ) yields the following representation for brownian motion @xmath140 where @xmath141 denotes the gaussian white noise which is defined here through the haar wavelet decomposition @xmath142 are obtained by dilations and translations of the mother function @xmath143 ( shown on the left ) . , width=302 ] it is easy to check that haar functions @xmath144 ( together with a constant function ) form an orthonormal basis in the space @xmath145)$ ] of measurable and square integrable functions that is @xmath146 moreover , this basis is known to be complete in @xmath145)$ ] . this means that any function from @xmath145)$ ] can be decomposed into a linear combination of the haar functions ( and a constant ) . from now on , we drop the superscript @xmath70 for getting simpler notations although the parameterization by @xmath70 remains valid for all discussed processes . the wavelet representation ( [ eq : bnoise2 ] ) can be extended to _ any _ complete orthonormal basis @xmath147 of @xmath145)$ ] . in fact , if the orthonormal basis @xmath147 is complete , one can decompose any function @xmath148 into a linear combination of @xmath149 : @xmath150 where the coefficients @xmath151 satisfy the orthogonality relation @xmath152 substitution of this decomposition into eq . ( [ eq : bnoise2 ] ) gives @xmath153 with new random weights @xmath154 the sum of independent gaussian variables is a gaussian variable , and its variance is simply the sum of the squared coefficients @xmath155 , which is equal to @xmath7 according to eq . ( [ eq : cmi ] ) . in other words , @xmath156 . moreover , the new random variables @xmath157 are independent due to the orthogonality of the functions @xmath149 . we have thus shown that the gaussian white noise can be decomposed into a linear combination with independent gaussian weights in _ arbitrary _ complete orthonormal basis of @xmath145)$ ] . the completeness of the basis @xmath147 yields the usual covariance of the gaussian white noise @xmath158 where @xmath0 is the dirac distribution ( or `` @xmath0-function '' ) . substituting eq . ( [ eq : dw_psi ] ) into eq . ( [ eq : bnoise ] ) , one gets @xmath159 for instance , one can consider the fourier basis on the unit interval , @xmath160 in order to get the karhunen - love expansion of brownian motion @xcite : @xmath161 we have explicitly constructed the haar wavelet decomposition ( [ eq : bomega1 ] ) and then general representation ( [ eq : bomega4 ] ) in order to reproduce the basic properties of brownian motion . alternatively , one could first postulate such a representation as a definition of brownian motion and then check that the basic properties are fulfilled . to illustrate this point , we check several properties . \(i ) brownian motion is a gaussian process with independent increments . first , @xmath162 is gaussian as being a linear combination ( [ eq : bomega4 ] ) of gaussian variables . let @xmath163 and @xmath164 define two increments , @xmath165 and @xmath166 . if @xmath167 ( i.e. , the increments do not `` overlay '' ) , then they are independent ( similar statement holds if @xmath168 by symmetry ) . to proof this statement , we decompose the unit interval as @xmath169 = [ 0,t_1)\cup [ t_1,t_2 ] \cup ( t_2,t_3 ) \cup [ t_3,t_4 ] \cup ( t_4,1],\ ] ] ( if one of subintervals @xmath170 , @xmath171 or @xmath172 $ ] is empty , it can be ignored ) . the basis @xmath147 in eq . ( [ eq : bomega4 ] ) can be chosen as the direct product of the haar eigenbases on each subinterval . for instance , @xmath173}\}$ ] is the haar basis of @xmath174)$ ] which is extended to @xmath97\backslash [ t_1,t_2]$ ] by zeros . in this particular representation , one has @xmath175 } ~ \psi_i^{[t_1,t_2]}(t ' ) = \hat{a}_0^{[t_1,t_2]}~ \sqrt{t_2 - t_1 } .\\ \end{split}\ ] ] in the second equality , the haar functions from other subintervals ( except @xmath176 $ ] ) vanished by construction . in turn , all the haar functions @xmath177}(t')$ ] on @xmath176 $ ] ( with @xmath178 ) vanished after integration due to their orthogonality to the constant . the only remaining contribution is the constant term which has the unit @xmath174)$ ] norm : @xmath179}(t ) = ( t_2 - t_1)^{-1/2}$ ] . integrating this term , one gets the right - hand side of eq . ( [ eq : b_increment_aux1 ] ) which shows that the increment @xmath180 is a gaussian variable with mean zero and variance @xmath181 , as expected . moreover , the same representation for @xmath182 $ ] yields @xmath183 } ~ \psi_i^{[t_3,t_4]}(t ' ) = \hat{a}_0^{[t_3,t_4]}~ \sqrt{t_3 - t_4 } , \\ \end{split}\ ] ] and the random weights @xmath184}$ ] and @xmath185}$ ] are independent by construction . as a consequence , the increments @xmath165 and @xmath166 are independent . \(ii ) the mean and covariance of brownian motion are : @xmath186 the first statement is obvious from eq . ( [ eq : bomega4 ] ) given that all weights have mean zero . to prove the second statement , one assumes that @xmath187 and considers @xmath188 ^ 2\ } . \\ \end{split}\ ] ] the first term vanishes due to the independence of increments ( and @xmath31 ) , while the second term is equal to @xmath189 according to eq . ( [ eq : b_increment_aux1 ] ) . \(iii ) brownian motion is continuous but nowhere differentiable almost surely . the proof relies on a simple fact that the gaussian weights @xmath60 can not be too large , e.g. , the probability that @xmath190 decays extremely fast ( as @xmath191 for large enough @xmath65 ) . in turn , the norm of tent functions decreases exponentially that ensures the continuity of brownian motion and the convergence of a partial sum approximation in eq . ( [ eq : bomega1 ] ) or similar expressions to brownian motion . moreover , the remainder of this approximation decreases exponentially fast with the truncated scale @xmath2 ( for technical details , see appendix [ sec : continuity ] ) . as we mentioned in sect . [ sec : physical ] , taking the particular orthonormal basis is equivalent to choosing a way to connect successive positions of brownian motion at a finite scale @xmath0 . the haar wavelets and the resulting tent functions present the simplest way of connection by linear segments . such a piecewise linear approximation of brownian motion introduces singularities at the connection nodes ( corners ) . for some problems , it is convenient to deal with a smooth approximation of brownian motion at finite scales ( although a true brownian trajectory , the limiting curve , remains nowhere differentiable ) . for this purpose , one can use the alpert - rokhlin multiwavelet basis @xcite . this basis is generated by a set of @xmath106 functions @xmath192 which are supported on the interval @xmath97 $ ] , are piecewise polynomials of degree @xmath193 on @xmath194 $ ] and on @xmath195 $ ] , and satisfy the moment cancellation conditions @xmath196 these mother functions generate the alpert - rokhkin multiwavelets of order @xmath106 by translations and dilations : @xmath197 the set of functions @xmath198 , completed by the set of orthonormal polynomials of order @xmath199 , forms a complete basis of @xmath145)$ ] . this completion is necessary because all mother functions @xmath200 ( and thus all @xmath201 ) are orthogonal by construction to all polynomials of order @xmath199 . similarly , the haar wavelets were completed by a constant function . when @xmath202 , there is only one mother function @xmath136 defined by eq . ( [ eq : phi11 ] ) which generates the haar wavelets by translations and dilations . for @xmath203 , there are two mother functions ( fig . [ fig : alpert ] ) , satisfying the moment cancellation conditions ( [ eq : moments ] ) : @xmath204 higher - order mother functions ( with @xmath205 ) can be constructed through an orthogonalization procedure ( see @xcite for details and examples ) . note that the wavelet representation of brownian motion involves the integral of wavelets @xmath206 for instance , one gets for @xmath203 ( fig . [ fig : alpert ] ) @xmath207 while the other functions @xmath208 and @xmath209 are obtained by dilations and translations . as a consequence , brownian motion gets a closed formula in terms of the alpert - rohklin multiwavelets of order @xmath82 : @xmath210 where all weights @xmath13 , @xmath211 , @xmath212 are independent @xmath14 variables , and the second term is the integral of the linear basis function @xmath213 . an extension of this representation to the alpert - rohklin multiwavelet basis of order @xmath106 is straightforward . ( on the right ) . the use of these wavelets corresponds to another type of connection ( not by linear segments ) between successive points of brownian motion in the refinement procedure . for comparison , the tent function @xmath36 is shown on the last plot by dashed line . note that horizontal and vertical scales are not matched here . , title="fig:",width=158 ] ( on the right ) . the use of these wavelets corresponds to another type of connection ( not by linear segments ) between successive points of brownian motion in the refinement procedure . for comparison , the tent function @xmath36 is shown on the last plot by dashed line . note that horizontal and vertical scales are not matched here . , title="fig:",width=158 ] ( on the right ) . the use of these wavelets corresponds to another type of connection ( not by linear segments ) between successive points of brownian motion in the refinement procedure . for comparison , the tent function @xmath36 is shown on the last plot by dashed line . note that horizontal and vertical scales are not matched here . , title="fig:",width=158 ] ( on the right ) . the use of these wavelets corresponds to another type of connection ( not by linear segments ) between successive points of brownian motion in the refinement procedure . for comparison , the tent function @xmath36 is shown on the last plot by dashed line . note that horizontal and vertical scales are not matched here . , title="fig:",width=158 ] the wavelet representation of brownian motion can be easily implemented in practice . to carry computations with a ( fixed ) desired precision @xmath78 , it is sufficient to truncate the first sum in eq . ( [ eq : bomega1 ] ) or equivalent relation up to @xmath214 , because higher - order terms describe geometrical details at smaller scales . the remainder of this series can be estimated using eq . ( [ eq : approx_remainder ] ) . for haar wavelets , such a truncation qualitatively corresponds to an approximation of brownian motion by a broken line composed of linear segments of length close to @xmath78 , while the alpert - rokhlin wavelets yield smoother approximations ( fig . [ fig : bm_iterates ] ) . this is a fascinating feature of wavelets , allowing one to capture geometrical details at different scales . a realization of a brownian path is completely determined by a set of random coefficients @xmath69 ( or @xmath127 ) . in sect . [ sec : weights ] , we discussed an explicit scheme to generate all these coefficients from a randomly chosen number @xmath70 from the unit interval . in practice , one can use standard routines to generate pseudo - random normally distributed weights @xmath69 ( or @xmath127 ) . the computation of tent functions @xmath59 can be easily implemented . consequently , the computation of @xmath162 at any time @xmath38 requires only @xmath79 operations , each of them consisting of finding @xmath59 , multiplying it by @xmath69 , and summing their contributions . ( dashed line ) , and @xmath215 ( solid line ) with haar wavelets * ( a ) * and alpert - rohkin wavelets with @xmath203 * ( b)*. , title="fig:",width=321 ] ( dashed line ) , and @xmath215 ( solid line ) with haar wavelets * ( a ) * and alpert - rohkin wavelets with @xmath203 * ( b)*. , title="fig:",width=321 ] it is instructive to compare the wavelet approach to conventional techniques . we consider the computation of all positions @xmath216 at equidistant times @xmath217 ( @xmath218 ) at some scale @xmath219 . in a classical scheme , brownian motion is modeled by a sequence of small random jumps @xmath220 with @xmath221 independent normally distributed random variables @xmath222 . similar computation relying on wavelet representations requires one random variable for a linear shift and @xmath223 random variables at each scale @xmath47 , @xmath47 ranging from @xmath6 to @xmath224 . the total number is then @xmath225 . it is not surprising that both schemes require the same degree of randomness to represent a brownian path at chosen scale . the wavelet representation does not reduce the complexity or randomness , but re - organize the data in a hierarchical structure to facilitate their use . for instance , formula ( [ eq : bomega1 ] ) accesses approximate positions of brownian motion at any time point @xmath38 , not necessarily @xmath226 . in a classical scheme , one could use a linear interpolation between two neighboring points to get the same result . again , the wavelets do not bring new features which are not available by conventional techniques , but provide another , structured and efficient , representation . throughout the above sections , brownian motion was constructed on the unit interval for convenience . the constructed process can be easily rescaled to any finite interval , while an extension to @xmath227 or @xmath228 is possible as well . finally , an extension to isotropic brownian motion in @xmath229 is obtained by taking @xmath5 independent samples of one - dimensional brownian motion . a similar technique can be applied to construct and study fractional brownian motion which is also known as random fractal velocity field @xcite . for instance , random fractal velocity field with the hurst exponent @xmath230 ( defined below ) corresponds to the kolmogorov spectrum in high reynolds number turbulence @xcite . fractional brownian motion , as a gaussian stochastic process with long - range correlations , has found numerous applications in different fields , ranging from transport phenomena in porous media @xcite to analysis of financial markets @xcite . p. lvy proposed the first extension of eq . ( [ eq : bnoise ] ) by using the riemann - liouville fractional integration which can be thought of as a moving average of a gaussian white noise @xcite @xmath231 where @xmath232 is the hurst exponent , and @xmath233 is the normalization factor ( @xmath234 being the gamma function ) . mandelbrot and van ness discussed the limitations of this definition ( e.g. , its strong emphasis on the origin ) and proposed to use the weyl fractional integral that yields @xcite @xmath235 dw(s ) \\ & + \int\limits_{0}^t ( t - s)^{h-\frac12 } dw(s ) \biggr\ } \\ \end{split}\ ] ] for @xmath236 ( and similar for @xmath237 ) . the last representation can also be written as ( see @xcite ) @xmath238 where @xmath239 and @xmath240 is the hypergeometric function . the ordinary brownian motion is retrieved at @xmath241 for all cases . using a wavelet representation ( [ eq : dw_psi ] ) for the gaussian white noise , one obtains @xmath242 the integrals can be evaluated using an appropriate basis @xmath147 . moreover , the moment cancellation property ( [ eq : moments ] ) for the alpert - rokhlin multiwavelets with a large enough order @xmath106 guarantees that the integrals in eq . ( [ eq : fbm4 ] ) are highly localized , yielding a rapid convergence of the above sum . this convergence is a key point for efficient numerical algorithms for simulation of fractional brownian motion ( see @xcite ) . among other numerical methods , we mention alternative wavelet representations @xcite ( e.g. , the method by arby and sellan is implemented in the matlab function ` wfmb ' ) , circulant embedding of the covariance matrix @xcite , and random midpoint displacement method @xcite which is often used in computer graphics to generate random two - dimensional landscapes . in general , the kernel @xmath243 can be replaced by any convenient kernel to extend this approach to various gaussian processes . for instance , setting @xmath244 yields the ornstein - uhlenbeck process @xcite . similarly , one can deal with various stochastic dynamics generated by langevin equations @xcite . brownian motion is the integral of the gaussian white noise which , in turn , is obtained as a linear combination of orthonormal functions @xmath245 forming a complete basis of the space @xmath145)$ ] , with standard gaussian weights . this construction can be extended to any separable hilbert space @xmath246 . however , whatever the functional space @xmath246 is taken , a linear combination of its orthonormal basis functions with standard gaussian weights does not belong to this space ( the argument is the same as for @xmath247 space , the norm of such a linear combination being infinite ) . in particular , the gaussian white noise is not a function but a distribution . in order to construct extensions of the gaussian white noise with desired properties , one needs to carefully choose the hilbert space @xmath246 . in this section , we briefly discuss two such extensions : gaussian free field ( gff ) @xcite and fractional gaussian field ( fgf ) with logarithmic correlations @xcite . the gff appears as the basic description of massless non - interacting particles in field theories . both gff and fgf constitute important models in different areas of physics , from astrophysics ( describing stochastic anisotropy in cosmic microwave background ) to critical phenomena , quantum physics , and turbulence @xcite . while the `` sequence '' of random variables of brownian motion @xmath162 was naturally parameterized by `` time '' @xmath38 ( a real number from the unit interval or , in general , from @xmath228 ) , random variables of a field can in general be parameterized by points from an euclidean domain , a manifold , or a graph . for instance , one can speak about random surfaces which can model landscapes ( e.g. , mountains ) or ocean s water surface , in which the height is parameterized by two coordinates . the geometrical structure of `` smooth '' random surfaces was thoroughly investigated @xcite , especially for gaussian fields which are fully characterized by their mean and covariance . important examples of smooth gaussian fields are the random plane waves and random spherical harmonics ( see @xcite ) . in turn , the gff and fgf are examples of highly irregular random fields . as brownian motion can be obtained as the limit of discrete random walks , the gaussian free field in two dimensions appears in the limit of discrete random surfaces @xcite . the gaussian free field is constructed by choosing the dirichlet hilbert space @xmath248 , in which the scalar product of two functions @xmath249 and @xmath250 is defined as @xmath251 for a given euclidean domain @xmath252 . when @xmath253 is bounded , an orthonormal basis of this space can be obtained by setting @xmath254 , where @xmath255 are the @xmath247-normalized dirichlet eigenfunctions of the laplace operator : @xmath256 in @xmath253 with @xmath257 at the boundary @xmath258 , and @xmath259 are the corresponding eigenvalues . the gaussian free field on @xmath253 is defined as @xmath260 where @xmath261 are independent gaussian weights . given that the eigenvalues @xmath259 asymptotically grow as @xmath262 according to the weyl s law @xcite , the sum in eq . ( [ eq : gff_2d ] ) is convergent for @xmath263 ( in which case the gff is simply a brownian bridge ) but diverges in higher dimensions . in the plane , this sum barely misses the convergence , being logarithmically diverging . in quantum field theory , it is related to the infra - red divergence for massless particles . as a consequence , @xmath264 is not a function but a distribution for @xmath265 . being a linear combination of normal variables , the field @xmath264 is gaussian and thus is fully characterized by its mean @xmath266 and the covariance @xmath267 where the right - hand side can be recognized as the green function @xmath268 of the laplace operator in the domain @xmath253 . alternatively , one could define the gff by setting the covariance equal to the green function ( in which case the definition holds even for unbounded domains ) . strictly speaking , since the sum in eq . ( [ eq : gff_2d ] ) diverges for @xmath269 , the gff should be treated as a distribution by its action on every fixed test function @xmath270 . in particular , the covariance should be given by covariance of actions of @xmath271 on two test functions @xmath272 and @xmath273 : @xmath274 the fractional gaussian fields can be obtained by replacing the gradient operators @xmath275 in the scalar product ( [ eq : hilbert ] ) by fractional laplacians @xcite , @xmath276 for a positive @xmath277 . the dirichlet hilbert space is retrieved for @xmath278 . in a similar way , the functions @xmath279 form a basis of the space @xmath280 , from which the fgf is constructed as in eq . ( [ eq : gff_2d ] ) . this sum is convergent for @xmath281 and divergent otherwise . in the particular case @xmath282 , the fgf is logarithmically divergent in all dimensions . note that the fgf coincides with the gff in the plane ( @xmath283 ) . in particular , the fgf with logarithmic correlations on the unit interval reads explicitly as @xmath284 more generally , one can take any orthonormal basis @xmath285 of @xmath145)$ ] and then apply the operator @xmath286 to get the basis of @xmath287)$ ] . the advantage of the laplacian eigenbasis is that the fractional integral operator @xmath288 ( which can be defined through the fourier transform ) is replaced by multiplication by @xmath289 . as a consequence , the fgf with logarithmic corrections in one dimension appears as the half - derivative of brownian motion : @xmath290 this formula reveals a very close relation between these two processes which are often considered as distinct objects . note that the series in eq . ( [ eq : auxil5 ] ) diverges logarithmically , while the derivative of order @xmath291 would lead to converging series . in other words , brownian motion belongs to the hlder space @xmath130 ( for any @xmath292 ) so that its derivatives of order less than @xmath293 exist , but @xmath293 and higher do not . as for brownian motion , the explicit closed formula ( [ eq : gff_2d ] ) allows one to sample random realizations of the fgf by picking up @xmath70 from the unit interval , i.e. , the probability space for this process is nothing else than the unit interval with uniform measure . in this section , we briefly discuss how multiscaling and dyadic decompositions help simulating restricted diffusion . this is an ubiquitous problem in physics ( e.g. , transport in porous media ) , chemistry ( e.g. , heterogeneous catalysis ) , biology and physiology ( e.g. , diffusion in cells , tissues , and organs ) . when brownian motion is restricted , physico - chemical or biological interactions between the diffusing particle and the interface of a confining medium should be taken into account . for instance , paramagnetic impurities dispersed on a liquid / solid interface cause surface relaxation in nuclear magnetic resonance ( nmr ) experiments @xcite ; cellular membranes allow for a semi - permeable transport through the boundary @xcite ; chemical reaction may transform the particle or alter its diffusive properties @xcite . while an accurate description of these processes at the microscopic level is challenging , the contact with the interface is very rapid at macroscopic scales , allowing one to resort to an effective description of the surface transport by an absorption / reflection mechanism @xcite . this mathematical process is known as ( partially ) reflected brownian motion @xcite . the presence of a boundary drastically changes the properties of brownian motion ( e.g. , the reflecting boundary forces the process to remain inside the domain ) so that earlier wavelet representations can not directly incorporate the effect of the boundary . since restricted diffusion is relevant for most physical , chemical and biological applications , various monte carlo methods have been developed for simulating this stochastic process , computing the related statistics ( e.g. , the first passage times @xcite ) , and solving the underlying boundary value problems @xcite . the slow convergence of monte carlo techniques ( typically of the order of @xmath294 in the number of trials ) requires fast generation of brownian paths . the simplest generation of a brownian path at successive times @xmath295 by adding normally distributed displacements and checking the boundary effects at each step becomes inefficient in multiscale media . in fact , tiny geometrical details of the medium require the use of comparably small displacements , resulting in a very large number of steps needed to model large - scale excursions . to overcome this limitation , the concept of fast random walks was proposed @xcite . the basic idea consists in adapting displacements to the local geometrical environment , performing as large as possible displacements without violating the properties of brownian motion . when the walker is at point @xmath81 , the largest displacement is possible at the distance @xmath296 between @xmath81 and the boundary @xmath297 of an euclidean domain @xmath252 . in fact , the ball @xmath298 of radius @xmath296 does not contain any `` obstacle '' ( e.g. , piece of boundary ) to the walker . since brownian motion is continuous , it must leave the ball before approaching the boundary of the confining domain . the rotation symmetry implies that the exit points are distributed uniformly over the boundary of the ball . instead of modeling the fully - resolved trajectory of brownian motion inside the ball , one can just pick up at random a point @xmath299 on the sphere of radius @xmath296 and move the random walker at this new position . the random duration of this displacement can be easily generated @xcite . from here , one draws a new ball @xmath300 , and so on , until the walker approaches the boundary @xmath297 closer than a chosen threshold . from this point , an appropriate boundary effect ( e.g. , absorption , relaxation , chemical transformation , permeation , reflection , etc . ) is implemented . due to its efficiency , fast random walk algorithms have been used to simulate diffusion - limited aggregates ( dla)@xcite , to generate the harmonic measure on fractals @xcite , to model diffusion - reaction phenomena in spherical packs @xcite , to compute the signal attenuation in pulsed - gradient spin - echo experiments @xcite , etc . in this section , we focus on multiscale tools to estimate the distance , while other aspects of fast random walk algorithms can be found elsewhere @xcite . the efficiency of fast random walk algorithms fully relies on the ability to rapidly estimate the distance between any point ( e.g. , the current position of the walker ) and the boundary . multiscale dyadic decompositions provide an efficient way to these estimates . to illustrate the idea , we first consider the one - dimensional case and then discuss its straightforward extension to the multidimensional case . in one dimension , the problem can be formulated as follows : given a set @xmath301 of @xmath2 `` boundary '' points on the unit interval , how the distance to this set from another point @xmath81 can be estimated in a rapid way ? successive computation of the distances @xmath302 and finding their minimum is of course the simplest but the slowest way ( of order of @xmath2 ) . instead of computing the distances to all boundary points , one can split the unit interval into two half - intervals , and check the distance to the points belonging to the half - interval that contains @xmath81 . if the distribution of points @xmath303 on @xmath97 $ ] is more or less uniform , this division approximately halves the number of computations . in the same spirit , splitting on subintervals of length @xmath28 , @xmath304 , etc . would reduce the number of computations roughly by factors @xmath305 , @xmath306 , etc . using such dyadic decompositions , one needs approximately @xmath307 splitting to attend the level when one ( or few ) point @xmath303 belongs to the same subinterval as @xmath81 . the number of computations is then of order of @xmath307 ( assuming the distribution of boundary points is more or less uniform ) . moreover , if computation is carried with a desired precision @xmath78 ( to consider @xmath303 as `` pointlike '' , one needs @xmath308 ) , one can continue splitting up to the level @xmath79 so that the length of subintervals becomes smaller than @xmath78 . since the points @xmath301 are stored with precision @xmath78 , one can not distinguish two points at any scale smaller than @xmath78 . consequently , any subinterval of length @xmath78 can be either vacant , or occupied by only one point @xmath303 ( two points from the same subinterval would be indistinguishable ) . in this case , the number of computations , @xmath79 , is actually independent of whether the distribution of points @xmath303 is uniform or not . in other words , this algorithm can be applied for any finite set of points @xmath303 that are all distinguishable at scale @xmath78 , i.e. , @xmath309 for any @xmath65 and @xmath86 . by its binary expansion . , width=321 ] for practical implementation , the boundary points @xmath303 are used to generate a dyadic tree of subintervals at the scales ranging from @xmath7 to @xmath79 . in turn , the binary expansion of the test point @xmath81 is used to `` navigate '' search on the tree ( fig . [ fig : dyadic ] ) . in fact , one can associate to a given point @xmath310 $ ] a sequence of dyadic intervals @xmath311 such that @xmath312 at any scale @xmath65 . at each scale @xmath65 , one chooses the left or the right subinterval depending on whether the @xmath313 bit is @xmath6 or @xmath7 . applying this procedure to all boundary points @xmath303 , one can generate a dyadic decomposition of the boundary . figure [ fig:1d_decomp]a shows an example with three boundary points @xmath314 at five scales , from @xmath66 to the smallest one @xmath315 . the dyadic decomposition is stored as a tree , where a vertex is associated with a subinterval . each vertex can be connected to one , two , or three other vertices ( see fig . [ fig:1d_decomp]a ) . the `` height '' of the vertex from the `` root '' determines the scale of the corresponding subinterval . ( shown by vertical dashed lines ) and the related tree of subintervals at five levels ( @xmath316 ) . for a test point @xmath317 ( shown by arrow ) , one uses its binary expansion @xmath318 to navigate over the tree . the descend is stopped at level @xmath319 since @xmath320 $ ] . however , the rough estimate , @xmath321 , of the distance between @xmath81 and the boundary , @xmath322 , obviously fails . to apply the @xmath323-trick , one constructs the dyadic decomposition for the boundary points shifted by @xmath323 * ( b)*. in this tree , the descent for the point @xmath324 is stopped at level @xmath325 . the combined lower estimate @xmath326 is valid.,title="fig:",width=158 ] ( shown by vertical dashed lines ) and the related tree of subintervals at five levels ( @xmath316 ) . for a test point @xmath317 ( shown by arrow ) , one uses its binary expansion @xmath318 to navigate over the tree . the descend is stopped at level @xmath319 since @xmath320 $ ] . however , the rough estimate , @xmath321 , of the distance between @xmath81 and the boundary , @xmath322 , obviously fails . to apply the @xmath323-trick , one constructs the dyadic decomposition for the boundary points shifted by @xmath323 * ( b)*. in this tree , the descent for the point @xmath324 is stopped at level @xmath325 . the combined lower estimate @xmath326 is valid.,title="fig:",width=317 ] once a dyadic decomposition for a given boundary is constructed , it can be used to estimate the distance to the boundary from any point @xmath81 . in fact , one can easily ( and very rapidly ) find the smallest subinterval @xmath327 containing simultaneously @xmath81 and some boundary point . for this purpose , one starts from the `` root '' vertex and descends on the tree using the bits of @xmath81 to choose left or right edges at each scale . the descend is stopped when there is no edge to follow ( fig . [ fig:1d_decomp]a ) . once the smallest common interval @xmath327 is found , there are two option : either @xmath328 so that the point @xmath81 is indistinguishable from some boundary point at scale @xmath78 , and the distance estimate is set to @xmath6 ; or @xmath329 , and the distance to the boundary can be roughly estimated as @xmath49 since the next subdivision must separate the point @xmath81 from the boundary points . however , this simplistic argument fails when the point @xmath81 and the closest boundary point lie on opposite sides of the midpoint of the smallest common interval @xmath327 , but very close to each other . although the next subdivision separates these two points , the distance between them can be arbitrarily small so that the rough estimate @xmath49 is wrong , as illustrated on fig . [ fig:1d_decomp]a . to get the correct lower estimate , one can apply the so - called `` 1/3-trick '' . the @xmath323-trick can be easily illustrated for two points @xmath81 and @xmath119 . let @xmath330 be the smallest common interval containing both points @xmath81 and @xmath119 . suppose that the distance between these points is smaller than @xmath331 . we consider the points @xmath332 and @xmath333 ( shifted by @xmath323 ) , and determine their smallest common interval @xmath334 . as shown in appendix [ sec : trick_proof ] , the distance between the points @xmath299 and @xmath335 ( and thus between the points @xmath81 and @xmath119 ) is larger than @xmath336 . it is thus sufficient to find the smallest common intervals for the pair @xmath337 and its shifted counterpart @xmath338 , and the distance between the points is bounded below as @xmath339 this simple fact allows one to rapidly estimate the distance to boundary points . let us consider a new boundary @xmath340 which is obtained by shifting the old one by @xmath323 : @xmath341 . for the new boundary , another dyadic tree of subintervals can be constructed ( fig . [ fig:1d_decomp]b ) in order to estimate the distance between @xmath340 and the shifted point @xmath342 at a given scale @xmath78 . while this construction may appear redundant at first thought because @xmath343 , the crucial point is that we search for a lower estimate at a finite scale at which two dyadic trees are different . performing the descend over both dyadic trees ( using the binary expasion of @xmath81 and @xmath299 , respectively ) , one identifies the level @xmath65 ( resp . @xmath344 ) of the smallest common interval of @xmath81 ( resp . @xmath299 ) and the closest boundary point ( resp . shifted closest boundary point ) . the lower estimate of the distance is then @xmath345 similar constructions are applicable in higher dimensions which are more relevant for applications . the lower estimate relies on the generalized mean inequality : @xmath346 where the generalized mean @xmath347 of @xmath65 positive numbers @xmath348 is @xmath349 setting @xmath202 , @xmath350 , and @xmath351 for two points @xmath352 and @xmath353 , eq . ( [ eq : generalized_mean ] ) yields the lower estimate of their euclidean distance : @xmath354 and thus the distance from any point @xmath355 to the boundary . in high dimensions ( @xmath356 ) , the right - hand side of the inequality ( [ eq : dist_d ] ) is strongly attenuated by the prefactor @xmath357 . this is an example of the so - called `` curse of dimensionality '' . to overcome this difficulty , one can implement random rotations and translations of the boundary . although these transformations preserve the distance , they can improve the lower bound at a finite scale . note that other multiscale constructions and the related searchable data structures can also be used such as whitney decompositions of the computational domain ( in which the size of each square ( or cube ) paving the domain is comparable to the distance to the boundary ) , quadtrees ( or q - trees ) , k - d trees , etc . @xcite . the advantages of multiscale dyadic trees are numerous : the simplicity of construction , the generality of boundary shapes , the rapidity of distance estimation , the flexibility for shape modifications , and low memory usage . in fact , for a given precision @xmath78 , the storage of the dyadic tree requires at worst @xmath358 intervals ( i.e. , @xmath79 levels for each boundary point , for @xmath2 points ) . in practice , this number is much smaller since many boundary points share the same interval at larger scales ( e.g. , the interval of size @xmath7 is shared by all boundary points ) . the crucial point is that geometrical structure of the boundary does not matter at all : the method works for a random cantor dust as well as for a circle . moreover , the tree - like representation is highly adaptive , allowing one to modify the boundary from one set of simulations to other ( or even from one run to the other ) . this feature can be very useful to study diffusion - controlled growth processes like dla or transport phenomena in domains with moving boundaries . in this paper , we revised the multiscale construction of gaussian processes and fields . first , the haar wavelet representation of brownian motion was explicitly constructed as a natural way to refine the geometrical features of a brownian path under magnification . since the haar functions form a complete basis in the space @xmath145)$ ] and their weights are gaussian , such a representation can be extended to any complete basis @xmath147 of @xmath145)$ ] , wavelet - like or not . in other words , the construction of brownian motion has two separate `` ingredients '' : deterministic functions @xmath149 capturing geometrical details , and their random weights @xmath157 . the choice of the functions @xmath149 is voluntary that gives a certain freedom and flexibility in dealing with different problems . qualitatively , this choice determines the way of connecting successive positions of brownian motion at a given scale . on the opposite , the random weights determine the intrinsic stochastic properties of brownian motion , independently of our choice of the basis @xmath147 . the multiscale construction gives not only a simple closed formula for brownian motion , but reveals its fundamental properties . for instance , continuity and non - differentiability of brownian paths naturally follow from this construction . in addition , we discussed a closed mapping from the sampling unit interval onto the space of brownian paths . sampling brownian paths can therefore be formally reduced to picking up a real number @xmath70 with the uniform measure . this construction does not require elaborate notions from modern probability theory such as wiener measures , sigma - algebras , filtrations , etc . although these notions are useful , the explicit multiscale construction is much easier for non - mathematicians . these concepts are not limited to brownian motion . we illustrated how fractional brownian motion , gaussian free field , and fractional gaussian fields can be constructed in a very similar way . extensions to other gaussian processes were also mentioned . finally , we briefly discussed how the multiscale concepts can be used for simulating restricted diffusion ( i.e. , brownian motion in confining domains ) . the dyadic subdivision and the related hierarchical ( multiscale ) tree of subintervals allow one to rapidly estimate the distance to the boundary and thus to generate random displacements adapted to the local geometrical environment . such fast random walk algorithms have found numerous applications . as for usual brownian motion , dyadic decompositions appear as natural tools to store and rapidly access the geometrical information , resulting in fast algorithms . dg acknowledges the financial support by anr project anr-13-jsv5 - 0006 - 01 . db was partially funded by epsrc fellowship ref . ep / m002896/1 . we explain why the distribution of the position of brownian motion @xmath359 at time @xmath360 under conditions @xmath10 and @xmath361 is given by the normal law @xmath362 . since the increments of brownian motion on the unit intervals @xmath363 and @xmath364 are independent , the joint probability density @xmath365 is simply equal to the product of the probability density @xmath366 to have the first increment equal to @xmath119 and the probability density @xmath367 to have the second increment equal to @xmath368 . these densities are given by normal laws with variance @xmath293 , yielding @xmath369 the second factor is simply the probability density for @xmath370 , while the first factor is the conditional probability density we are looking for : @xmath371 this is the normal distribution with the mean @xmath372 and the variance @xmath28 . in this section , we illustrate how the wavelet representation of brownian motion can be used to prove its basic properties such as continuity and non - differentiability . first , we show that a partial sum approximation converges to brownian motion in both @xmath373 and @xmath247 norms . let us denote @xmath374 so that @xmath375 according to eq . ( [ eq : bomega1 ] ) . each function @xmath376 is a sum of hat functions with disjoint supports . to estimate the size of @xmath376 we need some upper bound on @xmath60 . we are going to show that @xmath377 for all sufficiently large @xmath65 . since @xmath378 are standard normal variables , we observe that @xmath379 for sufficiently large @xmath65 . this inequality yields @xmath380 by borel - cantelli lemma this implies that @xmath377 for all but finitely many coefficients @xmath60 . in other words , almost surely , there is finite , but random , @xmath381 such that @xmath377 if @xmath382 . for such @xmath65 we have @xmath383 ) } = 2^{-n/2 - 1}\max_k \{|a_{nk}|\ } < n 2^{-n/2 - 1}.\ ] ] since the sum of these norms converges , the series @xmath384 converges to @xmath162 in @xmath373 norm ( uniformly ) , i.e. for any @xmath292 @xmath385 ) } > \epsilon \right\ } = 0 , \ ] ] where @xmath386 is a partial sum approximation of brownian motion at scale @xmath387 . this proves that brownian motion is almost surely continuous . moreover , the remainder of a partial sum approximation at scale @xmath387 is exponentially small : @xmath388 when @xmath2 is large enough . the @xmath247 convergence can be shown in the same way . we have @xmath389 ) } = \sum\limits_{k=0}^{2^n-1 } |a_{nk}|^2 \|h_{nk}\|^2_{l^2([0,1 ] ) } < 2^n ~ n^2 ~ 2^{-2n},\ ] ] where we used @xmath390 for large enough @xmath65 . this inequality implies the convergence of @xmath391)}$ ] with probability one and hence the series in eq . ( [ eq : bomega1 ] ) converges almost surely in @xmath247 norm : @xmath392 ) } \right\ } = 0 .\ ] ] both statements ( [ eq : bconv1 ] , [ eq : bconv2 ] ) can be extended to arbitrary spectral representation of brownian motion . note also that these statements are applicable pointwise , e.g. , @xmath393 for any @xmath394 $ ] . since brownian motion is a sum of hat functions it is easy to believe that @xmath162 it not differentiable almost everywhere . one can prove a much stronger statement that @xmath162 is nowhere differentiable with probability one . the proof goes along the same lines as for convergence ( the argument follows the proof from @xcite ) . we are going to show that almost surely for all @xmath38 at least one of the two limits , @xmath395 is infinite . this obviously implies that @xmath162 is almost surely nowhere differentiable . note that at local extrema , one of these limits can be finite , so it is not true that _ both _ of them are always infinite . let us assume that there is @xmath396 such that both limits are finite at @xmath396 . this implies that there is a random finite constant @xmath397 such that @xmath398 for all @xmath399 . for a given scale @xmath65 , let @xmath47 be such that @xmath396 is between dyadic points @xmath400 and @xmath401 , where @xmath402 . the triangle inequality implies that for any @xmath403 @xmath404 let @xmath405 be the event that this inequality holds for @xmath406 . since increments are independent normal variables , one gets @xmath407 , where @xmath408 is a constant . the probability that these inequalities hold for _ some _ @xmath47 from @xmath6 to @xmath409 is then bounded by @xmath410 . the sum of these probabilities over @xmath65 is finite , hence by borel - cantelli lemma , with probability one only finitely many of them will occur . on the other hand the assumed inequality ( [ eq : increment ] ) implies that infinitely many of @xmath405 will occur . this yields the contradiction and proves that ( [ eq : increment ] ) can not be true . although the 1/3-trick is classical in analysis , we provide some explanations which may be instructive for non - experts . the trick is based on a very simple result . let @xmath411 and its boundary @xmath412 . for any real @xmath81 and any integer @xmath65 , there exist two intervals @xmath72 and @xmath413 of the same length @xmath73 that @xmath81 belongs to @xmath72 and @xmath332 belongs to @xmath413 . if the distance from @xmath81 to the boundary of @xmath72 is smaller than @xmath331 , then the distance from @xmath299 to the boundary of @xmath413 is larger than @xmath331 , and vice - versa . in other words , the point @xmath81 and the shifted point @xmath299 can not be simultaneously close to the interval endpoints . suppose the opposite is true , so that @xmath414 where the integers @xmath415 $ ] and @xmath416 $ ] denote the closest endpoint to @xmath81 and @xmath299 , respectively . since @xmath417 with an integer @xmath418 , the point @xmath419 should be simultaneously within the distance @xmath420 to @xmath421 and @xmath422 that is impossible since the distance between these points is larger than @xmath323 ( fig . [ fig : trick ] ) : @xmath423 using this simple result , one can prove the estimate for the distance from a given point @xmath81 to the set of boundary points @xmath425 . let @xmath72 be the largest interval containing @xmath81 and not containing any boundary point @xmath426 . similarly , for the shifted point @xmath332 , let @xmath427 be the largest interval containing @xmath299 and not containing any shifted boundary point @xmath428 . then the distance from @xmath81 to the set of boundary points @xmath425 is larger than @xmath429 . suppose that @xmath430 . assume that the statement is false so there exists a boundary point @xmath431 such that @xmath432 . suppose that @xmath433 ( the opposite case is similar ) . since @xmath434 and @xmath435 , both points @xmath81 and @xmath119 should be close to the endpoint @xmath436 : @xmath437 the shifted boundary point @xmath335 belongs to some interval @xmath438 . according to the previous result , the second inequality implies that the shifted point @xmath335 can not be close to the endpoints of @xmath438 : @xmath439 since @xmath440 and @xmath433 , then @xmath441 , hence @xmath442 i.e. , the point @xmath299 belongs to the same interval @xmath438 as @xmath335 . at the same time , @xmath299 belongs to @xmath427 which is larger than @xmath438 due to @xmath430 . the dyadic structure implies that @xmath443 so that @xmath335 should belong to @xmath427 as well . but this is in contradiction with the initial assumption about @xmath427 . n. kobayashi , y. yamazaki , h. kuninaka , m. katori , m. matsushita , s. matsushita , and l .- y . chiang , `` fractal structure of isothermal lines and loops on the cosmic microwave background '' , j. phys . japan * 80 * , 074003 ( 2010 ) . r. fernandez , j. frhlich , and a. d. sokal , _ random walks , critical phenomena , and triviality in quantum field theory _ ( texts and monographs in physics . springer , berlin heidelberg , new york , 1992 ) . r. brown , `` a brief account of microscopical observations made in the months of june , july and august , 1827 , on the particles contained in the pollen of plants ; and on the general existence of active molecules in organic and inorganic bodies '' , edinburgh new phil . j. * 5 * , 358 - 371 ( 1828 ) . b. alpert , `` construction of simple multi - scale bases for fast matrix operations '' , in _ wavelets and their applications _ , ed . by ruskai , beylkin , coifman , daubechies , mallat , mayer , and raphael ( jones & bartlett , boston , 1992 ) , p. 211 . g. beylkin , r. coifman , and v. rokhlin , `` wavelets in numerical analysis '' , in _ wavelets and their applications _ , ed . by ruskai , beylkin , coifman , daubechies , mallat , mayer , and raphael ( jones & bartlett , boston , 1992 ) , p. 181 . p. abry and f. sellan , `` the wavelet - based synthesis for fractional brownian motion proposed by f. sellan and y. meyer : remarks and fast implementation '' , appl . comput . harm . anal . * 3 * , 377 - 383 ( 1996 ) . w. t. coffey , y. p. kalmykov , and j. t. waldron , _ the langevin equation : with applications to stochastic problems in physics , chemistry and electrical engineering _ ( world scientific publishing , singapore , 2004 ) . r. j. adler and j. e. taylor , _ random fields and geometry _ ( springer monographs in mathematics , 2007 ) . e. vanmarcke , _ random fields : analysis and synthesis _ ( world scientific publishing company , 2010 ) . d. s. grebenkov , `` partially reflected brownian motion : a stochastic approach to transport phenomena '' , in `` focus on probability theory '' , ed . l. r. velle , pp . 135 - 169 ( nova science publishers , 2006 ) . d. s. grebenkov , `` efficient monte carlo methods for simulating diffusion - reaction processes in complex systems '' , in `` first - passage phenomena and their applications '' , eds . r. metzler , g. oshanin , s. redner ( world scientific press , 2014 ) .

in two recent papers @xcite , a new calculation of the phase shift of a matter - wave interferometer caused by the passage of a gravitational wave was presented . this calculation was based not on the usual geodesic equation of motion ( eom ) , but rather on the geodesic _ deviation _ eom . contrary to expectations , these phase shifts have a form that is very much different from those calculated previously in the literature @xcite . while a direct comparison of the various expressions for the phase shift is difficult because of the different configurations of the matter - wave interferometers used for their calculations , it was noted in @xcite that the power - law dependence on the frequency of the gravitational wave , in both the low- and high - frequency limit , of the expressions calculated in @xcite are at times diametrically opposite to those calculated in @xcite . consequently , we expect that the differences between these expressions are _ not _ due simply to differences in the configurations of the interferometers chosen , but are rather caused by fundamental differences in the physics on which their derivations are based . in this paper our objective is to demonstrate that these fundamental differences do exist by presenting a series of calculations that follows the two basic approaches in the literature used to determine the phase shift caused by a gravitational wave . this will be done for the _ same _ configuration of matter - wave interferometer , and will allow us to compare the different expressions we obtain for the phase shift on an equal footing . we then determine their validity by comparing each of them , in the appropriate limit , with the well - known expression for the phase shift caused by _ stationary _ sources of spacetime curvature @xcite . the focus in this paper will be on a _ theoretical _ study of the underlying physics ; we leave issues of the practicality and feasibility of constructing the interferometer to @xcite . the various approaches taken to determine the phase shift of a matter - wave interferometer caused by a gravitational wave can be divided into two distinctly separate classes . the class of approaches that is more often found in the literature @xcite is done in transverse - traceless ( tt ) coordinates ( see sec . 9.2.3 of @xcite ) , either explicitly or implicitly . the tt coordinates are one specific choice of reference frame where the gravitation wave is in the tt gauge . in this frame the number of components of the deviations of the flat spacetime metric @xmath0 caused by the gravitational wave equals the number of physical degrees of freedom of the gravitational wave . consequently , @xmath1 , @xmath2 , and @xmath3 . in addition , the mirrors and the beam splitters are at fixed coordinate values in these coordinates , and they do not move . these coordinates have been used to study the properties of light - wave - interferometry - based detectors of gravitational waves as well @xcite . to calculate the phase shift for matter - wave interferometers , it is combined with either the geodesic eom ( and its action ) , if the calculation is done at the quantum - mechanical level @xcite , or the lagrangian ( and the wkb or stationary phase approximation ) for a quantum field in curved spacetime , if the calculation is done at the quantum field - theoretic level @xcite . not surprisingly , irrespective of the level of sophistication of the approach used , the general expressions calculated for the phase shift have the same final form . the other class of approaches to calculating the phase shift is taken by @xcite , and is done in thorne s `` proper reference frame '' . they follow the classical analysis in @xcite for light - wave - based interferometers . in this frame , the positions and velocities of test particles are measured relative to the worldline of an observer , and follow the motion of the observer in spacetime , in much the same way that the fermi - normal @xcite , fermi - walker coordinates @xcite , and the recently constructed general laboratory frame @xcite do . consequently , the motion of the test particle in this frame is _ not _ described by the usual geodesic equation of motion ( eom ) , but rather by the geodesic _ deviation _ eom ( see eq . ( 2 ) of @xcite ) , which was first applied to the analysis of physical systems in @xcite . as in @xcite , the tt gauge was taken for the gravitational wave . the action ( see @xcite ) for the geodesic deviation eom was then used in conjunction with the quasi - classical approximation of the feynman - path - integral formulation of quantum mechanics . this approximation is the lagrangian form of the wkb approximation in the schrdinger representation of quantum mechanics , and in @xcite it was applied to the calculation of the phase shift of the two matter - wave - interferometer configurations . the phase shift of a matter - wave interferometer is a physically measurable quantity . from the principle of general covariance , it should not matter whether the tt coordinates , or whether the proper reference frame is used in its calculation @xcite . indeed , thorne states the following : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` if @xmath4 then the [ gravitational wave ] detector can be contained entirely in the proper reference frame of its center , and the analysis can be performed using non - relativistic concepts augmented by the quadrupolar gravity - wave force field ( 3 ) , ( 5 ) . if one prefers , of course , one instead can analyze the detector in tt coordinates using general relativistic concepts and the spacetime metric ( 8) . the two analyses are guaranteed to give the same predictions for the detector s performance , _ unless errors are made _ [ emphasis added ] . '' ( from page 400 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ evidently , errors have indeed been made , as we shall see , since the resulting calculations often contradict each other for the same interferometer . the power - law dependence on frequency of the phase shifts calculated in @xcite using tt coordinates are fundamentally different from that of the phase shifts calculated in @xcite using the proper reference frame . since the principle of general covariance must hold , this means that either ( a ) the calculations in @xcite are wrong , ( b ) the calculations in @xcite are wrong , or ( c ) all of the calculations of the phase shifts are wrong . to determine which one of these three possibilities is correct , we will present side by side calculations of the phase shift caused by a gravitational wave for one specific matter - wave - interferometer configuration using the following three approaches . the first calculation uses the quasi - classical approximation in the tt coordinates and geodesic action , and is done at the nonrelativistic , quantum - mechanics level . it follows , as much as possible , the line of analysis given in @xcite . the second calculation is also done at the nonrelativistic , quantum - mechanics level using the same approximation , but now in the proper reference frame , and uses the geodesic _ deviation _ action . it follows the line of analysis given in @xcite . these two calculations would seem to exhaust the two different classes of approaches currently found in the literature . that a third calculation of the phase shift is needed is because of an error in @xcite . either because changes in the kinetic energy term in the action for the atom in @xcite were neglected , or because the turning points of the atom s phase at the interferometer s mirrors in the wkb - type of approach in @xcite were omitted , the end result is that important boundary terms are not included in the final expression for the phase shift in @xcite . for the sake of completeness , and to compare the calculations in @xcite with these previous calculations in the literature , this necessitates a _ third _ calculation of the phase shift , which will result in yet another expression for the phase shift based on the geodesic eom . we find that all three expressions for the phase shift calculated here are different from one another . although there are currently no _ direct _ experimental checks that can used to determine which one of these three expressions for the phase shift is correct , an _ indirect _ experimental check can be used by taking the low - frequency limit . all three expressions for the phase shift depend on the transit time of the atom through the interferometer , and the period of the gravitational wave . if the transit time of the atom is much shorter than the period , then , with respect to the atom , the gravitational wave does not oscillate appreciably as it traverses the interferometer . the gravitational wave will contribute a local riemannian curvature to the spacetime that is effectively static . the phase shift of matter - interferometers caused by static ( as well as stationary ) sources of curvature is well established @xcite , and has been shown to be proportional to components of the riemann curvature tensor . just as importantly , this phase shift can be calculated using only newtonian gravity [ eq . ( 2.6 ) of @xcite ] . the phase shift due to the newtonian gravitational potentialthrough the acceleration due to earth s gravity @xmath5has been measured , first by collela , overhauser and werner @xcite , and more recently by chu and kasevich @xcite . these more recent measurements of the phase shift induced by @xmath5 are sensitive enough that effect of _ spatial variations _ in @xmath5which are proportional to the static riemann curvature of the spacetime caused by the earthon the phase shift can be discerned . gradients in @xmath5 were seen through _ changes _ in the phase shift of an atomic fountain when its vertical position was varied @xcite . more recently , they were seen through a differential measurement of the phase shifts of two atomic fountain matter - wave interferometers operating in tandem @xcite . therefore , while the direct measurement of the phase shift of a matter - wave interferometer caused by a static riemann curvature tensor has not been made , its discovery is expected . indeed , indications are that the phase shift induced by the static riemann curvature of the earth has already been seen @xcite . as a consequence of the above , in the limit of low - frequency gravitational waves the phase shift calculated for the matter - wave interferometer _ must _ approach the well - known static result , and be proportional to the curvature tensor . we will see that in this limit , only the phase shift calculated along the lines of @xcite has the correct form . the other two do not , and _ must be wrong _ , since they do not yield the correct static limit . our focus here is on the phase shift that a _ gravitational wave _ will induce on a matter - wave interferometer . we will thus neglect the effects of all other gravitational effectssuch as the acceleration due to earth s gravity , and the local curvature of spacetime due to _ stationary _ sources such as the earth , sun and moonin addition to the phase shift due to the sagnac effect caused by the earth s rotation . we have shown in @xcite that for these types of stationary gravitational effects the description of the motion of test particles in the general laboratory frameof which the proper reference frame is a special casereduces to the usual geodesic eom description , and there is no controversy . the phase shifts caused by these stationary effects calculated in either the proper reference frame or the tt coordinates will be the same . we will in this paper use the same notation the is commonly found in the literature @xcite . in particular , it follows the notation in @xcite . while this has the benefit of consistency and simplicity , it means that we do not separate with notation the coordinates used for the tt coordinates from the coordinate used for the proper reference frame . the coordinates used in the two reference frames have different physical meanings , which we will make clear in the next section . these differences are outlined in detail in @xcite . in this section we will present two of the three calculations of the phase shift mentioned in the introduction . the first calculation will be done in the tt coordinates , and will be based on the geodesic eom . the second calculation will be done in the proper reference frame , and will be based on the geodesic deviation eom . a third calculation , following the approach in the literature , will be done in the next section where comparisons between the different results of the calculations are made . all three calculations are done for gravitational waves in the long - wavelength limit for the matter - wave interferometer configuration shown in fig . @xmath6 ; this configuration is described in detail in * appendix a*. to calculate the phase shifts , we will use the quasi - classical approximation of the feynman path integral representation of quantum mechanics . this approximation is the lagrangian version of the wkb approximation , or stationary phase approximation , in the schrdinger representation of quantum mechanics , and will be reviewed in * appendix a * as well . the presentation in this section is done in detail to ensure that all the underlying assumptions , approximations , and subtleties in our analysis are readily apparent . consider a gravitational wave in the long - wavelength limit with an amplitude @xmath7 incident perpendicularly to the plane of the interferometer in fig . @xmath6 . as usual , greek indices run from @xmath8 to @xmath9 , and latin indices run from @xmath10 to @xmath11 . the gravitational wave is represented by the tensor @xmath0 as a perturbation on the flat spacetime metric @xmath12 , so that the total metric of the spacetime is @xmath13 . we have chosen the signature of @xmath14 to be @xmath15 . in the tt - gauge , a coordinate system is chosen so that in these coordinates , @xmath16 , @xmath17 , and @xmath3 ; the only nonvanishing components of @xmath0 are thus @xmath18 . one choice of coordinates where the tt gauge is realized is the tt coordinates ( see @xcite and the general analysis in @xcite ) . in these coordinates the action for the atom with mass @xmath19 used in fig . @xmath6 is the usual geodesic action in the nonrelativistic limit , @xmath20dt , \label{geoaction}\ ] ] where the subscript `` geo '' stands for `` geodesic '' . since the rest mass @xmath21 only shifts the lagrangian for the atom by a constant , we drop this term from @xmath22 . the nonrelativistic hamiltonian for the motion is then , @xmath23 which results in the usual geodesic equation of motion @xmath24 here @xmath25 is christoffel symbol for the spacetime ; for gravitational waves in the tt gauge , @xmath26 . to calculate the phase shift of fig . @xmath6 , we follow the description in * appendix a * , and consider the phase shift due to the @xmath27th atom emitted from the atomic source , which at a time @xmath28 is diffracted by the initial beam splitter . there are two possible classical paths for the atom through the interferometer . in the absence of gravitational waves , these are straight - line paths @xmath29 ( see * appendix a * ) , which will be shifted when a gravitational wave is present . these shifts are expected to be very small , however , and we can use perturbation theory to solve for them . we thus take @xmath30 , and @xmath31 , where @xmath32 and @xmath33 are perturbations to the atom s path @xmath29 and velocity @xmath34 , respectively ; we will keep only terms linear in @xmath33 . to first order in @xmath18 , eq . @xmath35 becomes @xmath36 once @xmath29 is determined , @xmath37 is solved for by dividing the trajectory of an atom into two parts : from the initial beam splitter to the mirror , and then from the mirror to the final beam splitter . the two paths are then joined with the appropriate boundary conditions at the mirror . we begin by determining @xmath29 . to determine the unperturbed path of the atoms through the interferometerwhich is the same no matter which approach is used to calculate the phase shiftsconsider first the path of the atom that is diffracted into the @xmath38 diffraction order . we will assume that the velocity of the atoms is high enough that we can ignore the effect of earth s gravity on the atom . it is then clear that for @xmath39during which it travels between the initial beam splitter and the first mirror in a time @xmath40the path of the atom is @xmath41 where @xmath42 and @xmath43 are @xmath44 and @xmath45 components of the velocity of the atom immediately after it leaves the initial beam splitter . at @xmath46 the atom is reflected off the hard - wall mirror , and from the general expression of the boundary condition for the atom at the mirror eq . @xmath47 , @xmath48 , while @xmath49 . ( the superscripts @xmath50 denote @xmath51 respectively . ) for the subsequent time interval @xmath52 , the atom s path is @xmath53 during which it travels between the mirror and the final beam splitter . it is clear that @xmath54 and @xmath55 , and @xmath56 is the total transit time of the atom through the interferometer . the unperturbed path for the atom in the @xmath57 diffraction order is similar to eqs . @xmath58 and @xmath59 , and is obtained by taking @xmath60 and @xmath61 . to determine the boundary conditions for @xmath62 in eq . @xmath63 , we note that from eq . @xmath35 the acceleration of a test particle for the geodesic eom is proportional to its velocity . as such , a test particle which was initially at rest will stay at rest , _ even in the presence of a gravitational wave_. this led to the following conclusion by stodolsky : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` in this discussion [ of matter - wave interferometry ] we have assumed that the parts of the apparatus [ interferometer ] stay at fixed coordinate values . this can be achieved if these parts are free masses , since a metric of the gravity wave type ( with no `` 0 '' components ) does not affect the positions of nonmoving bodies . '' ( from page 399 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ thus , in the tt coordinates , the positions of beam splitters and the mirrors do not change when a gravitational wave passes through the system . the velocity of the atoms travelling between the supersonic source and the initial beam splitter , on the other hand , _ will _ be affected by the gravitational wave . since the interferometer in fig . @xmath6 is _ balanced _ , only _ differences _ in the atom s trajectory along the two paths through it will matter . we can thus neglect any shifts in the atom s velocity _ before _ it reaches the beam splitter . and , since the initial beam splitter is not moved by the gravitational wave , @xmath64 while @xmath65 ; thus we take @xmath66 . for the boundary conditions at the mirror , we note that like the beam splitter , the positions of the mirrors in the interferometer also do not change when a gravitational wave is present . thus the boundary conditions for @xmath33 across the hard - wall mirrors is @xmath67 integrating eq . @xmath63 is now straightforward , and for @xmath39 , @xmath68 while for @xmath52 , @xmath69 we now calculate the action for the @xmath27th atom as it travels through along the upper path through the interferometer . to first order in @xmath18 , @xmath70 & \equiv & \frac{1}{2}m\int_{t_r}^{t_r+t } \left\{[\vec{v}_0(t)+\vec{v}_1(t)]^2 + v_0^iv_0^jh_{ij}\right\ } dt , \nonumber \\ & \approx & \frac{1}{2}m\int_{t_r}^{t_r+t } \vec{v}_0^{\>2}\ > dt + m\int_{t_r}^{t_r+t}\left\{\vec{v}_0\cdot\vec{v}_1(t ) + \frac{1}{2}v_0^iv_0^jh_{ij}\right\}dt , \label{actiongeo}\end{aligned}\ ] ] where the superscript `` cl '' stands for `` classical '' , and denotes that the integration is over a classical path through the interferometer . the @xmath71 term in eq . @xmath72 is the same for both paths through the interferometer , and will not contribute to the phase shift . it can be neglected . the last term @xmath73 can be trivially integrated using eqs . @xmath58 and @xmath59 , while the @xmath74 term can be integrated using eqs . @xmath75 and @xmath76 . we then get @xmath77 = & t\left[v_\|^2h_{xx}(t_r)+v_\bot^2h_{yy}(t_r)\right ] \nonumber \\ & - \frac{1}{2}\int_{t_r}^{t_r+t } \bigg[v_\|^2h_{xx}(t)+v_\bot^2h_{yy}(t)\bigg]dt \nonumber \\ & + 2 v_\bot v_\|t\bigg[h_{xy}(t_r ) -h_{xy}(t_r+t/2)\bigg ] \nonumber \\ & - v_\bot v_\|\left(\int_{t_r}^{t_r+t/2 } h_{xy}(t)\>dt - \int_{t_r+t/2}^{t_r+t } h_{xy}(t)\ > dt \right ) , \label{actiongeofinal}\end{aligned}\ ] ] where the tilde denotes the fact that we have dropped the @xmath78 term from the action . we get the action @xmath79 $ ] for the atom if it took the lower path through the interferometer by taking @xmath60 in @xmath80 $ ] . the first two terms in eq . @xmath81 do not change , while the last two terms flip sign . thus , from eq . @xmath82 , the phase shift of the atom is @xmath83 \nonumber \\ & + \frac{1}{t^2 } \left(\int_{t_r}^{t_r+t/2 } h_{xy}(t)\>dt - \int_{t_r+t/2}^{t_r+t } h_{xy}(t)\>dt \right)\bigg\}. \label{phasegeo}\end{aligned}\ ] ] after taking the fourier transform , @xmath84 the calculated phase shift for fig . @xmath6 using tt coordinates simplifies to @xmath85 ^ 2 - 4e^{i\omega t/4 } \hbox{sinc}\,(\omega t/4 ) , \label{phasegeores}\end{aligned}\ ] ] where @xmath86 . the first result of our comparison of phase - shift calculations for the interferometer shown in fig . @xmath6 is given in eq . the derivation of these equations is based on the geodesic equation of motion using the tt coordinates . we will find that the phase shift calculated in all three approaches in this paper will have this same overall form , and will differ only in the resonance function @xmath88 . reference frames centered on the worldline of an observer , such as the proper reference frame and general laboratory frame , are less commonly used than the tt coordinates used in the previous subsection . we therefore begin the calculation of the phase shift for fig . @xmath6 in the proper reference frame with a brief review of this frame . how the proper reference frame is constructed is described as follows : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` consider an observer ( freely falling or accelerated , it does nt matter so long as the acceleration is slowly varying ) . let the observer carry with herself a small cartesian latticework of measuring rods and synchronized clocks ( a ` proper reference frame ' in the sense of sec . 13.6 of mtw , with spatial coordinates @xmath89 that measure proper distance along orthogonal axes ) . '' ( from page 339 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in this frame the coordinate @xmath90 is the position of a nonrelativistic test particle travelling along its worldline _ relative _ to the worldline of the observer at her propertime @xmath91 . the magnitude of @xmath90 is then the distance between the worldline of the test particle , and the worldline of the observer at the same instant of time @xmath91 . this frame is only valid when the test particle is moving nonrelativistically , and the size of the apparatus is small compared to the wavelength of the gravitational wave . the velocity @xmath92 is then the velocity of the test particle measured _ relative _ to the observer . in the absence of all other forces , both the test particle and the observer travel along geodesics in the spacetime , and thus their motion are each , separately , governed by geodesic eoms . by subtracting the geodesic eom of the observer from that of the test particle , and then expanding the resultant to first order in @xmath90 , the motion of the test particle _ as seen by the observer _ is governed by @xmath93 this is the geodesic _ deviation _ eom , and it holds as long as the local riemann curvature tensor of the spacetime @xmath94 varies so slowly with @xmath95 that it can be approximated as a constant . this condition holds in the long - wavelength limit of the gravitational wave . as observed in @xcite for ligo ( the _ _ l__aser _ _ i__nterferometer _ _ g__ravitational - wave _ _ o__bservatory ) , if the characteristic size of the interferometer @xmath96 is much smaller than the reduced wavelength @xmath97 of the gravitational wave the interferometer is contained entirely within the proper reference frame . one does not need to use tt coordinates . like ligo , the characteristic size of the interferometers considered in @xcite is also smaller than the reduced wavelength of the gravitational wave @xcite , and thus the use of the proper reference frame was also valid . we shall assume that this is true for the interferometer shown in fig . @xmath6 as well . we will also follow thorne @xcite , and take @xmath98 . this relation between @xmath94 and @xmath99 holds only when the `` tt gauge '' is taken . because of the naturalness of its construction , it should be expected that the tt gaugewhere the number of independent components of @xmath18 equals the number of degrees of freedom for the gravitational wavecan be taken for the proper reference frame as well as for the tt coordinates , but this has not been shown explicitly in the literature @xcite . as we shall see in * appendix b * , this can be done explicitly in the general laboratory frame . it can thus be done in the proper reference frame since it was shown in @xcite that the general laboratory frame reduces to the proper reference frame in the long - wavelength limit for the gravitational waves . for our calculation of the phase shift in the proper reference frame using the geodesic deviation eom , the observer is fixed on the initial beam splitter , and the motion of all objects , and thus the phase shift , is measured with respect to it ( see fig . @xmath6 ) . in this reference frame , the action for a test particle is ( see @xcite or @xcite for a derivation ) @xmath100 where the subscript `` gd '' stands for `` geodesic deviation '' , and we have dropped the @xmath101 term . as in eq . ( 4 ) of @xcite , we have taken the gravitational wave to be in the tt gauge in eq . the resulting hamiltonian ( see also @xcite ) in this case is @xmath103 to lowest order in @xmath18 . it is straightforward to see from eq . @xmath102 that the eom for the test particle is the geodesic _ deviation _ equation eq . we once again solve perturbatively for the classical trajectories , which are determined now by eq . @xmath104 , and take @xmath105 . _ unlike _ the geodesic eom [ eq . @xmath35 ] , in the geodesic _ deviation _ eom [ eq . @xmath104 ] the acceleration depends on the _ position _ of the test particle , and _ not _ on its velocity . consequently , when a gravitational wave passes through the system , it shifts the position of _ all _ parts of the interferometer _ except _ the initial beam splitter , where @xmath108 due to our choice of origin . we can thus take @xmath109 as an initial condition for @xmath33 once again . in addition , from the force lines drawn in fig . @xmath6 we see by symmetry that it is the @xmath110 polarization that contributes to the phase shift , and not the @xmath111 polarization . for this polarization , the motion induced on the final beam splitter by the gravitational wave also does not affect the overall phase shift . while the motion of the final beam splitter does not affect the phase shift of the atoms , the motion of the mirrors in response to the gravitational wave will . this motion changes the boundary condition for the atom at the mirrors from the simple expression eq . @xmath47 . let @xmath112 denote the position of the mirror placed in the upper path of the interferometer , and let @xmath113 denote its velocity . in absence of the gravitational wave the mirrors do not move , and we can take @xmath114 and @xmath115 , where @xmath116 are fluctuations in the positions of the mirror due to the gravitational wave . because of these fluctuations , the boundary condition eq . @xmath117 for @xmath118 at the mirror now becomes @xmath119 where @xmath120 . to determine the response @xmath121 of the mirror to the gravitational wave , we follow @xcite , and model the mirror , and its connection to the frame of the interferometer , as a spring with a quality factor @xmath122 and resonance frequency @xmath123 ( see also section 37.6 of @xcite ) this frequency depends on the size of the interferometer as well as its material properties . from the equivalence principle , the motion @xmath124 of the mirror , like that of the atom , is also a solution of eq . @xmath104 , but because the mirrors are connected to damped springs , @xmath125 the action for the atom along the upper path through the interferometer in the proper reference frame is @xmath126;\gamma_{+1 } ) & \equiv & \frac{1}{2}m\int_{t_r}^{t_r+t } \left\{[\vec{v}_0(t)+\vec{v}_1(t)]^2 - v_0^ix_0^j\dot{h}_{ij}\right\}\>dt \nonumber \\ & \approx & \frac{1}{2}m\int_{t_r}^{t_r+t } \vec{v}_0 ^ 2\ > dt + m\int_{t_r}^{t_r+t } \left\{v_{1i } - \frac{1}{2}x_0^j\dot{h}_{ij}\right\}\frac{dx_{0}^i}{dt}\>dt , \label{actiongeodev}\end{aligned}\ ] ] where once again `` cl '' stands for `` classical '' . the @xmath71 term is once again the same for both paths through the interferometer , and can be neglected . performing an integration by parts , and using eq . @xmath104 , we get @xmath127 = & x_0^i(t)\left(v_{1i}(t)-\frac{1}{2 } x_0^j(t ) \dot{h}_{ij}(t)\right ) \big\vert_{t_r}^{t_r+t/2 } \nonumber \\ & + x_0^i(t)\left(v_{1i}(t)-\frac{1}{2 } x_0^j(t ) \dot{h}_{ij}(t)\right)\big\vert_{t_r+t/2}^{t_r+t } \nonumber \\ & + \frac{1}{2}\int_{t_r}^{t_r+t/2}x_0^iv_0^j \dot{h}_{ij}\>dt+ \frac{1}{2}\int_{t_r+t}^{t_r+t } x_0^iv_0^j \dot{h}_{ij}\>dt . \label{actiongeodev2}\end{aligned}\ ] ] we integrate by parts once again , and then use eqs . @xmath58 , and @xmath59 to get @xmath127 = & l_\|\left\ { v_{1x}(t_r+t ) -\frac{1}{2 } l_\| \dot{h}_{xx}(t_r+t)\right\ } \nonumber \\ & + \frac{1}{2}l_\bot\left\{v_{1y}(t_r+t^{-}/2)-v_{1y}(t_r+t^{+}/2)\right\ } \nonumber \\ & + \frac{1}{2}x_0^i(t ) v_0^j(t ) h_{ij}(t)\bigg\vert_{t_r}^{t_r+t/2 } + \frac{1}{2}x_0^i(t ) v_0^j(t ) h_{ij}(t)\bigg\vert_{t_r+t/2}^{t_r+t } \nonumber \\ & - \frac{1}{2}\int_{t_r}^{t_r+t}v_0^iv_0^j \dot{h}_{ij}\>dt . \label{actiongeodev3}\end{aligned}\ ] ] obtaining @xmath128 $ ] the usual way from @xmath129 $ ] , we arrive at @xmath130 + l_\bot\left[v_{1y}^{(+1)}(t_r+t^{-}/2 ) + v_{1y}^{(-1)}(t_r+t^{-}/2 ) \right ] \nonumber \\ & - \frac{1}{2}l_\bot\left[v_{1y}^{(+1)}(t_r+t/2 ) + v_{1y}^{(-1)}(t_r+t/2 ) \right ] \nonumber \\ & - 2v_\| v_\bot \left(\int_{t_r}^{t_r+t/2 } h_{xy}(t)\>dt - \int_{t_r+t/2}^{t_r+t } h_{xy}(t)\ > dt \right)\bigg\ } , \label{gdphase1}\end{aligned}\ ] ] where we have used the boundary condition eq . the superscripts @xmath132 denote whether it is the velocity of the atom or mirror along the top [ represented by the superscript ( @xmath133 ) ] or the bottom [ represented by the superscript @xmath134 path through the interferometer . we note that to lowest order in @xmath18 eq . @xmath104 becomes @xmath135 we then take the component of eq . @xmath136 along the @xmath44 direction for an atom travelling along the lower path , and subtract it from the same equation for the atom travelling along the upper path . since @xmath137 , while @xmath138 , we integrate and find @xmath139 similarly , if we take the sum of these equations along the @xmath45 direction , we get @xmath140 consequently , @xmath141 \nonumber \\ & - \frac{1}{4}\dot{h}_{xy}(t_r+t/2)+ \frac{h_{xy}(t_r+t/2 ) - h_{xy}(t_r)}{t } \nonumber \\ & + \frac{1}{t^2 } \left(\int_{t_r}^{t_r+t/2 } h_{xy}(t)\>dt - \int_{t_r+t/2}^{t_r+t } h_{xy}(t)\>dt\right)\bigg\}. \label{gdphase2}\end{aligned}\ ] ] using eq . @xmath142 and the steady - state solution for eq . @xmath143 , then @xmath144 and we find that @xmath145 ^ 2 - \frac{1}{2 } \frac{\omega^2}{\omega^2-\omega_0 ^ 2 + i\omega\omega_0/q}. \label{f - gd}\end{aligned}\ ] ] the second result of our comparison of phase - shift calculations for the interferometer shown in fig . @xmath6 is given in eq . @xmath146 . the derivation of these equations is based on the geodesic deviation equation of motion using the proper reference frame . the phase has precisely the same form as eq . @xmath87 ; only the resonance function @xmath147 has changed . .a comparison between the geodesic eom and the geodesic deviation eom descriptions of test - particle dynamics , and the phase shifts calculated for fig . @xmath6 using the two eoms . [ cols="^,^,^",options="header " , ] a summary of the geodesic- and geodesic - deviation - based eom approaches to the description of the motion of test particles in the presence of a gravitational wave , along with the resonance functions for the phase shifts calculated with both approaches , is listed in table @xmath148 . it is readily apparent from the form of @xmath88 and @xmath147 that the corresponding phase shifts @xmath149 and @xmath150 _ do not agree with each other _ even if the lorentzian term in @xmath147 is neglected . moreover , as we will see in the next section , both phase shifts are different from the phase shift calculated previously in the literature as well . there are thus _ three _ different expressions for the phase shift for the _ same _ matter - wave interferometer . to determine the correct expression , we consider the low - frequency limit of these expressions , and compare their limiting forms to the phase shift of caused by a static riemann curvature tensor . the motion of test particles in _ stationary _ spacetimes has been well studied experimentally both directly , through measurements of the advance of the perihelion of mercury , the deflection of light , and the gravitational redshift @xcite in the weak gravity limit , and indirectly , through the slow - down of the taylor - hulse pulsar @xcite in the strong gravity limit . it has also been well studied theoretically through analyses of various black hole geometries @xcite . the motion of test particles in _ nonstationary _ spacetimes , such as when a gravitational wave is present , has not been established to the same level of certainty . on an experimental level , until ligo has detected a gravitational wave , there will have been no _ direct _ measurements of the motion of test particles in nonstationary spacetimes . on a theoretical level , while it would _ seem _ that the motion of test particles in nonstationary spacetimes has been well studied , a recent analysis @xcite based on the general laboratory frame has detailed differences in the dynamics of test particles in nonstationary versus stationary spacetimes . there is , however , is no disagreementon either on a theoretical or an experimental levelover the motion of text particles in _ stationary _ spacetimes . indeed , it was explicitly demonstrated that in this case the general laboratory frame @xcite ( and thus the proper reference frame ) goes over to the standard description of the motion of test particles in stationary spacetimes using the geodesic eom . the phase shift @xmath151 of a matter - wave interferometer due to a _ static _ curvature is well known , and can be written as @xmath152 , where @xmath153 is the area of the interferometer , @xmath56 is the transit time of the atom through the interferometer , and @xmath154 are the components of the static riemann curvature tensor in the plane of the interferometer ( see @xcite , eq . @xmath155 of @xcite , and eq . ( 2.23 ) of @xcite ) . while the calculation of the calculation of @xmath151 in @xcite and @xcite was based on general relativity , @xmath151 can be derived using only newtonian gravitywhere the static curvature is formed from gradients of @xmath5as was done in @xcite . newtonian gravity has been experimentally verified , of course , and the effect of @xmath5 on the phase shift of massive particles in matter - wave interferometers is well known for neutrons @xcite , and for atoms @xcite . the effect of the static riemann curvature of the earth has also been seen in measurements of the _ gradients _ in @xmath5 using differential measurements of the phase shift caused by @xmath5 at differing heights @xcite . while indirect , these all serve as experimental indications that @xmath151 holds . indeed , there are indications that the phase shift induced by the static riemann curvature of the earth has already been seen @xcite . consider now the limit where the period of a gravitational wave is very long compared to the transit time of the atom through the interferometer . the gravitational wave does not oscillate appreciably in the time it takes for the atom to traverse the interferometer , and to the atom , the curvature of the spacetime caused by the gravitational wave is effectively a static one . correspondingly , the phase shift induced by the gravitational wave should have the same form as the phase shift induced by a static riemann curvature . this correspondence between the gravitational wave - induced , and the static - curvature - induced phase shifts serves as a check of our results . when it is combined with the experimental results @xcite , we conclude that the _ correct _ phase shift must have a low - frequency limit that has the same _ form _ as @xmath151 . unlike @xmath88 , which depends only on the transit time @xmath56 of the atoms through the interferometer , and the period @xmath156 of the gravitational wave , @xmath147 depends on the period @xmath157 of the resonance frequency of the mirror - interferometer assembly as well . both @xmath56 and @xmath158 are fixed by the design of the interferometer , while @xmath159 can , in principle , take a broad range of values . we note , however , that the resonance frequency of the interferometer can be estimated as @xmath160 , where @xmath161 is the speed of sound of the material used to construct the interferometer . in almost all cases , @xmath162 , and since @xmath163 , @xmath164 . if the transit time of the atom is much shorter than the period of the gravitational wave so that @xmath165 , then @xmath166 as well . in this low - frequency limit , the gravitational wave will not oscillate in the time it takes the atom to traverse the interferometer , nor will the mirrors oscillate appreciably during this time . this , then , corresponds to the static - limit conditions considered above . from eq . @xmath87 , we see that in this low - frequency limit where @xmath167 , @xmath168 . then @xmath169 since @xmath170 . thus , @xmath171 is proportional to the fourier transform of the _ christoffel symbols _ for the gravitational wave in the low frequency limit , and _ not _ the riemann curvature tensor . it thus has an incorrect limiting form . if we now take the low - frequency limit of eq . @xmath146 , we find that @xmath172 to lowest order , so that @xmath173 since @xmath174 . thus , unlike @xmath171 , @xmath175 _ is _ proportional to the riemann curvature tensor , and _ has the correct limiting form_. the differences in the low - frequency behavior of @xmath149 and @xmath150 are shown fig . @xmath176 , where we have graphed @xmath177 versus @xmath159 for both expressions of the phase shift . for completeness , we have also included in this comparison the phase shift @xmath178 for the matter - wave interferometer fig . @xmath6 calculated using the linet - tourenc - stodolsky formula @xcite . as we mentioned in the introduction , a third calculation of the phase shift of fig . @xmath6 based also on the geodesic eom is necessary , and in the next section this phase shift @xmath178 is explicitly calculated for our interferometer using the liner - tourrenc - stodolsky formula . for now , we note that the graphs in fig . @xmath176 were plotted using the parameters of the `` earth - base horizontal migo '' described @xcite , which has the same configuration as the interferometer we consider here in fig . this interferometer has a @xmath179 m , @xmath180 km , @xmath181 s , @xmath182 , @xmath183 rad / s , and used @xmath184li with @xmath185 gm . the plots in fig . @xmath176 cover the same frequency range considered in @xcite . notice that for frequencies much less than @xmath186 rad / s , @xmath187 approaches a constant , indicating that @xmath188 for low frequencies , as expected , while both @xmath189 , and @xmath190 do not . the behavior of all three phase shifts also differ dramatically at high frequencies . we will consider the high - frequency behavior of @xmath150 in * appendix c*. versus @xmath159 for @xmath150 , @xmath149 , and @xmath178 , for the interferometer fig . @xmath6 using @xmath179 m , @xmath180 km , @xmath181 s , @xmath182 , @xmath191 rad / s , and used @xmath184li with @xmath192 gm . , scaledwidth=100.0% ] we now compare results of the above calculations with those done previously in the literature . calculations of the phase shift by linet and tourrenc @xcite , and stodolsky @xcite were both done at the quantum - mechanical level , but unlike us , they presented a completely general analysis , and did not take the nonrelativistic limit for the interfering particles they considered . their expressions for the phase shift were thus manifestly covariant . linet and tourrenc did not explicitly define the coordinate system that they used , while stodolsky only referred to the presentation by landau and lifschitz @xcite on synchronous reference frames . it is clear that since a gravitational wave in the tt gauge has @xmath17 , the tt coordinates is a synchronous reference frame as well . since both calculations were ultimately based on the action for the _ geodesic _ eom , they were implicitly done in the tt coordinates . linet and tourrenc used the wkb approximation in the quasi - classical limit , and replaced the mass - shall constraint @xmath193 ( note that the signature of their metric is different from ours ) with a differential equation in the phase of the wavefunction for the test particle by making the substitution @xmath194 . stodolsky worked directly with the action , and used the quasi - classical approximation we followed here to calculate the phase shift . the expression for the phase shift derived by both nevertheless have the same overall form . the phase shift ( given in eq . [ 3.2.3 ] of @xcite ) derived by linet and tourrenc does , however , differ from the original phase shift ( given in eq . ( 2.4 ) of @xcite ) derived by stodolsky by a factor of 2 . since stodolsky uses the linet and tourrenc phase shift expression ( the equation above eq . ( 6.1 ) of @xcite ) to calculate the phase shift for a model interferometer , we will do so also . in the nonrelativistic limit of the atoms that we are interested in , this phase shift reduces to @xmath195 which is different from @xmath149 , even though both were calculated in tt coordinates . to understand the origin of the difference between @xmath178 and @xmath149 , let us write @xmath196 , where @xmath197 and the superscript `` kin '' stands for `` kinetic '' , while the superscript `` pot''stands for `` potential '' . clearly , @xmath198 . since @xmath199 , contributions to the total phase shift @xmath200 from changes to the kinetic energy of the test particle as it traverses the interferometer was not taken into account in eq . this kinetic term can not be neglected , however . the passage of the gravitational wave through the system will cause shifts in the velocity of the atom as it traverses the interferometer , and thus to its kinetic energy . in fact , by using eqs . @xmath75 and @xmath76 , it is straightforward to show that @xmath202 , where @xmath203 are boundary terms for the atoms at the mirrors . these terms do not vanish , but instead contribute the @xmath204 term to @xmath88 . if we nonetheless use @xmath178 to calculate the phase shift for fig . @xmath6 , we once again find that @xmath178 has the same form as eqs . @xmath87 and @xmath146 , @xmath205 ^ 2 . \label{f - stod}\end{aligned}\ ] ] the final result of our comparison of phase - shift calculations for the interferometer shown in fig . @xmath6 is given in eq . @xmath206 . while these equations , like eq . @xmath87 , are also based on the geodesic equation of motion using the tt coordinates , @xmath207 nonetheless differs from @xmath88 by the term @xmath208 . this term comes directly from the boundary terms neglected in eq . ( when expressed in terms of resonance functions , eq . @xmath206 agrees , after the nonrelativistic limit is taken , with the phase shift calculated in @xcite up to overall factors . ) taking the low - frequency limit , we find that @xmath209 . thus , like @xmath149 , @xmath178 has the incorrect low - frequency limiting form . it also has a _ high_-frequency behavior that is dramatically different from either @xmath88 or @xmath147 , as can be seen in fig . @xmath176 . in this limit , while @xmath210 and @xmath211 ( see * appendix c * ) , @xmath212 , and dies off much more rapidly . as pointed out in @xcite , this rapid die off is due to three effects . firstly , the @xmath178 does not include the boundary terms at the mirror . these terms depend on the value of the velocity of the atom at the surface of the mirror , and the length @xmath213 , and they mitigate this rapid decrease for @xmath149 . secondly , @xmath178 depends on @xmath214 , and is constant as the atom travels through the interferometer , while @xmath150 depends on the product @xmath215 , and increases linearly with the distance the atom travels from the initial beam splitter . thirdly , since @xmath178 depends only on @xmath216 , discontinuous jumps in the velocity due to the hard wall mirrors were not taken into account in @xmath178 . from * appendix c * , these hard - wall jumps contribute directly to the high - frequency behavior of @xmath150 . calculations of the phase shift at the quantum - field - therapy level were first done by cai and papini @xcite . their starting point is the covariant version of the ginzburg - landau equation , but for a charged particle . this is a relativistic klein - gordon equation for a bosonic field @xmath217 that interacts with itself through a @xmath218 interaction ( eq . ( 2.1 ) of @xcite ) . by turning off the @xmath218 interaction , and setting the charge of the particle to zero , one recovers linet and tourrenc s starting point . if one extracts the single - particle hamiltonian from this equation , and take the correspondence principle limit of the eom , one recovers the geodesic eom used by stodolsky . thus , like linet and tourrenc , and stodolsky , cai and papini work in tt coordinates . by acting on @xmath217 with a phase operator @xmath219 , cai and papini was able to solve for a general expression for the phase shift due to gravitational effects in the weak - gravity limit through a stationary - phase type of analysis . this operator has the form [ eq . ( 3.1 ) of @xcite ] , @xmath220 ( using our notation ) for a closed path where @xmath221 is the generator of lorentz transformations , and @xmath222 is the generator of translations . we have dropped the vector potential term since the atoms we consider in this paper are not charged . as they mentioned , using stokes theorem the first term in @xmath219 can be written as @xmath223 in the weak - gravity limit ; the integral is over the timeline surface @xmath224 bounded by @xmath225 , and @xmath226 is a 2-form on @xmath224 . this expression has been explicitly verified in @xcite for the phase shift caused by riemann curvature tensor for a _ stationary _ spacetime . the second term agrees with stodolsky s expression for the phase shift . indeed , cai and papini uses eq . @xmath227 to calculate the phase shift of a model interferometer due to the passage of a gravitational wave [ eq . ( 3.15 ) of @xcite ] , and they found that up to terms of order @xmath228 , the following holds : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` the @xmath229 terms in ( 3.15 ) agrees with stodolsky s result [ 11 ] when a misprint in the latter is corrected by inserting an overall factor @xmath230 . '' ( from page 415 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the reference [ 11 ] cited above is the work by stodolsky cited as @xcite in this paper . consequently , their expression for the phase shift , like linet and tourrenc s , and stodolsky s , also has the incorrect low - frequency limiting form . the effects of the mirrors on the test particle s phase was not considered by them either . bord and coworkers @xcite considered the use of fermionic atoms in matter - wave interferometers to measure various gravitational effects , including the detection of gravitational waves . their starting point was the dirac equation for a two - level , spin 1/2 atom expressed in curved spacetimes using the tetrad formalism . similar to cai and papini , theirs was a hamiltonian - based approach , and they showed that the dominant contribution by the gravitational wave to the hamiltonian is from a term with the form @xmath231 , where @xmath19 is the mass of the atom , and @xmath232 its momentum [ see eq . ( 10 ) of @xcite ] . the other terms in the hamiltonian due to the gravitational wave are from very small spin - gravitation couplings . these terms can be neglected here since we are dealing with atoms without spin . when this is done , the hamiltonian they calculated has the same form as eq . @xmath233 for the geodesic eom , and is manifestly different from the hamiltonian eq . @xmath234 for the geodesic _ deviation _ eom . thus they , too , worked in the tt coordinates . although an expression for the phase shift was not calculated in @xcite , they concluded the following : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` the phase shift , that one can calculate from the first terms of eq . ( 10 ) , increases with the velocity of the atoms in contrast with what happens in the inertial field case and a gravitational wave detector using atomic interferometry should therefore use relativistic atoms . in this case , it can be shown directly from eq . ( 5 ) that the phase shift is proportional to @xmath235 where @xmath236 is the atomic de broglie wavelength , result which is similar to the optical case . '' ( from page 161 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in a subsequent paper @xcite , bord gave an explicit expression for the phase shift [ see eq . ( 98 ) of @xcite ] . it has , for its spin 0 component , the same expression , @xmath178 , for its phase shift as stodolsky s . thus , like stodolsky s approach , any phase shift calculated using this analysis would have the incorrect low - frequency limiting form as well . finally , all of the analyses that we outlined in this paper were done in the weak - gravity limit . in @xcite , a derivation of the phase shift of a quantum mechanical test particle was done without making this approximation . their analysis was also based on the geodesic - eom approach , and they arrived same expression for the phase shift that stodolsky does for spin 1/2 test particle with no approximations , while it holds only the first order in @xmath237 for spin 0 and spin 1 particles . consequently , phase shifts calculated with their expression will also have the same incorrect low - frequency limiting form as stodolsky s . there has been a preference among those who study matter - wave interferometry and gravitational waves to use tt coordinates to analyze the phase shift of an interferometer induced by gravitational waves instead of the proper reference frame . the expectation is that irrespective of which reference frame is used to calculate the phase shift , the final expression will be the same @xcite . we have shown in this paper that for matter - wave interferometry _ this expectation is wrong_. by presenting a calculation of the phase shift using the tt coordinates following the line of the analysis found in the literature @xcite , and also a calculation of the phase shift using the proper reference frame following the analysis found in @xcite , we have shown that the expressions for the phase shift obtained using the tt - coordinates _ is _ different from that obtained using the proper reference frame . moreover , we have shown that the commonly accepted expression for the phase shift eq . @xmath201which is the end result of all previous derivations of the phase shift in the literature @xcite__is incomplete _ _ , and gives yet a different expression for the phase shift of fig . this happens even though derivations of this expression were all done in tt coordinates . the focus of this paper has been on demonstrating in detail that differences exist between the various approaches to calculating the phase shift , and showing explicitly that different expressions for the phase shift are obtained . to determine which one , if any , is _ correct _ , we noted that in the limit of sufficiently low frequency gravitational waves , the correct expression for the phase shift _ must _ reduce to the well known expression for the phase shift induced by a static curvature . it must therefore be proportional to components of the riemann curvature tensor . this static phase shift can also be calculated within newtonian gravity , where the curvature is formed from gradients of the acceleration due to gravity @xmath5 . the phase shift due to @xmath5 has been measured @xcite , and recently has been used to measure gradients in @xmath5 as well @xcite . in addition , there are indications that the phase shift induced by the static riemann curvature tensor has already been seen @xcite . we have explicitly shown that when the phase shift is calculated using the proper reference frame and geodesic deviation eom , it has the expected limiting form in the low - frequency limit , and is therefore correct . the other two , both of which used the tt coordinates and the geodesic eom , do not , and are thus incorrect . the implications of this conclusion are immense . in @xcite , migo , the _ _ m__atter - wave _ _ i__nterferometric _ _ g__ravitational - wave _ _ o__bservatory , was proposed . the theoretical sensitivity predicted for migo was based on a calculation of its phase shift using the proper reference frame and the geodesic deviation eom . the conclusion arrived at in these papers was that by using _ atoms , it would be possible to construct gravitational wave observatories that are as sensitive as ligo or lisa ( the _ _ l__aser _ _ i__nterferometer _ _ s__pace _ _ a__ntenna ) , but would be orders of magnitude smaller , and have a much broader frequency response . this conclusion is diametrically opposite from the conclusions arrived previous in the literature that _ fast _ atomsindeed , ultrarelativistic onesshould be used in the matter - wave interferometer instead @xcite . the conclusion that relativistic atoms should be used in the detection of gravitational waves with matter - wave interferometers is the basis on which most of the atomic , molecular and optical physics community judges the suitability of using matter - wave interferometers as gravitational wave detectors . the physical origin of this conclusion by bord and his coworkers is the dependence of the phase shift @xmath178 on the _ square _ of the atom s velocity in the geodesic - eom - based approach ; clearly , the faster the atom , the larger the phase shift will be . even though our analysis of the phase shift was done for nonrelativistic atoms , we note that if the size of the interferometer is held constant , and only the velocity of the atom is increased , then in the high - velocity limit , @xmath238 , and eventually @xmath239 for the frequency of gravitational waves expected from astrophysical sources . the high - velocity limit is thus equivalent to the low - frequency limit for the gravitational waves , and we have shown that the low - frequency limit of @xmath178 is incorrect . thus , bord and coworker s assertion about relativistic atoms being more suitable for the detection of gravitational waves is also incorrect . all previous calculations and estimates of the phase shift in the literature also neglected the effect the mirrors of the interferometer have on the phase shift . indeed , from * appendix c * we see that it is precisely because hard wall mirrors are usedalong with the _ instantaneous _ impulsive force that they exert on the atoms when they reflect offthat the use of slow - velocity atoms leads to a sensitivity that _ increases _ with frequency is obtained . this dependence on frequency is the exact opposite for ligo and lisa @xcite , and is part of the reason that migo can be so much smaller than either one . we have not , in this paper , presented the underlying physical reasons _ why _ the current tt coordinates approach to calculating the phase shift fails , and the proper reference frame succeeds . we shall address this issue in detail later @xcite . here , we only note that one may try to modify the boundary conditions for the atom on the mirrors so that the phase shift calculated for fig . @xmath6 in tt coordinates will have the correct low - frequency limit . doing this is especially tempting in light of the following statement by thorne : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` however , errors are much more likely in the tt analysis than in the proper - reference - frame analysis , because our physical intuition about how experimental apparatus behaves is proper - reference - frame based rather than tt - coordinate based . as an example , we intuitively assume that if a microwave cavity is rigid , its walls will reside at fixed coordinate locations @xmath89 . this remains true in the detector s proper reference frame ( aside from fractional changes of order @xmath240 , which are truly negligible if the detector is small and which the proper - reference - frame analysis ignores ) . but it is not true in tt coordinates ; there the coordinate locations of a rigid wall are disturbed by fractional amounts of order @xmath241 , which are crucial in analyses of microwave - cavity - based gravity wave detectors . '' ( from page 400 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we find that while it may be possible to suitably modify the boundary conditions to bring @xmath149 in line with the low frequency limit , doing so will be _ ad hoc _ , and will not explain the underlying physical reason for the differences between @xmath149 and @xmath150 . what is at the heart of the physics are not questions of which boundary conditions to take and how they should be modified , but rather how measurement of physical quantities are made in general relativity when the background metric varies with time . when doing an experiment to measure some physical property of a test particleits phase shift , saywe start by choosing an origin for our coordinate system . it is natural to choose as this origin some point _ on _ our measuring apparatus , which in our case is the initial beam splitter . in the three spatial dimensions , this origin is represented by a _ point _ on a three - dimension space ; this point is what we use when measurements with a _ physical _ apparatus are actually made . if we wish to represent the origin in space _ and _ time , it would seem to be natural to represent it as an _ event _ in four - dimensional _ spacetime_. coordinate systems constructed by doing so are called quasi - cartesian coordinate systems by synge . he also noted who also noted that a suitable choice of origin in spacetime for is often difficult ( see section 10 of @xcite ) , however . this is especially true in the case of gravitational waves , where the metric is nonstationary , and varies with time . in fact , by identifying the origin as an _ event _ , we have ignored the fact that our measuring apparatus is a _ physical _ object . it too must travel along a _ worldline _ in the spacetime , as will any test particle whose physical properties we are measuring . as was argued in @xcite , only _ relative _ motions of a test particle with respect to the apparatus can thus be measured . these arguments were based in part on the following observation by hawking and ellis : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ `` in chapter 3 we saw that if the metric was static there was a relation between the magnitude of the timelike killing vector and the newtonian potential . one was able to tell whether a body was in a gravitational field by whether , if released from rest , it would accelerate with respect to the static frame defined by the killing vector . however , in general , space - time will not have any killing vectors . thus one will not have any special frame against which to measure acceleration ; the best one can do is to take two bodies close together and measure their _ relative acceleration _ [ emphasis added ] . '' ( from page 78 of @xcite . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this relative motion is not measured in the tt coordinates where the coordinates are defined relative to a fixed _ event _ , chosen as the origin , in the spacetime . the motion of the measuring apparatus through spacetime was not taking into account in the tt coordinates , so that the beam splitters and the mirrors are at _ fixed _ coordinate values . this relative motion , on the other hand , is measured in the proper reference frame where the coordinates are explicitly constructed to span the distance between the measuring apparatus , and the test particle whose properties are to be measured . the motion of the measuring apparatus has been built in from the beginning , and indeed , the positions of the beam splitters and the mirrors shift during the passage of the gravitational wave . failing to take into account the motion of the measuring apparatus in spacetime is the underlying _ physical _ reason why the wrong expression was obtain for fig . @xmath6 when the tt coordinates are used . these statements will be expanded upon , and justified in @xcite . in this * appendix * we set the framework used in our different calculations of the phase shift . like stodolsky @xcite , we use the quasi - classical approximation to calculate the phase shift of an atom in a matter - wave interferometer . this approximation is the lagrangian version of the wkb , or stationary phase approximation , in the schrdinger representation of quantum mechanics . we assume that the density of particles through the interferometer is low enough that they do not interact with one another appreciably in the time it takes for the atom to traverse the interferometer . within this approximation , the particles can be treated as test particles that interact only with the gravitational wave , and not with themselves . we can thus consider the phase shift of each atom individually , and use the feynman path integral method to calculate its phase shift . however , instead of integrating over all possible paths , we make the quasi - classical approximation , and consider only the most _ probable _ pathwhich is also the _ classical _ paththat the test particle takes through the interferometer ( see @xcite ) . as a result , @xmath242 -s^{cl}[x_a , x_b;\gamma_2]\bigg\ } , \label{def - phase - shift}\ ] ] where the superscript `` cl '' stands for `` classical '' , and @xmath243 and @xmath244 are the two possible paths a classical test particle can take through the interferometer that links a common point @xmath245 on the interferometer ( the initial beam splitter , say ) to a common point @xmath246 on the interferometer ( the final beam splitter , say ) . note that @xmath247 $ ] are functionals of the points @xmath245 and @xmath246 , along with the paths @xmath243 and @xmath244 connecting them . unlike @xcite , as well as the calculations of the phase shift done in @xcite , we will work in the nonrelativistic limit . the reason for doing so is two fold . first , the proper reference frame , as well as the general laboratory frame and all the other reference frames based on the worldline of an observer , are constructed for nonrelativistic test particles . second , as is well known , the phase shift of a matter - wave interferometer is proportional to the mass of the particle . this argues for the use of atoms in matter - wave interferometers to detect gravitational waves , as was proposed in @xcite , and in any conceivable matter - wave interferometer , these atoms will be moving nonrelativistically . the configuration of the interferometer that we use in our calculations is shown in fig . details of this interferometer can be found in @xcite . for our purposes , it is sufficient to note that the interfering test particle in the interferometer is an atom with mass @xmath19 . a beam of these atoms is emitted from an atomic source with velocity @xmath248 , and is beam - split into different diffraction orders using a diffraction grating with periodicity @xmath249 ( see @xcite ) . only the @xmath38 and @xmath57 diffraction orders are used in the interferometer . there are two possible paths , denoted as @xmath38 and @xmath57 in fig . @xmath6 , that any individual atom can follow through the interferometer ; it is not possible to tell _ a priori _ which of the two paths an atom will take . if the atom is diffracted into the @xmath38 diffraction order , then immediately after the beam splitter it will have a velocity of @xmath250 perpendicular to the longitudinal axis of the interferometer . similarly , if an atom is diffracted into the @xmath57 diffraction order , it will have a velocity of @xmath251 ( see fig . . since the diffraction grating does not impart energy to the atom , for both diffraction orders @xmath252 is the velocity of the atom parallel to the axis of the interferometer immediately after it leaves the beam splitter . in @xcite the use of hard - wall , specular reflection mirrors was considered for the matter - interferometer . mirrors of this type have been made from silicon wafers @xcite , and for near normal and near glancing angles , the reflection from these mirrors is close to 100% . these are not the mirrors that are currently used in matter - wave interferometry , however . mirrors based on diffraction gratings , standing waves of light , or stimulated raman pulses are used instead @xcite . we will nevertheless consider only the case of hard - wall mirrors in this paper , because of their simplicity . in the absence of gravitational waves , the effect of hard - wall mirrors , if they do not move , on the velocity of the atoms through the interferometer is @xmath253 where @xmath254 and @xmath255 are , respectively , the velocity of the atom immediately before it is reflected off the mirror , and immediately afterwards . in this appendix we will show how the tt gauge can be taken in the general laboratory frame in a treatment that follows @xcite . like the proper reference frame , the general laboratory frame is an orthonormal coordinate system fixed on the worldline of an observer . in our case it is fixed on the worldline of the initial beam splitter in fig . we choose a local tetrad @xmath256 fixed on the observer s worldline . thus , @xmath257 , while @xmath258 . in a slight break in the notation followed elsewhere in this paper , which follows the notational convention established in @xcite , we use in this * appendix * the convention taken in @xcite . thus , we use as the tetrad indices capital latin letters , which run from 0 to 3 . lower - case latin letters are used as the spatial tetrad indices , and they run from 1 to 3 . as usual , greek letters are used for the spacetime indices of a general coordinate @xmath259 . consider a the motion of a test particle @xmath260 close to the observer @xmath261 and her worldline . let @xmath262 be the position of @xmath263 at any proper time @xmath264 in the observer s frame . @xmath265 can also be considered as a coordinate transformation from the general coordinates @xmath259 to the tetrad frame at any time @xmath264 . thus in a small neighborhood @xmath266 of @xmath267 , @xmath268 so that @xmath269 integrating eq . @xmath270 , we find for the spatial coordinates @xmath271 where @xmath272 is a 1-form , and @xmath273 is a space - like geodesic linking @xmath261 to @xmath260 at a given proper time @xmath264 of the observer . next , after taking appropriate account of causality , the time component is @xmath274 where @xmath275 is a null - vector linking @xmath261 at time @xmath276 with @xmath260 at time @xmath264 ; the difference @xmath277 is the transit time of a light pulse emitted by @xmath261 , and scattered by @xmath260 . in the case of linearized gravity , @xmath278 . it is straightforward to see that in terms of tetrads @xmath279 this choice is _ not _ unique . if one does a _ local _ lorentz transformation @xmath280 on @xmath281 , then @xmath282 while @xmath283 . this leaves @xmath284 invariant . the usual derivation of the tt gauge makes use of a remnant of general coordinate transformation invariance still left in the linearized - gravity limit of general relativity . if @xmath285 where @xmath286 is a ` small ' , arbitrary vector , then @xmath0 transforms to @xmath287 one can choose @xmath286 such that @xmath288 can be chosen for gws in minkowski space . doing so defines a specific coordinate system , which is called the tt coordinates in @xcite . the tt gauge can also be implemented in the general laboratory frame as well . in terms of a tetrad frame , the linearized coordinate transformation eq . @xmath289 induces a lorentz transformation on the tetrads such that @xmath290 , where @xmath291 and @xmath292 . like the tt coordinates , the tt gauge can also be realized , but now through the appropriate lorentz transformation . from eq . @xmath270 , this in turn means we can choose the tt gauge in the construction of the general laboratory frame as well . in this * appendix * we consider the high - frequency limit of @xmath150 . because of its dependence on three timescales , the analysis of the high - frequency limit of @xmath150 is more involved than that for the low frequency limit . there are two cases to consider . the simplest case is when the transit time of the atom through the interferometer is very _ long _ in comparison with the period of the gravitational wave , so that @xmath293 , but @xmath166 still . to the atom , the mirrors still do not move in response to the gravitational wave , but the atom will experience many oscillations of the gravitational wave as it traverses the interferometer . in this limit the phase of the atom oscillates so fast that we would naively expect the phase shift to average to zero . it does not do so , however . we instead find that @xmath211 . that @xmath150 does not vanish in the limit of high - frequency gravitational waves is due to action of the mirrors on the atoms as they traverse through the interferometer . when the atoms reflect of the mirrors , they impart a impulsive force on the atoms that is effectively _ instantaneous _ .. since this is much shorter than any of the timescale considered here , the force can be considered as instantaneous . ] the effect of this force is readily apparent from the discontinuity in @xmath294 across the mirror in eqs . @xmath47 and @xmath117 . note in particular that the force is proportional to perturbed velocity @xmath295 of the atom _ caused by the gravitational wave _ , and is _ different _ depending on which path the atom travels through the interferometer , as can be seen from the force lines in fig . it is because of this net instantaneous force that @xmath150 does not average to zero in the high - frequency limit . when , @xmath296 as well , the mirrors themselves oscillate rapidly as the atom traverses the interferometer . this rapid motion of the mirror is not correlated with the motion of the atoms through the interferometer , however . when the atom reaches the mirror , roughly half the time it will be moving in the same direction as the mirror , which would , from eq . @xmath131 , tend to _ decelerate _ the atom . the other half of the time it will be moving in the opposite direction as the mirror , and would tend to _ accelerate _ the atom . not surprising , in this limit the resonance function decreases by a half , and @xmath297 . in this high - frequency limit @xmath150 is _ not _ proportional to @xmath298 . there is no reason to expect it to be , as there was in the low - frequency limit , however . because the mirrors exert a net forcewhich itself is the result of the passage of the gravitational wave through the systemon the atom as it passes through the interferometer , @xmath299 . m. j. snadden , j. m. mcguirk , p. bouyer , k. g. haritos , and m. a. kasevich , phys . lett . * 81 * , 971 ( 1998 ) ; j. m. mcguirk , g. t. foster , j. b. fixler , m. j. snadden , and m. a. kasevich , phys . rev . * a65 * , 033608 ( 2002 ) .

it is well known that anti - de sitter ( ads ) gravity coupled to a scalar field with mass at or slightly above the breitenlohner - freedman bound @xcite admits a large class of boundary conditions , defined by an essentially arbitrary real function @xmath0 . for all ( regular ) @xmath0 , the conserved charges are well defined and finite @xcite , despite the fact that the scalar field falls off slower such that @xmath5 corresponds to turning off the independent subleading term in the asymptotic expansion of the scalar field . the usual boundary conditions instead keep this term and turn off the leading term . ] than usual . theories of this type have been called designer gravity theories @xcite , because their dynamical properties depend significantly on the choice of @xmath0 ( see e.g. @xcite ) . in supergravity theories with a dual conformal field theory ( cft ) description , the ads / cft duality @xcite relates @xmath0 to a potential term @xmath6 in the cft action , where @xmath7 is the field theory operator that is dual to the bulk scalar for @xmath5 boundary conditions @xcite . this led @xcite to conjecture that ( a ) there is a lower bound on the gravitational energy in those designer gravity theories where @xmath0 is bounded from below , and that ( b ) the solutions locally minimizing the energy are given by the spherically symmetric , static soliton configurations found in @xcite . more recently , the stability of designer gravity theories has been studied using purely gravitational arguments . in particular , a lower bound on the conserved energy in terms of the global minimum of @xmath0 was rigorously proven within a specific ads gravity theory by relating the hamiltonian charges to spinor charges @xcite . arguments were given in @xcite suggesting that these bounds hold more generally . however , it was subsequently discovered @xcite that solutions with arbitrarily negative energy can be constructed numerically for certain theories with @xmath8 . this raised a puzzle , which we resolve in this paper . our resolution focuses on an auxiliary construct , the superpotential " @xmath9 built from the bulk scalar potential @xmath10 . this superpotential is an important ingredient in constructing the spinor charge . in @xcite , it was shown that any @xmath10 satisfying the breitenlohner - freedman bound _ perturbatively _ admits an appropriate superpotential @xmath9 . it was of course recognized that the global existence of @xmath9 was required for the proof of an energy bound , and that this global existence may impose constraints on @xmath10 . what is interesting about the counter - examples of @xcite is that they _ do _ admit a globally defined superpotential , but nevertheless violate the proposed bound . the issue turns out to be that superpotentials associated with a given @xmath10 can be of two types , which we call @xmath4 and @xmath3 . a particular @xmath10 may admit ( distinct ) superpotentials of both types , or it may admit only the @xmath4 type . the proof requires existence of a @xmath3-type superpotential , while the examples of @xcite admit only the @xmath4-type can admit both @xmath4 and @xmath3 superpotentials was used to study asymptotically ads domain wall solutions in @xcite . ] . we verify this claim below and also confirm various other details of the arguments of @xcite . after briefly listing our conventions in section [ prelim ] , we review the proof of the lower bound in section [ two ] , illustrating why a @xmath3-type superpotential is essential ( and why a @xmath4-type superpotential is not sufficient ) . section [ solitons ] confirms that the counter - examples of @xcite admit _ only _ @xmath4-type superpotentials and numerically explores the energy bound in a number of examples . we find evidence that the global existence of @xmath3 may be necessary ( as well as sufficient ) for any lower bound to hold . section [ spinors ] investigates other aspects of the proof from @xcite . arguments for positivity of the spinor charge require witten spinors , whose global existence can be difficult to demonstrate . questions about this global existence were raised in @xcite . however , we show that the argument for global existence given in @xcite for a particular @xmath3 extends to the general case . as a check , we also explicitly demonstrate the existence of witten spinors in the context of spherical symmetry by analyzing an associated ordinary differential equation . we evaluate these spinors numerically for a particular example and use the results to check the relation between the spinor and hamiltonian charges derived in @xcite . we close with some discussion in section [ disc ] . our conventions and our definition of asymptotically anti - de sitter spacetimes follow those of @xcite . in particular , we consider gravity theories minimally coupled to a scalar field with lagrangian density given by @xmath11 \ , , \ ] ] where we have set @xmath12 . here the scalar potential @xmath10 is of the form @xmath13 near @xmath14 . for simplicity we assume that @xmath15 is even . the constant @xmath16 is the cosmological constant , given by @xmath17 with @xmath18 a positive length that we may set to one by rescaling the metric and scalar field . the spacetime dimension is denoted @xmath19 , and we assume that @xmath20 . we will furthermore assume that the scalar field is tachyonic ( @xmath21 ) , with mass in the breitenlohner - freedman range @xcite @xmath22 where @xmath23 . the metric of exact ads space of unit radius ( and @xmath14 ) , given by @xmath24 is an exact solution of the theory , where @xmath25 is the unit - radius round metric on @xmath26 . as part of our boundary conditions , we assume that the metric of a general solution asymptotically approaches ( [ pureads ] ) in the manner described in @xcite , and that the scalar field is asymptotically of the form @xmath27 where @xmath28 to obtain a well - defined dynamics for the linearized theory , it is necessary to impose a boundary condition at @xmath29 on the scalar field , i.e. we must impose a relation between @xmath30 and @xmath31 in . for example , one can impose @xmath32 , leaving @xmath31 totally unspecified . we refer to this option as fast fall - off boundary conditions . " alternatively , one may set @xmath33 , leaving @xmath30 unspecified . one may also impose more general boundary conditions of the form @xmath34 where @xmath35 is an arbitrary smooth function . under the ads / cft duality , this function @xmath0 appears as an additional potential term in the action for the dual field theory @xcite . an elegant way to prove energy bounds is witten s spinor method @xcite , which proceeds by constructing a manifestly positive `` spinor charge , '' and then comparing it to the energy of the gravitational solution . witten s argument was originally given in the context of asymptotically flat spacetimes , but it can be generalized to the asymptotically ads situation . when the matter satisfies the dominant energy condition apart from the negative cosmological constant term ( regarded as a contribution @xmath36 to @xmath37 ) , witten s argument may simply be modified by the addition of a term @xmath38 to the covariant derivative which acts on the spinor , where @xmath1 is a constant proportional to @xmath39 . this term is needed to deal with the negative energy associated with the cosmological constant . our interest here is in tachyonic scalars @xmath40 , whose potential energy may in fact be unbounded below and does not satisfy the dominant energy condition . as shown by @xcite for @xmath41 ( based on @xcite , and extended to higher dimensions by @xcite ) , many such settings may be addressed by generalizing the constant @xmath1 to a real `` superpotential '' @xmath9 satisfying @xmath42 where @xmath10 is the scalar potential . taking @xmath43 to be the ads vacuum , we are interested in potentials for which @xmath44 . the value @xmath45 then determines the cosmological constant . with fast fall - off boundary conditions ( @xmath32 ) , the proof @xcite requires only that there be a solution to ( [ vtop ] ) with @xmath46 and @xmath47 . perturbatively , i.e. , in the sense of formal power series , a solution of this form always exists when @xmath48 sets the scalar field mass to satisfy the breitenlohner - freedman bound . in fact , there are two such perturbative solutions : @xmath49 where @xmath50 are given by ( [ roots ] ) . however , the proof of @xcite requires that @xmath9 be well - defined ( and real ) for all @xmath40 , and this imposes further restrictions on @xmath10 . in both the fast @xcite and slow fall - off cases @xcite , the proof proceeds by using @xmath51 to construct a spinor charge @xmath52 . in the fast fall - off case one may show ( e.g. , following the basic method outlined by @xcite ) that @xmath53 , where @xmath54 is the conserved energy . the proof of @xcite is identical no matter which superpotential is used , and it is sufficient that only @xmath4 existboth @xmath4 and @xmath3 exist near @xmath14 . however , a real solution ceases to exist if ( [ vtop ] ) forces @xmath55 to become imaginary ; i.e. , if @xmath56 . since @xmath57 near @xmath14 , and since this in turn implies @xmath58 , one finds @xmath59 for all @xmath60 and global existence of @xmath3 implies global existence of @xmath4 , but not vice versa . ] . however , in order to derive an energy bound with slower fall - off conditions , @xcite assumed the existence of @xmath3 . we verify in subsection [ critical ] below that this choice is critical in this context , and that @xmath4 alone does not lead to an energy bound . this turns out to resolve the issue raised in @xcite . a simple example is discussed in section [ simple ] . we now quickly repeat the derivation of the energy bound from @xcite , examining both the original argument ( using @xmath61 ) and an analogous argument based on @xmath62 . for either superpotential , the spinor charge is defined as @xmath63 where the integrand is the hodge dual of a suitably defined nester two - form @xcite @xmath64 } \widehat{\nabla}^e \psi + \textrm{h.c.}\ , , \ ] ] and @xmath65 is a surface at spatial infinity that bounds a spacelike surface @xmath66 . in ( [ nester ] ) , @xmath67 is a dirac spinor and the covariant derivative is defined in terms of @xmath1 ( @xmath68 ) as @xmath69 we require that the spinor field @xmath67 approaches a covariantly constant spinor ( i.e. a killing spinor ) of pure ads , @xmath70 , at infinity . we furthermore assume that asymptotically @xmath71 . using gauss s theorem we can rewrite the spinor charge @xmath72 ( @xmath73 ) as @xmath74 where @xmath75 is the integration element on @xmath66 , and @xmath76 the unit normal . letting @xmath77 denote directions in the tangent space of the surface @xmath66 , one can then show @xcite that the integrand of ( [ q2 ] ) is @xmath78 \ , , \ ] ] where @xmath79 the first and third terms in ( [ divb ] ) are manifestly nonnegative as written . a negative contribution from the second term can be avoided by imposing the witten condition @xcite @xmath80 which is essentially the spatial dirac equation . in section [ spinors ] , we recall the argument @xcite that globally smooth spinors satisfying with the above boundary conditions exist in designer gravity . this establishes @xmath81 . however , an energy bound can be derived only once we relate @xmath72 to the physical energy @xmath54 . while these coincide for fast fall - off boundary conditions , they differ in the slow fall - off case . one consequence of this is that @xmath72 is not in general conserved . another is that @xmath72 may now depend on whether the spinor charge is defined using @xmath4 or @xmath3 . that is , @xmath82 . in @xcite , the covariant phase space method of @xcite , was used to show ( following @xcite ) that energy in designer gravity takes the form ) describes the generic case . additional terms are present for special cases . see @xcite . ] : [ generale ] e = - _ c e_ab ^a ds^b - ( _ + - _ - ) _ c _ b ds^b , where @xmath83 is the time translation conjugate to @xmath54 , @xmath84 is the integration element on the cut @xmath85 of the ads conformal boundary @xmath86 , and @xmath87 is the suitably rescaled electric part of the weyl tensor , which is smooth at @xmath86 as a consequence of our boundary conditions . the numerical value of @xmath54 is independent of the particular choice of the cut . one can derive the relation between @xmath52 and the energy @xmath54 @xcite by expanding the metric and the spinors in an asymptotic series in @xmath88 , using the witten equation , and using einstein s equation . the result is : @xmath89 d\omega + \frac{1}{2 } \lim_{r \to \infty } ( \lambda_{\pm}-\lambda_- ) \ , r^{d-1 - 2\lambda_- } \int_{c_r } \alpha^2\ , d\omega \ , , \ ] ] where for simplicity we have chosen @xmath90 to be a cut of constant @xmath91 so that @xmath92 is the integration element of the unit sphere @xmath26 . in the final term , @xmath93 is a large sphere of radius @xmath94 in @xmath66 . choosing @xmath3 and using @xmath95 yields the bound @xmath96 where @xmath97 is the volume of the unit - radius @xmath26 . on the other hand , for @xmath98 choosing @xmath4 causes the final term in ( [ qew ] ) to diverge . since @xmath54 and @xmath0 are manifestly finite , it follows that @xmath99 is infinite . it is also infinitely larger than @xmath54 , and its positivity yields no lower bound on @xmath54 . thus , the energy bound ( [ bound ] ) has been established only if the theory admits a @xmath3 superpotential ; the existence of @xmath4 alone is not sufficient . let us consider a simple example shown in @xcite to have an energy @xmath54 which is unbounded below . we take @xmath41 and the superpotential @xmath100 this choice corresponds to @xmath101 ( that is , @xmath102 ) , and since the coefficient of @xmath103 is @xmath104 , we see from ( [ series ] ) that ( [ simpleex ] ) is of the @xmath4 type . to see that the corresponding @xmath3 does not exist , let us write ( [ vtop ] ) as @xmath105 for the example presented in the text . the dashed line corresponds to the @xmath4 solution , which is stationary at the global minimum of the potential , @xmath106 . the solid line corresponds to the @xmath3 solution , whose derivative vanishes at @xmath107 . hence , @xmath3 does not meet the global existence criterion . , width=336 ] one may solve ( [ dp ] ) by integrating out from @xmath108 and matching to the expansion ( [ series ] ) for small @xmath40 . such solutions exist until the quantity @xmath109 becomes negative . now , note that ( [ series ] ) implies @xmath110 for all nonzero @xmath40 and similarly for their derivatives . in our example , @xmath4 has a maximum at the global minimum of the potential occurring at @xmath111 , so @xmath112 vanishes there . but this means that @xmath113 must vanish at some @xmath114 . in fact , it must vanish linearly at @xmath115 , since @xmath116 would force @xmath117 by ( [ vtop ] ) . thus a real @xmath3 can not exist for @xmath118 . this is illustrated in figure [ pprime ] using numerical solutions of ( [ dp ] ) . since only @xmath4 exists for all @xmath40 , one does not expect ( [ bound ] ) to hold for solutions which explore large values of @xmath40 . indeed , @xcite found solutions with @xmath119 and arbitrarily negative energy for this potential . in order to confirm the above resolution of the puzzle raised in @xcite , we now numerically investigate energy bounds in a simple class of examples . for designer gravity boundary conditions that preserve the full ads symmetry group , the existence of solitons has proven to be a reliable indicator whether or not a theory satisfies a positive energy theorem @xcite . when the theory has a static spherical soliton solution , no such theorem can hold because ads - invariant gravitational solitons can always be rescaled to obtain solutions with arbitrarily negative energy that obey the same boundary conditions . in this section we consider a one - parameter class of scalar potentials and construct spherical solitons numerically . we find that if the boundary conditions are specified by a @xmath0 which is i ) ads - invariant and ii ) bounded below , then solitons exist _ if and only if _ @xmath3 does not exist . this provides a strong test of the above stability proof , and it suggests that the existence of @xmath3 is both a sufficient and a necessary condition for the energy to be bounded from below in designer gravity . we consider the following class of potentials in @xmath41 dimensions : [ pot ] v()=-3- ^2 -2 3 ^6 + a ^8 , where @xmath120 is a free parameter . these yield scalar potentials with a negative maximum at @xmath14 , and with two global minima at @xmath121 . small fluctuations around @xmath14 have @xmath122 , which is within the range ( [ range ] ) . hence asymptotically the scalar generically decays as [ genfall2 ] = r + r^2 + and the asymptotic behavior of the @xmath123 metric component reads [ asmetric2 ] g_rr = 1 r^2 - ( 1+^2/2 ) r^4 + o ( r^-5 ) . ( 0,0 ) ( 90,201)@xmath124 ( 388,22)@xmath125 designer gravity boundary conditions . the dashed line shows a potential that violates the positive energy theorem for designer gravity boundary conditions , but satisfies such a theorem for fast fall - off boundary conditions ( @xmath126).,width=336 ] to construct the corresponding superpotentials @xmath2 we solve ( [ vtop ] ) , starting with @xmath127 and with @xmath128 . a solution to ( [ vtop ] ) exists unless @xmath109 becomes negative . as we integrate out from @xmath14 , @xmath1 is initially increasing and @xmath109 remains positive because the scalar satisfies the breitenlohner - freedman bound . for sufficiently large values of @xmath129 the global minima at @xmath130 will not be very much lower than the local maximum at @xmath14 , so one expects global solutions for @xmath2 to exist , with @xmath131 . by contrast , if the global minima are too deep , the quantity under the square root may become negative before the global minimum is reached , and consequently a real superpotential @xmath1 will not exist . clearly there are two distinct critical potentials ( with parameters @xmath132 ) , corresponding to those where @xmath133 just vanishes as the global minimum is reached . numerically solving ( [ vtop ] ) for a range of values for @xmath129 , one finds @xmath134 , and @xmath135 . we plot the latter critical potential in figure 2 ( full curve ) . the dashed line in figure 2 shows a potential with @xmath136 . we now look for static spherical soliton solutions of the theory ( [ pot ] ) satisfying ads - invariant boundary conditions with @xmath0 bounded below . since @xmath137 and @xmath138 , ads - invariant boundary conditions are given @xcite by @xmath139 for real @xmath140 . we see that @xmath0 is bounded below only for @xmath141 , so that only solitons with @xmath33 lead to violations of ( [ bound ] ) . the metric of a general such solution takes the form [ metric ] ds^2=-h(r)e^-2(r)dt^2+h^-1(r)dr^2+r^2d_2 ^ 2 , and the field equations read [ hairy14d ] h_,rr+(+_,r^2h+h_,r ) _ , r = v _ , , [ hairy24d ] 1-h - rh_,r-_,r^2h = r^2v ( ) , [ hairy34d ] _ , r = -1 2r _ , r^2 . regularity at the origin requires @xmath142 and @xmath143 at @xmath144 . rescaling @xmath91 shifts @xmath145 by a constant , so its value at the origin is arbitrary . thus soliton solutions can be labeled by the value of @xmath40 at the origin and the set of all soliton solutions of a particular potential with a negative maximum is found by integrating the field equations ( [ hairy14d])-([hairy34d ] ) for different values of @xmath40 at the origin . for @xmath146 the scalar asymptotically behaves as ( [ genfall2 ] ) with constant @xmath147 . the soliton therefore defines a point in the @xmath148 plane for each such @xmath149 . repeating for all @xmath150 yields a curve @xmath151 . given a choice of boundary condition @xmath152 , the allowed solitons are simply given by the points where the soliton curve intersects the boundary condition curve : @xmath153 . here we are interested in the existence of @xmath154 solitons , for potentials of the form ( [ pot ] ) . figure 3 shows how the value of @xmath125 for the @xmath154 soliton changes when one increases the potential parameter @xmath129 , from @xmath155 to its critical value @xmath156 . this corresponds to deforming @xmath15 from the potential given by the dashed line in figure 2 , to the critical potential at which @xmath3 begins to exist . one sees that @xmath157 precisely when @xmath158 . furthermore , for @xmath159 the @xmath151 curve intersects the @xmath154 axis only at the origin , so no nontrivial soliton solution exists in this parameter regime is further decreased to @xmath160 , an intersection point at finite @xmath124 appears between @xmath161 and the @xmath126 axis . this supports the claim of @xcite that the theory admits a positive energy theorem for @xmath126 scalar boundary conditions only when @xmath15 can be derived from a superpotential ( @xmath162 ) . ] . it appears , therefore , that regular spherical @xmath5 soliton solutions cease to exist precisely when @xmath158 . ( 0,0 ) ( 90,201)@xmath129 ( 388,22)@xmath125 for the @xmath5 soliton in a range of theories with different values for @xmath129 , the coefficient of the @xmath163 term in ( [ pot ] ) . one sees that regular spherical @xmath5 soliton solutions cease to exist precisely when @xmath164 . , width=336 ] the existence of scalar solitons with ads - invariant boundary conditions for @xmath165 implies there are negative energy solutions in these theories . this can be seen as follows @xcite . starting with a static soliton @xmath166 , consider the one parameter family of configurations @xmath167 . because we chose conformally invariant boundary conditions , these are again satisfied by the rescaled data . it then follows from the constraint equations that the total energy of the rescaled configurations takes the form e_= ^-3 e_1 + ^-1 e_2 where @xmath168 is independent of the potential and is manifestly positive , and both @xmath169 are finite and independent of @xmath170 . furthermore , because the static soliton extremizes the energy @xcite , one has 0=d e_d|_=1 = -3 e_1 -e_2 and hence @xmath171 . therefore the contribution to the energy that scales as the volume , which includes the potential and scalar terms , is negative . this means that rescaled configurations are initial data for time dependent ( but time - symmetric ) solutions . for sufficiently small @xmath170 one has a large central region where @xmath40 is essentially constant and away from an extremum of the potential . hence one expects the field to evolve to a spacelike ` big crunch ' singularity @xcite . ] @xmath172 with @xmath173 must have negative total energy , and hence violate the energy bound ( [ bound ] ) . thus , the energy is unbounded below for potentials with @xmath174 and @xmath5 , indicating that existence of @xmath3 is necessary for energy in designer gravity to be bounded below . ( note that we have already proven that it is sufficient . ) as explained in section [ two ] , the derivation of the lower bound for the energy @xmath54 uses the positivity of the spinor charge . this requires the global existence of smooth spinors satisfying the witten condition l = 0 on a constant time surface @xmath66 with suitable asymptotic behavior , where @xmath175 is the operator in the witten equation , [ witteneq ] l = ^i _ i . here , @xmath176 is the first order covariant differential operator defined in eq . , taken in directions @xmath177 tangent to the spacelike surface @xmath66 . global existence of such solutions is non - trivial to establish , and indeed it was suggested in @xcite that some subtle failure of such spinors to exist might be responsible for the lack of energy bounds in the systems studied there . in contrast , we have now proposed that the failure of those examples to satisfy ( [ bound ] ) was due solely to the fact that such potentials do not admit a @xmath3-type superpotential . in order to dispel any remaining concerns about the global existence of spinors , we reexamine this issue below . section [ again ] essentially follows @xcite and proves global existence of the witten spinors in general , but we deviate slightly from @xcite to correct an error in this proof . in order to correct this proof , we need to work on a maximal slice @xmath66 ( i.e. , one for which the trace of the extrinsic curvature satisfies @xmath178 ) in the spacetime that intersects scri transversely . this proof is in fact quite abstract and draws on certain very non - trivial results in the mathematics literature . we therefore restrict attention in section [ sphere ] to the particular case of maximal ( e.g. , time - symmetric ) , spherically symmetric hypersurfaces , which is enough to address any concerns raised by @xcite . in this case we are able to give a more explicit construction of the witten spinors by reducing the problem to ordinary differential equations . existence of solutions is then straightforward to show . finally , we numerically solve for the witten spinors in section [ numerics ] and evaluate the corresponding spinor charges , verifying the relationship between @xmath72 and @xmath54 derived in section [ two ] . if @xmath66 is a maximal slice in the spacetime , then the witten operator @xmath175 in eq . can be written as [ witteneq1 ] l = ^i ( d_i + p ( ) _ i ) , where @xmath179 is now the spin - connection intrinsic to @xmath66 . on slices with @xmath180 , there would be another term proportional to @xmath181 in the above expression for @xmath175 . for the proof to work , we must assume that such a slice exists . below , we will briefly outline how one might construct such a @xmath66 . we would like to find a smooth spinor field @xmath67 on @xmath66 with the property that @xmath182 , and that @xmath67 tends to a suitable exact killing spinor of ads - space near infinity . following @xcite , the first step in the proof is to find a formal power series solution @xmath183 to the witten spinor equation of the form [ asympt ] ~_0 + r^-1 _ 1 + r^-2 _ 2 + + r^-n _ n + , where @xmath184 , and where @xmath70 is a killing spinor in exact ads - space ( i.e. , @xmath185 in exact ads with vanishing @xmath40 ) such that - ( ) ^a = _ 0 ^a _ 0^ . the spinor fields @xmath186 are determined recursively by the condition that @xmath183 be a formal power series solution to the witten equation . it can be proven that @xmath187 are uniquely determined for @xmath188 by making a split of the witten equation into a part containing an @xmath94-derivative and a part containing derivatives tangent to a sphere of constant @xmath94 . the explicit form of the first coefficients @xmath187 in terms of the spacetime curvature near infinity is given in @xcite . the formal power series solution satisfies the equation l= j , where @xmath189 is a smooth source vanishing faster than any inverse power of @xmath94 near @xmath86 . we may terminate the expansion at some large finite @xmath190 , thus obtaining a @xmath189 vanishing faster than @xmath191 . the next step is to obtain from the formal power series solution @xmath183 a global solution @xmath67 to the witten equation . this step requires some global analysis . the idea is to write the global solution @xmath67 that we seek as = + , where @xmath192 is a smooth spinor field to be determined . thus , @xmath192 should satisfy @xmath193 , and @xmath192 should vanish at least as fast as @xmath194 in order that @xmath67 have the same asymptotics as @xmath183 to leading order . the global existence of such a @xmath192 can be established using the following key inequality which holds whenever @xmath195 is a solution to the einstein equations satisfying the asymptotic conditions given in @xcite . to state the inequality , consider any smooth spinor field @xmath196 on @xmath66 that vanishes outside some compact region in @xmath66 , and introduce the following norm on such @xmath196 : [ normu ] u ^2 = _ ds . then there exists a positive constant independent of @xmath196 such that [ ineq ] const.^-1 u ^2 _ |l u|^2 ds u ^2 . the proof of the inequality uses a similar technique as in the `` hardy - inequality , '' combined with the key identity . it may be found in @xcite . we now use the inequality to establish a solution to the equation @xmath193 . we will do this by first constructing a distributional solution @xmath192 , and then showing that it is in fact smooth . let @xmath197 be the linear functional on smooth compactly supported spinors @xmath196 defined by f(u ) = -_(nj)^u ds , where @xmath198 , and @xmath76 is the unit timelike normal to @xmath66 . note that @xmath199 , and that [ hermitian ] n l^= l n due to the fact that @xmath178 . by the cauchy - schwartz inequality and the first half of the inequality , we get |f(u)| & & ( _ |u|^2 ( 1+r^2)^-1 ds ) ^1/2 ( _ |j|^2 ( 1+r^2 ) ds ) ^1/2 + & & const . ( _ |lu|^2 ds ) ^1/2 . the constant is finite because by construction @xmath189 drops off sufficiently rapidly at infinity . we interpret this inequality as saying that @xmath197 is a bounded functional with respect to the positive definite scalar product given by v | u = ( lv)^lu ds . let @xmath200 be the hilbert space defined by this inner product ( which , by the second half of inequality is identical to the hilbert space obtained from the norm @xmath201 ) . by the riesz representation theorem , there is hence an element @xmath202 such that @xmath203 for all @xmath204 . again by the inequality , every element in @xmath200 is locally square integrable , so in particular a distribution on @xmath66 . thus , @xmath205 is a solution in the distributional sense of the equation @xmath206 . next , we apply @xmath207 to this equation , use eq . , and put @xmath208 . it follows that this @xmath192 is the desired distributional solution to @xmath193 , and @xmath209 is a global solution to the witten equation . it remains to prove that @xmath67 is smooth , and that it satisfies the desired boundary conditions . this will follow if we can show that @xmath192 is smooth and vanishes sufficiently fast at infinity . it follows from our construction so far only that @xmath210 . but since @xmath189 is smooth and vanishes quickly , one can now use the mapping properties of the parametrix of the operator @xmath175 established in @xcite to prove that @xmath192 is indeed smooth and vanishes sufficiently fast . the details of this argument are given in @xcite . finally , we outline how one might construct the desired @xmath178 slice @xmath66 in the type of spacetimes that we consider . first , let @xmath211 be any ( weakly ) spacelike hypersurface in spacetime , @xmath212 its domain of dependence . if @xmath211 is such that the closure of @xmath212 is contained in a globally hyperbolic subregion of spacetime ( note that the full spacetime need _ not _ be globally hyperbolic ) , then a theorem of bartnik @xcite guarantees that there exists a sufficiently regular maximal slice @xmath66 with prescribed extrinsic curvature and boundary @xmath213 . unfortunately , it is not immediately clear how to apply this theorem when the surface @xmath211 touches the conformal boundary . however , by considering a sequence @xmath214 eventually touching the conformal boundary transversely , we obtain a corresponding sequence @xmath215 of maximal slices . presumably it can be shown that these have a subsequence that is convergent in a sufficiently strong sense . we now explicitly describe the solution to the witten equation for a slice @xmath66 which is both spherically symmetric and maximal ( i.e , @xmath178 ) though not necessarily static . such a slice automatically exists when the initial data are time - symmetric . since this case is sufficient to address the concerns of @xcite , we explain the analysis in some detail . our conventions will be as follows . for simplicity , we restrict attention in this subsection to the case @xmath41 . we denote by @xmath216 indices on a flat internal space ( while @xmath217 are spacetime indices ) . the gamma matrices satisfy @xmath218 and we choose the explicit matrix representation @xmath219 where @xmath220 is the @xmath221 identity matrix , and where @xmath222 are the standard pauli matrices . we wish to address spherical spacetimes . let @xmath66 be a maximal ( @xmath178 ) , spherically symmetric spacelike slice on which we set @xmath223 , e.g. , a surface of time - symmetry . though such spacetimes are not in general static , on @xmath66 they may nevertheless be written in the form ( [ metric ] ) , which may be described by the orthonormal frame @xmath224 because @xmath178 , the witten equation only depends on the geometry intrinsic to @xmath66 , i.e. , @xmath225 . we further assume that the scalar field on @xmath66 is a function of @xmath94 alone , and that @xmath226 , @xmath227 are known functions . the intrinsic covariant derivative acting on spinor fields is in general given by @xmath228 where the intrinsic spin connection is determined by the relation @xmath229 . for the metric ( [ metric ] ) , we find @xmath230 with the above ingredients in hand , the witten condition becomes the explicit equation @xmath231\psi + \frac{3}{2 } p(\phi ) \psi \,.\ ] ] writing @xmath67 in terms of two - component spinors as @xmath232 ( [ witcon ] ) implies @xmath233\psi_2 + \frac{3}{2 } p(\phi ) \psi_1 , \\ \label{witcon22 } 0 & = & i \left[\sigma_{3 } \sqrt{h } \left(\partial_r + \frac{1}{r } \right ) + \frac{1}{r } \ , \rlap/\!\nabla_{\ ! \!s^2 } \right]\psi_1 - \frac{3}{2 } p(\phi ) \psi_2 \,.\end{aligned}\ ] ] here the operator @xmath234 is the dirac operator on the 2-dimensional sphere @xmath235 . to decouple the differential equations ( [ witcon21 ] ) and ( [ witcon22 ] ) , we define @xmath236 then the @xmath237 satisfy @xmath238 \psi_+ & = & 0 \\ \label{witcon32 } \left [ \sigma_{3 } \left(\partial_r + \frac{1}{r } \right ) + \frac{1}{r \sqrt{h } } \ , \rlap/\!\nabla_{\ ! \!s^2 } - \frac{3}{2 \sqrt{h } } p(\phi ) \right ] \psi_- & = & 0 \ , .\end{aligned}\ ] ] next we wish to expand @xmath237 in spinor spherical harmonics , which are eigenfunctions of @xmath234 . these are given , for example in @xcite , by @xmath239 @xmath240 where @xmath241 are integers such that @xmath242 and @xmath243 . the @xmath244 are jacobi polynomials and the spherical harmonics satisfy @xmath245 separating variables as @xmath246 @xmath247 we find from ( [ witcon31 ] ) and ( [ witcon32 ] ) that the radial functions satisfy the following differential equations : @xmath248 @xmath249 to pick the desired solution to the witten equation , we now impose our boundary conditions . we demand that our solution be regular in the interior , and that it asymptotically approaches a suitable killing spinor @xmath70 of pure ads . as we will see shortly , this @xmath70 has only @xmath250 components . since the remaining modes are decoupled , they may be consistently set to zero . thus we consider only @xmath251 below . near the origin @xmath144 , the solutions behave as @xmath252 where @xmath253 are constants . regularity at the origin then requires boundary conditions of the form @xmath254 the analysis for the @xmath255 solutions near the origin is exactly the same . as @xmath256 , we require that the solution to the witten condition approaches a killing spinor in exact ads space . for large @xmath94 , solutions to ( [ radialde ] ) and ( [ tradialde ] ) behave as @xmath257 the solution which grows as @xmath258 matches the known asymptotic behavior of killing spinors in exact ads space . in particular , an explicit solution to the killing spinor equation is given by @xcite @xmath259 where @xmath260 is a constant spinor and @xmath261 the spinor @xmath260 is chosen so that @xmath70 satisfies the additional requirement @xmath262 , or equivalently @xmath263 . one can achieve this normalization by taking @xmath264 with the choice ( [ norm ] ) , one can rewrite ( [ ks ] ) in terms of the spinor spherical harmonics as @xmath265 hence , the solution for the witten spinor is @xmath266 \end{array } \right ) \,.\ ] ] furthermore , comparing to ( [ ks2 ] ) we see that the asymptotic conditions on the radial functions are @xmath267 comparing ( [ wcasym ] ) and ( [ wcbc ] ) with ( [ near0 ] ) and ( [ largerr ] ) demonstrates that the desired witten spinors exist . one first notes that @xmath268 decouple from @xmath269 . to construct the desired @xmath268 , one simply enforces the boundary condition ( [ wcbc ] ) at @xmath144 for some @xmath270 and integrates ( [ radialde ] ) outward . from ( [ largerr ] ) , any solution will grow asymptotically as @xmath258 , and in fact will satisfy ( [ wcasym ] ) up to an overall scale . rescaling @xmath270 yields the desired solutions @xmath268 . the analysis for @xmath271 is similar . thus , witten spinors exist for any spherically symmetric maximal hypersurface @xmath66 satisfying our boundary conditions . a final crucial aspect of the argument of section [ two ] is the detailed relation ( [ qew ] ) between the spinor charges @xmath52 and the energy @xmath54 . since ( [ qew ] ) was derived via a tedious calculation , it is useful to verify the relation numerically . let us consider the theory with potential @xmath272 which is just ( [ pot ] ) for @xmath273 . this theory admits both a @xmath4 and a @xmath3 superpotential . we take as initial data a soliton with @xmath274 and boundary condition @xmath275 for some constant @xmath276 ( so that @xmath0 is not conformally invariant ) . we find that a soliton exists with @xmath277 and @xmath278 . the energy of this soliton is @xmath279 . numerical results for @xmath3-type witten spinors are shown in figures [ rrm ] and [ irm ] . ) with initial data given by the rescaled configurations @xmath280 discussed in section [ solitons ] . ] ) confirm the existence of witten spinors that are regular at the origin and have the correct behavior at large @xmath94 . the solid line is @xmath281 , which by the arguments in the text asymptotically approach @xmath282 , shown here as the dashed line.,width=336 ] , which by the arguments in the text asymptotically approach @xmath283 , shown here as the dashed line . , width=336 ] using these solutions , we can also calculate the spinor charge and check ( [ qew ] ) . we obtain @xmath284 , thus confirming the relation @xmath285 . similarly , for the @xmath4 case we find @xmath286 , in agreement with ( [ qew ] ) . we have resolved the puzzle raised in @xcite concerning energy bounds in designer gravity . while the arguments of @xcite are correct as written , global existence of an appropriate superpotential is a subtle requirement . in particular , two types of superpotentials may ( or may not ) exist for a given scalar potential @xmath10 . the proof of @xcite requires the global existence of a @xmath3-type superpotential . if one attempts to follow the same argument with a @xmath4-type superpotential , one finds that the difference between the conserved energy @xmath54 and the associated spinor charge @xmath99 diverges , and in particular that @xmath99 diverges . thus , positivity of @xmath99 does not yield a lower bound for @xmath54 , and the existence of a @xmath4-type superpotential alone is not sufficient to yield a positive energy theorem . numerical explorations support this resolution . a specific question raised in @xcite concerned the global existence of witten spinors . we have demonstrated ( section [ spinors ] ) that no such difficulties arise , even for the models considered in @xcite . the existence theorem of @xcite was shown to hold in general and , for the special case of spherical symmetry , existence of witten spinors on maximal ( e.g. , time - symmetric ) hypersurfaces was again demonstrated using simple arguments based on ordinary differential equations . the spherical time - symmetric context is sufficient to address the concerns of @xcite . this reinforces our claim that , when @xmath0 has a global minimum , the existence of a @xmath3-type superpotential is sufficient for the energy to be bounded below . interestingly , our numerical explorations of particular examples ( section [ solitons ] ) suggest that global existence of @xmath3 may also be _ necessary _ for the energy to be bounded below . such a property would be analogous to the claim of @xcite that the existence of a superpotential is necessary for an energy bound to hold in the case of fast fall - off boundary conditions @xmath287 . however , in the fast fall - off case the relevant superpotential is @xmath4 , since either superpotential is sufficient and existence of @xmath3 implies existence of @xmath4 ( footnote [ exist ] ) . indeed , for @xmath32 the spinor charges @xmath52 satisfy @xmath288 and , in particular , @xmath289 when both potentials exist . thus we find that each type of superpotential provides a stability criterion for ads gravity - scalar theories , with @xmath4 controlling the @xmath32 case and @xmath3 controlling cases with slower fall - off conditions on the scalar . finally , a key tool in our investigation of particular examples was the argument from @xcite showing that the existence of designer gravity solitons satisfying ads - invariant boundary conditions implies that the energy is unbounded below . we may thus interpret the lower bounds of @xcite in terms of the soliton content of such theories . we conclude that designer gravity theories where @xmath3 exists can have no solitons when @xmath0 is both ads - invariant and bounded below . the authors would like to thank gary horowitz for useful discussions . s.h . would also like to thank piotr chrusciel and bob wald for discussions and the physics department at ucsb for its hospitality . d.m . also thanks jim isenberg and robert bartnik for discussions . a.a . and d.m were supported in part by nsf grant phy0354978 , and by funds from the university of california . p. breitenlohner and d. z. freedman , `` stability in gauged extended supergravity , '' annals phys . * 144 * ( 1982 ) 249 ; `` positive energy in anti - de sitter backgrounds and gauged extended supergravity , '' phys . b * 115 * ( 1982 ) 197 . m. henneaux , c. martinez , r. troncoso and j. zanelli , `` black holes and asymptotics of 2 + 1 gravity coupled to a scalar field , '' phys . rev . d * 65 * , 104007 ( 2002 ) [ arxiv : hep - th/0201170 ] . m. henneaux , c. martinez , r. troncoso and j. zanelli , `` asymptotically anti - de sitter spacetimes and scalar fields with a logarithmic branch , '' phys . d * 70 * , 044034 ( 2004 ) [ arxiv : hep - th/0404236 ] . t. hertog and k. maeda , `` stability and thermodynamics of ads black holes with scalar hair , '' phys . d * 71 * , 024001 ( 2005 ) [ arxiv : hep - th/0409314 ] . j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' adv . * 2 * ( 1998 ) 231 [ arxiv : hep - th/9711200 ] . d. z. freedman , c. nunez , m. schnabl and k. skenderis , `` fake supergravity and domain wall stability , '' phys . d * 69 * , 104027 ( 2004 ) [ arxiv : hep - th/0312055 ] . a. celi , a. ceresole , g. dallagata , a. van proeyen and m. zagermann , `` on the fakeness of fake supergravity , '' phys . d * 71 * , 045009 ( 2005 ) [ arxiv : hep - th/0410126 ] . m. zagermann , `` n = 4 fake supergravity , '' phys . d * 71 * , 125007 ( 2005 ) [ arxiv : hep - th/0412081 ] . k. skenderis and p. k. townsend , `` hidden supersymmetry of domain walls and cosmologies , '' phys . * 96 * , 191301 ( 2006 ) [ arxiv : hep - th/0602260 ] . i. papadimitriou , non - supersymmetric membrane flows from fake supergravity and multi - trace deformations , " [ arxiv : hep - th/0606038 ] . k. skenderis and p. k. townsend , `` pseudo - supersymmetry and the domain - wall / cosmology correspondence , '' arxiv : hep - th/0610253 . s. hollands , a. ishibashi and d. marolf , `` comparison between various notions of conserved charges in asymptotically ads - spacetimes , '' class . quant . * 22 * , 2881 ( 2005 ) [ arxiv : hep - th/0503045 ] . m. heusler , n. straumann , `` scaling arguments for the existence of static , spherically symmetric solutions of self - gravitating system , '' class . * 9 * ( 1992 ) 2177 ; m. heusler , `` a no hair theorem for selfgravitating nonlinear sigma models , '' j. math * 33 * ( 1992 ) 3497 . r. mazzeo : `` elliptic theory of differential edge operators i , '' commun . part . diff . equations * 16 * ( 10 ) ( 1991 ) , 1615 ; r. mazzeo : `` the hodge cohomology of a conformally compact metric , '' j. diff * 28 * ( 1988 ) 309

many alternatives to the standard model ( sm ) of particle physics have been proposed over the last thirty years . among these , supersymmetry ( susy ) , and particularly its minimal version dubbed the minimal supersymmetric standard model ( mssm ) @xcite , is one of the most popular sm extensions . it consists in a symmetry linking fields with opposite statistics , matching thus a bosonic ( fermionic ) superpartner with each fermionic ( bosonic ) sm degree of freedom . as a consequence , it predicts the stabilization of the gap between the electroweak and the planck scale , gauge coupling unification at high energies , and a lightest supersymmetric particle , which is weakly interacting and stable and thus a good dark matter candidate . since the superpartners of the sm particles have not yet been observed and in order to remain a viable solution to the hierarchy problem , susy must be softly broken at low energy , which makes the susy particles massive , with a mass lying in the tev range . therefore , susy searches at present hadron colliders , such as the tevatron at fermilab or the lhc at cern , are important topics of the current experimental high - energy physics program . within the standard model , flavour violation in the quark sector arises only through the rotation of the up- and down - type quark interaction eigenstates into the basis of physical mass eigenstates . four bi - unitary matrices are required to diagonalize the quark yukawa matrices , which renders the charged - current interactions proportional to the unitary ckm matrix @xmath0 . in the super - ckm basis @xcite , the squark interaction eigenstates undergo the same rotations as their quark counterparts , so that their charged - current interactions are also proportional to the ckm matrix . however , in non - minimal flavour - violating supersymmetric theories , the quark and squark fields can be misaligned due to additional sources of flavour violation which are related to the breaking of supersymmetry . as a consequence , this leads to flavour - violating ( non - diagonal ) entries in the squark mass matrices . in recent works , we have analysed the cases where such soft terms appear in ( non - minimal ) supergravity and gauge - mediated susy breaking scenarios @xcite . in the first case , supersymmetry is broken in a hidden sector and transmitted to the visible sector of squarks , sleptons , gauginos and gluinos through gravitational interactions . soft masses for sfermions are induced by direct khler interactions , which can in general be flavour non - diagonal @xcite . in the second case , the breaking of supersymmetry is mediated to the visible sector via gauge interactions with messenger fields in a flavour - conserving fashion @xcite . however , it has recently been shown that non - minimal versions of the gauge - mediated susy - breaking mechanism can yield important flavour - violation in the squark and slepton sectors @xcite . when susy is broken in a hidden sector , the soft masses also receive contributions from quantum effects due to the superconformal anomaly @xcite . in this work , we therefore consider this so - called anomaly - mediated susy - breaking ( amsb ) scenario , where those anomaly - mediated effects are large compared to all other sources of susy - breaking , which are subdominant . we extend our previous work on flavour violation @xcite by investigating possible non - minimal flavour violation within the amsb context . this paper is organized as follows : in sec . [ sec : model ] , we define anomaly - mediated supersymmetry breaking scenarios and show how non - minimal flavour violation can appear . in sec . [ sec : scan ] we impose current experimental constraints on the flavour - violating amsb scenario and perform scans of the parameter space . in addition , experimentally allowed benchmark points are defined . cross sections for the production of at least one gluino are analytically and numerically computed in sec . [ sec : xsec ] . we dedicate sec . [ sec : cosmo ] to an analysis of the possible cosmological constraints related to the presence of cold dark matter in our universe . our conclusions are presented in sec.[sec : conclusion ] . in amsb scenarios , the soft terms are related to the anomalous dimensions of the different fields and have the feature to be renormalization - group invariant @xcite . as a consequence , they are fully determined by the known low - energy gauge and yukawa couplings and an overall mass scale @xmath1 , the vacuum expectation value of the scalar auxiliary field of the gravitation supermultiplet . this scale is expected to be of the order of the gravitino mass @xmath2 , and we assume in the following , to simplify , @xmath3 . consequently , the model is highly predictive , with fixed mass ratios and distinctive signatures @xcite . among all the predictions , one finds , however , tachyonic sleptons . this problem must be cured in order to have a phenomenologically viable model . several solutions have been proposed @xcite , and we adopt here the phenomenological approach of assuming non - negligible contributions to the scalar soft masses , induced , _ e.g. _ , by supergravity , which makes their square positive at the weak scale . however , in this case , solving the tachyonic sfermion mass problem can also introduce non - minimal flavour violation in the theory , through possible non - diagonal flavour - violating soft mass terms . the squark mass matrices are written , in the super - ckm basis , as m_^2 = ( ccc|ccc m^2_l_q_1 & ^q_1 q_2_ll & ^q_1 q_3_ll & m_q_1 x_q_1 & ^q_1 q_2_lr & ^q_1 q_3_lr + ^q_1 q_2_ll & m^2_l_q_2 & ^q_2 q_3_ll & ^q_1 q_2_rl & m_q_2 x_q_2 & ^q_2 q_3_lr + ^q_1 q_3_ll & ^q_2 q_3_ll & m^2_l_q_3 & ^q_1 q_3_rl & ^q_2 q_3_rl & m_q_3 x_q_3 + m_q_1 x_q_1^ & ^q_1 q_2_rl & ^q_1 q_3_rl & m^2_r_q_1 & ^q_1 q_2_rr & ^q_1 q_3_rr + ^q_1 q_2_lr & m_q_2 x_q_2^ & ^q_2 q_3_rl & ^q_1 q_2_rr & m^2_r_q_2 & ^q_2 q_3_rr + ^q_1 q_3_lr & ^q_2 q_3_lr & m_q_3 x_q_3^ & ^q_1 q_3_rr & ^q_2 q_3_rr & m^2_r_q_3 ) , where the flavour - diagonal elements are given by m_l_q_i^2 = & m_q_i^2 + m_q_i^2 + 2 ^ 2 ( t_q^3 - e_q s_w^2 ) , + m_r_q_i^2 = & m_u_i^2 + m_q_i^2 + 2 ^ 2 e_q ^2 , + m_r_q_i^2 = & m_d_i^2 + m_q_i^2 + 2 ^ 2 e_q ^2 , + x_q_i = & a_q_i^- \ { l , + . . the weak isospin quantum numbers are @xmath4 for left - handed up - type and down - type ( s)quarks , their fractional electromagnetic charge is denoted by @xmath5 , and @xmath6 is the mass of the quark @xmath7 , @xmath8 being the flavour index , , @xmath9 , @xmath10 , @xmath11 , @xmath12 , @xmath13 , and @xmath14 . in addition , @xmath15 is the @xmath16-boson mass , and @xmath17 is the sine of the electroweak mixing angle . the soft supersymmetry - breaking mass terms are @xmath18 and @xmath19 for the left - handed and right - handed squarks , while the quantities @xmath20 are the trilinear couplings between the higgs bosons and the scalar susy particles . in the higgs sector , @xmath21 denotes the off - diagonal superpotential higgs mass parameter , and @xmath22 is the ratio of vacuum expectation values of the two higgs doublets . in our phenomenological approach , the off - diagonal parameters @xmath23 are arbitrary and can be normalized to the diagonal entries according to @xcite _ ab^q_i q_j = ^q_i q_j_ab m_q_i m_q_j . additional sources of quark flavour violation are then parameterized through the 21 dimensionless ( possibly complex ) new variables @xmath24 , since due to @xmath25 gauge invariance , the @xmath26 elements of the up- and down - type squark squared mass matrices are related to each other , [ eq : su2l ] m_,ll^2 = v_ckm m_,ll^2 v_ckm^ , as are the associated @xmath27-parameters . this equation shows that both squark mass matrices can not be simultaneously diagonal ( without neglecting the ckm matrix ) . the diagonalization of the mass matrices @xmath28 and @xmath29 requires the introduction of two @xmath30 matrices @xmath31 and @xmath32 , ( m_u_1 ^ 2 , , m_u_6 ^ 2 ) = r^u m_^2 r^u ( m_d_1 ^ 2 , , m_d_6 ^ 2 ) = r^d m_^2 r^d , where by convention the masses are ordered increasingly , @xmath33 . these mixing matrices relate the physical mass eigenbasis to the interaction eigenbasis through ( _ 1 , _ 2 , _ 3 , _ 4 , _ 5 , _ 6)^t = r^u ( _ l , _ r , _ r , _ r)^t ( _ 1 , _ 2 , _ 3 , _ 4 , _ 5 , _ 6)^t = r^d ( _ l , _ r , _ r , _ r)^t . recently , it has been shown that ( minimal ) amsb scenarios with an additional @xmath34 symmetry satisfy the requirements of the minimal flavour violation principles @xcite . in this case , the @xmath27-parameters are not free and directly dictated from the flavour structure of the ckm matrix . in our approach , we are going beyond this scheme , keeping the flavour - violation parameters independent . in this section , we discuss the most relevant experimental measurements that can be used to constrain the parameter space of the mssm . apart from direct searches for superpartners at collider experiments , and particularly the recent results of the atlas and cms experiments at the lhc @xcite , numerous low - energy and electroweak precision measurements @xcite constrain masses and mixings of the superpartners . they can often impose stronger limits on the flavour - violating entries introduced in sec . [ sec : model ] . extensive studies of the kaon sector , @xmath35- and @xmath36-meson oscillations , rare decays , and electric dipole moments suggest that only flavour mixing involving the second and third generations of squarks can be substantial , and this only in the left - left and right - right chiral sectors , which mix the superpartners of the left - handed and right - handed quarks @xcite . for this reason , we restrict ourselves to the simpler scenario where only the flavour - mixing parameters related to the second and third generations and non mixing squark chiralities can be non - vanishing , _ l _ ll^ct , _ u _ rr^ct _ d _ rr^sb . [ eq : lambda ] the parameter related to the mixing among second and third generation down - type squarks @xmath37 is not a free parameter of our simplified model , since it is connected to @xmath38 through eq . . the measurement of the rare @xmath39 decay represents one of the most stringent constraints on these squark mixings . the inclusive branching ratio is determined to be ( bs ) = ( 3.55 0.26_exp 0.23_theo ) 10 ^ -4 [ eq : bsg ] from babar , belle , and cleo data @xcite . for the theoretical error estimate , we refer to the discussion in refs . squark diagrams contribute already at the one - loop level , as do the sm particles . as a consequence , this measurement can lead to strong constraints on the squark masses and couplings , especially on the @xmath40 parameter of the left - left chiral sector , since the lightest neutralinos and charginos appearing in the loops are mostly winos coupling to the left - handed components of the squark mass - eigenstates . the same arguments apply to the @xmath41 branching fraction , experimentally measured as @xcite ( bs^+^- ) = ( 2.23 0.98_exp 0.11_theo ) 10 ^ -6 , the associated theoretical uncertainties being computed in ref.@xcite , or to the @xmath42-meson branching fractions to a muon pair , recently bounded from above by the cms and lhcb experiments @xcite , [ eq : bs0 ] br(b_s^0 ^+^- ) < 1.1 10 ^ -8 , at @xmath43 confidence level , which is , however , still about four or five times larger than the sm expectation . also , the @xmath44 oscillations , which have been recently observed , directly probe the mixing between squarks of the second and third generations . since nmfv contributions arise at the same loop level as the sm ones , this observable can again be sensitive to non - vanishing @xmath27-parameters . hence , the measured mass difference @xcite m_b_s^0 = ( 17.78 0.12_exp 3.3_theo ) ps^-1 , [ eq : dmbs ] where the theoretical uncertainty of 3.3 ps@xmath45 at the 95% confidence level has been derived in ref . @xcite , allows to constrain the magnitude of the above - mentioned flavour - violating parameters . another important consequence of nmfv mixing in the squark sector is a large splitting between squark mass eigenvalues . this influences the @xmath16- and @xmath46-boson self - energies at zero momentum , contributing hence to the electroweak @xmath47-parameter = - = ( m_z ) t = ( 2.36 8.65 ) 10 ^ -4 , the experimental value arising from combined fits of the @xmath16-boson mass , width , and pole asymmetry as well as of the masses of the @xmath46-boson and the top quark @xcite . furthermore , recent measurements of the anomalous magnetic moment of the muon @xmath48 indicate a discrepancy of about @xmath49 between the data and the standard model predictions @xcite , a_^exp - a_^sm = ( 25.5 7.98 ) 10 ^ -10 a _ ( g-2)_/2 . this gap could be explained by new physics . in the case of supersymmetric scenarios , the leading contributions , depending on the smuon , sneutrino , chargino and neutralino masses , have been found to be proportional to the sign of the @xmath21-parameter @xcite . since negative values would then increase the discrepancy , we limit ourselves to positive values of @xmath21 . moreover , since squarks contribute only at the two - loop level , the dependence on flavour violation in the squark sector is expected to be considerably reduced . finally , from direct searches of the higgs boson , we ask for the mass of the lightest higgs - boson to fulfil 111 gev m_h 130 . [ eq : higgsmass ] the lower bound is based on the exclusion limit of @xmath50 gev from lep @xcite , after accounting for a theoretical uncertainty of about @xmath51 gev @xcite , whilst the upper bound corresponds to the non - observation of a higgs boson by the atlas and cms experiments @xcite . for a restricted set of universal scalar mass @xmath52 , introduced to solve the tachyonic slepton problem , smaller than a few tev , the upper bound is almost always satisfied . considering quantum corrections , the higgs mass @xmath53 depends on the squark masses , and thus on the flavour - violating entries in the associated mass matrices . in order to study the phenomenology of amsb models including non - minimal flavour violation , we have scanned over the parameter space of the model , generalizing the squark mass matrices by including the three flavour - violation parameters presented in eq . . our procedure starts at a high - energy scale , where we define our input parameters as the gravitino mass @xmath2 , the universal scalar mass @xmath52 introduced to solve the tachyonic slepton problem , the ratio of the two neutral higgs field vacuum expectation values @xmath54 , and the sign of the higgs mixing parameter @xmath55 . the soft supersymmetry - breaking terms at the electroweak scale are then obtained through renormalization group running using the spheno package version 3.0 @xcite , which solves the renormalization group equations numerically to two - loop order and extracts the particle spectrum and mixings at the electroweak scale at the one - loop level for the matter and gauge sectors and at the two - loop level for the higgs sector . it also computes the electroweak precision and low - energy observables presented in sec.[sec : constraints ] . for the numerical values of the standard model parameters , we fix the top quark pole mass to @xmath56 gev @xcite , the bottom quark mass to @xmath57 gev and the @xmath16-boson mass to @xmath58 gev . the fermi constant has been taken as @xmath59 gev@xmath60 , and the strong and electromagnetic coupling constants at the @xmath16-pole as @xmath61 and @xmath62 . at this stage , the only source of flavour violation lies within the ckm - matrix , which is calculated using the wolfenstein parametrization . the corresponding four free parameters are set to @xmath63 , @xmath64 , @xmath65 , and @xmath66 @xcite . plane at fixed @xmath67 ( left panel ) , in the @xmath68 plane at fixed @xmath69 tev ( central panel ) and in the @xmath70 plane at fixed @xmath71 tev ( right panel ) . we show regions where there is no physical solution to the renormalization group equations ( grey ) , or which are excluded by the constraints related to the @xmath39 branching ratio ( blue ) and the higgs mass ( red ) . the regions , where the agreement between theory and experiment for the anomalous magnetic moment of the muon is restored , are presented in beige . on the left panel , we also indicate the region where the mass of the lightest higgs boson is close to 125 gev ( green ) and the exclusion limit obtained ( crosses ) from the reinterpretation of the lhc results on supersymmetric particle searches in the context of amsb scenarios ( see ref . @xcite).,title="fig : " ] plane at fixed @xmath67 ( left panel ) , in the @xmath68 plane at fixed @xmath69 tev ( central panel ) and in the @xmath70 plane at fixed @xmath71 tev ( right panel ) . we show regions where there is no physical solution to the renormalization group equations ( grey ) , or which are excluded by the constraints related to the @xmath39 branching ratio ( blue ) and the higgs mass ( red ) . the regions , where the agreement between theory and experiment for the anomalous magnetic moment of the muon is restored , are presented in beige . on the left panel , we also indicate the region where the mass of the lightest higgs boson is close to 125 gev ( green ) and the exclusion limit obtained ( crosses ) from the reinterpretation of the lhc results on supersymmetric particle searches in the context of amsb scenarios ( see ref . @xcite).,title="fig : " ] plane at fixed @xmath67 ( left panel ) , in the @xmath68 plane at fixed @xmath69 tev ( central panel ) and in the @xmath70 plane at fixed @xmath71 tev ( right panel ) . we show regions where there is no physical solution to the renormalization group equations ( grey ) , or which are excluded by the constraints related to the @xmath39 branching ratio ( blue ) and the higgs mass ( red ) . the regions , where the agreement between theory and experiment for the anomalous magnetic moment of the muon is restored , are presented in beige . on the left panel , we also indicate the region where the mass of the lightest higgs boson is close to 125 gev ( green ) and the exclusion limit obtained ( crosses ) from the reinterpretation of the lhc results on supersymmetric particle searches in the context of amsb scenarios ( see ref . @xcite).,title="fig : " ] in fig . [ fig : mfv ] , we present typical scans of the _ minimal _ amsb parameter space , , when flavour - violation is induced by the ckm - matrix alone and all @xmath27-parameters vanish . we show the examples of the @xmath72 plane at fixed @xmath67 ( left panel ) , the @xmath68 plane at fixed @xmath69 tev ( central panel ) as well as the @xmath70 plane at fixed @xmath71 tev ( right panel ) . all experimental limits described in sec.[sec : constraints ] are imposed at the @xmath73-level . we observe that the low - mass regions with a relatively small @xmath52 , attractive from a collider point of view , are strongly disfavoured by both the measurements of the @xmath39 branching ratio and the too low predicted mass for the lightest higgs boson . the latter also excludes small values of @xmath74 for a large range of @xmath52 and @xmath2 masses . in addition , a significant part of the regions where the predicted value for the @xmath75 branching ratio lies outside the @xmath76-range deduced from eq . is also excluded by the constraints associated to the @xmath41 branching ratio . therefore , these regions have not been shown on fig . [ fig : mfv ] , for clarity . predictions for the other considered @xmath35-physics observables , @xmath77 and the branching ratio @xmath78 , are mostly compatible with data and hence absent from the figure , as is the computed value for the @xmath79 parameter . this quantity only restricts the parameter space at very heavy masses , which lie outside the mass range presented in the figures . on the left panel of fig . [ fig : mfv ] , we also show the exclusion limits on the parameter space of amsb scenarios obtained from a reinterpretation of the results of the direct searches for superparticles at the lhc in the amsb context @xcite , as well as the region where the mass of the lightest higgs boson is close to 125 gev , a value favoured by the recent observations of the atlas and cms experiments @xcite . however , in the absence of a confirmed signal for a higgs boson in that mass range , we do not consider this last limit for our analysis in the sequel . the size of the regions , where the gap between the data and the predictions for the anomalous magnetic moment of the muon @xmath80 is closed , is relatively small . moreover , a significant fraction of it is excluded by the @xmath81 constraint , as it has also been found in refs . hence , a possible explanation of the discrepancy between theory and data for this observable by contributions related to amsb scenarios is rather difficult . however , the dominant supersymmetric contributions to @xmath80 highly depend on the slepton masses , entering into the theoretical calculation already at the one - loop level . since in the context of amsb scenarios , any prediction associated to the slepton sector is tightly linked to the employed solution to solve the tachyonic slepton problem , we choose to relax the constraint associated to the anomalous magnetic moment of the muon . this is also motivated by the fact that non - minimal flavour violation in the squark sector , , the scope of this paper , contributes subdominantly , at the two - loop level , to @xmath80 , so that it becomes almost independent of the @xmath27-parameters as it was already noticed in the discussions of refs.@xcite . we now turn to non - minimal flavour violation in amsb scenarios , where non - vanishing @xmath27-parameters can arise , , from non - trivial khler interactions as in supergravity . in our phenomenological approach , we introduce non - minimal flavour violation at the weak scale , generalizing the squark mass matrices by introducing the three parameters defined in eq . . we use again the spheno 3.0 program to diagonalize the squark sector and compute the flavour and weak observables presented in sec . [ sec : constraints ] . in figs.[fig : tb10 ] and [ fig : m60 ] , we depict the impact of the considered nmfv parameters on these observables , showing the associated constraints on the non - minimal amsb parameter space . we impose the above - mentioned limit for the higgs - boson mass , , @xmath82 gev ( see the discussion in section [ sec : constraints ] ) , as well as all the other constraints at the 2@xmath83 confidence level . we present different typical amsb planes for different values of the flavour - violating @xmath27-parameters , keeping only the most constraining observables for clarity and neglecting the anomalous magnetic moment of the muon for the reasons mentioned above . plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] + plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] + plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] plane with a fixed value of @xmath67 . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameters @xmath40 ( upper panels ) , @xmath84 ( central panels ) and @xmath85 ( lower panels ) . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) and the higgs mass ( red).,title="fig : " ] typical scans in the @xmath86 planes of the amsb parameter space are shown in fig . [ fig : tb10 ] . we use a fixed value of @xmath67 and choose a positive sign for the @xmath21-parameter . comparing with the corresponding minimal amsb results ( left panel of fig . [ fig : mfv ] ) , we find that the most sensitive observable is the @xmath39 branching ratio , which directly probes squark mixing in the left - left chiral sector , , the @xmath38-parameter . this strong dependence is due to squark and neutralino - chargino loops , involving @xmath87 interactions between squarks , quarks and neutralinos or charginos , proportional to the squark mixing matrices @xmath31 and @xmath32 ( see eq . below ) . this also explains the less pronounced sensitivity to nmfv mixing in the right - right chiral sectors ( @xmath84 and @xmath85 ) , where the results do not show any strong dependence on the @xmath27-parameters . at intermediate values of @xmath88 , the interplay between squark masses and mixings is such that almost all the parameter space accessible by renormalization group running remains allowed by the @xmath89 branching ratio compared to the minimal results of fig.[fig : mfv ] . similar conclusions hold for the @xmath90 branching ratio , not presented on the figures for clarity . the other considered observables , , the constraint on the lightest higgs boson mass , are barely sensitive to non - minimal flavour violation , as the dependence on the squark sector is subdominant . plane with a fixed value of @xmath69 tev ( upper panels ) and in the @xmath91 plane with a fixed value of @xmath71 tev ( lower panels ) . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameter @xmath38 . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) , the @xmath42-meson branching to a muon pair @xmath92 ( green ) and the lightest higgs - boson mass ( red).,title="fig : " ] plane with a fixed value of @xmath69 tev ( upper panels ) and in the @xmath91 plane with a fixed value of @xmath71 tev ( lower panels ) . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameter @xmath38 . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) , the @xmath42-meson branching to a muon pair @xmath92 ( green ) and the lightest higgs - boson mass ( red).,title="fig : " ] plane with a fixed value of @xmath69 tev ( upper panels ) and in the @xmath91 plane with a fixed value of @xmath71 tev ( lower panels ) . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameter @xmath38 . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) , the @xmath42-meson branching to a muon pair @xmath92 ( green ) and the lightest higgs - boson mass ( red).,title="fig : " ] + plane with a fixed value of @xmath69 tev ( upper panels ) and in the @xmath91 plane with a fixed value of @xmath71 tev ( lower panels ) . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameter @xmath38 . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) , the @xmath42-meson branching to a muon pair @xmath92 ( green ) and the lightest higgs - boson mass ( red).,title="fig : " ] plane with a fixed value of @xmath69 tev ( upper panels ) and in the @xmath91 plane with a fixed value of @xmath71 tev ( lower panels ) . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameter @xmath38 . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) , the @xmath42-meson branching to a muon pair @xmath92 ( green ) and the lightest higgs - boson mass ( red).,title="fig : " ] plane with a fixed value of @xmath69 tev ( upper panels ) and in the @xmath91 plane with a fixed value of @xmath71 tev ( lower panels ) . the sign of the @xmath21-parameter is chosen positive and we present the results for different values of the nmfv - parameter @xmath38 . we show regions excluded due to the absence of physical solutions to the renormalization group equations ( grey ) , by the constraints associated to the @xmath39 branching ratio ( blue ) , the @xmath42-meson branching to a muon pair @xmath92 ( green ) and the lightest higgs - boson mass ( red).,title="fig : " ] in fig . [ fig : m60 ] , we investigate the dependence of the predictions for weak and flavour observables on @xmath74 . we first fix the universal scalar mass @xmath52 to 1 tev and scan over the @xmath93 plane . secondly , we fix the gravitino mass @xmath2 to 60 tev and scan over the @xmath91 plane . in both cases , the sign of the @xmath21-parameter is again chosen to be positive . as for the analysis in fig . [ fig : tb10 ] , the most sensitive observables to non - minimal flavour violation are the @xmath39 and @xmath94 branching ratios . since the constraints related to the second decay are weaker and overlapping with those of the first , they are again omitted from the figures . the regions of the parameter space excluded by the constraints associated to the @xmath75 branching ratio increase with the values of the flavour - violating parameter @xmath95 . for small values , squark mixings and mass effects compensate , which leaves again almost all the parameter space open . in fact , the exclusion contours related to the @xmath75 branching ratio even vanish , when @xmath40 increases slightly . such contours reappear only at much larger @xmath40 values . as a consequence , for @xmath96 , possible flavour violating entries in the squark mass matrices are reduced to be rather modest , , @xmath97 . on the other hand , very low values of @xmath74 are excluded by the bounds on the lightest higgs boson mass for all values of @xmath40 . it is also interesting to remark that the predictions for the @xmath98 observable lie above the current experimental bounds of eq . in the rather large @xmath74 region , for moderate values of the mixing parameter of the left - left chiral sector ( @xmath99 $ ] ) and for relatively light gravitino and universal scalar masses . let us finally note that the results for non - minimal flavour violation in the right - right chiral sector are unchanged with respect to the minimal case shown in fig . [ fig : mfv ] , and we subsequently do not present the corresponding figures . inspecting the various amsb planes presented in sec . [ sec : planes ] , we select three benchmark scenarios allowed by the present low - energy and electroweak precision constraints , which permit a sizeable mixing between second and third generation squarks and which are collider - friendly in the sense that one or several superpartners could be produced with a large rate at the lhc . this means that at least some of the superpartners should not be too heavy . the chosen benchmark points are presented in tab . [ tab : points ] , together with the sps9 benchmark point @xcite which is shown as a reference . as a generic feature for all the four points , the gravitino mass is taken as @xmath71 tev and the sign of the off - diagonal higgs mixing parameter is positive , for the reasons discussed in sec.[sec : constraints ] . due to the renormalization group running invariance of the gaugino mass parameters @xmath100 , @xmath101 and @xmath102 , the dependence of the masses of the gluino , the two lightest neutralinos and the lightest chargino , all mainly gaugino - like , on @xmath52 and @xmath74 is reduced and subdominant . therefore , they are roughly identical for all scenarios , with @xmath103 gev , @xmath104 gev , and @xmath105 gev . in contrast , the scalar spectrum is quite different . .three amsb benchmark scenarios allowing for sizeable flavour - violating entries in the squark mass matrices with respect to the second and third generation mixing in the left - left and/or the right - right chiral squark sectors . the sps9 scenario is also presented for comparison . we indicate the susy - breaking parameters at the high scale and the resulting masses at low energy ( after renormalization group running ) for the gluino , the lightest squarks and sleptons , the two lightest neutralinos , the lightest chargino , and the lightest higgs - boson . the values are presented assuming constrained minimal flavour - violation , , when @xmath106 . [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^ " , ] for gluino pair production as well as for the associated production of a chargino or a neutralino with a gluino , only the @xmath107- and @xmath108-channel diagrams depend on the flavour - violating parameters in the squark sector since they contain a squark propagator ( see the feynman diagrams shown in figs . [ fig:01 ] and [ fig:03 ] ) . however , all squark eigenstates contribute to the total cross section and the corresponding diagrams must be summed over , leading subsequently to production cross sections insensitive to non - minimal flavour - violation . we present these results , therefore independent of the @xmath27-parameters , in tab . [ tab : xsec ] , both for our scenarios i , ii and iii as well as for the sps 9 benchmark point as a reference . strong gluino pair production is clearly dominant , and the luminosity required to observe possible signal events is not so high . in contrast , the cross sections related to the semi - weak production of a gluino and a chargino / neutralino vary from @xmath109 fb for the lightest chargino and neutralino case to the barely visible level of @xmath110 fb for the heavier superpartners . , @xmath84 , and @xmath85 for the reference scenario i of tab . [ tab : points].,title="fig : " ] , @xmath84 , and @xmath85 for the reference scenario i of tab . [ tab : points].,title="fig : " ] + , @xmath84 , and @xmath85 for the reference scenario i of tab . [ tab : points].,title="fig : " ] , @xmath84 , and @xmath85 for the reference scenario i of tab . [ tab : points].,title="fig : " ] for reference scenario ii.,title="fig : " ] for reference scenario ii.,title="fig : " ] + for reference scenario ii.,title="fig : " ] for reference scenario ii.,title="fig : " ] for reference scenario iii.,title="fig : " ] for reference scenario iii.,title="fig : " ] + for reference scenario iii.,title="fig : " ] for reference scenario iii.,title="fig : " ] contrary to the previous cases , associated squark and gluino production shows an interesting dependence on the non - minimal flavour - violation @xmath27-parameters , as illustrated in figs . [ fig : xsec1 ] , [ fig : xsec2 ] and [ fig : xsec3 ] for our scenarios i , ii and iii , respectively . moreover , the cross sections are fairly high , reaching the level of several hundereds of fb for many channels , which makes them nice candidates to study non - minimal flavour violation in supersymmetry at the lhc . analyzing the dependence of the cross section on the @xmath27-parameters , sharp and smooth transitions can be observed , the first ones being related to the presence of an avoided crossing and the second ones to a smooth change in the flavour content of the relevant eigenstate ( see section [ sec : benchmark ] ) . indeed , at the point where the mass of two eigenstates should cross , , at @xmath111 on the top - left panel of fig . [ fig : xsec1 ] or at @xmath112 on the lower - right panel of fig . [ fig : xsec2 ] , the flavour content of both squarks is exchanged and the same sharp transition is observed at the level of the cross section . smooth increases and decreases in the cross section with the values of the @xmath27-parameters have two sources . first , second and third generation squark mixing induces larger mass splitting , as illustrated in figs . [ fig : squarks1 ] , [ fig : squarks2 ] and [ fig : squarks3 ] . this renders certain channels phase - space favoured and other channels phase - space suppressed . secondly , the magnitude of the cross section is connected to the flavour content of the squark produced in association with the gluino , since producing a given flavour of squark requires an initial quark of the same flavour , as shown in the feynman diagrams of fig . [ fig : gosq ] . as an example , the lightest down - type squark mass eigenstate is a pure sbottom state for small values of @xmath113 , whilst it becomes a mixed state for @xmath114 , with a larger and larger strange squark component with increasing values of @xmath85 , as it can be seen from the upper - right panel of fig.[fig : squarks1 ] . consequently , as presented in lower - right panel of fig . [ fig : xsec1 ] , the cross section related to the process @xmath115 is small and constant for @xmath113 ( @xmath116 fb ) and gets larger and larger for increasing @xmath85 , even reaching a couple of hundreds of fb for @xmath117 . the opposite effects can be observed for the process @xmath118 in the context of the scenario iii , as shown in the lower - left panel of fig.[fig : xsec3 ] . among the most compelling evidences for physics beyond the standard model is the presence of cold dark matter ( cdm ) in our universe . its relic density is today constrained to be _ cdmh^2 = 0.1123 0.0035 [ eq : omh2 ] from recent wmap data combined with measurements related to supernov and baryonic acoustic oscillations @xcite . here , @xmath119 is the present hubble expansion rate in units of @xmath120 . new physics models should therefore include a viable dark matter candidate that can account for the above amount of dark matter . in amsb models , the lightest of the four neutralinos is the lightest superpartner and therefore the dark matter candidate , if r - parity is assumed to be conserved . after the renormalization group evolution from the high - scale parameters to the weak scale , the wino mass parameter @xmath101 turns out to be smaller than the bino and gluino masses @xmath100 and @xmath102 . in consequence , in amsb scenarios the lightest neutralino is wino - like . this is in contrast to , e.g. , models based on minimal supergravity where usually @xmath121 leading to a bino - like lsp . since also the chargino mass is governed by @xmath101 , the mass difference between the lightest neutralino and the lightest chargino is rather small . for our reference scenarios , the mass difference is less than a gev , as can be seen in tab . [ tab : points ] . due to the larger pair annihilation cross section as compared to the bino and due to efficient co - annihilations with the chargino , the resulting relic density of the thermally produced neutralino is usually one or two orders of magnitude below the range given in eq.([eq : omh2 ] ) @xcite . using the public programme darksusy @xcite , we obtain the values @xmath122 , and @xmath123 for the scenarios of tab . [ tab : points ] , respectively . however , thermal production of neutralinos is not the only mechanism to be considered . possible non - thermal production modes include the decay of heavy fields such as moduli or gravitinos in the early universe @xcite . moreover , axions and axinos can contribute to the dark matter relic abundance @xcite . the contribution to the neutralino relic density from moduli decay can be estimated as @xcite _ ^0_1h^2 0.1 ( ) ( ) ^1/4 ( ) ( ) ^3/2 . it depends on the neutralino mass @xmath124 , the mass of the moduli fields @xmath125 , the effective number of degrees of freedom @xmath126 , and the thermally averaged annihilation cross - section @xmath127 . in fig . [ fig : cosmo ] , we depict isolines of the neutralino relic density in the @xmath124@xmath125 plane , assuming @xmath128 @xcite and different values of the annihilation cross - section @xmath127 . our typical scenarios lead to a neutralino relic density of about @xmath129 , which is consistent with a neutralino annihilation cross - section of @xmath130 . for such cross - sections , a neutralino mass of @xmath131 gev and moduli masses comparable with the gravitino mass , @xmath132 tev , which is consistent with gravity and anomaly mediation @xcite , this yields nearly the measured abundance of eq . ( [ eq : omh2 ] ) . in the @xmath125@xmath124 planes for different values of @xmath133 . the dotted lines correspond to @xmath134 gev and @xmath135 tev . right : the neutralino relic density @xmath136 from gravitino decay as a function of the reheating temperature @xmath137 for a typical amsb scenarios with @xmath138 gev , @xmath71 tev , as it is the case for the scenarios of tab . [ tab : points ] . the solid line includes a contribution of @xmath139 from thermal neutralinos , while the dashed line indicates the contribution from gravitino decay only.,title="fig : " ] in the @xmath125@xmath124 planes for different values of @xmath133 . the dotted lines correspond to @xmath134 gev and @xmath135 tev . right : the neutralino relic density @xmath136 from gravitino decay as a function of the reheating temperature @xmath137 for a typical amsb scenarios with @xmath138 gev , @xmath71 tev , as it is the case for the scenarios of tab . [ tab : points ] . the solid line includes a contribution of @xmath139 from thermal neutralinos , while the dashed line indicates the contribution from gravitino decay only.,title="fig : " ] the neutralino relic abundance from gravitino decay is directly related to the thermal abundance @xmath140 , that the gravitino would have if it did not decay . the latter is obtained from computing the thermal production of gravitinos in the early universe . more precisely , the resulting neutralino relic density can then be evaluated according to @xcite _ ^0_1h^2 = _ 3/2h^2 ( ) ( ) _ i=1 ^ 3 _ i g_i^2 ( 1 + ) , [ eq : omh2grav ] where the ratio of neutralino and gravitino mass expresses the fact that each gravitino decays into one stable neutralino . the thermal gravitino production depends linearly on the gravitino mass @xmath2 and the reheating temperature @xmath137 of the universe after inflation . the sum runs over the three gauge groups @xmath34 , @xmath25 , and @xmath141 , @xmath142 are the coupling constants of the three gauge groups , and @xmath143 the mass parameters of the associated gauginos . note that in eq . ( [ eq : omh2grav ] ) , @xmath142 and @xmath143 are evaluated at the reheating scale @xmath137 . the constants @xmath144 and @xmath145 are given by @xmath146 and @xmath147 for @xmath148 , respectively @xcite . the right panel of fig . [ fig : cosmo ] shows the resulting neutralino relic density as a function of the reheating temperature for the situation corresponding to the scenarios of tab . [ tab : points ] . for low values of @xmath137 , the thermal neutralino production dominates , leading to the value of @xmath149 already mentioned above . for @xmath150 gev , the contribution from gravitino decay becomes dominant and @xmath151 grows linearly with @xmath137 . as can be seen , the observed relic density of @xmath152 is obtained for a reheating temperature of @xmath153 gev , which is well compatible with thermal leptogenesis @xcite . the relic abundance of the neutralino may also depend on flavour violating entries of the squark ( or slepton ) mass matrices . in ref . @xcite this has been studied for the case of minimal supergravity scenarios , where flavour violating couplings can modify the annihilation and coannihilation modes that enter the boltzmann equation in the typical scenario with thermal production of neutralinos . similar conclusions have been found for flavour non - diagonal entries in the leptonic soft matrices @xcite . flavour - violating effects are , of course , also present in the discussed cases of moduli or gravitino decay . a full study of their impact within this context is , however , clearly beyond the scope of this work . in this paper , we have studied the consequences of non - minimal flavour violation in minimal anomaly - mediated supersymmetry breaking models , where tachyonic sleptons are avoided by introducing a common scalar mass similar to the one introduced in minimal supergravity . in these scenarios , new sources of flavour violation are in general introduced at high scales , leading to different flavour mixings for sm particles and their superpartners at the weak scale . using the conventional parameterization of squark mixing at the weak scale , we analyzed the viable amsb parameter space in the light of the latest limits on low - energy observables and of the latest results from direct searches for higgs and susy particles at the lhc . we found that intermediate values of @xmath154 and relatively large scalar masses of @xmath155 tev , increasing with @xmath74 , were preferred and allowed for sizeable flavour violation in the left - left and essentially unconstrained flavour violation in the right - right squark sectors . we completed our analytical calculations of flavour - violating supersymmetric particle production at hadron colliders with those related to gluino pair production and to the associated production of gluinos with charginos and neutralinos as well as with squarks . the corresponding cross sections were expected to be large due to the strong coupling of gluinos to the initial quarks and gluons . flavour violation effects were expected to be only significant for the associated production of gluinos and squarks , since the other processes involved ( almost ) complete sums over internal squark exchanges . this was confirmed in our numerical analysis for the high - energy phase of the lhc , where phenomena such as avoided crossings or smooth flavour dependences known from our previous studies could again be observed . for the experimental analysis of the ensuing cascade decays , leading to final states with second and third generation quarks and missing transverse energy , we referred the reader to previously published dedicated studies performed , , in supergravity models . the corresponding analysis in amsb models was beyond the scope of this paper and is left for future work . finally , we briefly addressed the related cosmological aspects , showing that the well - known problem of dark matter underabundance in minimal amsb models could be solved with moduli or gravitino decays also in the presence of flavour violation . the authors would like to thank w. porod for helpful discussions . this work has been supported by the theory - lhc france - initiative of the cnrs / in2p3 and by the `` helmholtz alliance for astroparticle phyics hap '' funded by the initiative and networking fund of the helmholtz association . g. bozzi , b. fuks , b. herrmann and m. klasen , nucl . b * 787 * ( 2007 ) 1 . b. fuks , b. herrmann and m. klasen , nucl . b * 810 * ( 2009 ) 266 . a. h. chamseddine , r. l. arnowitt , p. nath , phys . * 49 * ( 1982 ) 970 . m. dine and a. e. nelson , phys . d * 48 * ( 1993 ) 1277 ; + m. dine , a. e. nelson and y. shirman , phys . d * 51 * ( 1995 ) 1362 ; + m. dine , a. e. nelson , y. nir and y. shirman , phys . d * 53 * ( 1996 ) 2658 . l. randall , r. sundrum , nucl . phys . * b557 * ( 1999 ) 79 - 118 . g. f. giudice , m. a. luty , h. murayama _ et al . _ , jhep * 9812 * ( 1998 ) 027 . t. gherghetta , g. f. giudice , j. d. wells , nucl . phys . * b559 * ( 1999 ) 27 - 47 . a. pomarol , r. rattazzi , jhep * 9905 * ( 1999 ) 013 . i. jack , d. r. t. jones , phys . lett . * b465 * ( 1999 ) 148 - 154 . n. arkani - hamed , d. e. kaplan , h. murayama _ et al . _ , jhep * 0102 * ( 2001 ) 041 . j. l. feng , t. moroi , l. randall _ et al . _ , phys . rev . lett . * 83 * ( 1999 ) 1731 - 1734 . j. l. feng , t. moroi , phys . rev . * d61 * ( 2000 ) 095004 . r. rattazzi , a. strumia , j. d. wells , nucl . phys . * b576 * ( 2000 ) 3 - 28 . a. j. barr , c. g. lester , m. a. parker _ et al . _ , jhep * 0303 * ( 2003 ) 045 . z. chacko , m. a. luty , i. maksymyk _ et al . _ , jhep * 0004 * ( 2000 ) 001 . e. katz , y. shadmi , y. shirman , jhep * 9908 * ( 1999 ) 015 . i. jack , d. r. t. jones , phys . lett . * b482 * ( 2000 ) 167 - 173 . i. jack , d. r. t. jones , nucl . * b662 * ( 2003 ) 63 - 88 . b. murakami , j. d. wells , phys . * d68 * ( 2003 ) 035006 . r. kitano , g. d. kribs , h. murayama , phys . * d70 * ( 2004 ) 035001 . m. ibe , r. kitano , h. murayama , phys . * d71 * ( 2005 ) 075003 . r. hodgson , i. jack , d. r. t. jones _ et al . _ , nucl . * b728 * ( 2005 ) 192 - 206 . d. r. t. jones , g. g. ross , phys . lett . * b642 * ( 2006 ) 540 - 545 . r. hodgson , i. jack , d. r. t. jones , jhep * 0710 * ( 2007 ) 070 . f. gabbiani , e. gabrielli , a. masiero and l. silvestrini , nucl . b * 477 * ( 1996 ) 321 . b. c. allanach , g. hiller , d. r. t. jones _ et al . _ , jhep * 0904 * ( 2009 ) 088 . k. nakamura _ et al . _ [ particle data group ] , j. phys . g * 37 * ( 2010 ) 075021 . d. asner _ [ heavy flavor averaging group ] , arxiv:1010.1589 [ hep - ex ] . j. s. hagelin , s. kelley and t. tanaka , nucl . b * 415 * ( 1994 ) 293 . p. brax and c. a. savoy , nucl . b * 447 * ( 1995 ) 227 . m. ciuchini , e. franco , d. guadagnoli , v. lubicz , m. pierini , v. porretti and l. silvestrini , phys . b * 655 * ( 2007 ) 162 . t. hurth , e. lunghi and w. porod , nucl . b * 704 * ( 2005 ) 56 . m. misiak _ et al . _ , phys . * 98 * ( 2007 ) 022002 . t. huber , t. hurth and e. lunghi , nucl . b * 802 * ( 2008 ) 40 . cms - pas - bph-11 - 019 . p. ball and r. fleischer , eur . j. c * 48 * ( 2006 ) 413 . t. moroi , phys . d * 53 * ( 1996 ) 6565 [ erratum - ibid . d * 56 * ( 1997 ) 4424 ] . w. porod , comput . commun . * 153 * ( 2003 ) 275 . atlas - conf-2011 - 163 . m. lancaster [ tevatron electroweak working group and for the cdf and d0 collaborations ] , arxiv:1107.5255 [ hep - ex ] . b. c. allanach , t. j. khoo and k. sakurai , arxiv:1110.1119 [ hep - ph ] . s. p. martin , j. d. wells , phys . * d64 * ( 2001 ) 035003 . b. c. allanach _ et al . _ , eur . j. c * 25 * ( 2002 ) 113 . t. hurth and w. porod , jhep * 0908 * ( 2009 ) 087 . a. bartl , k. hidaka , k. hohenwarter - sodek , t. kernreiter , w. majerotto and w. porod , phys . b * 679 * ( 2009 ) 260 . a. bartl , h. eberl , b. herrmann , k. hidaka , w. majerotto and w. porod , phys . b * 698 * ( 2011 ) 380 [ erratum - ibid . b * 700 * ( 2011 ) 390 ] m. bruhnke , b. herrmann and w. porod , jhep * 1009 * ( 2010 ) 006 . a. bartl , h. eberl , e. ginina , b. herrmann , k. hidaka , w. majerotto and w. porod , arxiv:1107.2775 [ hep - ph ] . c. gross and g. hiller , phys . d * 83 * ( 2011 ) 095015 . w. beenakker , r. hopker , m. spira and p. m. zerwas , nucl . b * 492 * ( 1997 ) 51 . t. gehrmann , d. maitre and d. wyler , nucl . b * 703 * ( 2004 ) 147 . j. pumplin , d. r. stump , j. huston , h. l. lai , p. nadolsky and w. k. tung , jhep * 0207 * ( 2002 ) 012 . e. komatsu _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 192 * ( 2011 ) 18 . c. h. chen , m. drees and j. f. gunion , phys . rev . d * 55 * ( 1997 ) 330 [ erratum - ibid . d * 60 * ( 1999 ) 039901 ] . t. moroi and l. randall , nucl . phys . b * 570 * ( 2000 ) 455 . p. gondolo , j. edsj , p. ullio , l. bergstrom , m. schelke and e. a. baltz , jcap * 0407 * ( 2004 ) 008 ; p. gondolo , j. edsj , p. ullio , l. bergstrm , m. schelke , e.a . baltz , t. bringmann and g. duda , http://www.darksusy.org . s. rajagopalan , arxiv:1010.3384 [ hep - ph ] . l. covi , j. e. kim and l. roszkowski , phys . * 82 * ( 1999 ) 4180 . l. covi , h. b. kim , j. e. kim and l. roszkowski , jhep * 0105 * ( 2001 ) 033 . b. s. acharya , g. kane , s. watson and p. kumar , phys . d * 80 * ( 2009 ) 083529 . j. pradler and f. d. steffen , phys . d * 75 * ( 2007 ) 023509 . w. buchmuller , p. di bari and m. plumacher , annals phys . * 315 * ( 2005 ) 305 . b. herrmann , m. klasen and q. le boulch , phys . d * 84 * ( 2011 ) 095007 . d. choudhury , r. garani and s. k. vempati , arxiv:1104.4467 [ hep - ph ] .

the theory of high - temperature superconductivity remains one of the most challenging problems in solid - state physics . in many metallic compounds , over a broad range of compositions and temperatures , only two phases are encountered , i.e. the normal ( fermi liquid ) and a magnetic or superconducting phase . in sharp contrast , the high - temperature superconductors ( htsc ) and other strongly correlated electron systems , such as heavy - fermion and a variety of transition - metal oxide systems , exhibit many ordered phases , which appear to compete and sometimes coexist.@xcite in the htsc , besides ordered antiferromagnetic ( af ) and superconducting ( sc ) phases , compelling evidence exists for charge- and spin-``stripe '' phases and phases with coexisting sc and af order.@xcite the common denominator and the underlying reason for these competing orders is certainly the presence of strong electronic correlations and mott - hubbard physics : the interplay between kinetic - energy and coulomb - correlation effects induces an extreme sensitivity to external parameters ( doping , temperature , pressure , etc . ) and a rather difficult to predict `` outcome '' , i.e. the characteristic low - energy excitations and the phase diagram at low temperatures . the central challenge in the field of high-@xmath1 superconductivity is , therefore , the connection of the ( known ) microscopic interactions at the level of electrons and ions , which are at high energy ( @xmath2ev ) and temperature @xmath3 , with the `` emerging phenomena '' at @xmath4 , i.e. competing and nearly degenerate orders . ideally , one should employ a systematic renormalization - group approach to integrate out the irrelevant degrees of freedom and , thereby , correctly bridge high to low energies and eventually go to @xmath4 . it is , however , by no means obvious how to do this when strong correlations are present , such as in the htsc . in this context cluster techniques , which systematically approach the infinite - size ( low - energy ) limit , provide a powerful alternative . @xcite in this paper , we will discuss and apply a variational - cluster approach ( vca ) which was proposed recently . @xcite it is based on the self - energy - functional theory ( sft ) , @xcite which provides a general variational scheme to use dynamical information from an exactly solvable `` reference system '' ( in our case , an isolated cluster ) to go to the infinite - size lattice fermion problem at low temperatures and at @xmath4 , in particular . in our earlier work @xcite this scheme was formulated to study phases with spontaneously broken symmetry . for the cluster sizes used , it was shown that the vca correctly reproduces long - range af order for the two - dimensional ( @xmath5 ) hubbard model and the absence of this order in one dimension . this non - trivial `` test '' implies that the vca goes well beyond ordinary mean - field theory . another crucial test is provided by the dynamical information contained in the one - particle green s function @xmath6 . compared to variational schemes based on wave functions , @xcite an important advantage of the vca consists in the fact that it quite naturally gives the one - electron green s function @xmath6 . for the @xmath5 hubbard model , it was recently demonstrated @xcite that the vca , with the lattice tiled by ( @xmath7 ) clusters , correctly reproduces low - temperature quantum monte - carlo ( qmc ) data , in particular , the coherent and incoherent `` bands '' experimentally known from arpes data . @xcite these tests provide also the foundation for attacking the question whether the `` minimal '' microscopic model , namely the @xmath5 one - band hubbard model , reproduces the essential features of the electron- and hole - doped htsc phase diagram . we will not go into a lengthy discussion of what interactions should be retained at the electron - ion level . but , when choosing the @xmath5 one - band hubbard model , @xcite i.e.@xmath8 where @xmath9 , @xmath10 are the usual annihilation and creation operators , @xmath11 denote the hopping - matrix elements , @xmath12 the density at site @xmath13 with spin @xmath14 and @xmath15 the local coulomb repulsion , one has introduced gross simplifications , leaving out other orbital ( e.g. @xmath16 ) degrees of freedom , long - range coulomb interaction , electron - phonon coupling , etc . nevertheless , this choice appears to be legitimate , last not least in view of the amazing agreement achieved between numerical simulations and experimental results for the normal - state properties of the cuprates ( see , for example , refs . ) . the ground - state phase diagram of the model was recently investigated using the vca by snchal et al . @xcite and by two of us . @xcite there are important technical differences , but the `` upshot '' of the two works is as follows : for the cluster sizes used in the vca , the @xmath4 phase diagram of the hubbard model ( including hopping terms up to second or third - nearest neighbors ) turns out to be qualitatively similar to that of the electron- and hole - doped cuprates . the model correctly describes the overall phase diagram , such as the occurrence of the af and sc phases and predicts the corresponding doping ranges in qualitative agreement with the experiments for the cuprate materials . the present paper has several purposes : first , we would like to stress that for an application of the vca to the high-@xmath1 problem it is of crucial importance to treat the on - site energies in the reference system as variational parameters . we will show that this ensures a thermodynamically consistent determination of the average particle number . compared to the study of ref . this represents an important methodical extension . on the other hand , without the variational optimization of the on - site energy , one has to tolerate an inconsistency in the determination of the average particle number . the effects of this error shall be demonstrated by model calculations . the issue of thermodynamic consistency is also discussed for the off - diagonal elements of the one - particle density matrix and , in case of spontaneous symmetry breaking , for the respective order parameter . it is interesting to note that there are no such problems in the dynamical mean - field theory ( dmft ) @xcite and its cluster extensions . @xcite here the on - site energies are kept at their `` physical '' values from the very beginning rather than being determined from the self - consistency condition . the ( cellular ) dmft , however , can be considered as a special approximation within the general sft framework . @xcite hence , the question arises why a fixed on - site energy does _ not _ spoil thermodynamic consistency in the case of ( cellular ) dmft . it is interesting to note that , in case of interacting bosons , the issue of consistency has also been shown to be very important , recently . @xcite second , an accurate analysis of the behavior of the chemical potential as a function of the particle density close to the transition to a non - magnetic state , as well as a corresponding maxwell construction , indicate the presence of an inhomogeneous ground state with macroscopically large regions of low- and high - particle density . using @xmath17-site clusters , we can get a rough estimate of the ground - state phase diagram and investigate the instability of the homogeneous ( af , sc ) phases against charge inhomogeneities . this represents an important complement to the work of ref . , where larger cluster sizes up to @xmath18 sites have been considered but without an appropriate analysis via a maxwell construction . finally , our numerical results give valuable insights into different questions of the high-@xmath1 problem , as they provide direct access to the single - particle excitation spectrum in the strong - coupling regime at zero temperature . this has to be contrasted with the work of maier et al . @xcite who have been able to treat clusters with @xmath19 sites within the dynamical cluster approximation ( dca ) , but are restricted to finite temperatures , intermediate coupling @xmath20 and imaginary time . a review comparing the application of different cluster ( sft , dca , and cellular dmft ) as well as weak - coupling methods for the hubbard model has appeared recently . @xcite our paper is organized as follows : we start with a summary of the central ideas of the sft in general ( sec . [ sec : sft ] ) . thermodynamic consistency with respect to the average particle number is discussed in sec . [ sec : n ] . consistency with respect to the off - diagonal elements of the density matrix is addressed in appendix [ sec : cicj ] and the case of the c - dmft is discussed in appendix [ sec : cdmft ] . our theoretical considerations are completed by describing some computational details of the vca calculations in sec . [ sec : vca ] . section [ spe ] then presents the results for the @xmath4 spectral function @xmath21 for electron- and hole - doping . we discuss how the characteristically different doping dependencies of the spectra give rise to different fermi - surface evolutions upon doping . these fermi - surface evolutions can then be tied up with the characteristic differences in the electron- and hole - doped phase diagrams , such as the enhanced robustness of the af order in the electron - doped case . the ground - state phase diagram will be presented and discussed in detail in sec . finally , sec . [ summ ] contains our main conclusions and a summary . the central idea of the self - energy - functional theory ( sft ) @xcite is to make use of the universality of the luttinger - ward functional @xmath22 $ ] @xcite or of its legendre transform @xmath23 $ ] : for a system with hamiltonian @xmath24 , where @xmath25 are the one - particle and @xmath26 the interaction parameters , the functional dependence @xmath27 $ ] or @xmath28 $ ] is independent of @xmath25 . this universality is obvious as the luttinger - ward functional is defined via a skeleton - diagram expansion involving dressed propagators and vertices only . @xcite concentrating on the self - energy @xmath29 instead of the single - particle green s function @xmath6 , the grand potential of the system at temperature @xmath3 and chemical potential @xmath30 can be written as a functional of @xmath29 : @xmath31 = { \mbox{tr}}\ln ( { { \bm g}}_{0 , { { \bm t}}}^{-1 } - { { \bm \sigma}})^{-1 } + f_{{{\bm u } } } [ { { \bm \sigma } } ] \ : , \label{eq : sf}\ ] ] where @xmath32 is the free green s function and @xmath33 with the usual trace @xmath34 and the matsubara frequencies @xmath35 for integer @xmath36 . at the physical self - energy @xmath37 , the grand potential is stationary : @xmath38 = 0 $ ] . why is it more advantageous to express @xmath39 as a functional of the self - energy @xmath29 rather than @xmath6 ? this has to do with the `` short - range '' character of @xmath29 as a function of its real - space coordinates , @xcite which in general is due to the fact that @xmath29 is qualitatively related to a dynamically screened particle - particle interaction . for the hubbard model , in particular , one believes that important effects are sufficiently accounted for by a local @xcite or short - ranged @xcite self - energy , at least for high lattice dimensions . a local or short - ranged self - energy , however , can be well generated by a cluster of finite size , and for the subsequent optimization of the cluster trial self - energy , the self - energy functional ( [ eq : sf ] ) can be used . this concept allows for the construction of a class of conceptually clear and thermodynamically consistent approximations , including the dynamical mean - field theory ( dmft ) @xcite and a cluster extension of the dmft @xcite ( see ref . for a detailed discussion ) . due to the universality of @xmath40 $ ] , we have @xmath41 = f_{{{\bm u } } } [ { { \bm \sigma } } ] + { \mbox{tr}}\ln ( { { \bm g}}_{0,{{\bm t}}'}^{-1}-{{\bm \sigma}})^{-1 } \label{eq : sfp}\ ] ] for the self - energy functional of a so - called `` reference system '' , which is given by a hamiltonian with the same interaction part @xmath26 but modified one - particle parameters @xmath42 : @xmath43 . although it has different microscopic parameters , the reference system is assumed to be in the same macroscopic state as the original system , so it has the same temperature @xmath3 and the same chemical potential @xmath30 . by a proper choice of its one - particle part , the problem posed by the reference system @xmath44 can be much simpler than the original problem posed by @xmath45 , such that the self - energy of the reference system , @xmath46 , can be computed exactly within a certain subspace of parameters @xmath42 . combining eqs . ( [ eq : sf ] ) and ( [ eq : sfp ] ) , we can eliminate the functional @xmath40 $ ] . inserting as a trial self - energy the self - energy of the reference system , then yields : @xmath47 = \omega_{{{\bm t } } ' , { { \bm u } } } + { \mbox{tr}}\ln ( { { \bm g}}_{0,{{\bm t}}}^{-1}-{{\bm \sigma}}_{{{\bm t}}',{{\bm u}}})^{-1 } - { \mbox{tr}}\ln { { \bm g}}_{{{\bm t}}',{{\bm u } } } \ : , \label{eq : ocalc}\ ] ] where @xmath48 and @xmath49 are the grand potential and the green s function of the reference system . stationary points are obtained , and this is the approximation , by restricting the variation to the subspace of trial self - energies @xmath46 : @xmath50 } { \partial { { \bm t } } ' } = 0 \qquad \mbox{for } \quad { { \bm t } } ' = { { \bm t}}'_{\rm s } \ : . \label{eq : stat}\ ] ] varying the trial self - energy means to vary the one - particle parameters @xmath42 of the reference system . for further details of the approach see ref . . here , we will focus on the hubbard model , eq . ( [ eq : hub ] ) , as the original model given by @xmath45 . different possible choices for @xmath44 and the corresponding systematics of approximations generated in this way are discussed in ref . . in the following , we will concentrate on two cluster approaches : ( i ) the variational cluster approach ( vca ) @xcite and ( ii ) the cellular dmft ( c - dmft ) . @xcite the vca can be seen as a variational generalization of the cluster - perturbation theory . @xcite it is obtained by partitioning the infinite lattice into disconnected ( identical ) clusters of @xmath51 sites each and choosing @xmath44 to consist of the intra - cluster parts only , i.e. the inter - cluster hopping is switched off in @xmath44 . the c - dmft is a cluster variant of the dmft . in the context of the sft , it is obtained in the same way as the vca but with an additional coupling of each of the @xmath51 correlated cluster sites to a continuous bath , i.e. to an infinite number of uncorrelated additional bath sites . the on - site energies of the bath sites as well as their coupling to the original sites are treated as variational parameters . once a reference system is specified , one should , in principle , vary _ all _ one - particle parameters of @xmath44 . this procedure would give the optimal result but requires a search for a stationary point in a high - dimensional parameter space . from a pragmatic point of view it is thus advisable to concentrate on a few parameters only which have to be selected by physical arguments . here , we argue that the variation of the on - site energies is important to achieve thermodynamic consistency with respect to the average particle number . in case of the hubbard model , this means to consider the site - independent energy @xmath52 as ( one of the ) variational parameter(s ) . the average particle number @xmath53 can be calculated in two different ways : on the `` zero - particle level '' by differentiation of the grand potential with respect to the chemical potential @xmath30 : @xmath54 and on the `` one - particle level '' by frequency integration of the one - particle excitation spectrum : @xmath55 here , @xmath56 is the local ( possibly spin - dependent ) spectral density with @xmath57 for short , and @xmath58 is the fermi function . for simplicity , we exclude non - collinear magnetic states and assume all expectation values to be diagonal in the spin index . thermodynamic consistency means that both ways of calculating @xmath53 yield the same result . since @xmath59 $ ] is the _ approximate _ sft grand potential at the stationary point @xmath60 , and since the spectral density or , equivalently , the one - particle green s function is the _ approximate _ green s function given by @xmath61 , the equivalence of ( [ eq : n1 ] ) and ( [ eq : n2 ] ) is by no means understood _ a priori_. to prove thermodynamic consistency , we start from eq . ( [ eq : n1 ] ) . according to eq . ( [ eq : ocalc ] ) , there is a twofold @xmath30 dependence of @xmath62 $ ] : ( i ) an _ explicit _ @xmath30 dependence due to the chemical potential in the free green s function of the original model , @xmath63 , and ( ii ) an _ implicit _ @xmath30 dependence due to the @xmath30 dependence of the self - energy @xmath64 , the green s function @xmath65 and the grand potential @xmath66 of the reference system : @xmath67 note that the implicit @xmath30 dependence is due to the chemical potential of the reference system which , by construction , is in the same macroscopic state as the original system ( with the same temperature @xmath3 and the same chemical potential @xmath30 ) _ as well as _ due to the @xmath30 dependence of the stationary point @xmath68 itself . this is a subtlety which , however , can be ignored since @xmath69 for @xmath60 because of stationarity condition ( [ eq : stat ] ) . ( actually , only those elements of @xmath68 show up a @xmath30 dependence that are treated as variational parameters . according to the chain rule , however , the derivative of @xmath39 has to be performed just with respect to those elements , with a vanishing result due to the stationarity condition . ) the self - energy , the green s function and the grand potential of the reference system are defined as grand - canonical averages . hence , their @xmath30 dependence due to the grand - canonical hamiltonian @xmath70 is ( apart from the sign ) the same as their dependence on @xmath71 : @xmath72 , etc . consequently , we have : @xmath73 the first derivative is readily calculated : @xmath74 here , the contour @xmath75 encloses the first - order poles of the fermi function at @xmath76 in counter - clockwise direction . using cauchy s theorem , we can proceed to an integration over real frequencies . inserting into eq . ( [ eq : exim ] ) , we get : @xmath77 for @xmath60 . the first term on the right - hand side is just the expression for the average particle number given by ( [ eq : n2 ] ) . the second term on the right - hand side vanishes _ provided that _ the variational condition ( [ eq : stat ] ) is satisfied , i.e. _ provided that _ @xmath71 is included in the set of variational parameters . in this case one has thermodynamic consistency . if @xmath71 was not treated as a variational parameter but kept at the value given by the original system , @xmath78 , one would have a finite @xmath79 in ( [ eq : nn ] ) , and the two expressions ( [ eq : n1 ] ) and ( [ eq : n2 ] ) for the average particle number would yield different results . this completes the proof . ( [ eq : n1 ] ) for the average particle number of the lattice model can be compared with @xmath80 which gives the average particle number of the reference system . again , there are two ways to get @xmath81 : either as the derivative of the reference system s grand potential @xmath82 or by frequency integration of the reference system s spectral density @xmath83 . as the reference system is solved exactly , both ways must yield the same result . note , however , that @xmath84 in general . the above reasoning can straightforwardly be generalized to the off - diagonal ( @xmath85 ) elements of the one - particle density matrix @xmath86 . this is discussed in appendix [ sec : cicj ] . the effect of thermodynamic ( in)consistency is illustrated for the single - band hubbard model of eq . ( [ eq : hub ] ) in fig . [ checkn ] . the figure shows @xmath87 as a function of @xmath30 as obtained from vca calculations described in sec . [ sec : vca ] below . solid lines display the result obtained by frequency integration of the spectral density , eq . ( [ eq : n2 ] ) , while dashed lines show the result of the numerical @xmath30 derivative of @xmath39 , eq . ( [ eq : n1 ] ) . we compare the results obtained by considering @xmath71 as a variational parameter ( a ) , with the ones obtained by setting @xmath88 ( b ) . in case ( a ) the two curves coincide within numerical accuracy , as expected , while in case ( b ) a considerable discrepancy is observed . this discrepancy increases upon approaching the transition to the non - magnetic state , i.e. precisely in the interesting region , where it reaches about @xmath89 doping . filling @xmath87 as a function of chemical potential @xmath30 obtained by integration of the spectral density ( eq . ( [ eq : n2 ] ) , solid lines ) and via the derivative of @xmath39 ( eq . ( [ eq : n1 ] ) , dashed lines ) . results are obtained , respectively , by considering @xmath71 as a variational parameter ( a ) , and by setting @xmath88 ( b ) . calculations for the electron - doped case with @xmath90 , nearest - neighbor hopping @xmath91 and next - nearest - neighbor hopping @xmath92 . ] while thermodynamic consistency with respect to the particle number is an issue for most approximations within the sft , there is one exception : the cellular dmft . this is discussed in appendix [ sec : cdmft ] . in detail , the calculation proceeds as follows : we consider the @xmath5 single - band hubbard model of eq . ( [ eq : hub ] ) with nearest- ( @xmath93 ) and next - nearest ( @xmath94 ) neighbor hoppings and the hubbard @xmath15 . for the purpose of our qualitative description of the htsc , it is sufficient to take parameters which are typical for both hole- and electron - doped high-@xmath1 cuprates , i.e. @xmath95 and @xmath96 . the energy scale is set by choosing @xmath91 . different values of the parameters ( e.g. a change of @xmath15 or @xmath94 within @xmath97 ) or a third - nearest neighbor hopping have been also incorporated , checked and found not to qualitatively change our conclusions ( for example , concerning the weak phase separation detected in the electron - doped case ) . the hamiltonian of the reference system @xmath44 is given by a set of decoupled clusters of finite size . for an individual cluster , the hamiltonian reads : @xmath98 it consists of the hubbard hamiltonian confined to the finite cluster @xmath99 plus two symmetry - breaking terms ( weiss fields ) @xmath100 and @xmath101 with @xmath102 and @xmath103 where @xmath104 is the strength of the staggered and @xmath105 the strength of the nearest - neighbor @xmath0-wave pairing field . @xmath106 is the af wave vector , and @xmath107 denotes the @xmath0-wave form factor which is non - vanishing for nearest - neighbor lattice sites only and is equal to @xmath108 ( @xmath109 ) for @xmath110 in x ( y ) direction . the sum in eq . ( [ hsc ] ) is restricted to sites @xmath13 and @xmath111 belonging to the same cluster . according to the discussion in sec . [ sec : n ] , the site - independent energy @xmath112 shall be treated as a variational parameter . it shows up in the local term @xmath113 which is already included in @xmath99 . the optimization of @xmath71 has to be done simultaneously with the optimization of the parameters @xmath104 and @xmath105 . due to the optimization of the weiss field strengths @xmath104 and @xmath105 , one can account for spontaneous af and @xmath0-wave sc symmetry breaking . the respective af and sc order parameters , @xmath114 and @xmath115 , are defined as @xmath116 in the limit @xmath117 where @xmath118 are the strengths of external _ physical _ staggered and pairing fields . these physical fields should not be confused with the ficticious weiss fields with strengths @xmath104 and @xmath105 . adding the respective physical field terms to the hamiltonian @xmath45 and performing the derivative with respect to @xmath119 of the sft grand potential ( at the respective optimal ficticious field strengths @xmath120 ) , yields @xmath114 and @xmath115 _ consistently _ with their representations ( on the `` one - particle level '' ) as frequency integrals over the usual and anomalous one - particle spectral density . this consistency is shown in the same way as in sec . [ sec : n ] for the average particle number and is a consequence of treating the ficticious fields as variational parameters . the quality of the approximation is decisively influenced by the cluster size used . on the one hand , for an appropriate characterization of the phase transition ( as considered in sec . iv , below ) , one needs a sufficient accuracy in the grand potential @xmath39 . this accuracy is , first of all , determined by the requirement that the clusters chosen must be large enough to fully account for the `` short - range '' spatial dependence of the self - energy as has been discussed below eq . ( [ eq : sf ] ) . on the other hand , for a given cluster size , an as accurate as possible numerical evaluation has to be employed . with respect to the latter , we found it both convenient and accurate to evaluate the frequency integrals contained in the trace in eq . ( [ eq : ocalc ] ) not by numerical integration ( where the required accuracy is difficult to achieve ) , but by converting these integrals to a sum over the poles of the green s function ( see ref . for details ) . this , however , requires the computation of all many - body eigenstates in a given sector of the cluster hamiltonian @xmath121 ( eq . ( [ eq : ham ] ) ) including symmetry - breaking fields . this technically limits the cluster sizes to be considered . therefore , in the present work , we have chosen an infinite lattice tiled with @xmath122 clusters . cluster consisting of @xmath122 sites , as well as larger cluster sizes have recently been systematically studied by kyung et al . @xcite in their influence on various physical quantities of the @xmath5 hubbard model by means of the c - dmft . already the smallest , i.e. @xmath122 , cluster has been found to account for more than 95% of the `` correlation effect '' in the single - particle spectrum ( for a precise definition see eq . ( 16 ) of ref . ) . this suggests that at least some of the relevant questions in a strongly correlated electron system , modelled by a hubbard - type hamiltonian , may be studied , even rather accurately , with a such a small reference cluster . nevertheless , for a more accurate determination of the sc state and of the phase transition , larger clusters have to be considered eventually . only in this way , phase - fluctuation effects between different d - wave cooper pairs ( which determine the spatial dependence of the self energy in the sc state ) can be accounted for . we expect , for example , that calculations for larger clusters would display a smaller sc gap than the one seen in fig . [ spectrelhol]b around @xmath123 in the electron - doped case . on the other hand , we expect our results on the global phase diagram to be quite robust . this is corroborated by the only weak dependence on ( moderate ) model changes , such as the next - nearest hopping or the hubbard interaction.@xcite and @xmath124 ) case at @xmath90 ( @xmath91 ) . for comparison with the qmc data , the next - nearest - neighbor hopping is set to @xmath125 . , title="fig : " ] + and @xmath124 ) case at @xmath90 ( @xmath91 ) . for comparison with the qmc data , the next - nearest - neighbor hopping is set to @xmath125 . , title="fig : " ] + the evolution of the single - particle excitation spectrum @xmath126 as a function of hole doping , and , in particular , the transition from a `` small '' fermi surface ( hole pockets around ( @xmath127 ) ) to an lda - like fermi surface closed around @xmath128 are key observations in unlocking the mystery of the cuprates . earlier qmc calculations for the @xmath5 hubbard model @xcite found that the single - particle spectral weight @xmath21 semi - quantitatively reproduces both the momentum ( @xmath129-symmetry ) and , in particular , doping dependence of the `` high - energy '' pseudogap of the order of the exchange energy @xmath130 as found in photoemission experiments around ( @xmath131 ) . the corresponding qmc data are reproduced in fig . [ qmc ] for comparison with the results of the vca shown in fig . [ spectrtp0 ] for hole dopings @xmath132 and @xmath124 . we first discuss the qmc data . in the under - doped regime , the `` pseudogap '' feature near ( @xmath131 ) moves to lower binding energy as doping is increased . at about @xmath133 ( full circle in fig . [ qmc ] ) the pseudogap vanishes in overall accordance with the experimental findings . @xcite in the experiments , the `` high - energy '' pseudogap is identified with the centroids of spectral weight near ( @xmath131 ) . however , the experimental arpes spectra also display a `` low - energy '' pseudogap in the normal ( superconducting ) state above ( below ) @xmath1 with energy @xmath134 , inferred from the leading edge in the spectral density , which also opens up in the under - doped regime and vanishes in the over - doped regime . @xcite this empirical correlation between the disappearance of the `` high - energy '' pseudogap and the decrease of the sc gap and , therefore , pairing strength suggested already several years ago that the high - energy features at ( @xmath131 ) ( which are correctly described by the `` high - temperature '' qmc simulations ) are closely related to the pairing interaction . @xcite it is clear , however , that there is a strong need to perform calculations at much lower temperatures , i.e. temperatures below @xmath135 . only then the `` low - energy '' sc or normal state pseudogap can be detected and only then the question `` where do holes enter first ? '' can be answered correctly . according to the results of the `` high - temperature '' ( @xmath136 ) qmc simulations shown in fig . [ qmc ] for @xmath132 , holes enter first around ( @xmath137 ) . this can be understood by referring to the idea @xcite that a higher temperature @xmath3 effectively acts as an increased doping which destroys the magnetic brillouin zone . this allows holes to first enter into the `` arc '' of single - particle excitations spanned around @xmath106 . and @xmath92 ( @xmath91 ) in the hole - doped ( a ) and in the electron - doped ( b ) case . results are shown for dopings in the mixed af+sc state ( see sec . [ gspd ] ) , i.e. for @xmath138 ( a ) and @xmath139 ( b ) , respectively . , title="fig : " ] and @xmath92 ( @xmath91 ) in the hole - doped ( a ) and in the electron - doped ( b ) case . results are shown for dopings in the mixed af+sc state ( see sec . [ gspd ] ) , i.e. for @xmath138 ( a ) and @xmath139 ( b ) , respectively . , title="fig : " ] however , a portion of the corresponding `` large '' fermi surface seems to disappear already at the lower temperature @xmath140 . this is indicated in fig . [ qmc](a ) by the downturn of the quasi - particle - like band between ( @xmath131 ) and ( @xmath137 ) . in the experiment ( fig . [ qmc](b ) ) , this behavior in the under - doped regime has been interpreted as the opening of a pseudogap in the underlying fermi surface near the ( @xmath141 ) to ( @xmath137 ) line . @xcite this opening is obviously intimately related to the above question , namely where doped holes first enter , i.e. to the possibility of hole pockets developing at very small temperatures and low dopings around ( @xmath127 ) in the hole - doped case . in contrast to the finite-@xmath3 qmc result , the corresponding vca calculations for @xmath4 ( see fig . [ spectrtp0 ] ) show that holes first go into the coherent band around ( @xmath142 ) forming `` hole pockets '' consistent with experiments . @xcite note that for the dopings considered in fig . [ spectrtp0 ] , the system is in a mixed af+sc state with non - vanishing af order parameter ( see also fig . [ phas - hole ] ) . therefore , although the low - lying spectral weight is found around @xmath143 , the spectrum still shows an af gap at this wave vector . related to this , it appears that even for @xmath124 the quasi - particle - like dispersion does not cross the fermi energy along the nodal direction ( between ( @xmath144 ) and ( @xmath145 ) in contrast to the experimental situation in fig . although some ( weak ) accumulation of spectral weight close and above @xmath146 can be seen in fig . [ spectrtp0 ] , we believe that this non - crossing is ( partly ) a shortcoming of our @xmath147 calculation and that the self - energy in the nodal direction requires larger clusters . to address this issue for more realistic model parameters , i.e. including a next - nearest - neighbor hopping @xmath148 , as well as to study the corresponding doping evolution in the electron - doped case , we discuss the vca results for @xmath126 shown in fig . [ spectrelhol ] . the calculations have been performed for the hole- ( a ) and for the electron - doped system ( b ) in the mixed af+sc phase ( see sec . [ gspd ] ) , i.e. for @xmath149 and @xmath139 , respectively . let us concentrate on the very small ( @xmath150 ) hole doping first . in agreement with the corresponding experiments @xcite in hole - doped cuprates , holes indeed first enter at ( @xmath127 ) . although the system is in a mixed af+sc phase ( see sec . [ gspd ] ) , the sc gap is zero at this nodal point , so that doping into nodal states , i.e. normal metallic screening , apparently destabilizes the af solution already for rather low doping values ( @xmath151 ) . this explains the phase diagram of fig . [ phas - hole ] . evolution of the low - energy spectrum upon hole doping . parameters are as in fig . [ spectrelhol ] . the weight is obtained by integrating the low - energy @xmath152-spectrum down to @xmath153 below the fermi energy . ] the same as in fig . [ fermis - holdop ] but for the electron - doped case . ] a similar picture can be inferred by looking at the evolution of the fermi surface as a function of doping . this can be extracted from fig . [ fermis - holdop ] where the low - energy spectral weight is plotted in the brillouin zone . in qualitative agreement with experiments , hole pockets start forming around @xmath143 for low doping , while a large fermi surface centered about @xmath128 starts building up at higher doping . in contrast , in electron - doped systems , doped electrons initially form pockets around ( @xmath131 ) ( see figs . [ spectrelhol](b ) and [ fermis - eldop ] , see also refs . ) , in agreement with experiments . @xcite fig . [ fermis - eldop ] shows the spectral weight obtained by integrating @xmath154 down to @xmath153 below the fermi energy ( @xmath146 ) . one has to be careful when interpreting fig . [ fermis - eldop ] , since for @xmath155 there is still an af gap at the fermi surface near @xmath143 besides the sc one near @xmath123 . therefore , the excitation spectrum is completely gapped , and one has to go away from @xmath146 in order to find some weight . since we integrate only in a small energy window around the fermi level , the scale in fig . [ fermis - eldop ] is an order of magnitude smaller than in the hole - doped case , fig . [ fermis - holdop ] , for which the af gap near @xmath143 is shifted away from the fermi surface . nevertheless , one finds that the lowest - lying states in the electron - doped case are around @xmath123 . here , the density of states is large and provides a large `` reservoir '' for electron doping . this , in combination with the fact that the chemical potential lies in the sc gap ( of the af+sc phase ) , stabilizes the af solution for a larger doping range than in the hole - doped case , allowing for the af gap to decrease more gradually ( see sec . [ gspd ] ) . also in this case , the breakdown of the magnetic solution occurs as soon as the chemical potential reaches the bottom of the band at ( @xmath127 ) . the observation that doping into ( @xmath127 ) generically makes the af phase unstable , suggests that also the occurrence of phase separation , to be discussed in sec . [ gspd ] , is rather generic and quite independent of model details . the low - energy spectral weight , as plotted for different dopings in fig . [ fermis - eldop ] , agrees quite well with the arpes measurements of armitage et al . @xcite ; doped electrons away from half - filling first form pockets around @xmath123 . at higher doping , fermi - surface segments start developing along the @xmath156 line . finally , these segments connect and form a large fermi surface around @xmath128 near optimal doping . in this section the ground - state phase diagram and the af to sc transition are presented and discussed ( see also refs . ) . in particular , we will focus on the similarities and differences between the hole- and electron - doped system as well as their relation with the evolution of the single - particle spectrum . in order to determine the @xmath4 phase diagram , we proceed as described in sec . [ sec : vca ] , i.e. two symmetry - breaking terms ( weiss fields ) @xmath100 and @xmath101 are included and their respective strengths , @xmath104 and @xmath105 are treated as variational parameters in addition to @xmath71 . as discussed in sec . [ sec : n ] , the use of the additional variational parameter @xmath71 is required in order to have a consistent determination of the particle density . the phase diagram for the hubbard model with @xmath90 and next - nearest - neighbor hopping @xmath92 ( we set @xmath91 ) as obtained from our calculations , is shown in fig . [ phas - hole ] for the hole - doped and in fig.[phas - elec ] for the electron - doped case . in the upper part of each figure , the chemical potential @xmath30 is plotted as a function of @xmath157 . in the corresponding lower parts , we display the af @xmath158 and sc @xmath159 order parameter as a function of doping @xmath157 . note that the order parameters @xmath114 and @xmath115 are different from the weiss fields @xmath104 and @xmath105 , respectively . quite generally , however , a nonvanishing stationary value for a weiss field produces a nonvanishing order parameter , respectively , although the latter can be much smaller . antiferromagnetic and superconducting order parameters , @xmath114 and @xmath115 , and chemical potential @xmath30 as functions of hole doping @xmath157 . @xmath115 and @xmath30 are plotted for the af+sc ( green , @xmath160 , @xmath161 ) as well as for the pure sc homogeneous solutions ( blue , @xmath162 , @xmath163 ) . note that @xmath115 is scaled by a factor 5 for convenience . for @xmath164 the system exhibits a coexistence of af and d - wave sc order . phase separation occurs between the doping levels @xmath165 and @xmath166 . for @xmath167 pure d - wave sc is realized . in the phase separation region @xmath168 , the homogeneous solution becomes unstable , and the system prefers to separate into a mixture of two densities corresponding to @xmath165 and @xmath166 . the chemical potential @xmath169 is determined by the maxwell construction shown in the upper figure . at @xmath170 the slope of the af+sc solution changes sign . ] same as fig . [ phas - hole ] but for electron doping . note the enhanced robustness of the af state and the strongly reduced scale @xmath171 as compared to hole doping . ] let us discuss hole doping first ( see fig . [ phas - hole ] ) . for low dopings @xmath157 we find a homogeneous symmetry - broken phase in which both the af as well as the sc order parameter @xmath114 and @xmath115 are non - zero . this corresponds to a phase where af and sc order microscopically and coherently coexist . a homogeneous phase with pure sc ( @xmath172 and @xmath173 ) is obtained for larger dopings . the behavior of @xmath114 vs. @xmath157 in fig . [ phas - hole ] seems to suggest that the transition to the non - magnetic state is continuous ( second order ) as a function of doping . however , a glance at the non - monotonous behavior of the chemical potential @xmath30 plotted as a function of @xmath157 in the upper part of the figure , indicates the occurrence of a charge instability . the system tends to separate into a hole - poor ( @xmath165 ) and a hole - rich ( @xmath166 ) phase . the two dopings @xmath165 and @xmath166 , as well as the chemical potential @xmath169 in the phase - separated region , are identified by the maxwell construction shown in the upper part of fig . [ phas - hole ] . @xmath170 is the point where the slope of @xmath174 changes sign . of course , this phase separation is obtained within a treatment that , in the present work , just considers for homogeneous order parameters only and that neglects surface effects . furthermore , the vca is a mean - field - type approach on length scales beyond the size of the individual cluster . therefore , the above result has to be interpreted with care . we expect it to signal a _ tendency _ towards the formation of microscopic inhomogeneities , such as stripes , checkerboard patterns , etc . @xcite this tendency will be further studied in future work by considering larger clusters and/or by allowing for a more general variational solution which explicitly describes stripe inhomogeneities ( for example , by considering coupled clusters with different dopings , as in ref . ) . in this case , one might expect the phase transition to become more continuous and phase separation to disappear eventually . let us now discuss the electron - doped case and the similarities and differences as compared to hole - doping . the first observation is that , while the phase diagrams in figs . [ phas - hole ] and [ phas - elec ] are _ qualitatively _ similar , the phase in which long - range af order is realized is spreading to significantly larger doping values in the electron - doped case , in overall agreement with the experimental situation . [ phas - elec ] shows that phase separation occurs in the electron - doped case as well , although the associated energy scale @xmath171 is smaller with respect to the hole - doped case by about an order of magnitude . in ref . it is argued that this can be related to the different pseudogap and sc transition scales in hole- and electron - doped materials . this may give support to theories @xcite which are based on the notion that fluctuations of competing phases , or of the related order parameters , are responsible for the pseudogap phenomenon . vs. @xmath104 ( with the two other parameters @xmath105 and @xmath71 fixed at their stationary - point values ) for different @xmath30 in the hole - doped case . parameters : @xmath90 , @xmath92 ( @xmath91 ) . ] depending on the value of @xmath30 , there may be two solutions , an af+sc and a pure sc one , corresponding to two stationary points of @xmath39 . in order to gain insight into the first - order transition between these two solutions , it is instructive to observe the behavior of the grand potential @xmath39 as a function of the variational parameter @xmath104 associated with this transition for different @xmath30 . for simplicity , we consider here the hole - doped case only . results are plotted in fig . [ omegamin ] . note that @xmath104 is varied , while the other two variational parameters , @xmath105 and @xmath71 , are fixed at their values at the respective stationary point . at low doping , there is a single minimum of the grand potential @xmath39 at a finite value @xmath104 only ( fig . [ omegamin ] , @xmath175 ) . in contrast , the paramagnetic ( @xmath176 ) solution is given by a local maximum . upon further doping , @xmath39 _ additionally _ develops a local minimum at vanishing af variational parameter @xmath176 ( fig . [ omegamin ] , @xmath177 ) . for @xmath178 , the minimum at @xmath179 becomes lower than the one at finite @xmath104 indicating a first - order phase transition to a non - magnetic state ( fig . [ omegamin ] , @xmath180 ) . eventually , the local maximum lying between the two minima merges with the minimum at finite @xmath104 and _ the af solution disappears _ ( fig . [ omegamin ] , @xmath181 ) . this disappearance just occurs when the chemical potential @xmath30 enters the quasi - particle band around @xmath143 ( @xmath182 ) . although fig . [ omegamin ] displays the results for the hole - doped case , this behavior occurs qualitatively in the electron - doped case as well , although with a much smaller characteristic energy scale as discussed above . again , these results reflect a qualitatively similar behavior for electron and hole doping . by means of a recently developed quantum - cluster approach , we have carried out a detailed analysis of the phase transition from the antiferromagnetic to the superconducting phase in the hubbard model at zero temperature . the main results concerning the nature of the af to sc transition are summarized in figs . [ phas - hole ] and [ phas - elec ] . at low dopings the af phase actually mixes with a weak @xmath0-wave sc component . a similar coexistence phase is observed , for example , in prcecuo , @xcite and was also obtained in previous mean - field and cluster calculations . @xcite upon further increasing doping , we find a transition to a pure @xmath0-wave sc phase . the phase - separation scenario , which is found for hole but also for electron doping within our approximation , should carefully be interpreted as a general _ tendency _ of the system to form microscopically inhomogeneous phases . indeed , by allowing for a more general spatial dependence of the order parameter , the macroscopic phase separation will probably be replaced by other microscopically inhomogeneous phases , such as stripes , checkerboard order , etc . in particular , this might be expected to be the case if long - range coulomb interaction is taken into account additionally . @xcite the situation is subtle in the electron - doped case . here , in contrast to previous theoretical calculations , our results also suggest phase separation , although the corresponding energy scale ( see fig . [ phas - elec ] ) is one order of magnitude smaller than in hole - doped compounds . our results for the single - particle excitations presented in sec . [ spe ] support the idea that it is , in fact , the single - particle spectrum which holds the key for understanding the qualitative differences seen in the ground - state phase diagram between the hole- and electron - doped cuprates . this concerns , for example , the robustness of the af state in the electron - doped case . here , electrons initially form pockets around ( @xmath131 ) in accordance with experiments . @xcite the fact that the density of states is large there , as well as the presence of a sc `` gap '' at this nodal point , stabilizes the af state for a larger doping regime as compared to the hole - doped case . the breakdown of the magnetic solution appears as soon as `` normal '' metallic screening sets in , i.e. when @xmath30 touches the band at ( @xmath142 ) . @xcite one should note that the variational cluster approach ( vca ) is able to treat the fluctuations correctly up to the range of the cluster size only . therefore , the question arises whether the sc solution we ( and also others @xcite ) obtain within the af phase is a true long - range sc phase or whether it is only a signal of strong pairing fluctuations within the af phase leading to a sc pseudogap . the latter hypothesis could be supported by the fact that results obtained with different cluster sizes @xcite seem to indicate a size dependence of the sc order parameter , and by the fact that the sc order parameter is about a factor three smaller in the af+sc phase than in the pure sc one . the presence or not of such a microscopic coexistence phase may depend on material details . certainly , our results suggest that the sc gap @xmath115 ( or pseudogap in the case of fluctuations ) is important in order to stabilize the af phase in electron - doped materials , as discussed in sec . [ spe ] . similar vca calculations have recently been carried out by snchal et al . @xcite using clusters up to 10 sites but without the variation of the on - site energies . these authors show that the single - band hubbard model is sufficient to explain the different overall shapes of the phase diagrams for hole- and electron - doped cuprates . however , their results seem to suggest that in the electron - doped case the af to sc transition is continuous and associated with a quantum - critical point , in contrast to our results . this clearly shows the importance for a consistent determination of the average particle number . the recent substantial progress in relating the `` high - energy '' physics of the hubbard model and its variants to the low - energy physics of competing phases is to a large extent due to the development of different quantum - cluster theories . apart from the vca , these are the cluster extensions of the dmft , such as the cellular dmft ( c - dmft ) and the dynamical cluster approximation ( dca ) . kyung et al.@xcite have shown that some of the important problems in strongly correlated electron systems may be studied highly accurately using comparatively small clusters . maier et al . @xcite performed a systematic cluster - size study of the @xmath5 hubbard model using rather large clusters ( up to @xmath183 sites ) . converged results suggest a finite-@xmath3 instability to @xmath0-wave sc state . because of the qmc minus - sign problem , however , results were limited to @xmath184 where the typical correlation energy @xmath15 and the magnetic energy scale of the htsc is not yet achieved . on the other hand , the present vca studies are clearly not yet converged with respect to the cluster size , as one can read off from , for example , the relatively large sc gap in the single - particle excitations , displayd in fig . [ fermis - holdop ] . an extension to larger clusters is , at least in principle , also possible within the vca . this , however , necessarily implies the use of stochastic ( qmc ) methods as solvers for the cluster reference system . we would like to thank d. j. scalapino for many useful discussions . one of us ( wh ) would additionally like to acknowledge the hospitality of the kavli institute for theoretical physics in santa barbara ( supported by nsf grant no . phy99 - 0794 ) . the work was supported by the dfg forschergruppe 538 , by the konwihr supercomputing network in bavaria and by the fwf project n. p18505-n16 . the reasoning in sec . [ sec : n ] can straightforwardly be generalized to the off - diagonal ( @xmath85 ) elements of the one - particle density matrix @xmath86 . in this case , thermodynamic consistency means that , for a selected pair of sites @xmath185 , the derivative @xmath186 is equivalent with the integral @xmath187 where , for convenience , the hopping is formally assumed to be spin - dependent . starting with eq . ( [ eq : hop1 ] ) , we note that the @xmath188 dependence of @xmath59 $ ] is due to the explicit @xmath188 dependence of the free green s function @xmath189 in eq . ( [ eq : ocalc ] ) , and due to the implicit @xmath188 dependence of @xmath68 : @xmath190}{\partial t_{ij\sigma } } \nonumber \\ & = & t \sum_{\omega_n } e^{i\omega_n 0^+ } \left ( \frac{1}{{{\bm g}}^{-1}_{0,{{\bm t}}}(i\omega_n ) - { { \bm \sigma}}_{{{\bm t}}'_{\rm s } , { { \bm u}}}(i\omega_n ) } \right)_{ji\sigma } \nonumber \\ & + & \frac{\partial \omega_{{{\bm t } } , { { \bm u } } } [ { { \bm \sigma}}_{{{\bm t}}',{{\bm u } } } ] } { \partial { { \bm t } } ' } \bigg|_{{{\bm t } } ' = { { \bm t}}'_{\rm s } } \cdot \frac{\partial { { \bm t}}'_{\rm s}}{\partial t_{ij\sigma } } \ : .\end{aligned}\ ] ] because of the stationarity of the grand potential , the second term on the r.h.s . can be ignored . we are thus left with the first term which , after transforming the matsubara sum into an integration over real frequencies , exactly yields the desired expression ( [ eq : hop2 ] ) . in case that not all elements of @xmath42 are treated as variational parameters , the derivative of @xmath39 in the second term on the r.h.s . is a derivative with respect to those elements only . if , for the given pair of sites @xmath13 and @xmath111 , @xmath191 is selected as a variational parameter , the contribution due to the implicit @xmath188 dependence vanishes again . if , on the other hand , @xmath191 is not treated as a variational parameter but fixed at @xmath192 from the beginning , the second term will give a finite contribution which spoils the thermodynamic consistency . the c - dmft is obtained by choosing as a reference system disconnected clusters with a continuous non - interacting bath coupled to each cluster site . carrying out the @xmath42 partial derivatives in eq . ( [ eq : stat ] ) and using eq . ( [ eq : ocalc ] ) for the sft grand potential , we get the euler equation in the form : @xmath193 here @xmath13 and @xmath111 run over the sites of the original lattice , excluding bath sites . the one - particle bath parameters , namely the on - site energies of the bath sites and the hybridization of the bath sites with the correlated ( physical ) sites , have to be treated as ( a continuous set of ) variational parameters . in the c - dmft it is assumed that bath parameters can be found such that the first factor in ( [ eq : euler ] ) vanishes : @xmath194 for arbitrary @xmath195 and for sites @xmath13 and @xmath111 belonging to the same cluster . note that , by construction of the reference system , @xmath196 and also @xmath197 vanish if @xmath13 and @xmath111 belong to different clusters . consequently , if bath parameters can be found such that eq . ( [ eq : cdmft ] ) holds , the euler equation ( [ eq : euler ] ) will be satisfied , too : the self - energy functional is stationary at the c - dmft self - energy . ( [ eq : cdmft ] ) is just the self - consistency equation of the c - dmft ( see refs . ) . it is easy to see that @xmath198 within the c - dmft : this simply follows by comparing eqs . ( [ eq : nn ] ) and ( [ eq : nnp ] ) for @xmath53 and @xmath81 and by using the c - dmft self - consistency equation ( [ eq : cdmft ] ) . consider now the high - frequency expansions of the green s function and of the self - energy of the reference system . using @xmath199 , @xmath200 , and @xmath201 for short , we have ( see ref . for high - frequency expansions in the hubbard model ) : @xmath202 here , @xmath203 is the average occupation in the reference system . using @xmath204 , inserting the expansions into the c - dmft self - consistency equation ( [ eq : cdmft ] ) , and expanding in powers of @xmath205 once more , we immediately find : @xmath206 for ( correlated ) sites @xmath13 and @xmath111 within the same cluster , and , in particular , @xmath78 where @xmath207 and @xmath208 . this means that , within the c - dmft , consistency with respect to the particle number is assured by setting the on - site energies within a cluster of the reference system to their `` physical '' values . it goes without saying that a consistent determination of @xmath53 requires an _ exact _ solution of the reference system which , in case of the c - dmft , is by no means trivial . an obvious idea is to replace the continuous bath by a few uncorrelated sites only to allow for an application of the lanczos method ( see ref . , for an example ) . this , however , immediately implies that the self - consistency equation ( [ eq : cdmft ] ) can not be fulfilled exactly any longer and , strictly speaking , the determination of the particle number becomes inconsistent _ unless _ @xmath71 is treated as a variational parameter within the sft framework . c. dahnken , m. aichhorn , w. hanke , e. arrigoni and m. potthoff , phys . b * 70 * , 245110 ( 2004 ) ; c. dahnken , m. potthoff , e. arrigoni , and w. hanke , to appear in fizika nizkikh temperatur ( cond - mat/0504618 ) . n. p. armitage , f. ronning , d. h. lu , c. kim , a. damascelli , k. m. shen , d. l. feng , h. eisaki , z .- x . shen , p. k. mang , n. kaneko , m. greven , y. onose , y. taguchi , and y. tokura , phys . rev . * 88 * , 257001 2002 . n. p. armitage , d. h. lu , c. kim , a. damascelli , k. m. shen , f. ronning , d. l. feng , p. bogdanov , z. x. shen , y. onose , y. taguchi , y. tokura , p. k. mang , n. kaneko , and m. greven , phys . * 87 * , 147003 2001 .

a self - consistent description of electrons , magnons , phonons and other fermionic and bosonic degrees of freedom in solids is still a challenging problem for strongly correlated systems where the simplest approximations are usually not sufficient . one of the classical examples is the theory of itinerant electron magnetism , where the d - electrons form both the correlated energy bands as well as the quasi - localized magnetic moments and the magnon spectrum @xcite . in this problem the conventional random phase approximation ( rpa ) gives a wrong description of the finite temperature effects whereas the proper corrections to rpa can be successfully calculated only in a very close vicinity of the stoner instability point @xcite . the main difficulty is the coexistence of localized ( atomic - like ) and itinerant features in the behavior of d - electrons in solids @xcite . recent progress in the theory of correlated electron systems is related to the development of dynamical mean field theory ( dmft ) @xcite , which maps the problem of fermions with local interactions on a crystal lattice onto a self - consistent effective impurity problem . the latter can be solved very efficiently within continuous time quantum monte carlo ( ct - qmc ) @xcite or exact diagonalization ( ed)@xcite schemes . as a result , both atomic - like features such as multiplet formation and band dispersion can be taken into account simultaneously @xcite . the combination of the dmft approach with realistic electronic structure calculations @xcite , in particular , for transition metal systems @xcite opens a new way for the description of electronic degrees of freedom in solids . a generalization of the dmft approach to non - local interactions and consistent description of collective bosonic excitations @xcite boost first realistic applications of a so - called gw+dmft approach @xcite . here we construct a theory , that is capable to describe systems with strongly correlated electrons interacting with collective bosonic modes . these modes can appear explicitly in an original hamiltonian ( for instance , phonons ) or arise from electron - electron interactions ( like plasmons or magnons ) . we assume that the bose excitations can not be considered as an ideal bose gas and boson - boson correlations are important . among numerous examples of such systems are the cases of weak itinerant ferromagnets @xcite and , more generally , any materials near quantum phase transition points @xcite . the feynmann diagrammatic technique became the most popular tool in the quantum many body theory as a compact and elegant representation of a perturbation expansion in powers of interaction @xcite . the main advantage of the diagrammatic approach is an opportunity to sum up infinite series of relevant diagrams going beyond the formal perturbation expansion . the mainstream paradigm is that summing up more sophisticated series expands the area of applicability . in some cases leading series of diagrams can be separated and summed up with the use of small parameters which are not directly related to the interaction strength . important examples are the gell - mann bruekner theory of the high - density electron gas @xcite , the gaseous approximations of low density fermi @xcite and bose @xcite systems , the @xmath0-expansion in theory of critical phenomena @xcite , and the @xmath1 expansion in the theory of magnetic impurities @xcite . historically , the dmft was introduced as a leading order approximation in a formal parameter @xmath2 , where @xmath3 is the dimensionality of the space . dmft is exact in the limit of @xmath4 , at the same time attempts to construct a regular expansion in @xmath2 turn out to be not practical @xcite . nevertheless , dmft appears to be quite a good approximation for many three- and even two - dimensional systems @xcite , since it provides reasonable interpolations between non - interacting and atomic limits for any dimensions . in terms of the diagrammatic approach the dmft corresponds to a summation of all local in space contributions to the self - energy @xcite . recently we developed the dual fermion scheme @xcite which introduces new variables in a path integral generating an alternative diagram technique such that the bare propagators ( free dual fermions ) correspond to dmft . this means that the main local physics is taken into account by the change of variables while the new diagrams describe only non - local corrections to the self - energy . interaction terms in the dual variables look more complicated than the original ones : the interactions become retarded and contain not only pair terms but also multi - fermion contributions . however , an advantage of dual fermion transformation is a faster convergence of the diagram series . for example , a simple ladder summation gives quite accurate description for the single band hubbard model in strongly correlated regime @xcite . thus , the dual fermion approach is an example of an alternative strategy in the quantum many - body theory , when a proper choice of variables allows one to consider a simpler series of diagrams . another example of such strategy is the use of auxiliary fermions and bosons in the hubbard model which allows one to describe the subtle effect of the growth of the effective mass near metal - insulator transitions , already on the level of the mean - field @xcite . the consideration of collective ( bosonic ) degrees of freedom in systems of interacting fermions within the conventional technique necessarily involves summations of sophisticated diagrams . the dispersion law of bosonic excitations ( e.g. , magnons or plasmons ) is determined by the poles of the corresponding susceptibilities ( the two particle fermionic green s functions ) and the minimal approximations which give rise to these poles , even in the case of a weak interaction corresponding to the ladder summation ( random phase approximation , rpa @xcite ) . in order to take into account magnon - magnon interactions one needs to consider more complicated , parquet - like diagrams @xcite . for strongly correlated systems one can consider the same procedure within the dual fermion approach . up to now this is only done at the level of ladder summations @xcite . as we pointed out , the above considerations of complicated diagrams can be replaced by a proper choice of variables in the path integral . following this strategy we propose an extension of dual fermion scheme explicitly adding bosonic variables . it is important to note that the present theory is not a direct extension of the dual - fermion approach @xcite . we will see that , even for the hubbard - like models with on - site interactions only , it gives a different result because of the account of collective modes ( magnons ) . a simultaneous description of fermionic and bosonic correlations on the equal footing was started within the extended dynamical mean field ( edmft ) framework @xcite . the strong - coupling generalization of edmft @xcite gives a useful extension of the hubbard - i like perturbation expansion for lattice models @xcite . edmft properly includes the physics of _ local _ fermionic and bosonic correlations . at the same time , a dispersion of collective excitations within edmft is determined by the non - local part of the interactions _ only_. in particular , to describe the dispersive magnons within edmft one needs to explicitly add direct heisenberg exchange interactions to the hamiltonian . obviously , this is not sufficient since the _ indirect _ exchange interactions ( superexchange , double - exchange , rkky , etc . ) contribute as well to the the magnon dispersion . moreover , typically , they give the main contribution . in the present paper we construct a formally exact diagram technique where the bare fermionic and bosonic propagators correspond to edmft , whereas the indirect exchange is given by diagrammatic corrections . therefore , the direct and indirect interactions enter the magnon dispersion law on equal footing . we consider a lattice fermionic problem with the intersite hopping and generic non - local interactions . for simplicity , we present only formulae for the single - band periodic lattice , but a generalization of all expressions to a multiorbital case is straightforward . the action is @xmath5+\sum_{r , r\neq 0 , \omega , \sigma } \varepsilon_r c^\dag_{r \omega \sigma } c_{r+r \omega \sigma } + \sum_{r , r\neq 0 , \omega } v_{r \omega } \rho_{r \omega}^ * \rho_{r+r \omega}\ ] ] here @xmath6 are grassmann variables , the indices @xmath7 run over the lattice sites and translations , @xmath8 stands for the wave - vector , @xmath9 is the imaginary time , @xmath10 and @xmath11 are , respectively , the fermionic and bosonic matsubara frequencies ( @xmath12 is integer and @xmath13 is the inverse temperature ) , and @xmath14 is the spin index . prefactors @xmath15 and @xmath16 are implied for sums over frequencies and wavevectors , respectively ( @xmath17 is the number of lattice sites ) . the `` bosonic variable '' @xmath18 in general is a vector quantity , whose components for single band models can be the charge - density or the spin - density operators : @xmath19 , where @xmath20 with @xmath21 are pauli matrices ( @xmath22 ) , and the quantity @xmath23 is introduced to satisfy the condition @xmath24 . to simplify the notations , we skip the index @xmath25 throughout the paper , but one should bear in mind that relevant components of @xmath18 are to be accounted in a practical calculation . for example , if the physics is determined by charge fluctuations only ( as may be the case for the coulomb interaction or an electron - electron attraction mediated by phonons ) , only the density - density term @xmath26 is important in the interaction part @xmath27 of eq.([lattice ] ) . the heisenberg isotropic exchange interaction corresponds to @xmath28 , etc . an important case of the hubbard model corresponds to @xmath29 , where the spin degrees of freedom @xmath30 do not show up in the bare expression part of eq.([lattice ] ) , but will appear in the following formulation , since the low - energy behaviour of system is governed by the spin fluctuations . within the path integral representation the atomic action @xmath31 is given by : @xmath32 where @xmath33 is the local coulomb interaction . we are interested in the single - electron green s function @xmath34 and the bosonic correlator @xmath35 . it is an important point that an account of the bosonic quantity is fundamental : we suppose that collective modes are relevant for the physics of the system . the general idea of effective - medium approximations is to reduce the lattice problem ( [ lattice ] ) to the so - called impurity model , describing a single atom ( or small cluster ) in a gaussian environment . for a system with bosonic and fermionic degrees of freedom , a single - site impurity problem is described by the action @xmath36 obtaining a numerical or analytical solution for the impurity problem is supposed to be much simpler . we later assume that its green s functions : @xmath37 and @xmath38 are known , as well as higher - order correlators . in practice , they can be calculated by a ct - qmc scheme @xcite or by exact diagonalization if the dimension of the hilbert space is not too large . normally @xmath39 is called the full polarizability . we will call it the bosonic green s function , to emphasize the formal analogy between the equations for @xmath40 and @xmath39 in our formalism . for the so - called local approximations , only the green s functions of the impurity model @xmath41 are taken into account . in a nutshell , the initial problem ( [ lattice ] ) is replaced by an auxiliary lattice with gaussian nodes ; those nodes are characterized by the green s functions @xmath41 . such a consideration gives us the well - known expressions for lattice green s functions @xcite , which are denoted with calligraphic letters in this paper : @xmath42 where @xmath8 is the wave vector and fourier transforms of the hopping and interaction parameters are introduced in the obvious way . one can see that the fermionic and bosonic self - energies of these expressions are @xmath8-independent , that is , local in space . the hybridization functions @xmath43 and @xmath44 should be defined in such a way , that the impurity problem provides us the on - site ( local ) properties of the solution for the lattice problem . such a requirement is physically quite reasonable . also , it is closely related to feynman s variational approach in dual fermion variables @xcite . most common are so - called static and dynamic mean - field theories . in the static approach , the hybridization functions @xmath45 and @xmath46 are chosen to be frequency independent . their values are defined from the equalities for the _ static parts _ of green s functions : @xmath47 in the dynamical edmft approach @xmath48 and @xmath49 are adjusted so that local parts of the lattice green s functions coincide with the corresponding quantities of the impurity problem _ for all frequency ( or time ) arguments _ : @xmath50 it is worthwhile to note that the standard definition of the edmft @xcite introduces an auxiliary bosonic field called @xmath51 , and the second line of eq.([dmftcondition ] ) is re - expressed as a condition for the green s function @xmath52 of that bosonic field . at the edmft level , the conditions for @xmath53 and @xmath52 are equivalent . we discuss this issue in detail in appendix and show that similar ( although not the same ) bosonic field can be introduced in non - local approximations beyond the edmft . some `` mixed '' local approximations can also make sense . for instance , if the hopping is absent ( or negligible ) , it is natural to consider a scheme with @xmath54 and to adjust bosonic hybridization only . contrary , @xmath55 corresponds to the `` purely fermionic '' dmft scheme @xcite . it is easy to make a connection between the lattice and impurity actions : @xmath56+\sum_{\omega , k , \sigma } \left ( \varepsilon_k - \delta_{\omega\sigma } \right ) c^\dag_{\omega k \sigma } c_{\omega k \sigma } + \sum_{\omega , k } \left ( v_{\omega k}-\lambda_{\omega } \right ) \rho_{\omega k } ^ * \rho_{\omega k}\ ] ] further on , we use a general matrix form of the hubbard - stratonovich transformation to decouple the term @xmath57 with @xmath58 into the pair of dual fermionic grassmann - fields @xmath59 and the term @xmath60 with @xmath61 into the pair of dual bosonic complex - fields @xmath62 ( a matrix notation is used to shorten the expressions ) : @xmath63e^{c^\dag e c } = \int d\left[c^\dag , c \right ] \det\left(\alpha^{-1}_f e \alpha^{-1}_f \right ) \int d\left[f^\dag , f \right ] e^{-f^\dag \alpha_f e^{-1 } \alpha_f f + f^\dag \alpha_f c + c^\dag \alpha_f f } \\ \\ \int d\left[\rho^ * , \rho \right]e^{\rho^ * w \rho } = \int d\left[\rho^ * , \rho \right ] \det\left(\alpha_b w^{-1 } \alpha_b \right ) \int d\left[\eta^ * , \eta \right ] e^{-\eta^ * \alpha_b w^{-1 } \alpha_b \eta + \eta^ * \alpha_b \rho + \rho^ * \alpha_b \eta } , \end{array}\end{aligned}\ ] ] where @xmath64 are arbitrary , but local in space scaling matrices . note that whereas the decoupling of grassmanian variables is mathematically straightforward , more comments are needed on the second line of eq.([hs ] ) . the first comment is that @xmath18 is hermitian , so that @xmath65 , and the integral @xmath66 $ ] is taken in fact over the variables from a half - space . the dual variables inherit this property , @xmath67 . the second comment is that the matrix @xmath68 is not positively defined , and integrals over @xmath69 $ ] should be re - defined to ensure convergence . here we describe the corresponding procedure , following , in particular , refs . @xcite . for illustration purposes , we suppose @xmath70 to be just a real constant factor ; the generalization to a complex matrix @xmath70 is straightforward . let us introduce a basis where @xmath71 is diagonal . let us number the states in this basis with an index @xmath72 , and consider some matrix element @xmath73 . for a positive real part of @xmath74 , the corresponding part of the action can be decoupled as follows : @xmath75 where we explicitly introduced the hermitian and anti - hermitian parts of @xmath76 . one can see that for @xmath77 and @xmath78 being interpreted as real and imaginary parts of a complex number @xmath79 , this decoupling becomes an explicit form of the second line of eq . ( [ hs ] ) , written for a particular state @xmath72 . the integral limits for positive @xmath74 can be chosen along the real axis , from @xmath80 to @xmath81 for both @xmath77 and @xmath82 . the key observation is that for @xmath74 having a _ negative _ real part , the decoupling ( [ hsplus ] ) still holds , up to the integration path along imaginary axis , from @xmath83 to @xmath84 for @xmath77 and in the opposite direction , that is @xmath84 to @xmath83 , for @xmath82 . thus , the integration over @xmath85 $ ] can be considered as a symbolic notation for the integration over @xmath86 $ ] with @xmath87 substituted in the integrand . the integration path over each component of vectors @xmath88 is explicitly defined for the basis where @xmath89 is diagonal : the integration goes along the real or imaginary axis , corresponding to the sign of the real part of the diagonal matrix element @xmath74 . one can check that such a definition of @xmath90 $ ] still allows one to use the standard algebraic manipulations of path integrals . in particular , to obtain the green s function of dual bosonic variables @xmath91 one uses the standsrd procedure of differentiation with respect to @xmath92 , as can be explicitly shown from eq . ( [ hsplus ] ) . on the other hand , the statistical properties of dual variables can be unusual because of the imaginary axis integration path . for instance , similarly to the fermionic dual green s function @xcite , @xmath93 is not necessary positive - defined and consequently the imaginary part of its fourier transform is not always negative . let us continue with the derivation of the dual formalism . the hubbard - stratonovich decoupling in the partition function with the action ( [ lattice ] ) gives ( in matrix form ) @xmath94e^{-\sum_{r}s_{imp}[c_{r}^{\dag } , c_{r}]+c^{\dag } ( \delta -\varepsilon ) c+\rho ^{\ast } ( \lambda -v)\rho } = \det ( \alpha _ { f}^{-1}(\delta -\varepsilon ) \alpha _ { f}^{-1})\det ( \alpha _ { b}(\lambda -v)^{-1}\alpha _ { b})\times \\ \\ \times \int d[c^{\dag } c]\int d[f^{\dag } f]\int d[\eta ^{\ast } \eta ] e^{-\sum_{r}s_{imp}[c_{r}^{\dag } , c_{r}]+f^{\dag } \alpha _ { f}c+c^{\dag } \alpha _ { f}f+\rho ^{\ast } \alpha _ { b}\eta + \eta ^{\ast } \alpha _ { b}\rho -f^{\dag } \alpha _ { f}(\delta -\varepsilon ) ^{-1}\alpha _ { f}f-\eta ^{\ast } \alpha _ { b}(\lambda -v)^{-1}\alpha _ { b}\eta } . \end{array}\end{aligned}\ ] ] integrating out the `` localized '' fermionic grassmann variables @xmath95 with @xmath96 for the fermionic and @xmath97 for the bosonic part , we arrive at the following action in dual variables @xmath98 \\ \\ k}=(g_{\omega}^{-1}+\delta _ { \omega } -\varepsilon _ { k})^{-1}-g_{\omega } \\ \\ \tilde{\mathcal{x}}_{\omega k}=(\chi _ { \omega}^{-1}+\lambda _ { \omega } -v_{k } ) ^{-1}-\chi _ { \omega } \\ \\ \tilde{u}[\eta , f , f^{\dag } ] = \sum_{\omega \omega } \left ( \lambda _ { \omega \omega } \eta _ { \omega } ^{\ast } f_{\omega + \omega } ^{\dag } f_{\omega } + \lambda _ { \omega \omega } ^{\ast } \eta _ { \omega } f_{\omega } ^{\dag } \ f_{\omega + \omega } \right ) + \frac{1}{4}\sum_{\omega \omega ^{\prime } \omega } \gamma _ { \omega \omega ^{\prime } \omega } f_{\omega + \omega } ^{\dag } f_{\omega ^{\prime } -\omega } ^{\dag } f_{\omega ^{\prime } } f_{\omega}+ ... , \end{array } \label{db_action}\ ] ] where @xmath99 is a full four - point fermionic vertex of the impurity model @xmath100 ( @xmath101 is the kronecker symbol ) , and @xmath102 is a `` mixed '' quantity @xmath103 the second term in the nominator typically equals zero as @xmath104 vanishes . the values of @xmath99 and @xmath102 are related , since @xmath105 , @xmath106 it is important to note that for the gaussian impurity problem @xmath99 vanishes , but @xmath102 takes a finite value of @xmath107 . the following exact relations between fermionic and bosonic green s functions of original and dual variables can be proven : @xmath108 the relation for the fermionic green s functions is derived in ref . main idea is to substitute the right - hand side of eq.([smix ] ) in the formula @xmath109 . the last term @xmath110 appears from the differentiation of the fermionic determinant . the bosonic relation is derived quite similarly , but the signs appearing during the derivation should be discussed . first , the expression for the bosonic green s function has a different sign , @xmath111 . this gives factors of @xmath112 at @xmath113 and @xmath114 . second , since the bosonic hubbard - stratonovich transformation contains an inverse of the determinant , the same different sign appears for the last term @xmath115 , so that the signs of all the terms are changed , compared to the fermionic relation between the real and dual fermionic green s functions . therefore the entire resulting relationship between the real and dual bosonic green s functions obeys exactly the same form , similar to the results of the supersymmetric hubbard - stratonovich transformation @xcite . it is useful to introduce the self - energy ( for fermions ) and polarization - operator ( for bosons ) corrections to the effective - medium approximation , defined as @xmath116 we remind that @xmath117 are given by the expressions ( [ mfg ] ) . the self - energy and polarization operator of the dual ensemble are given by the following expressions : @xmath118 being re - expressed in these quantities , the exact relations ( [ exact ] ) have a particularly simple form : @xmath119 which results in the following expression for the fermionic and bosonic green s functions : @xmath120 one can immediately see from these formulas that the effective - medium expressions ( [ mfg ] ) correspond to the gaussian approximation for the dual variables : @xmath121 ( and consequently @xmath122 ) vanish . following the effective medium paradigm we introduce a generalized self - consistent condition for @xmath123 and @xmath124 similar to the eq.([dmftcondition ] ) , but for `` exact '' fermionic and bosonic green s functions : @xmath125 the simplest non - local physics can already be described in the edmft . in our formalism this corresponds to a zero - th order approximation ( `` no diagrams '' ) @xmath126 and @xmath127 , and condition ( [ dmftcondition ] ) is satisfied . we intend to preserve the advantages of the edmft , but also include the formation of collective modes and their interaction with the electrons . for a `` purely fermionic '' hubbard - like theory this means an account of the ladder series in dual space @xcite . for the present theory with the dual action of eq.([db_action ] ) , there are two basic green s function lines ( fig . [ basic_diagram ] ) : a fermionic ( full - line with arrow ) and a bosonic ( wavy - line ) as well as two interaction vertices : a triangle ( corresponding to the dual boson - fermion interaction - @xmath128 ) and a square ( describing the dual - fermion interaction - @xmath129 ) . in this paper , we restrict ourselves to a polarization operator , given by the electron ladder ended by two triangles ( fig . [ fig_ladder ] ) . equally , one can consider this diagram as just a two - lines bubble , with a renormalized vertex ( fig . [ fig_pi ] ) . @xmath130 here a symbolic notation is used , so that @xmath131 is a tensor , and @xmath132 is a vector with components corresponding to different @xmath133 , and the dots in the first line denote a scalar product vector - tensor - vector . the bare dual polarization operator ( an empty loop of the dual green s function ) @xmath134 is a tensor having only diagonal components . later we also use bare polarization operators for the initial variables and for the impurity model , so that @xmath135 scalar polarizabilities , respectively denoted by @xmath136 , @xmath137 , and @xmath138 , can be found by summating over fermionic frequencies , so that , for example , @xmath139 . the simplest approximation one can make is just to neglect the vertex part @xmath99 . in this case @xmath140 . simple manipulations with the second lines of eqs . ( [ mfg ] ) , ( [ selfenergy ] ) and ( [ exact2 ] ) then give @xmath141 a similar non - local contribution to the fermionic self - energy can be introduced to describe the renormalization of the fermionic green s function @xmath142 . the corresponding diagram contains one fermionic and one bosonic line ( fig . [ fig_sig ] ) . we will however concentrate on the analysis of bosonic quantities and for simplicity use the edmft result @xmath143 instead of the renormalized @xmath144 propagator . this allows for a further simplification . indeed , in the local approximation @xmath145 and @xmath146 differ in their local part only @xcite : @xmath147 further , since we suppose that @xmath148 and since @xmath24 , the bosonic green s function for the impurity problem becomes equal @xmath149 . substituting the expression ( [ ggprime2 ] ) into eq . ( [ drpa ] ) , we obtain @xmath150 for @xmath151 , this expression is clearly related to the random phase approximation ( rpa ) . to get the `` textbook '' rpa , one should completely neglect the interaction terms of the impurity problem , so that @xmath152 and @xmath153 . the scheme with an exactly calculated @xmath154 can be called rpa+dmft . in this case the rpa is a simple case of fig . [ fig_pi ] . to go beyond the rpa , one should be careful . as we will discuss in sec . [ conserv ] in general it is wrong to work with the renormalized green s functions but to neglect the vertex @xmath99 . because this may lead to a violation of the conservation laws . a theory using the single - bubble approximation ( [ buble ] ) with @xmath155 approximated by @xmath102 ( that is , without putting @xmath102 to @xmath156 ) is essentially the gw+edmft approach @xcite . there is only a slight difference in the self - consistency condition , as we explain in appendix . it is also important that for the diagram calculation of the gw+edmft approach that the local part of the bosonic green s function is excluded _ ad hoc_. as one can see from eq . ( [ edmftggprime ] ) , for the edmft choice of the bose variables , it actually does not vanish exactly . so , our consideration proposes a certain a modification of gw+edmft . anyhow , such modification should not strongly affect the results of the gw+edmft . the real advantage of our consideration is the possibility to go beyond the domain of applicability of the gw+edmft , i.e. in is the situation when the single - bubble approximation becomes incorrect . it is considered in the next sections . the diagrams shown in figs . [ bold3 ] , [ fig_pi ] , [ fig_sig ] form the minimal consistent set beyond the rpa to describe the formation of collective modes with electron - boson coupling taken into account explicitly . so far , we did not consider whether the constructed approximations conserve particle number , total spin , etc . however , the conservation laws are particularly important for bosonic excitations . for example , the charge ( particle number ) conservation resulting in the equation @xmath157 immediately leads to the requirement @xmath158 at any finite frequency ( provided that the current - current correlator @xmath159 is finite ) . for 3d systems with coulomb interaction the long - wavelength asymptotic behaviour for the density - density correlator should be @xmath160 , where @xmath161 is the plasma frequency . the rpa for free electrons is proven to obey this property , since @xmath162 vanishes with @xmath163 , but this is not true for the renormalized green s function @xcite . indeed , following the standard proof @xcite one writes @xmath139 as @xmath164 for free electrons of dispersionless @xmath165 ( that is , in static mean - field ) the summation over @xmath133 produces a difference of occupation numbers , @xmath166 , in the nominator , that vanishes at @xmath167 . however this reasoning clearly breaks down for the frequency - dependent @xmath168 in dmft approximation . it was known for a long time that the neglection of vertex corrections together with the use of _ renormalized _ green s functions , violate charge and spin conservations @xcite . we found that @xmath169 with the dmft green s function does not vanish at small @xmath170 and therefore it follows form eq . ( [ rpa2 ] ) that _ edmft+gw is not a valid description of plasmons_. all the more , _ plasmons do not appear in simple edmft _ , as one can easily observe from the second line of eq . ( [ mfg ] ) : the local quantity @xmath171 can not behave like @xmath172 . this is the reason why the gw approach is normally used only for calculating the single - particle green s function , while any two - particle properties are investigated within the bethe - salpeter equations @xcite . nevertheless , such a scheme can not describe self - consistent effects of collective excitations on the single - particle spectral function . to construct a conserving approximation , it is useful to consider the following reformulation of the theory . let us return to the initial lattice action ( [ lattice ] ) and do the hubbard - stratonovich decoupling for the bosonic variables only . it will give an analog of eq . ( [ smix ] ) with the action @xmath173+\rho^ * \alpha_b \eta + \eta^ * \alpha_b \rho-\eta^ * \alpha_b ( \lambda - v)^{-1 } \alpha_b \eta,\\ s^f=\sum_r s_{at}[c^\dag_r , c_r]+\sum_{r , r\neq 0 , \omega , \sigma } \varepsilon_r c^\dag_{r \omega \sigma } c_{r+r \omega \sigma } + \sum_{r , \omega } \lambda_{\omega } \rho_{r \omega}^*\rho_{r \omega}. \end{array}\ ] ] physically , @xmath174 describes a lattice with the interaction being local in space but retarded . now we formally integrate over the fermionic variables . the gaussian approximation in dual bosons for such a theory corresponds to the expression @xmath175 where @xmath176 is an exact @xmath170-dependent two - particle correlator for the lattice problem @xmath174 . formally , the calculation of @xmath177 corresponds to a summation of all parts of the diagrams with fermionic lines in series for eq.([db_action ] ) . comparing eqs . ( [ greenf ] ) and ( [ exact ] ) we obtain @xmath178 this way , we need a conserving approximation for @xmath176 . such an approximation is actually known within the framework of dmft . because it is a conserving theory in the baym - kadanoff sense @xcite , with the self energy being just the sum of all local diagrams . in such a theory , the conserving approximation for two - particle quantities can be obtained as a green s function variational derivative with respect to the dispersion law , @xmath179 an important property of the baym s criterion of consistency is that the variation should be considered self - consistently , so that the @xmath43 and @xmath40 of dmft vary with @xmath0 to preserve the self - consistency condition ( [ dmftcondition ] ) . one can also explicitly see that such a self - consistent variation preserves the charge conservation . indeed , the charge conservation follows from gauge invariance under the transformation @xmath180 with an arbitrary dependence of the infinitesimal phase shift @xmath181 on time and space arguments . in an exact theory , such a transformation just results in similar phase rotations for averages , for example @xmath182 . a conservative approximation must inherit this property . dmft is entirely defined by the self - consistency condition ( [ dmftcondition ] ) and impurity action ( [ impurity ] ) . if all the quantities entering the green s function @xmath183 ( such as @xmath0 , @xmath43 , @xmath40 ) evolve simultaneously under the gauge transformation , the self - consistency ( [ dmftcondition ] ) clearly stays fulfilled . the entire green s function @xmath183 obeys the gauge phase rotation , which proves that the theory is conserving . we can prove that the ladder approximation for dual bosons results in such a conserving theory . we introduce a variation of the local part of the dispersion law @xmath184 , and corresponding variations of local quantities @xmath185 , and @xmath186 . in the tensorial notation used previously , the variations @xmath187 are vectors with a single index @xmath188 ; their components are defined as fourier - transforms : @xmath189 etc . the straightforward variation of the self - consistency condition with the dmft green s function ( [ dmftcondition ] ) gives : @xmath190 it is worth to remind that @xmath191 and @xmath192 are tensors with double indices @xmath193 , as well as the quantities @xmath194 and @xmath195 to appear in the following formulas . now , let us consider the full lattice susceptibility @xmath194 as a fourier - transform of eq.([susc4 ] ) with pairwise - equal space arguments : @xmath196 the self - consistency requires @xmath197 , and therefore the above - introduced variations correspond to @xmath198 similarly , for the susceptibility of the impurity model @xmath199 although the right - hand side formally includes the wavevector @xmath170 , @xmath200 is of course @xmath170-independent . with the above two formulas formulas , it immediately follows from ( [ suscdraft ] ) that @xmath201 such a self - consistent condition ( [ susc ] ) is equivalent to locality of the irreducibly vertex and it is used in the dmft scheme for calculations of non - local susceptibilities @xcite . it is worthwhile to remind that the susceptibilities with a single frequency argument , which we defined previously , can be obtained by summation over the fermionic frequencies , so that @xmath202 and @xmath203 to establish a connection with the dual formalism , we substitute this expression into eq . ( [ xpi ] ) and express @xmath204 using eq . ( [ susc ] ) . simple algebraic manipulations give @xmath205 one can recognize that @xmath206 and @xmath207 are , respectively , the previously introduced quantities @xmath102 and @xmath99 with proper indices . furthermore , it follows from the self - consistency condition ( [ dmftcondition ] ) that @xmath208 so we conclude that the obtained expression for @xmath209 is exactly the result of the dual ladder summation ( [ buble ] ) . finally , let us consider the plasmon dispersion given by the equation @xmath210 in the long - wavelength limit , one considers an expansion in @xmath211 , in particular @xmath212 . the zeroth - order term in the left - hand side vanishes due to charge conservation . the second - order term results in the dispersion equation for plasmons in a 3d correlated lattice : @xmath213 for the 2d case , the right - hand side equals @xmath214 . while deriving the last equation , we supposed that @xmath49 stays finite and can be therefore neglected in comparison with @xmath215 at small @xmath170 . note however that this quantity still enters @xmath174 and therefore affects the left - hand side of the dispersion relation , so that the physics of local plasmonic fluctuations is taken into account . the single - bubble approximation becomes invalid when vertex @xmath99 is large . there is a general situation of this kind : the four - point vertex of an impurity problem with _ magnetic moment formed _ appears to have a part displaying a slow dynamics that give a large contribution at low frequencies . this simply means that for isolated atoms the magnetic moments are free and , thus , for _ almost _ isolated atoms ( when the energy of interatomic effective interactions is much smaller than the intraatomic one ) they can be considered as _ almost _ integrals of motion . to illustrate this statement , let us consider the atomic limit of the single - band hubbard model . the single - electron green s function of the hubbard atom at half - filling @xmath216 has a time - scale @xmath217 , which we call fast . contrary , the spin - spin correlator @xmath218 is independent , as one can easily check , of time arguments . so , a time - scale for @xmath218 is the inverse temperature @xmath13 . it means that , apart from a small time domain @xmath219 , the value of @xmath220 is determined by its non - gaussian part , which does not fall as @xmath221 increases . the much slower dynamics of @xmath222 in comparison with @xmath40 just reflects the fact that a rotation of the spin does not change the energy of the atom , whereas the change of the particle number does . direct calculation of @xmath99 for the hubbard atom does support this observation . the finite - temperature expressions @xcite for @xmath223 contain a `` singular '' term proportional to @xmath224 . so , the first two time arguments of @xmath99 in time - domain must coincide , as well as the last two : @xmath225 , but the difference @xmath226 can be large . the properties of the impurity problem are different from those of the isolated atom . however , for the strong - coupling limit the hybridization is small . in this case the fast dynamics roughly stay unchanged , so that the atomic green s function ( [ gatomic ] ) can be used , and the four - point vertex obeys the form @xmath227 . the hybridization can introduce its own slow timescale @xmath228 , so that @xmath99 does depend only on @xmath226 . in the frequency representation , this means @xmath229 and at small frequencies @xmath230 . since @xmath222 includes a large factor , we can keep only the irreducible term in the expressions for @xmath102 and @xmath39 . note that this is completely opposite to the case considered in section [ barediagsec ] . the terms proportional to @xmath231 in eq . ( [ gamma ] ) are small in @xmath232 and therefore can be neglected . then we obtain the following relation from eq . ( [ gamma ] ) : @xmath233 in the above equation we introduced a `` bare susceptibility '' of the impurity problem @xmath234 next , we use the relation ( [ lambdagamma ] ) and keep only the large factor @xmath99 : @xmath235 note that in the present approximation @xmath102 , like @xmath99 , only has single frequency argument . let us consider the half - filled hubbard lattice with the nearest - neighbour hopping @xmath236 , at large @xmath33 . the low - frequency behaviour of this system is essentially the dynamics of the heisenberg model . the effective - medium description of spin waves in this case should result in the expression @xmath237 with @xmath238 for nearest neighbors . although this expression is rather simple ( in our notation , it is just @xmath239 ) , it was not reproduced by any dmft - like approximations so far . the superexchange interaction containing @xmath240 can easily be obtained from a single loop of the fermionic lines , but the rpa and gw+edmft denominators contain an inverse of this quantity ( see section [ barediagsec ] ) , so the structure of the formulas is drastically different . in our formalism , the desired expression appears from the ladder summation , that is the calculation for the eq . ( [ buble ] ) and corresponding diagram shown in fig . ( [ fig_ladder ] ) . in this summation , we take into account the `` slow '' vertex ( [ gammaslow ] ) only . since @xmath99 depends only on a single time argument , the calculation is quite simple . with the exact relation ( [ exact2 ] ) , we obtain @xmath241^{-1 } .\ ] ] finally , we substitute the formulas ( [ irreducc ] ) , ( [ irreduc ] ) for @xmath39 and @xmath102 . this results in the cancellation of @xmath99 out of the expressions : @xmath242 where @xmath132 from eq . ( [ irreduc ] ) has a meaning of an effective local stoner parameter . both @xmath243 and @xmath244 obey the `` fast '' dynamics only . to determine the low - energy properties , in which we are interested , it is enough to calculate it at @xmath245 , and to replace the summation over frequency by an integration . a straightforward calculation of the expression ( [ chi0 ] ) with @xmath246 gives @xmath247 . to calculate @xmath248 , we switch to real - space and use the strong - coupling limit . it gives the green s function @xmath249 for the nearest neighbors and @xmath250 elsewhere . now , the integration gives @xmath251 for the nearest - neighbors and we end up with the desired expression ( [ heizenberg ] ) . this derivation can be considered as a generalization of the effective exchange formula for the magnetically ordered case @xcite to the paramagnetic state . the standard theory of magnons in itinerant - electron systems @xcite is based on the rpa which is supposed to be exact at zero temperature ; at finite temperatures , magnon - magnon and electron - magnon interactions should be taken into account @xcite . electron - hole excitations are primary objects in such an approach , spin waves are coherent superposition of such excitations and , in general , are only well - defined in a restricted part of the brillouin zone , due to stoner damping . in the case of strong - coupling hubbard model , we have an `` anti - rpa '' situation when local magnetic moments are primary objects . therefore , completely different approaches should be used , e.g. those based on the hubbard x - operators @xcite . it is very non trivial that the dual boson approach describes these two opposite cases within the same formalism , thus providing a practical way to interpolate between weak and strong coupling limits . in summary , the general scheme for non - local correlation effects within the dual boson approach is presented . this method gives a direct means to obtain the higher - order diagrammatic contributions beyond the edmft scheme free from the double - counting contributions and it can be useful to describe the magnon - spectrum renormalization in strongly correlated systems . the method combines the numerically exact solution of an effective single - cite impurity model with frequency - dependent electron - electron interactions as well as an analytical diagrammatic expansion of non - local fermionic and bosonic self - energies . this approach is designed to describe of correlated systems with strong local fermion correlations and fluctuating non - local bosonic modes on equal footing . in this paper we focused on the applications to the description of bosonic degrees of freedom . additional contributions to the electron self energy also arise , in comparison with the dual fermion approach @xcite , e.g. , the diagram shown in fig . [ fig_sig ] . similar diagrams are important in the cases of coexistence of disorder with electron - electron @xcite or electron - phonon @xcite interactions . it is important to calculate these additional contributions within the hubbard model and make a detailed comparison with dual fermion theory . this will be the subject of our further studies . this work was supported by dfg grant no . 436 113/938/0-r and rffi 11 - 02 - 01443 . mik acknowledges a support from the eu - india fp-7 collaboration under monami . ail acknowledges a support from sfb 668 and for-1346 . the formalism presented here has an important difference compare to the common introduction of the edmft : whereas we introduce dual bosons @xmath79 by the hubbard - stratonovich decoupling of the quadratic form @xmath252 , the term decoupling @xmath253 is decoupled in the standard edmft scheme @xcite . as one can see , this does not result in any physical consequences , as the two approaches yield the identical results for the averages @xmath254 over original variables . however , properties of the _ dual _ bosons , of course , do depend on their definition . in particular , the difference is about local properties of the dual propagator : for the present formalism , its local part vanishes in the edmft approximation , @xmath255 whereas the `` effective - medium '' condition @xmath256 holds for the decoupling ( [ vdecoupl ] ) . here @xmath257 is the green s function of the dual variables @xmath51 arising from the decoupling of @xmath253 : @xmath258 and @xmath259 is defined for the impurity problem with decoupled bosonic hybridization , so that @xmath260 to maintain a similarity with our formalism , we introduced in the above expressions a scaling factor @xmath70 , that is absent in the standard edmft scheme @xcite . all physical quantities do not depend on a particular value of this factor . we will now consider a general case of some approximation utilizing a decoupling of @xmath263 with an arbitrary @xmath264 . the approximation can be non - gaussian , so that there is a correction @xmath265 to the edmft polarization operator : @xmath266 the green s function for the dual bosons can be found from the exact relation @xmath267 one can realize that there are two special cases . first , let us take @xmath268 , and use @xmath269 , as in the main body of the paper . for this choice , using ( [ w1 ] , [ exactw ] ) one can straightforwardly show that @xmath270 summing over @xmath8 and taking eq.([dbcondition ] ) into account we conclude that the condition ( [ dual0 ] ) is fulfilled @xmath271 it turns out that _ the dual propagator vanishes for variables obtained from the hubbard - stratonovich decoupling of the renormalized interaction , with the hybridization part * excluded * _ , if the condition ( [ dbcondition ] ) is fulfilled . for the edmft approximation , @xmath265 is absent and the condition ( [ dual0 ] ) holds for the decoupling with @xmath272 used in the present paper . now , let us consider the case @xmath273 . it is suitable to take @xmath274 in this case . we obtain @xmath275 again taking a summation over @xmath8 and substituting eq.([dbcondition ] ) , we obtain @xmath276 . the second line of ( [ exactv ] ) with @xmath274 gives the same answer for @xmath261 , therefore we obtain @xmath277 _ the `` effective - medium '' condition holds for variables obtained from the hubbard - stratonovich decoupling of the renormalized interaction , with the hybridization part * included * _ , if condition ( [ dbcondition ] ) is fulfilled . for edmft , this corresponds to the decoupling of purely @xmath253 part used in ref . @xcite . we stress that , strictly speaking , apart from the simple edmft case the conditions ( [ dual0 ] ) , ( [ dualem ] ) are not fulfilled simultaneously with ( [ dual0 ] ) . this does not cause serious problems for the gw+edmft calculations , because one typically does self - consistent calculations of the edmft problem and then calculates diagram corrections for a pure edmft hybridization . on the other hand , for higher order diagrammatic approximations the difference can be essential . finally , we note that a specific choice of the decoupling procedure is physically irrelevant . as we have seen , edmft can be constructed using @xmath278 or @xmath279 , and the result for initial ensemble is the same . it can be checked that any other decoupling will also work ; what matters is the gaussian approximation for dual variables and the value of the hybridization . the latter is defined by the condition ( [ dual0 ] ) , that does not include any dual quantities . we use the theory based on @xmath280 decoupling because it allows a simple construction of the dual diagrammatic series without a double counting of the local contributions . this important property is based on the fact that the edmft a dual green s function has a very simple physical meaning : according to ( [ ggprime ] ) , it is just the green s function of initial variables with local part subtracted . e. gull , a. j. millis , a. i. lichtenstein , a. n. rubtsov , m. troyer , p. werner , rev . 83 ( 2011 ) 349 . e. gorelov , j. koloren , t. wehling , h. hafermann , a. b. shick , a. n. rubtsov , a. landa , a. k. mcmahan , v. i. anisimov , m. i. katsnelson , a. i. lichtenstein , phys . b 82 ( 2010 ) 085117 . e. gorelov , t. o. wehling , a. n. rubtsov , m. i. katsnelson , a. i. lichtenstein , phys . rev . b 80 ( 2009 ) 155132 . g. kotliar , s. y. savrasov , k. haule , v. s. oudovenko , o. parcollet , c. a. marianetti , rev . 78 ( 2006 ) 865 . a. i. lichtenstein , m. i. katsnelson , phys . b 57 ( 1998 ) 6884 . m. i. katsnelson , v. y. irkhin , l. chioncel , a. i. lichtenstein , r. a. de groot , rev . 80 ( 2008 ) 315 . r. chitra , g. kotliar , phys . b 63 ( 2001 ) 115110 .

a large class of physical and engineering problems can be described in terms of lagrangian and hamiltonian systems , such as controlled mechanical systems , externally forced and dissipative systems , collisions , electric circuits , stochastic systems , and field theories such as electromagnetism and elasticity . such systems are usually categorized into lagrangian or hamiltonian systems with constraints , whether holonomic or nonholonomic ; however , one may be faced with serious problems on ever larger scales becoming more and more heterogeneous , involving a mixture of mechanical and electrical components , with flexible and rigid parts , and elastic or magnetic couplings ( see , for instance , @xcite and @xcite ) . dirac structures are known as a powerful tool for providing a natural geometric framework for describing the structure of such diverse lagrangian and hamiltonian systems on the product of configuration spaces by introducing nontrivial constraints that model interactions between subsystems , which is called an _ interconnection_. the notion of an interconnection was originally developed by @xcite in his book of `` diakoptics '' for modeling complicated physical systems . the word `` diakoptics '' denotes a procedure of _ tearing _ a dynamical system into disconnected subsystems as well as _ interconnecting _ them to reconstruct the original system . the original system may be regarded as an interconnected system of subsystems , with constraints induced from the interconnection . interconnection has been studied through the lens of power conservation motivated by the bond graph theory established by @xcite . in electric circuit theory , the interconnection network has been widely employed through nonenergic multiports ( see @xcite ) , where kirchhoff s current law gives a constraint distribution on a configuration charge space such that tellegen s theorem holds . in mechanics , it was shown by @xcite that kinematical constraints due to mechanical joints , coordinate transformations as well as force equilibrium conditions in dalembert principle can be represented by the interconnection modeled by nonenergic multiports using the bond graph theory in the context of lagrangian mechanics . it was demonstrated by @xcite that the interconnection can be represented by dirac structures induced from poisson structures and also that nonholonomic systems and l - c circuits can be represented by _ implicit hamiltonian systems_. on the lagrangian side , it was shown by @xcite that nonholonomic mechanical systems and l - c circuits ( as degenerate lagrangian systems ) can be formulated by _ implicit lagrangian systems _ associated with the dirac structures on the cotangent bundles induced from given constraint distributions . from the view point of variational structures , it was shown by @xcite that the standard implicit lagrangian systems , namely , _ implicit euler - lagrange equations _ for unconstrained systems can be derived from the _ hamilton - pontryagin principle _ and also that the lagrange - dirac dynamical systems can be formulated in the context of _ lagrange - dalembert - pontryagin principle_. from the viewpoint of control theory , a notion of _ implicit port - controlled hamiltonian ( ipch ) systems _ ( an implicit hamiltonian system with external control inputs ) was developed by @xcite ( see also @xcite , @xcite and @xcite ) and much effort has been devoted to the passivity based control for the interconnected mechanical systems ( @xcite ) . for the case of regular lagrangian systems , the equivalence between _ controlled lagrangian ( cl ) systems _ and _ controlled hamiltonian ( ch ) _ systems was shown by @xcite and the reduction theory for the cl systems was shown by @xcite . for the general case in which a given lagrangian is degenerate , an implicit lagrangian analogue of ipch systems , namely , a notion of _ implicit port - controlled lagrangian ( ipcl ) systems _ for electrical circuits was constructed by @xcite and @xcite , where it was shown that l - c transmission lines can be represented in the context of the iplc system by employing induced dirac structures with lagrange multipliers . recently , it was shown by @xcite that any power - conserving interconnection of ipch systems is another port - hamiltonian system , with the dirac structure being the _ composition _ of the dirac structures of its constituent parts , hamiltonian the sum of the hamiltonians , and resistive relations determined by the individual resistive relations . as a control application , the feedback interconnection of a plant port - hamiltonian system with a controller port - hamiltonian system can be represented by the composition of a plant dirac structure with a controller dirac structure . [ [ goals - and - organizations - of - the - paper . ] ] goals and organizations of the paper . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + one of the main goals in this paper is to develop a simple expression for the interconnection of distinct dirac structures and their associated mechanical systems ; namely , we will show how to obtain an interconnection of distinct dirac structures @xmath4 on distinct manifolds @xmath5 as another dirac structure @xmath6 on @xmath7 by `` interconnecting '' @xmath4 . to do this , we will introduce two mathematical ingredients , namely , the _ dirac sum _ @xmath8 and the _ bowtie product _ @xmath2 of dirac structures and we will show how two distinct dirac structures can be interconnected to yield another dirac structure via an _ interconnection _ dirac structure ( usually labeled @xmath1 in this paper ) under an assumption of transversal intersections . in particular , the interconnection of dirac structures is @xmath9 . the interconnection dirac structure @xmath1 provides an interface to interconnect `` distinct '' dirac structures @xmath4 to yield another dirac structure @xmath6 . we will also show how the associated interconnection of lagrange - dirac dynamical systems can be developed in the context of variational structures . we will demonstrate our theory by some examples of circuits and mechanical systems . this paper consists of the following parts : 1 . we briefly review dirac structures and we show how associated lagrange - dirac dynamical systems can be developed by the _ lagrange - dalembert - pontryagin principle _ ( see @xcite ) . we show how distinct dirac structures can be interconnected to yield another dirac structure on the product manifold through an interconnection dirac structure . to do this , we develop a map called the dirac sum @xmath8 and a tensor product called the bowtie product @xmath2 between dirac structures . we explore the variational structure of interconnection of lagrange - dirac dynamical systems ; namely , the interconnected lagrange - dirac dynamical system can be formulated in the context of the lagrange - dalembert -pontryagin principle . we demonstrate some illustrative examples of circuits and nonholonomic systems , which can be generalized to more complex systems . [ [ acknowledgements . ] ] acknowledgements . + + + + + + + + + + + + + + + + + we are also very grateful to henrique bursztyn , mathieu desbrun , melvin leok , tomoki ohsawa , arjan van der schaft and alan weinstein for useful remarks and suggestions . the research of h. j. is supported by caltech and jst - crest in part . the research of h. y. is partially supported by jsps grant - in - aid 20560229 , jst - crest and waseda university grant for sr 2010a-606 . [ [ linear - dirac - structures . ] ] linear dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + first , let us recall the definition of a linear dirac structure , namely , a _ dirac structure on a vector space _ @xmath10 ; we assume that @xmath10 is finite dimensional for simplicity ( see , @xcite ) . let @xmath11 be the dual space of @xmath10 , and @xmath12 be the natural paring between @xmath11 and @xmath10 . define the symmetric paring @xmath13 on @xmath14 by @xmath15 for @xmath16 . a _ dirac structure _ on @xmath10 is a subspace @xmath17 such that @xmath18 , where @xmath19 is the orthogonal of @xmath6 relative to the pairing @xmath20 . [ [ dirac - structures - on - manifolds . ] ] dirac structures on manifolds . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath21 be a smooth manifold and let @xmath22 denote the whitney sum bundle over @xmath21 , namely , the bundle over the base @xmath21 and with fiber over the point @xmath23 equal to @xmath24 . a subbundle @xmath25 is called an almost dirac structure on @xmath21 , when @xmath26 is a dirac structure on the vector space @xmath27 at each point @xmath28 . a given two - form @xmath29 on @xmath21 together with a distribution @xmath30 on @xmath21 determines an _ ( almost ) dirac structure _ on @xmath21 as follows : for each @xmath28 , @xmath31 where @xmath32 is the restriction of @xmath29 to @xmath30 . we can also restate the dirac structure as , for each @xmath28 , @xmath33 where @xmath34 is an annihilator of @xmath30 . we call @xmath6 an _ integrable dirac structure _ if the following integrability condition @xmath35 is satisfied for all pairs of vector fields and one - forms @xmath36 , @xmath37 , @xmath38 that take values in @xmath6 , where @xmath39 denotes the lie derivative along the vector field @xmath40 on @xmath21 . [ [ remark . ] ] remark . + + + + + + + let @xmath41 be a space of local sections of @xmath22 , which is endowed with the skew - symmetric bracket @xmath42 : \gamma(tm \oplus t^{\ast}m ) \times \gamma(tm \oplus t^{\ast}m ) \to \gamma(tm \oplus t^{\ast}m)$ ] defined by @xmath43 & : = \left ( \left[x_{1 } , x_{2 } \right ] , \pounds_{x_1 } \alpha_{2}- \pounds_{x_2 } \alpha_{1 } + \mathbf{d}\left < \alpha_{2 } , x_{1 } \right>\right)\\ & \;=\left ( \left[x_{1 } , x_{2 } \right ] , \mathbf{i}_{x_1 } \mathbf{d}\alpha_{2}- \mathbf{i}_{x_2 } \mathbf{d}\alpha_{1 } + \mathbf{d } \left < \alpha_{2 } , x_{1 } \right>\right ) . \end{split}\ ] ] this bracket is the one that @xcite originally developed and it does not necessarily satisfy the jacobi identity . it was shown by @xcite that the integrability condition of the dirac structure @xmath44 given in equation can be expressed as @xmath45 \subset \gamma(d),\ ] ] which is the closedness condition of the courant bracket ( see also @xcite and @xcite ) . the following lemma is useful and the proof may be found in @xcite and @xcite . [ lem : isotropic ] a subbundle @xmath46 is isotropic with respect to @xmath20 if and only if @xmath47 . [ [ induced - dirac - structures . ] ] induced dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + one of the most important and interesting dirac structures in mechanics is an _ induced dirac structure _ from kinematic constraints , whether holonomic or nonholonomic . such constraints are generally given by a distribution on a configuration manifold ( as to the details , refer to @xcite ) . let @xmath48 be a configuration manifold . let @xmath49 and @xmath50 be the tangent and cotangent bundles of @xmath48 . let @xmath51 be a regular distribution on @xmath48 and define a lifted distribution on @xmath50 by @xmath52 where @xmath53 is the cotangent projection so that its tangent is a map @xmath54 . let @xmath29 be the canonical two - form on @xmath50 . the _ induced dirac structure _ @xmath6 on @xmath50 is the subbundle of @xmath55 , whose fiber is given for each @xmath56 as @xmath57 this is , of course , a special instance of the construction . notice that the induced dirac structure can be restated by using the bundle map @xmath58 as follows : @xmath59 where @xmath60 is the annihilator of @xmath61 . [ [ remark.-1 ] ] remark . + + + + + + + if there exists no constraint , namely , the case in which @xmath62 , the dirac structure @xmath6 corresponds to the canonical dirac structure , which is given by a graph of the bundle map @xmath63 associated to @xmath29 . [ [ local - expressions . ] ] local expressions . + + + + + + + + + + + + + + + + + + let @xmath10 be a model space for @xmath48 and @xmath64 be the range in @xmath10 of a chart on @xmath48 . then , the range of the induced chart on @xmath49 is represented by @xmath65 , while @xmath50 is locally given by @xmath66 . in this representation , @xmath67 are local coordinates of @xmath49 and @xmath68 are local coordinates of @xmath50 . further , @xmath69 is represented by @xmath70 , while @xmath71 is locally given by @xmath72 . in this local representation , @xmath73 are the corresponding coordinates of @xmath69 , while @xmath74 are the local coordinates of @xmath71 . using @xmath75 locally denoted by @xmath76 and @xmath77 , it follows that @xmath78 and the annihilator of @xmath61 is locally represented as @xmath79 since we have the local formula @xmath80 , the condition @xmath81 reads @xmath82 and @xmath83 thus , the induced dirac structure is locally represented by @xmath84 where @xmath85 is the polar of @xmath86 . in this section , let us recall the definition of implicit lagrangian systems or lagrange - dirac dynamical systems by following @xcite . [ [ lagrangedirac - dynamical - systems . ] ] lagrange dirac dynamical systems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath87 be a lagrangian , possibly degenerate . let @xmath6 be a dirac structure on @xmath50 induced from a given distribution @xmath51 . recall that a _ partial vector field _ @xmath88 is defined as a mapping such that @xmath89 , where @xmath90 and @xmath91 is the tangent projection . as a mapping @xmath92 such that @xmath93 the partial vector field @xmath94 is a map that assigns to each point @xmath95 , a vector in @xmath69 at the point @xmath96 ; we write @xmath40 as @xmath97 where we note that @xmath98 and @xmath99 are understood to be functions of @xmath100 . let @xmath101 be the _ generalized energy _ given by @xmath102 an implicit lagrangian system or a lagrange dirac dynamical system is the triple @xmath103 that satisfies , for each @xmath95 and with @xmath104 , namely , @xmath105 , @xmath106 where @xmath107 is the legendre transformation . in the above , @xmath108 is a linear function on @xmath109 at each point @xmath95 and hence the restriction @xmath110 may be regarded as a linear function on @xmath111 such that @xmath112 . [ [ local - expressions.-1 ] ] local expressions . + + + + + + + + + + + + + + + + + + it follows from that the lagrange dirac dynamical system @xmath113 is locally given by @xmath114 [ [ remark.-2 ] ] remark . + + + + + + + if a partial vector field @xmath115 satisfies the condition for a lagrange dirac dynamical system @xmath113 , then the legendre transformation @xmath112 is consistent with the equality of base points and the condition itself gives the second - order condition @xmath116 . in other words , the partial vector field of the lagrange dirac dynamical system is uniquely given on the graph of the legendre transformation . for the case in which no kinematic constraint is imposed , i.e. , @xmath62 , we can develop the standard implicit lagrangian system , which is expressed in local coordinates as @xmath117 which we shall call implicit euler lagrange equations . note that the implicit euler lagrange equations include the euler lagrange equations @xmath118 , the legendre transformation @xmath119 as well as the second - order condition @xmath116 . [ [ example - harmonic - oscillators . ] ] example : harmonic oscillators . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we illustrate the implicit euler lagrange equations by the example of a linear harmonic oscillator . in this case , the configuration space is @xmath120 representing the position of a particle on the line . the lagrangian is given by @xmath121 . we find the generalized energy to be @xmath122 . the canonical dirac structure @xmath123 can be easily constructed by using the canonical symplectic structure @xmath29 on @xmath50 . the lagrange - dirac system is given by @xmath103 , which satisfies @xmath124 and it follows @xmath125 . we see @xmath126 and @xmath127 . so , the dynamics is equivalent to the following equations : @xmath128 together with the legendre transform @xmath129 . [ [ lagrange - dirac - dynamical - systems - with - external - forces . ] ] lagrange - dirac dynamical systems with external forces . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + here , we consider the case of a system with external force fields . let @xmath130 be a natural projection . an external force field @xmath131 induces a horizontal one - form @xmath132 on @xmath133 such that , for @xmath95 , @xmath134 where @xmath135 and @xmath136 . given a lagrangian @xmath87 ( possibly degenerate ) , a lagrange - dirac dynamical system with an external force field is defined by a _ quadruple _ @xmath137 , which satisfies , in coordinates @xmath95 , @xmath138 it follows that the local expression of a lagrange - dirac system in equation may be given by @xmath139 and with @xmath140 the curve @xmath141 , @xmath142 in @xmath133 that satisfies the condition is a solution curve of @xmath143 . [ [ power - balance - law . ] ] power balance law . + + + + + + + + + + + + + + + + + + the time derivative of @xmath144 restricted to the solution curve @xmath141 reads @xmath145 where we have employed @xmath146 and @xmath147 . [ [ example - the - harmonic - oscillator - with - damping . ] ] example : the harmonic oscillator with damping . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as before , consider the harmonic oscillator in the setting of @xmath120 , @xmath148 , @xmath149 and @xmath150 and add a damping force @xmath151 defined by @xmath152 , where @xmath153 is a positive damping constant . then , @xmath154 . the formulas in equation give us the equations : @xmath155 with the legendre transformation @xmath129 . [ [ the - hamilton - pontryagin - principle . ] ] the hamilton - pontryagin principle . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + lagrange - dirac dynamical systems can describe general scenarios such as non - holonomic degenerate lagrangian systems because of a firm basis in variational structures . given a lagrangian @xmath87 , unconstrained equations of motion are described by hamilton s principle : @xmath156 from the viewpoint of dirac geometry , we start rather from the hamilton - pontryagin principle , which is given by the stationary condition of the action integral on the space of curves @xmath157 $ ] in @xmath133 as @xmath158 where @xmath159 . the above variational principle can be restated by using the generalized energy @xmath160 as @xmath161 keeping the endpoints @xmath162 and @xmath163 of @xmath164 fixed , we can obtain the _ implicit euler lagrange equations_. [ [ the - lagrange - dalembert - pontryagin - principle . ] ] the lagrange - dalembert - pontryagin principle . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider a mechanical system with kinematic constraints that are given by a constraint distribution @xmath165 on @xmath48 . then , the motion of the mechanical system @xmath166 \to q$ ] is said to be constrained if @xmath167 for all @xmath168 . assume that the distribution @xmath169 is not involutive ; that is , @xmath170 \notin \delta_{q}(q)$ ] for any two vector fields @xmath171 on @xmath48 with values in @xmath169 . further , let @xmath172 be a ( possibly degenerate ) lagrangian on @xmath49 and let @xmath131 be an external force field . the lagrange - dalembert - pontryagin principle for a curve @xmath141 , @xmath173 in @xmath133 is represented by @xmath174 for a given variation @xmath175 and with the constraint @xmath176 . keeping the endpoints of @xmath164 fixed , it follows @xmath177 for a chosen variation @xmath175 , for all @xmath178 and @xmath179 , and with @xmath180 . the lagrange - dalembert - pontryagin principle gives the local expression of equations of motion for nonholonomic mechanical systems with external forces such that @xmath181 from equation , it reads @xmath182 which is satisfied for a given variation @xmath175 , for all @xmath178 and @xmath179 , and with the constraint @xmath180 . thus , we obtain equation . [ [ coordinate - representation . ] ] coordinate representation . + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath48 be a finite dimensional smooth manifold whose dimension is @xmath183 . choose local coordinates @xmath184 on @xmath48 so that @xmath48 is locally represented by an open set @xmath185 . then , @xmath49 is locally given by @xmath186 . similarly @xmath50 may be locally denoted by @xmath187 . the constraint set @xmath169 defines a subspace on each fiber of @xmath49 , which we denote by @xmath188 at each point @xmath189 . if the dimension of @xmath190 is @xmath191 , then we can choose a basis @xmath192 of @xmath193 . recall that the constraint sets can be also represented by the annihilator @xmath194 , which is spanned by @xmath195 one - forms @xmath196 . it follows that equation can be represented , in coordinates , by employing the lagrange multipliers @xmath197 , as follows : @xmath198 where we employ the local expression @xmath199 . there exist various types of the power conserving interconnections between distinct dynamical systems such as two rigid bodies connected by ball - socket joints , mechanical systems connected by transmission gears , transducers between electric circuits and mechanical systems such as d - c motors , fluid and rigid body interactions , etc . all the examples listed above may be represented by dirac structures . in this section , we will show how distinct dirac structures can be interconnected via an _ interconnection _ dirac structure to yield another dirac structure of a fully interconnected dynamical system . [ [ the - direct - sum - of - dirac - structures . ] ] the direct sum of dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + before going into details on the case of @xmath183 distinct dirac structures , we will begin with the case of @xmath200 to introduce the direct sum of dirac structures on manifolds . let @xmath201 be the set of ( almost ) dirac structures over a manifold @xmath21 . is called `` daleth '' , which is the fourth letter of the semitic alphabet corresponding to `` d '' . ] let @xmath202 and @xmath203 be distinct smooth manifolds . to transition from systems on @xmath204 and @xmath205 to one on @xmath206 , we shall define a _ direct sum _ associated to distinct dirac structures . let @xmath207 and @xmath208 be given distinct dirac structures on @xmath202 and @xmath203 respectively . define the dirac sum of @xmath209 and @xmath210 by @xmath211 we now show that @xmath8 is a mapping from @xmath212 . if @xmath213 , then @xmath214 . let @xmath215 and let @xmath216 . notice that @xmath217 for @xmath218 . by definition , @xmath219 and @xmath220 for some @xmath221 and @xmath222 such that @xmath223 and @xmath224 . then , one has , for each @xmath225 , @xmath226 since @xmath223 and @xmath224 . thus , noting that @xmath195 is arbitrary , one can prove the theorem by lemma [ lem : isotropic ] . [ [ the - dirac - sum - of - induced - dirac - structures . ] ] the dirac sum of induced dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath227 and @xmath228 be distinct configuration spaces . let @xmath229 and @xmath230 be constraint distributions , and we can define the induced dirac structures @xmath209 and @xmath210 , where we assume that @xmath231 and @xmath232 . let @xmath233 be an extended configuration manifold and @xmath234 be a distribution on @xmath48 . as before , define the induced distribution @xmath61 on @xmath50 by @xmath235 where @xmath54 is the tangent map of the cotangent bundle projection @xmath236 , while the annihilator of @xmath61 is defined by , for each @xmath96 , @xmath237 let @xmath238 be the canonical symplectic structures on @xmath239 and @xmath240 be the associated bundle maps . let @xmath241 be the natural projections with @xmath242 . ] . denote by @xmath243 the dirac structures on @xmath239 induced from @xmath244 . then , the _ dirac sum _ @xmath245 can be represented by the induced dirac structure on @xmath50 from @xmath246 , which is given by , for each @xmath96 , @xmath247 where @xmath248 is the associated bundle map with @xmath249 . let @xmath250 for each @xmath96 . let @xmath251 . now , we may decompose @xmath252 as @xmath253 and @xmath254 such that @xmath255 and @xmath256 . define the distributions @xmath257 and @xmath258 . then , it follows that @xmath259 and @xmath260 , where @xmath261 and @xmath262 . noting that @xmath263 and @xmath264 , one has @xmath265 . additionally , @xmath266 . it follows on each fiber @xmath267 that @xmath268 and it is obvious that @xmath269 . putting these all together gives us @xmath270 and @xmath271 , for each @xmath96 . since @xmath272 is the canonical symplectic form of @xmath50 we find @xmath245 is the induced dirac structure from @xmath165 . [ [ an - interconnection - dirac - structure . ] ] an interconnection dirac structure . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in order to intertwine distinct dirac structures , we will introduce a special dirac structure called _ interconnecting dirac structure_. a simple case may be given by a distribution and its annihilator . suppose there exists a constraint distribution @xmath273 due to the interconnection , such that @xmath274 where @xmath275 is a one - form on @xmath48 associated with the interconnection . the annihilator @xmath276 is defined by @xmath277 we can define the lifted distribution @xmath278 on @xmath50 by @xmath279 where @xmath54 is the tangent map of the cotangent bundle projection @xmath236 , while the annihilator of @xmath280 is given by , for each @xmath96 , @xmath281 then , the interconnection dirac structure associated to the distribution @xmath280 on @xmath50 can be defined by , for each @xmath96 , @xmath282 [ [ example - two - particles - moving - with - contact . ] ] example : two particles moving with contact . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider two masses moving with contact in a vector space @xmath283 , whose velocities are given by @xmath284 . since the two particles are in contact and their velocities are common , it follows @xmath285 where @xmath286 is a constraint subspace of @xmath10 . this constraint is enforced through the associated constraint forces @xmath287 at the contact point , where @xmath288 is the force associated with the @xmath289-th particle . this implies newton s third law and it follows @xmath290 where @xmath291 is the annihilator of @xmath278 . note that we can define a dirac structure on @xmath10 by @xmath292 . then , it follows that @xmath293 . [ [ remark.-3 ] ] remark . + + + + + + + we can consider a more general interconnection dirac structure than stated in the above , such as those given by gyrators , transformers and so on , which are expressed by two - forms or poisson structures . in this paper , we will mainly focus on the simple case of the interconnection dirac structure @xmath1 induced from a given distribution @xmath278 [ [ strategy - for - the - interconnection - of - distinct - dirac - structures . ] ] strategy for the interconnection of distinct dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + before going into details , let us outline our strategy for interconnecting dirac structures . let @xmath209 and @xmath210 be two distinct dirac structures on distinct manifolds @xmath202 and @xmath203 . we will interconnect @xmath209 and @xmath210 by introducing an _ interconnection _ dirac structure @xmath1 on @xmath294 , by which constraints due to the interconnection are imposed . to do this , we first make the direct sum of dirac structures , namely @xmath245 , which is a single dirac structure on @xmath21 . then , by introducing the _ bowtie product _ @xmath295 of dirac structures on @xmath21 , we define an interconnected dirac structure as @xmath296 . the bowtie product will be shown to make @xmath297 a commutative category with identity element @xmath298 . under the condition that the fiber dimension of @xmath299 is constant , it will be shown that @xmath300 is indeed a dirac structure on @xmath21 . [ [ the - bowtie - product . ] ] the bowtie product . + + + + + + + + + + + + + + + + + + + let @xmath301 and let @xmath302 be a submanifold given as @xmath303 . given dirac structures @xmath304 , @xmath305 on @xmath21 , define the direct product of dirac structures @xmath304 and @xmath305 by @xmath306 on @xmath307 . recall from @xcite that one may pull back the dirac structure @xmath308 on @xmath307 to @xmath309 as @xmath310 which is a lagrangian submanifold of @xmath311 . assume that @xmath312 is a smooth sub - bundle by either of the following conditions : * @xmath313 has constant dimension , * @xmath314 has constant dimension . the subbundle @xmath312 is a dirac structure on @xmath309 . let @xmath315 be the map @xmath316 ( this is a diffeomorphism between @xmath21 and @xmath309 ) . then @xmath317 given by @xmath318 is called the _ diagonal embedding_. given dirac structures @xmath319 , we may define the dirac structure @xmath320 . noting @xmath321 , it follows that @xmath322 \in \daleth(m).\ ] ] let @xmath304 and @xmath323 be dirac structures over @xmath21 . then , we define a tensor product called the bowtie product of @xmath304 and @xmath323 by @xmath324 or alternatively given by : @xmath325 under the assumption that @xmath326 has constant dimension , @xmath327 is a dirac structure on @xmath21 . [ [ remark.-4 ] ] remark . + + + + + + + in this paper , we focus on how to use the bowtie product @xmath2 to make a dirac structure for an interconnected lagrange dirac dynamical system . while the bowtie product was initially proposed in @xcite for this purpose using formula , it was revealed that an equivalent _ tensor product _ ( denoted by @xmath328 ) was also developed in the context of generalized complex geometry ( see @xcite and @xcite ) using formula . we will make explicit the equivalence of @xmath2 and @xmath328 ( i.e. and ) in the appendix . [ [ remark.-5 ] ] remark . + + + + + + + for standard lagrange - dirac mechanics , we usually employ @xmath329 , @xmath304 is the canonical dirac structure on @xmath21 , and @xmath330 for some distribution @xmath331 is an interconnection dirac structure . though we will not explore this in detail , one of the particularly interesting interconnection dirac structures is the case in which @xmath305 can be developed by lagrangians that are linear in velocities of the form @xmath332 , where @xmath333 and @xmath334 is a one - form on @xmath48 causing magnetic terms to appear . in order to show interconnection of dirac structures is associative , we will use a special restricted two - form @xmath32 induced by a dirac structure @xmath6 on @xmath21 with a distribution @xmath335 , where @xmath336 . this two - form will be employed in the proofs for theorems [ thm : bowtie ] and [ associative ] . [ lem : two - form ] let @xmath337 and set @xmath335 . on each fiber of @xmath338 at @xmath28 there exists a bilinear anti - symmetric map @xmath339 defined as @xmath340 this two - form was initially introduced by @xcite for linear dirac structures ( see also @xcite ) . we can easily generalize it to the case of general manifolds since @xmath32 is defined fiber - wise , and hence we have that @xmath26 is a linear dirac structures on each fiber @xmath341 for all @xmath28 . here , we present a proof for completeness . we desire to show that equation defines a unique bi - linear anti - symmetric map @xmath32 . first we prove linearity . let @xmath342 and let @xmath343 . since @xmath26 is a linear space at each @xmath28 we have @xmath344 . by definition of @xmath32 , it follows , for all @xmath345 , @xmath346 thus , @xmath347 is linear in the first argument . second , in order to prove that @xmath32 is anti - symmetric , take any @xmath348 . since @xmath6 is isotropic , we find @xmath349 and we compute @xmath350 thus , we have shown anti - symmetry from which bi - linearity follows . given a dirac structure @xmath337 , we call the two - form @xmath32 of lemma [ lem : two - form ] the dirac two - form . recall from equation that , for each @xmath28 , @xmath26 may be given by @xmath351 which may be also stated by @xmath352 where @xmath353 is the skew - symmetric bundle map that is a natural extension of the skew - symmetric map @xmath354 , which is defined by @xmath355 on @xmath356 . [ thm : bowtie ] let @xmath304 and @xmath357 . let @xmath358 and let @xmath359 and @xmath360 be the dirac two - forms for @xmath304 and @xmath305 respectively . if @xmath361 has constant rank , then @xmath362 is a dirac structure with distribution @xmath363 and with the dirac two - form @xmath364 . [ pf : bowtie ] let @xmath365 for each @xmath28 . then , there exists @xmath366 such that @xmath367 for each @xmath28 . this is equivalent to @xmath368 where @xmath369 and @xmath370 . this means @xmath371 and @xmath372 . but @xmath373 . therefore , upon setting @xmath374 and @xmath375 , we can write @xmath376 and @xmath377 ; namely , @xmath378 , where @xmath379 is a dirac structure with @xmath278 and @xmath380 . then , it follows that @xmath381 . equality follows from noting that both @xmath382 and @xmath383 are subspaces of @xmath384 with the same dimension . [ cor : bowtie2 ] if @xmath385 , then it follows that @xmath386 and also that @xmath387 is induced by the constraint @xmath388 and the two - form @xmath359 . [ associative ] let @xmath389 with distributions @xmath390 , @xmath391 , and @xmath392 . assume that @xmath361 , @xmath393 and @xmath394 have constant ranks . then the bowtie product @xmath2 is associative and commutative ; namely we have @xmath395 and @xmath396 first we prove commutativity . recall that any dirac structure may be constructed by its associated constraint distribution @xmath335 and the dirac two - form @xmath347 . let @xmath397,and @xmath398 be the dirac two - forms corresponding to @xmath399 , and @xmath400 respectively . then we find by theorem [ thm : bowtie ] that @xmath401 is defined by the distribution @xmath402 and the dirac two - form @xmath403 . by commutativity of @xmath404 and @xmath405 , we find the same distribution and the two - form for @xmath406 , we have @xmath407 . now we prove associativity . let @xmath408 and @xmath409 and it follows @xmath410 if @xmath411 and @xmath412 are respectively the dirac two - forms for @xmath413 and @xmath414 , we find @xmath415|_{\delta_{(ab)c } } \\ & = ( \omega_{\delta_{a } } + \omega_{\delta_{b } } + \omega_{\delta_{c } } ) |_{\delta_{(ab)c } } \\ & = ( \omega_{\delta_{a } } + \omega_{\delta_{b } } + \omega_{\delta_{c } } ) |_{\delta_{a(bc)}}\\ & = \omega_{\delta_{a(bc)}}.\end{aligned}\ ] ] thus , we obtain @xmath416 [ [ remark.-6 ] ] remark . + + + + + + + we have shown that @xmath2 acts on pairs of dirac structures with clean intersections to give a new dirac structure and also that that it is an associative and commutative product . it is easy to verify from equation that the dirac structure @xmath417 satisfies the property of the identity element as @xmath418 for every @xmath337 . thus we find that @xmath297 is a commutative category ( it lacks totality and an inverse element , so it is neither a monoid nor a groupoid ) . [ [ interconnection - of - two - distinct - induced - dirac - structures . ] ] interconnection of two distinct induced dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath419 and @xmath420 be dirac structures on distinct manifolds @xmath227 and @xmath228 induced from constraint distributions @xmath421 and @xmath422 as before . let @xmath278 be a given distribution on @xmath423 due to the interconnection of @xmath209 and @xmath210 . we have mathematical ingredients for _ interconnecting _ the induced dirac structures @xmath209 and @xmath210 through the interconnection dirac structure @xmath1 on @xmath48 . recall that an interconnection dirac structure is given by @xmath424 , where @xmath425 is the constraint distribution associated to the interconnection . recall that @xmath426 and @xmath427 . it follows from theorem [ thm : bowtie ] that @xmath428 may be constructed from the constraint @xmath429 and @xmath430 , where @xmath431 and @xmath432 are respectively the canonical two - forms on @xmath433 and @xmath434 . [ prop : velocity_constraint_bowtie ] assuming @xmath30 has constant rank , the interconnection of two distinct induced dirac structures @xmath209 and @xmath210 through @xmath435 , namely , @xmath436 is represented by , @xmath437 this may be viewed as a corollary to theorem [ thm : bowtie ] and proposition [ induced_ds_dirac_sum ] . [ [ interconnection - of - n - distinct - dirac - structures . ] ] interconnection of @xmath183 distinct dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us consider the interconnection of @xmath183 distinct dirac structures @xmath438 on distinct manifolds @xmath439 , in which @xmath440 . recall that the dirac sum @xmath8 is associative as @xmath441 so that we may define the iterated direct sum by @xmath442 by choosing the appropriate interconnection dirac structure @xmath443 and assuming that @xmath444 is constant , we can simply develop the interconnected dirac structure for @xmath445 distinct dirac structures as @xmath446 the condition that @xmath447 is constant is equivalent to the assumption that @xmath448 is constant , which can be shown by resorting to local coordinate expressions . [ [ a - link - between - composition - and - interconnection - of - dirac - structures . ] ] a link between composition and interconnection of dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the notion of _ composition _ of dirac structures was introduced in @xcite in the context of port - hamiltonian systems . let @xmath449 and @xmath450 be vector spaces . let @xmath419 be a linear dirac structure on @xmath451 and @xmath420 be a linear dirac stucture on @xmath452 . the _ composition _ of @xmath419 and @xmath420 is given by @xmath453 where @xmath454 and @xmath455 denote the dual space of @xmath456 and @xmath457 . it was also shown that the set @xmath458 is itself a dirac structure on @xmath459 . in this section , we shall see the link between the composition and the interconnection of dirac structures . the composition developed in @xcite is constructed on vector - spaces . so , we also focus on the vector space case in this paper . let @xmath460 and @xmath461 . before going into details , we establish an important fact on the projection from @xmath10 to @xmath462 . set @xmath463 to be the projection @xmath464 . the mapping @xmath465 dual to @xmath466 is given by @xmath467 next , we recall the push - forward map associated to a dirac structure . let @xmath64 and @xmath468 be vector spaces . let @xmath469 be a linear map and let @xmath470 be a linear dirac structure on @xmath64 . the push - forward of @xmath470 to @xmath468 by @xmath471 is the set @xmath472 it is worth noting that the push - forward of a dirac structure @xmath473 is itself a dirac structure @xmath474 ; namely , we write it as @xmath475 see , for instance , @xcite and @xcite . we are now ready to provide the link between the bowtie product and the composition of dirac structures . [ thm : composition ] let @xmath476 . set @xmath477 . set @xmath463 to be the projection @xmath464 . for linear dirac structures @xmath419 on @xmath478 and @xmath420 on @xmath479 , set @xmath480 and set also @xmath481 then , one has @xmath482 first , we observe that @xmath483 . we also observe @xmath484 utilizing the fact that @xmath485 if and only if @xmath486 and @xmath487 , we may restate the above as @xmath488 finally , we have @xmath489 but this is nothing but the composition of dirac structures , namely , @xmath490 . given distinct mechanical systems , one can develop dynamics of each mechanical system in the context of both the lagrange - dalembert - pontryagin variational structures and the induced dirac structures over separate configuration manifolds . in this section , we will show how dynamics of an interconnected systems can be formulated in the context of the induced dirac structures as well as in the context of the variational structures . [ [ variational - structures - for - interconnected - mechanical - systems . ] ] variational structures for interconnected mechanical systems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us introduce an interconnection of the lagrange - dalembert - pontryagin principles for distinct mechanical systems . let @xmath491 be distinct configuration manifolds corresponding to distinct mechanical systems @xmath492 . let @xmath493 be the generalized energies associated to given lagrangians ( possibly degenerate ) @xmath494 on @xmath495 and @xmath496 be external force fields . let @xmath497 be a distribution on @xmath491 for @xmath492 . the lagrange - dalembert - pontryagin principles for curves @xmath498 in @xmath499 is denoted by @xmath500 for @xmath501 and with @xmath502 . the interconnection of distinct lagrange - dalembert - pontryagin principles may be given by imposing some distribution @xmath278 on the extended configuration manifold @xmath503 such that @xmath504 it follows from equations and that , by using variations @xmath505 vanishing at the endpoints , one can obtain @xmath506 by direct computations and integration by parts , it follows that @xmath507 for @xmath508 and with @xmath509 . imposing the interconnection condition , we obtain implicit lagrange - dalembert equations for the interconnected system . [ [ an - alternative - lagrange - dalembert - pontryagin - principle . ] ] an alternative lagrange - dalembert - pontryagin principle . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + here , we show that there exists an alternative lagrange - dalembert - pontryagin principle . let @xmath503 and let @xmath510 , @xmath511 and @xmath512 . define the generalized energy for the interconnected system by @xmath513 , where @xmath514 is the lagrangian for the interconnected system . set the constraint distribution @xmath515 . the lagrange - dalembert - pontryagin principle for a curve @xmath516 in @xmath133 may be given by @xmath517 for the chosen variation @xmath518 , for all @xmath178 and @xmath179 , and with @xmath519 . + keeping the endpoints of @xmath164 fixed , it follows that one can simply derive the lagrange - dalembert - pontryagin equations as follows : @xmath520 the lagrange - dalembert - pontryagin principles in for curves @xmath521 @xmath522 , @xmath523 in @xmath499 together with the interconnection condition in is equivalent with the lagrange - dalembert - pontryagin principle in . it follows from the lagrange - dalembert - pontryagin principles that one can obtain equation as @xmath524 imposing the interconnection condition @xmath525 and @xmath526 , one has @xmath527 and @xmath528 where we note that the final distribution is given by @xmath529 and hence that its annihilator is denoted by @xmath530 the following statements are equivalent : 1 . a curve @xmath531 in @xmath133 satisfies the lagrange pontryagin principle : @xmath532 with respect to variations @xmath533 and @xmath534 arbitrary and with fixed endpoints of @xmath164 . curves @xmath535 satisfy the lagrange dalembert pontryagin principles @xmath536 under variations @xmath537 , arbitrary @xmath538 and @xmath539 , together with @xmath502 and fixed endpoints of @xmath540 , such that the force vector @xmath541 and the velocity vector @xmath542 satisfy @xmath543 where @xmath544 denote force fields . [ [ interconnected - lagrange - dirac - dynamical - systems . ] ] interconnected lagrange - dirac dynamical systems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + next , consider the interconnection of distinct lagrange - dirac dynamical systems corresponding to the above variational structures . to do this , we first consider the interconnection of distinct dirac structures @xmath243 on @xmath545 , each of which is induced from given distinct distributions @xmath546 . let @xmath547 be @xmath183 distinct lagrange - dirac systems , where @xmath548 denote the generalized energies on @xmath499 associated to given lagrangians @xmath494 on @xmath495 , @xmath549 are the partial vector fields , and the external force fields @xmath550 . recall that @xmath551 induce the horizontal one - forms @xmath552 on @xmath499 such that @xmath553 for all @xmath554 , where @xmath555 are natural projections . as before , for an interconnected system , let @xmath503 and @xmath278 be a given distribution on @xmath48 associated to the interconnection . let @xmath515 . then , define an interconnection dirac structure by @xmath556 as in , where @xmath557 . set a generalized energy @xmath513 associated to the lagrangian @xmath558 . furthermore , define a partial vector field on @xmath133 by @xmath559 define also an external force field by @xmath560 which induces the horizontal one - form @xmath561 on @xmath133 such that @xmath562 for all @xmath563 , where @xmath564 is the natural projection . recall from that the interconnection of distinct dirac structures @xmath565 through @xmath1 is given by @xmath566 the interconnected lagrange - dirac dynamical system is a quadruple @xmath103 that satisfies @xmath567 where @xmath568 . using local coordinates @xmath510 , @xmath569 , and @xmath570 , one has @xmath571 [ prop : ldds_splitting ] the following statements are equivalent : 1 . a curve @xmath572 satisfies @xmath573 where @xmath574 . 2 . curves @xmath575 satisfy @xmath576 where @xmath577 , such that @xmath578 where @xmath579 . recall that @xmath580 , the condition @xmath581 is equivalent to the existence of some covector @xmath582 such that @xmath583 and @xmath584 it follows from the condition that @xmath585 and also from condition it follows that @xmath586 and @xmath587 . furthermore , noting @xmath588 , it follows that the curves @xmath589 satisfy the conditions @xmath590 together with @xmath591 . one can easily check the converse . thus , we obtain the following theorem : [ thm : ldds_splitting ] the following statements are equivalent : 1 . a curve @xmath592 in @xmath593 satisfies @xmath594 where @xmath595 . 2 . curves @xmath596 satisfy @xmath597 where @xmath598 and @xmath599 is the horizontal one - forms on @xmath600 of force fields @xmath601 , such that the condition of interconnection holds @xmath602 3 . a curve @xmath592 satisfies the hamilton pontryagin principle : @xmath603 with respect to variations @xmath604 and @xmath534 arbitrary , with @xmath605 and with fixed endpoints of @xmath164 . curves @xmath596 satisfy the lagrange dalembert pontryagin principles : @xmath606 where @xmath501 , and @xmath607 are arbitrary , @xmath502 and with fixed endpoints of @xmath164 , such that the interconnection condition for the force vector @xmath541 and the velocity vector @xmath542 holds as @xmath608 in this section we provide several examples of how interconnection can be carried out with the bowtie product . we have chosen rather simple examples , however the advantage of using the bowtie product is based on the procedure of tearing and interconnecting systems , which enables us to treat more complicated systems in a hierarchical structure . in future , we will seek more complex uses of the bowtie product . * ( i ) example : mass - spring mechanical systems . * + let us consider an illustrative example of a mass - spring system in the context of the interconnection of dirac structures and associated lagrange - dirac systems . in this example , there exists _ no external force_. let @xmath609 and @xmath610 be the @xmath289-th mass and spring @xmath611 . [ [ tearing - and - interconnection . ] ] tearing and interconnection . + + + + + + + + + + + + + + + + + + + + + + + + + + + + inspired by the concept of _ tearing and interconnection _ originally developed by @xcite , the mechanical system can be torn apart into two subsystems as in fig [ tornapart ] , each of which can be regarded as a modular unit of the whole system . let us call each disconnected system a `` primitive system '' by following @xcite . note that the procedure of tearing inevitably yields _ interactive boundaries _ associated to the two disconnected subsystems , where it induces the following condition : @xmath612 without the above condition , there exists no energetic interaction between the disconnected systems . in other words , _ the original mechanical system can be reconstructed by interconnecting the subsystems through the continuity conditions_. the continuity conditions in imply the continuity of power flow ; namely , the _ power invariance _ holds as @xmath613 where @xmath614 and @xmath615 needless to say , equation can be understood as the defining condition for an interconnection dirac structure . [ [ primitive - lagrangian - systems . ] ] primitive lagrangian systems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us consider how dynamics of the disconnected subsystems can be formulated in the context of lagrange - dirac dynamical systems . the configuration space of the subsystem 1 may be given by @xmath616 with local coordinates @xmath617 , while the configuration space of the subsystem 2 is @xmath618 with local coordinates @xmath619 . we can naturally define the canonical dirac structures @xmath620 and @xmath621 in this example . for primitive system 1 , the lagrangian @xmath622 is given by @xmath623 while the lagrangian @xmath624 for primitive system 2 is given by @xmath625 then , we can define the generalized energy @xmath626 on @xmath627 by @xmath628 and also define the generalized energy @xmath629 on @xmath630 by @xmath631 . though the original system has no external force , each disconnected primitive system has an interconnection constraint force at the interactive boundary . when viewing each system separately , the constraint force acts as an external force . this is because _ tearing always yields constraint forces at the boundaries associated with the disconnected primitive systems _ , as shown in fig . [ tornapart ] the constraint force fields at the boundaries @xmath632 and @xmath633 are given by @xmath634 an @xmath635 . further , let @xmath636 be the partial vector field , which is defined at each point @xmath637 as @xmath638 similarly , let @xmath639 be the partial vector field , which is defined at each point @xmath640 as @xmath641 together with the _ consistency condition _ @xmath642 . primitive system 1 : we can formulate dynamics of primitive system 1 in the context of the lagrange - dirac dynamical system @xmath643 as @xmath644 which may be given in coordinates by @xmath645 and with @xmath646 and @xmath647 . primitive system 2 : similarly , we can also formulate dynamics of primitive system 2 in the context of the lagrange - dirac dynamical system @xmath648 as @xmath649 which may be given in coordinates by @xmath650 together with @xmath651 and @xmath652 as well as @xmath642 . recall that each primitive system is physically independent of the other , which implies that there exists no energetic interaction between them . next , let us see how the disconnected primitive systems can be interconnected through a dirac structure . [ [ interconnection - of - distinct - dirac - structures . ] ] interconnection of distinct dirac structures . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath653 be an extended configuration space with local coordinates @xmath654 . recall that the dirac sum of the dirac structures is given by @xmath245 on @xmath50 . now , the constraint distribution due to the interconnection is given by @xmath655 where @xmath656 is a one - form on @xmath48 . on the other hand , the annihilator @xmath657 is defined by @xmath658 it follows from this codistribution that @xmath659 , @xmath660 and @xmath661 . hence , we obtain the conditions for the interconnection as in equation ; namely , @xmath662 and @xmath663 . recall that the distribution @xmath280 on @xmath50 is obtained by @xmath664 and hence the dirac structure @xmath665 can be defined as in . further , recall that the interconnected dirac structure @xmath6 on @xmath50 is given by @xmath666 [ [ interconnected - lagrange - dirac - systems . ] ] interconnected lagrange - dirac systems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + now , let us see how two primitive lagrange - dirac systems , namely , @xmath667 and @xmath668 can be interconnected to be a lagrange - dirac dynamical system . define the lagrangian @xmath669 for the interconnected system by @xmath670 and hence the generalized energy is given by @xmath671 . . set a partial vector field by @xmath673 , which is defined at points @xmath674 , where the consistency condition @xmath642 holds . finally , the interconnected lagrange - dirac system is given by @xmath675 , which satisfies , for each @xmath676 , @xmath677 and with @xmath678 . [ [ the - lagrange - dalembert - pontryagin - principle.-1 ] ] the lagrange - dalembert - pontryagin principle . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + alternatively , we can develop the interconnected lagrange - dirac system by the lagrange - dalembert - pontryagin principle : @xmath679 for all @xmath680 , for all @xmath681 and @xmath682 , and with @xmath683 . thus , we can obtain the dynamics of the interconnected lagrange - dirac system . * ( ii ) example : electric circuits * + consider the electric circuit depicted in figure [ vrlc_circuit_pic ] , where @xmath684 denotes a resistor , @xmath172 an inductor , and @xmath685 a capacitor . as in figure [ torn_circuit_pic ] , we decompose the circuit into the disconnected two circuits , each of which we shall call `` primitive circuit 1 '' and `` primitive circuit 2 '' , in which @xmath686 and @xmath687 denote external ports . in order to reconstruct the original circuit in figure [ vrlc_circuit_pic ] , the external ports may be connected by equating currents across each . [ [ primitive - circuit-1 . ] ] primitive circuit 1 . + + + + + + + + + + + + + + + + + + + + the configuration manifold for circuit 1 is denoted by @xmath688 with local coordinates @xmath689 , where @xmath690 and @xmath691 are the charge fluxes through the resistor @xmath684 , inductor @xmath172 and port @xmath686 respectively . there exists the kcl constraint distribution @xmath229 in circuit 1 , which is given by for each @xmath692 , @xmath693 where @xmath694 denotes the current vector at each @xmath695 , while the kvl constraint is given by its annihilator @xmath696 . then , we can naturally define the induced dirac structure @xmath419 on @xmath697 from @xmath698 as before . the lagrangian for circuit 1 , namely , @xmath699 on @xmath700 is given by @xmath701 which is degenerate . define the generalized energy @xmath702 on @xmath627 by @xmath703 on @xmath627 . circuit 1 also has the external force field due to the resistor and the external port , @xmath704 as @xmath705 so , we can set up the lagrange - dirac dynamical system @xmath706 . [ [ primitive - circuit-2 . ] ] primitive circuit 2 . + + + + + + + + + + + + + + + + + + + + the configuration manifold for circuit 2 is @xmath707 with local coordinates @xmath708 , where @xmath709 is the charge flux through the port @xmath710 and @xmath711 is the charge stored in the capacitor . the kcl constraint distribution is given by , for each @xmath712 , @xmath713 and so the kvl space is given by its annihilator @xmath714 . this gives us the dirac structure @xmath210 on @xmath715 . set the lagrangian @xmath716 for circuit 2 as @xmath717 and set the generalized energy @xmath718 on @xmath630 . circuit 2 has the external force field due to the resistor and the external port @xmath687 , namely , @xmath719 as @xmath720 then , we can formulate the lagrange - dirac dynamical system @xmath721 . [ [ the - interconnection - dirac - structure ] ] the interconnection dirac structure + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + set @xmath722 . the interconnection constraint is given by @xmath723 by the tangent lift @xmath724 , we can define the interconnection dirac structure as @xmath725 which is denoted , in local coordinates , by @xmath726 [ [ the - interconnected - system . ] ] the interconnected system . + + + + + + + + + + + + + + + + + + + + + + + + + + now , we can define the interconnected dirac structure by @xmath727 which may be viewed as the induced dirac structure from the constraint space @xmath728 which is given , in coordinates @xmath729 , by @xmath730 set the lagrangian for the interconnected system as @xmath731 and the external force field @xmath732 . set also @xmath733 . let @xmath734 be a partial vector field . the interconnected lagrange - dirac dynamical system is given by the quadruple @xmath735 , which satisfies @xmath736 where @xmath737 . * ( iii ) example : a rolling ball on rotating tables * + consider the mechanical system depicted in fig . [ fig : gear_system ] , where there are rotating tables via a gear and a ball is rolling on one of the tables without slipping . we assume that there is no loss of energy and non - slipping in the gear and also that the external torque is constant . let @xmath738 be moments of inertia for the table @xmath739 . let us decompose the system into distinct three systems ; ( 1 ) a rotating ( small ) table , ( 2 ) a rotating ( large ) table , and ( 3 ) a rolling ball on a table . [ [ primitive - system-1 . ] ] primitive system 1 . + + + + + + + + + + + + + + + + + + + for primitive system 1 the configuration manifold is the circle , @xmath740 . the lagrangian is the rotational energy of the system . @xmath741 and set the generalized energy @xmath742 on @xmath627 . we employ the canonical dirac structure on @xmath697 as @xmath743 there is a constant torque acting on primitive system 1 , given by a map @xmath744 defined as @xmath745 , where @xmath746 is constant in @xmath747 . [ [ primitive - system-2 . ] ] primitive system 2 . + + + + + + + + + + + + + + + + + + + for primitive system 2 , the configuration manifold is also the circle , @xmath748 and the lagrangian is again the rotational energy @xmath749 and set the generalized energy @xmath750 on @xmath630 . again , we have the canonical dirac structure @xmath751 [ [ primitive - system-3 . ] ] primitive system 3 . + + + + + + + + + + + + + + + + + + + primitive system 3 is a rolling sphere of uniform density and radius @xmath752.the sphere moves in space by changing its position and orientation relative to a reference configuration . the configuration manifold is given by the special euclidean group @xmath753 , which we parametrize as @xmath754 where @xmath755 . following @xcite , let @xmath756 be the set of points of the sphere in the reference configuration . for configuration @xmath757 , a point @xmath758 is transformed into @xmath759 by the action @xmath760 . the lagrangian is given by the kinetic energy as @xmath761 where @xmath762 . we use body coordinates such that the center of the sphere in the reference configuration is at the origin so that @xmath763 . substituting these relations , the above lagrangian is to be @xmath764 setting @xmath765 and noting that @xmath766 when @xmath767 , one finally obtains @xmath768 and with the generalized energy @xmath769 on @xmath770 . since the motion along the @xmath771-direction is constrained so that the ball does not leave the plane of table 2 , we get the holonomic constraint @xmath772 this yields the induced dirac structure @xmath773 [ [ interconnection - constraints . ] ] interconnection constraints . + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath774 . in order to interconnection the three primitive systems , we need to impose the constraints due to the non slip conditions . first , consider the mapping @xmath775 for @xmath776 . the map @xmath777 sends the tangent fiber @xmath778 to the lie - algebra @xmath779 , where @xmath780 is the identity element on @xmath781 viewed as a group . from this point on , we do not explicitly invoke @xmath777 but use it to interpret @xmath782 as @xmath783 . the interconnection constraint between system 1 and system 2 is given by @xmath784 and with its annihilator @xmath785 where @xmath786 . this constraint ensures that the gears rotate ( without slipping ) at the same speed in opposite directions . next , we consider the interconnection constraint between systems 2 and 3 . note that the velocity of a point located at the bottom of the sphere is given by @xmath787 note also that a point rotating on the gear of system 2 with the axle taken to be the origin has velocity @xmath788 so the non slip condition between system 2 and 3 is given by @xmath789 where @xmath790 are the basis on @xmath759 . setting the interconnection constraint distribution @xmath791 let @xmath792 and with its annihilator @xmath793 . then , the interconnection dirac structure is given by @xmath794 [ [ the - interconnected - lagrange - dirac - system . ] ] the interconnected lagrange - dirac system . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the dirac structure for the interconnected system is given by @xmath795 and we use the lagrangian @xmath796 . the dynamics of the system are implied by the constraint : @xmath797 where @xmath798 is a partial vector field , and @xmath101 is the generalized energy of @xmath172 . this gives us the interconnected lagrangian system @xmath103 . by theorem [ thm : bowtie ] , these dynamics use the constraint distribution @xmath799 and the canonical two - form on @xmath50 restricted to @xmath30 as the dirac two - form . noting the annihilator is given by @xmath800 we finally obtain the dynamics as @xmath801 in this paper , we provided a notion of interconnection for dirac structures and their associated lagrange - dirac dynamical systems . the principal idea lies in the fact that the _ interconnection of distinct dirac structures itself is a dirac structure_. we also clarified the interconnection of the associated lagrange - dirac dynamical systems with the variational structure by the lagrange - dalembert - pontryagin principle . to do this , we developed the bowtie product , @xmath2 , which serves us well in the case of interconnection by constraints . we then showed how the bowtie product can effectively carry the lagrange - dalembert - pontryagin principle to the interconnected system . lastly , we demonstrated our theory with some illustrative examples of a mass - spring mechanical system , an interconnection in circuits , and a non - holonomic mechanical system . we raise the following topics as arena s for further investigation : * applications to complicated systems such as guiding central motion problems , multibody systems , fluid - rigid body interactions , passivity controlled interconnected systems , etc . ( see , for example , @xcite and @xcite ) . * reduction and symmetry for interconnected lagrange - dirac systems ( @xcite , @xcite ) . * the use of more general interconnection dirac structures with nontrivial dirac two - forms such as the one given with symplectic gyrators ( in this paper , we restrict @xmath1 to the form @xmath802 ) . as to examples of such general interconnections in physical systems , see for instance @xcite and @xcite . * the integrability condition for the bowtie product of dirac structures . as to the integrability condition for dirac structures , see @xcite and @xcite . the link with _ symplectic categories _ ( see @xcite ) . * discrete version of interconnected lagrange - dirac mechanics ; namely , discretizing the hamilton - pontryagin principle one arrives at a discrete mechanics version of dirac structures ( see @xcite and @xcite ) . we will show the equivalence between the bowtie product @xmath2 and the dirac tensor product @xmath328 developed by @xcite , where the definition of the tensor product @xmath328 is given as follows . let @xmath803 . let @xmath804 be the diagonal embedding in @xmath805 . the tensor product of @xmath304 and @xmath305 is defined as @xmath806 where @xmath807 and @xmath808 is the orthogonal complement in @xmath809 . first note that @xmath812 so that @xmath813 . then we see that @xmath814 and the equivalence class @xmath815 is @xmath816 where we chose the element @xmath817 as the representative for the equivalence class @xmath818 = \ { ( v , v,\alpha + \xi,\alpha-\xi ) , \forall \xi \in t^{\ast}p \ } \in d_a \boxtimes d_b.\ ] ] splitting the expression @xmath306 into @xmath819 and @xmath323 , we obtain @xmath820 we see that the assignment @xmath821 is a bijection and also that @xmath822 this is just the definition of @xmath401 . 150 afshari , e. , h. s. bhat , a. hajimiri , and j. e. marsden [ 2006 ] , extremely wideband signal shaping using one- and two - dimensional nonuniform nonlinear transmission lines , _ journal of applied physics _ , * 99 * , 05401116 . bloch , a. m. and p. e. crouch [ 1997 ] , representations of dirac structures on vector spaces and nonlinear l - c circuits . differential geometry and control ( boulder , co , 1997 ) , _ proc . pure math . _ * 64*. amer . soc . , providence , ri , pages 103117 . bloch , a. m. [ 2003 ] , _ nonholonomic mechanics and control _ , volume 24 of _ interdisciplinary applied mathematics_. springer - verlag , new york . with the collaboration of j. baillieul , p. crouch and j. marsden , and with scientific input from p. s. krishnaprasad , r. m. murray and d. zenkov . chang , d. , a. m. bloch , n. leonard , j. e. marsden , and c. woolsey [ 2002 ] , the equivalence of controlled lagrangian and controlled hamiltonian systems , _ control and the calculus of variations ( special issue dedicated to j.l . lions ) _ , * 8 * , 393422 . courant , t. and a. weinstein [ 1988 ] , beyond poisson structures . in _ action hamiltoniennes de groupes . troisime thorme de lie ( lyon , 1986 ) _ , volume 27 of _ travaux en cours _ , pages 3949 . hermann , paris . ortega , r. , j.a.l . perez , p. j. nicklasson , h. j. sira - ramirez and h. sira - ramirez [ 1998 ] , _ passivity - based control of euler - lagrange systems : mechanical , electrical , and electromechanical applications_. communications and control engineering series , springer verlag . ortega , r. , a. j. van der schaft , b. maschke , and g. escobar [ 2002 ] , interconnection and damping assignment passivity - based control of port - controlled hamiltonian systems , _ automatica _ , * 38*(4 ) , 585596 . yoshimura , h. and j. e. marsden [ 2006c ] , dirac structures and implicit lagrangian systems in electric networks , proc . of the 17th international symposium on mathematical theory of networks and systems , paper wea08.5 , 2006 , pp . 16 , july 24 - 28 , 2006 , kyoto . yoshimura , h. and j. e. marsden [ 2007a ] , dirac structures and the legendre transformation for implicit lagrangian and hamiltonian systems , _ lagrangian and hamiltonian methods in nonlinear control 2006 _ , volume * 366 * of _ lecture notes in control and information science series _ , pages 233247 , springer - verlag . yoshimura , h. and j. e. marsden [ 2008 ] , representations of dirac structures and implicit port - controlled lagrangian systems , _ proc.of international symposium on mathematical theory of networks and systems 2008 , blacksburg , virginia _ , pages 112 . yoshimura , h. , h. jacobs and j. e. marsden [ 2010 ] , _ interconnection of dirac structures and lagrange - dirac dynamical systems _ , _ proceedings of mathematical theory on networks and systems in 2010 , budapest _ , july 59 . pages 16 .

a massive star ends its life as a supernova ( sn ) explosion . this leaves behind a supernova remnant ( snr ) and sometimes creates a rapidly rotating pulsar . a pulsar can produce an outflow of a magnetic field and relativistic particles referred to as a pulsar wind . the interaction between such a pulsar wind and the ambient medium , i.e. the sn ejecta for pulsars embedded in snrs , results in a termination shock , beyond which the shocked wind materials inflate a broadband synchrotron - emitting bubble known as a pulsar wind nebula ( pwn ) . the evolution of pwne can be divided into several stages ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? . a pwn first expands supersonically inside its associated snr . the next stage starts when the sn reverse shock , which is driven inward by the interaction between the ejecta and the interstellar medium , crushes the pwn . the interplay between the shockwave and the pwn is complex and will cause the pwn to reverberate ( e.g. * ? ? ? after the effect of the reverse shock fades away , the pwn will expand subsonically into the snr . since a pulsar is generally born with high space velocity , it can be significantly off - centered with respect to the snr during the reverberation stage and can protrude from its own pwn because of the reverse shock interaction @xcite . this can result in a complicated pwn morphology consisting of a `` relic '' component and an elongated `` head '' bridging between the relic and the pulsar @xcite . as the pulsar continues to travel toward the edge of the snr , its motion will eventually become supersonic because the local sound speed in an snr decreases moving outward . the ram pressure acting on the pulsar wind due to the pulsar s supersonic motion will deform the pwn into a bow shock . for even older systems , the pulsar can escape from its associated snr shell and drive a bow - shock pwn into the interstellar medium . broadband emission from pwne can be observed from the radio to @xmath1-rays . in the radio regime , a pwn emits through the synchrotron process , and the emission is often highly linear polarized because of the ordered magnetic field configuration . a radio pwn is characterized by a centrally filled morphology and a flat spectrum , with a spectral index of @xmath2 ( @xmath3 ) . the synchrotron spectrum can extend to the x - ray band where the spectrum is usually steeper than that in the radio due to synchrotron cooling . gev and tev @xmath1-ray emission from pwne has also been detected , which is due to inverse - compton scattering ( see the review by * ? ? ? * ) . one interesting area to explore in the study of pwne is the interaction with the sn reverse shocks . theoretically , hydrodynamical ( hd ) modeling shows that the interplay can give rise to turbulence in pwne @xcite . however , the role of the magnetic field is unclear since magnetohydrodynamic ( mhd ) efforts are lacking . observationally , we aim to probe the magnetic field structure of these pwne to expand the currently inadequate sample size . the results from our study can serve as inputs to mhd modeling . one of the few examples of such systems is , which is an snr system containing a pwn that is believed to have been recently crushed by the reverse shock . it was discovered as a non - thermal radio source @xcite . its peculiar radio morphology was revealed by the molonglo observatory synthesis telescope ( most ) snr survey @xcite . the snr shell is @xmath4 in diameter and contains an off - centered pwn . the latter consists of a circular main body with a @xmath5 diameter together with a @xmath6-long finger - like structure protruding northwest from the main body . the first x - ray detection of g327.1@xmath01.1 was reported by @xcite using the _ einstein observatory_. further x - ray studies were conducted with _ rosat _ @xcite , _ asca _ @xcite , _ bepposax _ @xcite , _ chandra _ @xcite , and _ xmm - newton _ @xcite . the x - ray remnant consists of a diffuse thermal component covering the radio shell and a non - thermal component coinciding with the radio pwn . an x - ray compact source at the tip of the finger was discovered along with two prong - like structures extending northwest from the finger into a diffuse , arc - like structure @xcite . the compact source is interpreted as the neutron star powering the pwn , but the nature of the prongs and the arc is not well understood . @xcite estimated the distance to g327.1@xmath01.1 to be @xmath7 based on an empirical relationship between the x - ray column density @xmath8 and the color excess @xmath9 , and the observed relationship between @xmath9 and the distance @xmath10 along the line of sight to g327.1@xmath01.1 . we will adopt this number ( @xmath11 ) in our analysis . in @xmath1-rays , a preliminary report of tev detection of g327.1@xmath01.1 with h.e.s.s . was described by @xcite . as the photon index in tev was found to be similar to that in kev , they suggested that the emission in the two bands could originate from the same population of particles through inverse - compton and synchrotron processes , respectively . this source has not been detected in the gev range by _ fermi _ lat @xcite . by considering the current position of the presumed neutron star relative to the sn shell , @xcite suggested that the neutron star is moving northward . the main body of the pwn is interpreted as the relic , and the overall pwn morphology could be caused by the reverse shock first crushing the pwn from the northwest , thus pushing the relic to the southeast . the asymmetry of the reverse shock interaction is attributed to both the inhomogeneous ambient interstellar medium and the pulsar s motion . this picture is supported by recent hd simulations @xcite . it is believed that the pwn has not yet evolved into the bow - shock stage @xcite . in this paper , we present radio observations of the pwn in g327.1@xmath01.1 with the australia telescope compact array ( atca ) . we obtained polarimetric measurements at 6 and 3 cm to investigate the magnetic field structure of the pwn . the observations and data reduction procedures are described in section [ sec : obs ] . we show the results in section [ sec : results ] and discuss our findings in section [ sec : discussion ] . we close by summarizing our work in section [ sec : conclusion ] . radio observations of the pwn in g327.1@xmath01.1 were performed with atca at 6 and 3 cm on 2008 december 09 and 2009 february 19 with array configurations of 750b and ew352 , respectively . the observation parameters are listed in table [ table : para ] . all of the observations were carried out prior to the compact array broadband backend ( cabb ) upgrade @xcite . the data were taken in continuum mode with all four stokes parameters recorded over a usable bandwidth of 104mhz at each frequency . we observed for a total integration time of @xmath12 in each band as a five - pointing mosaic . pks b1934@xmath0638 was adopted as the primary calibrator to set the flux density scale , and pks 1613@xmath0586 was observed at 20 minute intervals to determine the antenna gain solution . we also processed archival atca pre - cabb data at 20 and 13 cm , for which the observation parameters are also listed in table [ table : para ] . the 13 cm observations were performed in continuum mode with the same usable bandwidth of 104mhz . however , the 20 cm observations were performed in spectral line mode on the h i line , such that only the total intensity ( i.e. no polarization ) within a narrow usable bandwidth of 4mhz was recorded . the simultaneous observations at 20 and 13 cm consist of a single pointing at g327.1@xmath01.1 , with array configurations of 1.5d and 750d for a total integration time of @xmath13 per band . pks b1934@xmath0638 was also used as the primary calibrator for these two bands , and pks 1610@xmath0771 was observed every 45 minutes for antenna gain calibration . cccccc 2000 apr 17 & 1.5d & 1420 , 2496 & 4 , 104 & 1025 , 13 & 9.1 + 2000 apr 29 & 750d & 1420 , 2240 & 4 , 104 & 1025 , 13 & 9.9 + 2008 dec 09 & 750b & 4800 , 8640 & 104 , 104 & 13 , 13 & 10.4 + 2009 feb 19 & ew352 & 4800 , 8640 & 104 , 104 & 13 , 13 & 10.4 + cccc 36 & @xmath14 & @xmath15 & @xmath16 + 20 & @xmath17 & @xmath18 & @xmath19 + 13 & @xmath20 & @xmath21 & @xmath22 + 6 & @xmath21 & @xmath23 & @xmath24 + 3 & @xmath25 & @xmath26 & @xmath27 + @xmath28 ( @xmath3 ) & @xmath29 & @xmath29 & @xmath30 we used the miriad package @xcite for all of the data reduction . first , edge channels and channels known to be affected by radio frequency interference were discarded . we then examined the data and flagged bad data points during periods of poor atmospheric stability . next , bandpass , gain , polarization , and flux calibration solutions were determined . in order to obtain a uniform _ u - v _ coverage , we excluded data from the 6 km baseline while forming radio maps . we formed radio images for the four bands , adopting the multifrequency synthesis technique @xcite which can further improve the _ u - v _ coverage . in particular , the 20 cm continuum image was formed by extracting line - free channels from the data . uniform weighting was used for the 20 , 13 , and 6 cm images to reduce sidelobes and to improve the spatial resolution . for the 3 cm image , due to the lower signal - to - noise ( s / n ) ratio in this band , we used a weighting parameter of ` robust ` @xmath31 @xcite to optimize the balance between the noise level and spatial resolution . we then applied a maximum entropy algorithm @xcite to deconvolve all of the dirty maps . for 20 and 13 cm , the task ` maxen ` was employed ; for 6 and 3 cm , we used the task ` pmosmem ` @xcite to deconvolve the stokes _ i _ , _ q _ , and _ u _ maps simultaneously . after that , we convolved the resultant models with synthesized beams of fwhm @xmath32 for 20 cm , @xmath33 for 13 cm , @xmath34 for 6 cm , and @xmath35 for 3 cm . the final maps have an rms noise of @xmath36 at 20 cm , @xmath37 at 13 cm , @xmath38 at 6 cm , and @xmath39 at 3 cm . these all agree well with the theoretical values . we also formed images with visibilities from the 6 km baselines only in an attempt to identify point sources . identical data reduction and imaging procedures as above were used , except that natural weighting was used to form the images in order to reduce the noise . for polarimetry , we formed a new set of low - resolution 3 cm images matching that at 6 cm in order to make a direct comparison between the two bands and generate a rotation measure ( rm ) map ( see section [ sec : rm ] ) . a gaussian taper has been applied to the visibility data by setting the parameter of ` fwhm ` @xmath40 , @xmath41 in the task ` invert ` . we then followed the same imaging and cleaning procedures as outlined above , except we convolved the output model of the task ` pmosmem ` with a gaussian beam of identical size as the 6 cm beam ( @xmath42 ) in the task ` restor ` . we obtained polarized intensity and polarization position angle ( pa ) maps from the stokes _ q _ and _ u _ maps at the two bands . the ricean bias was corrected for when the polarized intensity was computed @xcite . we blanked out areas where either the polarized intensity has s / n @xmath43 5 or the total intensity has s / n @xmath43 12 . note that polarization was not measured at 20 cm and the 13 cm data are not useful , since the stokes _ q _ and _ u _ maps are corrupted by severe polarization leakage from a nearby bright h ii region g327.3@xmath00.6 . this is due to a flaw in the feedhorn design of the old atca receiver . for the h i data , we used the task ` uvlin ` to separate line emission from continuum emission within the 20 cm data . we then smoothed the line spectrum to a resolution of @xmath44 in velocity . to filter out large - scale structures , only data with _ u - v _ distances shorter than @xmath45 were selected to form the line images . we found that the source is too weak to determine any h i absorption . a composite radio image of snr g327.1@xmath01.1 at 36 cm ( most ; * ? ? ? * ) and 6 cm ( atca ) is shown in figure [ fig : g327]a , featuring both the snr shell and the pwn . radio intensity maps of the pwn at 20 , 13 , 6 , and 3 cm ( all from atca ) are shown in figure [ fig : int ] . the nebula exhibits similar morphology over the observed wavelengths . it has a main circular structure with a diameter of @xmath46 and a finger - like structure of length @xmath47 protruding toward the northwest . together with the two x - ray prongs ahead of the radio finger , the whole pwn resembles a snail . we thus refer to the entire pwn as `` the snail '' : the circular structure is the `` body '' and the protrusion is the `` head . '' a more detailed look at the body revealed two features `` the bay '' to the southwest and `` the apex '' to the northwest ( see figure [ fig : structure ] ) . the bay is an indented area to the body , while the apex is a sharp corner at the body s boundary . faint emission is detected outside of the bay at 20 and 6 cm ( about @xmath48 and @xmath49 , respectively ) , and was also found in the most image @xcite . the emission is not obvious in the 13 and 3 cm images . this could be attributed to the contamination of the 13 cm image by the snr , while at 3 cm it could be too faint to be detected , or could have been resolved out because of insufficient _ u - v _ coverage ( see section [ sec : spec ] ) . on a larger scale , hints of an snr shell of diameter @xmath50 are seen in the 20 and 13 cm images ( not shown in the field of view of figure [ fig : int ] ) . since the shortest baseline of the 6 and 3 cm observations is 31 m , which translates to angular scales of about @xmath51 and @xmath52 , respectively , we do not expect these observations to be sensitive to the snr shell . the 13 , 6 , and 3 cm images clearly reveal small - scale structure within the body . we discovered filaments running through the pwn with an unresolved width by our observations . these filaments seem to form loop - like structures that are typically @xmath53 diameter . there are two point sources near the northern edge of the body . the eastern one is located at ( @xmath54 , @xmath55 ; j2000.0 ) and the western one is at ( @xmath56 , @xmath57 ; j2000.0 ) . these are marked in figure [ fig : structure ] . both point sources are unresolved in the 6 km baseline images , and therefore are less than @xmath58 in extent . the head has an extent of about @xmath59 long and @xmath60 wide , and shows uniform surface brightness ( @xmath61 at 6 cm , and @xmath62 at 3 cm ) , except near the tip ( @xmath63 , @xmath64 ; j2000.0 ) where it shows a peak with @xmath65 at 6 cm and @xmath66 at 3 cm . the radio peak is extended , with a size of @xmath67 elongated along the head , and is offset from the x - ray point source by @xmath68 to the southeast . no radio counterparts of the x - ray prongs and arc outside of the head are detected . careful examination of the 6 km baseline images shows no counterparts of the x - ray point source with a @xmath69 upper limit of @xmath70 at 6 cm . to measure the pwn flux density , we employed a source region enclosing both the body and the head as shown in figure [ fig : int ] . the eastern point source is believed to be a background source while the western point source might be a compact structure within the pwn ( see section [ sec : ewsrc ] ) . therefore , we defined a circular exclusion region of @xmath71 diameter centered at the former . we tried different background regions within the snr shell , and the difference gives us a handle on the systematic uncertainty . the results are shown in table [ table : flux ] . the flux densities of the pwn are @xmath72 , @xmath17 , @xmath20 , @xmath73 , and @xmath25 at 36 , 20 , 13 , 6 , and 3 cm , respectively . the uncertainties are dominated by the systematic uncertainties associated with background subtraction , as the statistical uncertainties are relatively small ( a few mjy ) . we note that the value we find at 36 cm ( 2.3jy ) is slightly larger than that ( 2.0jy ) reported by @xcite . this could be attributed to our different choice of regions . the radio spectrum is shown in figure [ fig : radiospec ] . fitting the flux densities at 36 , 20 , 13 , and 6 cm with a power law ( @xmath74 ) gives a spectral index of @xmath75 , and extrapolation to 3 cm suggests a flux density of @xmath76 , which is almost @xmath77 higher than the observed value of @xmath78 . if we directly join the 6 and 3 cm data points , then we obtain @xmath79 between the two bands . the shortest _ u - v _ spacing at 3 cm is about @xmath80 , corresponding to an angular scale of @xmath81 . this sets a rough upper limit on the size of structures that our observations are sensitive to . in order to see if the missing flux problem is significant , we formed a 6 cm radio image only with _ u - v _ distance larger than @xmath80 , using identical procedure as described in section [ sec : obs ] . this results in about @xmath82 loss in flux density of the pwn , suggesting that the 3 cm observations may not be sensitive to the larger - scale emission of the pwn . we also determined the flux densities of the body and the head of the snail separately , with the results also listed in table [ table : flux ] . the flux densities and fitted spectra are shown in figure [ fig : radiospec ] . for the body , the flux densities are @xmath15 , @xmath18 , @xmath21 , @xmath23 , and @xmath26 at 36 , 20 , 13 , 6 , and 3 cm , respectively , and fitting the flux densities between 36 and 6 cm with a power - law spectrum gives a spectral index of @xmath75 . for the head , the flux densities are @xmath16 , @xmath83 , @xmath84 , @xmath85 , and @xmath86 at 36 , 20 , 13 , 6 , and 3 cm , respectively . the spectrum of the head appears to be peculiar , as the observed flux density is lower at 36 cm than that at 20 cm . we believe that the flux density measurement of the head could be affected by faint sidelobes from g327.3@xmath00.6 at 36 cm . we fitted the flux densities of the head between 20 and 3 cm , as the size of the head is only about @xmath59 long and @xmath60 wide , and should be well sampled by our 3 cm observations . the resulting power - law spectrum has a spectral index of @xmath87 . the flux densities of the eastern point source are found to be @xmath88 at 6 cm and @xmath89 at 3 cm , while those for the western point source are @xmath90 at 6 cm and @xmath91 at 3 cm . these are very rough estimates , as the diffuse emission from the pwn precludes precise measurements . the results suggest an inverted spectrum with @xmath92 for the eastern point source and @xmath93 for the western point source . maps of the polarization fraction of the snail at 6 and 3 cm are shown in figure [ fig : pm ] . overall , the pwn is highly linearly polarized . the typical polarization fraction of the body is about @xmath94 and @xmath82 at 6 and 3 cm , respectively , while that along the head is correspondingly about @xmath82 and @xmath95 . at a smaller scale , we found a highly polarized core inside the body as enclosed by an inner contour in figure [ fig : pm ] , with polarization fractions of about @xmath95 and @xmath77 at 6 and 3 cm , respectively . near the northern edge of the body , the eastern point source is unpolarized , while the western point source is about @xmath96 polarized at both bands . note that the actual polarization fraction at 3 cm could be lower due to the missing flux problem as mentioned . finally , the polarization fraction at the edge of the body may be overestimated because we have used only a single rms value for the debiasing procedure . we generated an rm map by comparing between the polarization pa maps at 6 and 3 cm : @xmath97 where @xmath98 and @xmath99 are the observed pas in the respective bands , and @xmath100 and @xmath101 are the center wavelengths of the bands . the resulting rm map is shown in figure [ fig : rm ] , and the rm is found to vary smoothly between @xmath102 and @xmath103 across the pwn with a typical uncertainty of @xmath104 . the average rm value is about @xmath105 . one problem of determining the rm this way is the so - called @xmath106 ambiguity . this ambiguity arises because an rm of larger magnitude can also be compatible with the observed polarization vectors if it provides an addition of multiples of @xmath107 radian to the relative rotation between the two bands . for example , adding an extra rm of @xmath108 can rotate the polarization vectors at 6 and 3 cm by @xmath109 relative to each other . in order to eradicate this problem , we picked the first and last channels of the 6 cm data , which are at frequencies of @xmath110 and @xmath111 , respectively , and formed pa maps separately using procedures identical to those of the full - band data . by applying the pa maps , we generated an rm map from the two channels . because the individual pixels of the rm map have large uncertainty , we did not attempt to investigate the spatial distribution of the rm from it . instead , we averaged the rm across the entire pwn to boost the s / n ratio , and found an average rm for the snail of @xmath112 from the two channels , which is identical to the value obtained above . since we expect the rotation of the pa within the 6 cm band to be small ( less than @xmath109 , otherwise the bandwidth depolarization would have been significant ) , this rules out the @xmath106 ambiguity in our measurements . we employed the rm map obtained from section [ sec : rm ] to derotate the polarization vectors at 6 and 3 cm in order to infer the intrinsic magnetic field orientation of the pwn . since the rm was derived from the pas at 6 and 3 cm only , the derotated vectors at the two bands have identical orientation . the derotated vectors have a typical uncertainty of about @xmath113 , contributed by both the error in pa and in rm . figure [ fig : bfield ] shows the projected intrinsic magnetic field orientation of the snail . overall , the field is highly ordered . the magnetic field at the head aligns with the pwn elongation , while for the body it is tangential to the boundary , except at the apex where it becomes radial . the field configuration of the interior of the body is complex with the magnetic field generally following the filamentary loops . the snail is believed to have first encountered the reverse shock about @xmath114 ago , as suggested by hd simulations @xcite . this could have resulted in a turbulent environment in the pwn and given rise to a tangled magnetic field geometry . the strong polarization signal we found from the snail indicates the presence of a highly ordered magnetic field . this could reflect a turbulence scale larger than the beam size of our observations . in this case , the beam depolarization will be insignificant and the observed magnetic field geometry will appear ordered at small scales . for further investigation , mhd efforts are required to fully understand the highly ordered magnetic field found in the snail . the magnetic field in the head of the snail shows good alignment with the elongation of the structure ( see figure [ fig : bfield ] ) . this is consistent with the evolutionary picture suggested by @xcite , assuming that the magnetic field lines are frozen in the newly generated pulsar wind being pushed to the southeast . similar magnetic field geometry is also found in some bow - shock pwne , including the `` handle '' of the frying pan ( g315.78@xmath00.23 ; * ? ? ? * ) and the tail of the mouse ( g359.23@xmath00.82 ; * ? ? ? * ) . however , a distinctly different example is the bow - shock system g319.9@xmath00.7 , which shows a helical magnetic field which then aligns with the flow further downstream @xcite . the cause of such a peculiar configuration is not well understood , though it may be related to the mach number of the pulsar with respect to its ambient medium or the relative inclination of the spin axis and the elongation of the pwn ( e.g. , * ? ? ? * ; * ? ? ? as the head of the snail is believed to be a non - bow - shock system ( i.e. the mach number of the neutron star @xmath115 ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , it serves as an example showing that magnetic field can align with subsonic comet - like pwne . we also compare the field geometry around the presumed pulsar with those of a few evolved pwne , including vela x @xcite , the boomerang ( g106.6@xmath1162.9 ; * ? ? ? * ) , and da 495 ( g65.7@xmath1161.2 ; * ? ? ? these pwne could also have been crushed by the reverse shocks . the boomerang and da 495 have diameters of about @xmath117 and @xmath118 , respectively . they are considerably smaller than the snail , which has a length of the head spanning @xmath118 and the diameter of the body stretching @xmath119 . the radio polarimetric observation of vela x @xcite only covered the immediate vicinity ( @xmath120 ) around the vela pulsar . in the boomerang and vela x , toroidal magnetic fields can be observed , while in da 495 it was suggested to present a toroidal component superimposed on the apparently dipolar field structure . the magnetic field structures in these three pwne could be driven by the pulsar itself @xcite , while in the head of the snail the magnetic field lines parallel to the nebulae elongation can be shaped by the passage of the reverse shock . note that the physical scale for which we have resolution of in the snail is very different from that of the three presented above . future high - resolution studies may potentially probe the pulsar - driven field near the presumed pulsar of the snail . the high spatial resolution atca images in figure [ fig : int ] revealed the filamentary structure inside the pwn . the filamentary loops generally align with the magnetic field ( see figure [ fig : bfield ] ) , suggesting that they could be magnetic loops similar to those observed in young pwne 3c 58 @xcite and g54.1@xmath1160.3 @xcite . in the snail , the filamentary loops have a typical diameter of @xmath53 , which corresponds to @xmath121 at a distance of @xmath122 . the scale is a few times larger than those in 3c 58 and g54.1@xmath1160.3 , which were estimated to have diameters of about @xmath123@xmath124 and @xmath125@xmath126 , respectively @xcite . it was suggested that kink instabilities can be the origin of magnetic loops in pwne the toroidal field near the termination shock could be torn away and eventually become a magnetic loop as observed @xcite . however , it is unclear if the interaction of the snr reverse shock could destroy these structures . in this case , the loops would be formed through another mechanism . we identified two structures of the snail near the edge of the body referred to as the apex and the bay . the former could be a protrusion from the body , but it is difficult to confirm because its proximity to the head complicates the interpretation of the morphology . the magnetic field vectors at the apex show a radial instead of tangential orientation as seen in other parts of the pwn boundary . similar protrusions and chimney - like structures are seen in the crab nebula @xcite and in kes 75 @xcite . the apex could be formed by the leakage of pwn materials into the surrounding shocked sn ejecta , similar to the chimney of the crab nebula . however , one difference between the crab nebula and the body of the snail is that the boundary magnetic field of the crab nebula is primarily radial @xcite , and so the physical processes appear to be different . the nature of the faint emission outside of the bay , which is outlined by the outermost contour in figure [ fig : structure ] , could be important for understanding the evolution scenario of g327.1@xmath01.1 . it has been suggested as part of the pwn @xcite , implying that the true boundary of the pwn should lie outside the bay , such that the head would be pointing closer to the center of the body as suggested by simulations @xcite , instead of being tangential to its edge . as pointed out in section [ sec : morph ] , our radio maps revealed two point sources in the northern edge of the pwn . we found that the eastern one is unpolarized ( figure [ fig : pm ] ) and has an inverted spectrum of @xmath127 , which differs significantly from the typical value of @xmath128 for pwne . these points suggest that it is likely an unrelated background source . on the other hand , the western source has a spectral index of @xmath129 and is about @xmath96 polarized at 6 and 3 cm . the polarized emission allows us to determine the rm of the source and its value is consistent with that of the surrounding pwn ( see figure [ fig : rm ] ) , suggesting a similar distance . it also spatially coincides with the filaments in the pwn . therefore , we argue that it is associated with the snail , possibly an unresolved compact knot as part of the filaments . as mentioned earlier , the highly ordered magnetic field found in the snail could imply a large characteristic scale of the turbulence . to get a handle on the scale in this case , we consider a toy model of the body of the snail ( i.e.excluding the head ) as a non - evolving spherical nebula . this assumption is based on the fact that we do not expect significant multipolar anisotropy in the pressure exerted by the reverse shock , except for the dipolar component , which will displace the pwn instead of deforming it . the volume of the nebula is divided into patches of uniform size that represent the characteristic turbulence scale in the pwn . the turbulent magnetic field is implemented in our model by assigning a magnetic field with random orientation but uniform strength to each of the patches . we argue that this assumption is justified , as the magnetic field inferred from simulations @xcite is close to the equipartition value we derived in section [ sec : bfield ] below , and therefore the fluctuations in the magnetic field strength should be small . otherwise , the fluctuation will rapidly fade away over a timescale of the order of the sound crossing time ( which is much shorter than the age of the system ) . the electron density is also assumed to be uniform for the same reason , and we assume a power - law distribution with the index given by the measured radio spectrum . we further allow for the possibility that a patch is filled with cold snr material instead of non - thermal pulsar wind particles . this mixing of sn ejecta in the snail was suggested based on x - ray observations @xcite and can also be seen in other reverse shock crushed pwne such as vela x @xcite . the probability that a patch is filled by synchrotron - emitting pulsar wind particles is simply the filling factor in our model . if there is no significant mixing of materials in the pwn , then the modeling results with a filling factor of @xmath130 should fit the observations best , while a lower filling factor would suggest substantial mixing . we integrated the synchrotron emission along different lines of sight through the volume of the nebula in all of the stokes parameters to compute the radio maps for comparisons with our observations . for this , we convolved the simulation results with the radio beam . note that the randomness introduced by the filling factor and the magnetic field orientations make it possible to exactly reproduce the observations if we run the model for a considerable number of times ( albeit that will be meaningless ) , and therefore it is more sensible to only produce one set of maps per combination of parameters , and then inspect by eye the scales of the structures , the number of features in the pwn , and the contrast in radio intensity instead of looking for a perfect match . this allows us to place a rough constraint on the filling factor and the characteristic turbulence scale of the snail . the simulated intensity and fluctuation maps are shown in figures [ fig : mhd1 ] and [ fig : mhd2 ] . a filling factor of 5075% and a patch size of 1/81/6 of the pwn radius ( @xmath131 ) appear to give the best match between the simulations and the data , such that they have comparable numbers of small - scale features and contrast in the intensity maps . in figure [ fig : mhd3 ] we show the simulated polarization maps . a large filling factor or small patch size generally give a lower polarization fraction . this is because , as there are more patches with different @xmath132-field orientation along a line of sight , depolarization becomes more severe . we find that the same range of filling factor and patch size as above provide the best fit to the observation . if this simple model is true , then the simulation results suggest that mixing in this pwn is significant and the turbulence scale is about @xmath133 @xmath134 . the former conclusion is in line with the speculation of @xcite , and should be taken into account for further theoretical work with the snail . finally , we note that the randomly oriented magnetic field in this simple model does not satisfy the solenoid condition ( i.e. , @xmath135 ) . therefore , it fails to reproduce the magnetic loops in the pwn interior and the tangential field structure at the boundary as observed . instead , it is more useful to compare the coherence scales of the models with the observations . we determined a spectral index of @xmath75 for the snail between 36 and 6 cm . such a flat spectrum is typical for pwne . the flux density at 3 cm is not well determined as the observations are only sensitive to smaller structures in the pwn but not the whole nebula . the head of the snail has a steeper spectral index of @xmath87 . this is unexpected , as x - ray observations suggest that the particles are injected from the head ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and therefore the materials in the head should be younger than those of the body . this means that the spectral difference probably can not be explained by synchrotron cooling . we also note that this spectral index is considerably steeper than the generally expected value for pwne ( @xmath2 ) , which is rather unusual and requires further theoretical modeling to explain . we computed the minimum - energy magnetic field strength by imposing the conventional assumption , i.e. that the total energy @xmath136 is distributed between magnetic energy @xmath137 and particle energy @xmath138 in a way that @xmath136 is minimized . from synchrotron theory @xcite , we have @xmath139 where @xmath140 is the volume filling factor of the magnetic field , @xmath141 is the volume of the emission region , @xmath142 is a constant weakly depending on the spectral index and frequency range considered , @xmath143 is the energy ratio of ions to electrons , and @xmath144 is the synchrotron luminosity of the pwn . minimizing @xmath136 gives a minimum - energy field , @xmath145^{2/7}{\rm , } \label{eqn : beq}\ ] ] which is also referred to as the `` equipartition field '' in some literature . we integrated the simple power - law radio spectrum for the entire pwn ( as in the fit of figure [ fig : radiospec ] ) between @xmath146 and @xmath147 ( e.g. * ? ? ? * ; * ? ? ? * ) , and found @xmath148 , where @xmath149 is the distance to the snail in units of @xmath122 . to estimate the volume @xmath141 , we modeled the body as a sphere of diameter @xmath52 and the head as a cylinder with a diameter of @xmath60 and a length of @xmath59 . the volume of the pwn is then @xmath150 . substituting these parameters into equation [ eqn : beq ] gives @xmath151 assuming @xmath152 , @xmath153 , and @xmath154 , we find a minimum - energy field of @xmath155 in the snail , which is of the same order of magnitude as that estimated by modeling the broadband emission of the pwn ( @xmath156 ; * ? ? ? note that the computed minimum - energy field strength can depend on the frequency range chosen . for instance , if we only integrate up to @xmath157 instead of @xmath147 , then we obtain a slightly lower value of @xmath158 , and it is much less sensitive to the lower frequency limit . it is also worth noting that the energy in the pulsar wind is believed to be particle - dominated upon injection , and hence the minimum - energy condition is not guaranteed to hold . we also computed the minimum - energy field strength of the body and the head separately , using identical procedures as above . for the body , we have @xmath159 and @xmath160 , which gives @xmath161 . for the head , we have @xmath162 and @xmath163 , which gives @xmath164 . the higher minimum - energy field strength in the head as compared to that in the body could suggest that the reverse shock has compressed the former but not yet the latter . figure [ fig : g327]b shows that the radio peak of the head is offset from the x - ray point source by @xmath68 ( corresponding to @xmath165 at a distance of @xmath122 ) , which is larger than the astrometric uncertainty of @xmath166 of _ chandra _ and the radio beam size ( @xmath167 ) . a similar offset is found in other systems , including g319.9@xmath00.7 @xcite and the lighthouse nebula @xcite powered by psr j1101@xmath06101 @xcite , in which the radio and x - ray peaks are displaced by @xmath118 and @xmath168 , respectively . on the other hand , there is no detectable radio emission at the location of the x - ray prongs and arc of the snail . this is analogous to the x - ray jet - like structures seen in the guitar nebula @xcite and the lighthouse nebula @xcite . this could be the result of the diffusion of high - energy particles , but the density is too low for the radio emission to be detected . we also note that the prong structures are aligned with the edge of the radio head . the broadband synchrotron spectra of pwne often steepen at high frequency due to synchrotron cooling . for a single population of particles injected into the system , the spectral break frequency @xmath169 and magnetic field strength @xmath132 could be used to estimate the pwn age @xmath170 by @xcite @xmath171 using the @xmath172 obtained through extrapolation of radio and x - ray spectra ( figure [ fig : radiospec]b ) and the minimum - energy field strength obtained above ( @xmath173 ) gives @xmath174 for the snail . if we instead use the magnetic field strength estimate from modeling ( @xmath156 ; * ? ? ? * ) , then we acquire an even larger @xmath175 . these values differ significantly from the age estimates of @xmath176@xmath177 from previous studies @xcite . we argue that the simple estimate from the spectral break may not reflect the true age of the system . the change in spectral index by a simple extrapolation of the radio and x - ray spectra is @xmath178 . this suggests that the actual spectral energy distribution ( sed ) of the system can be much more complicated than expected , as the result of a broken power - law injection spectrum and enhanced synchrotron burnoff induced by the reverse shock interaction ( see * ? ? ? this can be verified by future observations between the radio and x - ray bands . figure [ fig : radiospec]b shows the broadband sed of the snail from radio to @xmath1-rays , with the @xmath1-ray data from _ fermi _ and h.e.s.s . while the tev @xmath1-ray spectrum from h.e.s.s . appears to align with the x - ray spectrum , this is probably just coincidence ( see * ? ? ? in this paper , we presented radio observations of the snail pwn in snr g327.1@xmath01.1 using atca and compared the results with the previous most study @xcite . the pwn shows the same morphology in all the five bands between 36 and 3 cm , consisting of a main circular body believed to be the `` relic '' pwn and a protruding head . the 6 and 3 cm polarization maps reveal a highly ordered magnetic field structure , in contrast with the turbulent interior with a tangled magnetic field structure expected for a reverse shock crushed pwn . this may suggest that the characteristic turbulence scale is larger than the radio beam size . to explore this scenario , we developed a toy model to simulate the emission from a turbulent pwn and found that a characteristic turbulence scale of @xmath179 @xmath134 with a filling factor of @xmath180@xmath181 can most closely match the observations . we showed that the magnetic field at the head of the snail aligns with the nebular elongation . this serves as a good example of subsonic cometary pwn systems with magnetic field parallel to the tail . in addition , we discovered filamentary structures in the body which could be magnetic loops . it is , however , unclear whether they are formed by kink instabilities as suggested for 3c 58 and g54.1@xmath1160.3 , or by turbulence during the reverse shock interaction . we also determined a spectral index of @xmath75 for the overall pwn in the observed radio bands , which gives a minimum - energy magnetic field strength of @xmath182 . the radio spectrum of the head appears to be steeper than that of the body , which is difficult to explain through synchrotron cooling . the authors thank j. lim for fruitful discussions . we also thank an anonymous referee for helpful comments and suggestions which improved the paper . the australia telescope compact array is part of the australia telescope national facility which is funded by the commonwealth of australia for operation as a national facility managed by csiro . most is operated by the university of sydney with support from the australian research council and the science foundation for physics within the university of sydney . this work is supported by ecs grant of hong kong government under hku 709713p . p.o.s . acknowledges partial support from nasa contract nas8 - 03060 .

magnetic fields are now generally acknowledged to be essential to accretion @xcite . amplified exponentially from weak seed fields by the magneto - rotational instability ( mri ) , they are stretched in a consistent direction so as to produce a net stress @xmath0 capable of carrying enough angular momentum outward through the disk to support sizable accretion rates . because the nonlinear state of the mri is fully - developed magnetohydrodynamic ( mhd ) turbulence , it is often thought ( or at least hoped ) that its properties have little to do with the character of the seed magnetic field brought to the disk with the accretion flow . however , there are certain properties of the magnetic field that are preserved despite the development of nonlinear turbulence . because magnetic fields are divergence - free and the plasma in accretion flows has very little electrical resistivity , the global topology of the field persists , no matter what happens . it is therefore natural to ask whether any interesting properties of the accretion flow may depend on the global field topology . for example , could the field topology alter the relation between surface density and accretion rate ? or how the stress varies through the marginally stable region ? in addition to swallowing matter , accreting black holes often expel relativistic jets . although the mechanisms by which these jets are launched have long been the subject of investigation and speculation , the current consensus focuses on a combination of large - scale magnetic fields and the rotation of either the central hole , the accretion disk , or both . two specific mhd models have received the greatest attention : the blandford - znajek mechanism @xcite and the blandford - payne mechanism @xcite . in the blandford - payne model a large - scale vertical magnetic field is anchored in the disk , rotating with the orbital frequency . above and below the disk magnetic tension dominates , and the field lines force the matter to rotate with this same frequency . if the fieldlines are angled outward sufficiently with respect to the disk , there can be a net outward force , accelerating the matter along the rotating fieldlines . the blandford - znajek model also depends upon rotating field lines , but the source of that rotation ( and the source of the power in the jet ) is the black hole rather than the disk : any field lines connecting the immediate vicinity of the black hole to infinity are forced to rotate by frame - dragging , so that poynting flux carries off the energy . given the strong dependence of both the blandford - znajek and the blandford - payne mechanisms on the nature of the magnetic field in the disk and near the black hole , it is equally natural to inquire whether the topology of the field embedded in the accretion flow has any influence on their operation . it is the goal of this paper to begin to investigate these questions , both as they relate to accretion disk dynamics and as they relate to jet properties . unfortunately , it is difficult even to scratch the surface by purely analytic techniques ( although there have been some attempts using highly - simplified models : * ? ? ? * ; * ? ? ? on the other hand , direct numerical simulations provide a means by which aspects of these problems can be investigated in detail . @xcite made a first exploration of the dependence of accretion properties on field topology in a highly - simplified approximation : newtonian dynamics in a pseudo - newtonian potential . they found that a purely toroidal initial field led to a systematically smaller accretion rate than one whose initial field was dipolar . more recently , fully general relativistic mhd ( grmhd ) simulations have become possible @xcite . these works have focused primarily on a single initial field structure : a set of nested dipole field loops contained entirely within the initial gas torus . other field geometries have been examined , but only in rather limited fashion . for example , @xcite simulated the evolution of a torus with an initial toroidal field around a schwarzschild black hole and noted the lack of any resulting jet , but did not consider why no jet formed . @xcite , using axisymmetric simulations , compared the electromagnetic luminosity of the standard dipole jet with that resulting from a few alternate magnetic field topologies . in particular , they studied cases in which the initial field was arranged in loops of alternating sign arranged either vertically or horizontally within the initial torus . the principal effect they reported was a factor of 23 reduction in the ratio of the time - averaged jet poynting flux to the mean mass accretion rate . they also noted that an initial field that is purely vertical led to significant enhancements in both the mean mass accretion rate and the time - averaged jet poynting flux . axisymmetric simulations of tori threaded by net vertical field led @xcite to concur that such a field can greatly enhance jet power . on the basis of these results , @xcite proposed that black holes inevitably become threaded by an organized vertical field if the initial disk contains a large - scale poloidal field with few parity changes . the intriguing character of these results motivates a more systematic study , one that employs three - dimensional simulations , considers the accretion flow as well as the jet , and investigates time - dependence . the question immediately arises : how can one best attempt to describe the very complex mixture of topologies likely to be found in real accretion flows ? the magnetic field near a black hole can be thought of as simply what the accretion process itself has self - consistently brought there from whatever source supplies it . in order to gain understanding of what , in realistic circumstances , must be a truly complex combination of structures , in this paper we do not try to construct a single realistic " field geometry ; instead , we study the effects due to individual members of a basis set " of field topologies . our hope is that by understanding the action of pure types , we can gain some understanding of how realistic mixtures behave . to be specific , we will begin with a carefully chosen set of different kinds of closed field loops that we expect to be generic . closed field loops have the advantage of requiring many fewer free parameters to describe them fully within a simulation than do field geometries with net flux . net flux geometries may well exist in nature and have important effects ; we will consider them in the second stage of this project . we begin with the conjecture that , at a qualitative level , the different sorts of closed field loop behavior can all be represented by a combination of those due to : dipolar loops , or fieldlines initially entirely poloidal and which cross the equatorial plane at well - separated radii ; toroidal loops , which encircle the black hole ; and quadrupolar loops , which ( again , in the initial state ) are poloidal , but close without crossing the equatorial plane . for each of these cases , we have run a lengthy three - dimensional grmhd simulation whose results we hope characterize the accretion flow structure created by that field topology . we have also run a series of high resolution axisymmetric simulations of each of the initially poloidal topologies ( i.e. dipole and quadrupole ) . in addition , in order to understand better the implications of the finite lifetimes of closed loops , we have performed an axisymmetric simulation whose initial magnetic field structure was a series of dipolar loops , each contained within a span of radii narrow enough that the difference in inflow times across a loop is comparable to the equilibration time for the magnetic field in the jet . the rest of this work is structured as follows . in [ numerics ] we briefly review the numerical scheme employed to solve the equations of grmhd and give a detailed description of the initial conditions used to generate each topology . in [ flowstructure ] we examine the late time accretion flow structure that evolves from each of the initial field topologies , paying attention to each of the structures ( main disk body , magnetized corona ) that were identified by @xcite as arising from the evolution of the dipole field topology . in [ jets ] we examine the properties of the third structure identified by @xcite , the unbound outflow , and examine how the initial evolution of the field during the linear growth phase of the mri gives rise to the components of the accretion flow at late times . finally in [ conclusion ] we summarize our results and highlight their importance for jets in astrophysical settings . in this work we undertake a study of both two- and three - dimensional simulations to investigate the influence of magnetic field topology on both the evolved accretion flow structure and relativistic jet formation . in doing so we continue a program of black hole accretion disk simulations begun in @xcite , @xcite , @xcite , @xcite and @xcite . three - dimensional simulations , such as those presented in the previous work , are the only means by which one can gain insight into the evolved accretion flow properties due to the fundamental restriction posed by the anti - dynamo theorem @xcite . further , toroidal field models can not be studied in axisymmetry , since the toroidal mri requires nonzero azimuthal wavevectors . on the other hand , because jet formation occurs promptly during the evolution of the flow @xcite , restrictions due to the anti - dynamo theorem are less critical for jet studies . in this case the higher resolution afforded by the two - dimensional simulations is particularly valuable . the simulation code we use is described in @xcite . this code solves the equations of ideal non - radiative mhd in the static kerr metric of a rotating black hole using boyer - lindquist coordinates . values are expressed in gravitational units @xmath1 with line element @xmath2 and signature @xmath3 . since we are focusing in this paper on the influence of the magnetic field topology , we fix the black hole spin to @xmath4 for all of the new simulations reported here . the relativistic fluid at each grid zone is described by its density @xmath5 , specific internal energy @xmath6 , @xmath7-velocity @xmath8 , and isotropic pressure @xmath9 . the relativistic enthalpy is @xmath10 . the pressure is related to @xmath5 and @xmath6 via the equation of state for an ideal gas , @xmath11 . the magnetic field is described by two sets of variables . the first is the constrained transport magnetic field @xmath12 f_{jk}$ ] , where @xmath13 $ ] is the completely anti - symmetric symbol , and @xmath14 are the spatial components of the electromagnetic field strength tensor . from the constrained transport ( ct ) magnetic field components , we derive the magnetic field four - vector , @xmath15 , and the magnetic field scalar , @xmath16 . the electromagnetic component of the stress - energy tensor is @xmath17 . in all of the simulations , the initial condition for the matter is an isolated , hydrostatic torus orbiting near the black hole . the initial state of the three - dimensional simulations was chosen so as to be consistent with that used in earlier simulations in this series ( see * ? ? ? the matter s adiabatic index , @xmath18 was @xmath19 and the angular momentum distribution was slightly sub - keplerian , with a specific angular momentum at the inner edge of the torus ( located at @xmath20 m ) @xmath21 ( we define @xmath22 ) . for this choice of @xmath23 , the pressure maximum is at @xmath24 . in the two - dimensional simulations , our fiducial torus was chosen to have a constant angular momentum distribution with @xmath25 , @xmath18 was 4/3 , and the inner edge was again located at @xmath20 m . this torus is significantly thicker than the nearly - keplerian torus used for the three - dimensional simulations ; its density scale height at the pressure maximum , measured along surfaces of constant radius , is a factor @xmath26 greater than that of the three - dimensional torus . this thicker configuration makes it easier to resolve the multiple field loop model , where several field loops must fit into a restricted space . this specific combination of @xmath27 and @xmath18 was also necessary to manipulate the shape of the torus so that there was an appropriate balance in magnetic flux between the different components of the multiple dipole loop simulation . in the quadrupolar field case , in order to study how the evolution of the field depends on its detailed geometry , we examined what happened as @xmath27 varied from 4.9 to 5.2 in steps of 0.05 . the initial torus is overlaid with a variety of magnetic field configurations , all with zero net flux . the initial poloidal magnetic field is determined from the curl of the four - vector potential , i.e. , @xmath28 with only @xmath29 . for a single dipolar loop , we set @xmath30 for a quadrupole , the azimuthal component of the vector potential is @xmath31 in both of these cases @xmath32 is set to zero where the density is less than @xmath33 ; this device confines the initial field to lie well inside the torus . in all the poloidal simulations , the initial field intensity ( i.e. , the parameter @xmath34 ) is determined by setting the volume - averaged ratio of gas to magnetic pressure @xmath35 . the vector potential for @xmath36 dipolar loops ( used only for the two - dimensional simulations ) is more complicated . it is given by @xmath37,\ ] ] where the total vector potential is the superposition of the ( non - relativistic ) vector potentials from @xmath36 current loops ( see * ? ? ? 182 , eqn . 5.37 ) . here @xmath38 is the current in the @xmath39th loop , located at radius @xmath40 in the equatorial plane . @xmath41 and @xmath42 are the complete elliptic integrals of the first and second kinds , and @xmath43 . the @xmath44 are chosen so that the difference between the inflow times of adjacent loops is approximately equal to one jet equilibration time , @xmath45@xmath46 . the signs of the @xmath47 alternate with increasing radius so that the senses of the dipolar field loops alternate . the magnitudes of the currents were chosen with the goal of ensuring that the evolution of the two radially innermost loops was governed by the field local to those loops . the purpose of this simulation was to study the interaction of the two inner loops as they accrete ; in that context , the only function of the two outer loops is to supply the outer part of the torus with magnetic field , so that it could accept the angular momentum brought to it from the true accretion flow . thus , the most appropriate outer field structure for this particular numerical experiment is one that has the least influence on dynamics inside the pressure maximum . after some experimentation , we found that this goal could be achieved only by giving the two outer loops relatively small currents . when that was done , the outer two loops moved outward with the outer portion of the disk and were quickly destroyed ( i.e. forced to reconnect ) as the second loop from the inside expanded toward them . we make no pretense that this choice is a fair representation of generic four - loop evolution , but it does support our study of what happens when two complete loops of opposite sense are brought into the inner disk one after the other . for a purely toroidal initial field , all that is necessary is to set @xmath48 and determine @xmath49 by a choice of volume - averaged @xmath50 . here we use @xmath51 , as , in the purely toroidal case , accretion can occur only once mri - induced turbulence on the large scales has grown to sufficiently large amplitude , and long wavelength modes grow comparatively slowly . beginning with a comparatively strong field ensured that accretion activity started soon after the simulation began . after saturation , the field ( in relative terms ) actually weakened somewhat : the volume - averaged @xmath50 grew to @xmath52 ( see section [ diskbody ] ) . as before , the initial magnetic field is set to zero for densities below @xmath33 . we take the dipole simulation kdpg ( originally presented in * ? ? ? * where further discussion of our boundary conditions and grids may be found ) as our fiducial three - dimensional run . the three - dimensional quadrupole simulation is designated qdpa and the three - dimensional toroidal simulation is tdpa . kdpg and qdpa were run for @xmath53 in time , which corresponds to approximately @xmath54 orbits at the initial pressure maximum . the toroidal field model , tdpa , evolves initially at a slower rate ; it was run to @xmath55 in time . for each simulation , the time step @xmath56 was determined by the minimal light crossing time for a zone on the spatial grid and remained constant for the entire simulation @xcite . each of the three - dimensional simulations used @xmath57 @xmath58 grid zones . the radial grid extends from an inner boundary located at @xmath59 , which lies just outside the black hole event horizon , to an outer boundary located at @xmath60 . for the two poloidal topologies , the radial grid was graded using a hyperbolic cosine distribution ; for the toroidal case a logarithmic distribution was used . an outflow condition is applied at both the inner and outer radial boundary . the @xmath61-grid spans the range @xmath62 using an exponential distribution that concentrates zones near the equator . a reflecting boundary condition is used along the conical cutout surrounding the coordinate axis . the @xmath63-grid spans the quarter plane , @xmath64 , with periodic boundary conditions applied in @xmath63 . the use of this restricted angular domain significantly reduces the computational requirements of the simulation . the two - dimensional simulations were performed on a @xmath65 @xmath66 grid in which the radial cells are graded logarithmically , and the polar angle cells are concentrated toward the equator in the same way as for the three - dimensional simulations . the @xmath61 grid , however , goes much closer to the axis than in the three - dimensional simulations , with a cut - out of only @xmath67 radians . in addition to allowing higher poloidal resolution at a reasonable computational cost , two - dimensional simulations have the additional advantage that field - lines can be easily visualized as level surfaces of the vector potential component @xmath32 . using this approach , we illustrate the magnetic field topologies for the three different poloidal configurations in figure [ topologies ] . we begin by examining the average properties of the accretion flows in simulations kdpg , qdpa and tdpa . @xcite described the accretion flow structure in terms of three distinct regions : the main disk body , the coronal envelope , and the unbound outflow . in this section , we study the first two , paying special attention to contrasts ( or lack of contrast ) when the magnetic topology changes . we will discuss the unbound outflow in the following section . much of the analysis will be in terms of various time- and space - averaged quantities . the radial profile of the time - averaged shell integral of @xmath68 is defined as @xmath69 similarly , the shell average of a quantity @xmath68 at a radius @xmath70 is defined to be : @xmath71 the angular profile of the time - average of @xmath68 at radius @xmath70 is defined by @xmath72 lastly , the time- and volume - average of @xmath68 is : @xmath73 in these equations @xmath74 is the time over which the integral is computed , and @xmath75 is the usual metric determinant . typically @xmath76 ; for kdpg and qdpg this is the last @xmath77 of the full evolution , while for tdpa we choose a @xmath77 window in the middle of the simulation after the accretion flow is established . the spatial extent of the shell integration is the full @xmath61 and @xmath63 computational domain . during a given simulation , various shell integrals and radial fluxes are computed and stored every @xmath78 in time . this data can then be integrated over time to obtain quantities such as the total or time - averaged jet outflow or accretion rate . it is also useful to divide the shell and volume integrals into two parts , one for bound and one for unbound flow . for simplicity we define a particular zone to be unbound " if @xmath79 . unbound outflow can further be defined as those unbound cells with @xmath80 . in these simulations only the outflow near the axis ( the jet outflow ) is unbound ; the coronal backflow from the disk itself remains bound . the initial evolution of the accretion disks in kdpg and qdpa is qualitatively similar . both field configurations begin with considerable radial field within the torus . this is sheared out , generating toroidal field which , by @xmath81 , is sufficiently strong that the resulting poloidal gradient in @xmath82 begins to drive the inner edge of the torus ( initially located at @xmath83 ) inward . the inner edge of the disk arrives at the black hole at @xmath84 . within the disk body , the mri generates the turbulence that will determine the subsequent evolution of the disk , and by @xmath85 a statistically stationary turbulent accretion flow has been established inside the radius of the inner edge of the initial torus . the toroidal field model tdpa evolves more slowly than the two poloidal field cases , consistent with the results from earlier toroidal field pseudo - newtonian simulations @xcite . as discussed at length in @xcite , this behavior stems both from the absence of an initial radial field ( which means that there is no toroidal field amplification due to shear ) and from the fact that long - wavelength modes which are the most effective in driving accretion grow relatively slowly . inflow can begin only when the mri has produced turbulence of sufficient amplitude , which occurs by @xmath86 , corresponding to about 5 orbits at the radius of the torus pressure maximum . the accretion rate into the hole increases until about @xmath87 , after which it shows large fluctuations without an overall trend . figure [ bndavgplt1 ] shows time - averaged , shell - integrated radial profiles of a number of quantities relevant to the accretion flow : accretion rate @xmath88 , surface density @xmath89 , the net accreted angular momentum per unit rest mass , @xmath90 , @xmath91 , the electromagnetic poynting flux @xmath92 and the em angular momentum flux @xmath93 . the subscripts fl and em denote the fluid and electromagnetic contributions to the stress - energy tensor , respectively . all quantities are computed in the coordinate frame after the turbulent accretion flow is established , and in all cases the volume integral was restricted to cells where the matter was bound . the poloidal field simulations were averaged over time @xmath94@xmath95 while the toroidal field data were averaged over @xmath96@xmath97 . these six plots may be divided into two groups : those with little dependence on initial field topology ( surface density and accreted angular momentum per unit mass ) , and those with a stronger dependence ( accretion rate , magnetic field strength , poynting flux , and electromagnetic angular momentum flux ) . although the radial distribution of the surface density in all three cases is fairly similar , there is a slight distinction between the two poloidal models on the one hand and the toroidal on the other . the surface density in the toroidal model rises somewhat more steeply with radius , with lower density near the black hole and higher values outside of @xmath98 . for all three topologies , the specific angular momentum carried into the hole is close to the angular momentum of the marginally stable orbit . for tdpa , the value is essentially equal to that of the isco , while both qdpa and kdpa have smaller values , indicative of continued stress operating near or inside the isco . in all three cases , the time - averaged accretion rate is constant with radius out to @xmath9910 ; in this sense , all three have achieved a quasi - steady state within @xmath100 , where @xmath101 is the radial coordinate of the innermost stable circular orbit , @xmath102 ( the rise in @xmath103 and @xmath104 at large radius in tdpa is likely an artifact of the stronger magnetic field with which that simulation was begun ; equilibration takes place on timescales proportional to the local orbital period , which is , of course , @xmath105 , so inflow equilibration is slower at large radius ) . however , the actual value of the accretion rate is significantly different in the three simulations . the time - averaged mass accretion rate is highest in qdpa , exceeding that of kdpg by about @xmath106 , while @xmath107 is lower in tdpa by more than a factor of 2 compared to qdpa . accretion is , of course , driven by angular momentum transport , which is largely due to magnetic torques . to support these torques , the shell - integrated electromagnetic angular momentum flux must increase outward , as the net torque on the matter is the divergence of the em angular momentum flux . the dependence of the electromagnetic angular momentum flux on initial field topology is therefore similar to that of the mass accretion rate : it is consistently greater for kdpg and qdpa than for tdpa . the most significant difference between the kdpg and qdpa curves is that in the region near the isco , the electromagnetic angular momentum flux in kdpg becomes almost constant with decreasing radius ; this effect has previously been seen in other simulations that begin with dipole magnetic field loops @xcite . in the other two cases , the em angular momentum flux continues to decline inward . the mean magnetic field strength and poynting flux show a very similar dependence on initial field topology : qdpa and kdpg closely resemble each other , tdpa is somewhat weaker . in the toroidal case , the disk body average magnetic energy density is weaker by a factor of several than in the other two simulations , and decreases inward toward the horizon rather more rapidly . correspondingly , the time- and volume - averaged @xmath50 parameter was 11 in qdpa , 16 in kdpg , and 30 in tdpa ( because this is a volume - average , the corona is weighted heavily in this statistic ) . the electromagnetic energy flux shows the greatest variation between models . @xmath108 in kdpg is consistently larger than the value in qdpa : by a factor of two at @xmath109 and a factor of six at @xmath110 . the value in tdpa is smaller still by a factor of @xmath111 at these same radii . all these relations can be understood in a straightforward fashion : there is a nearly - constant ratio between the em angular momentum flux ( and therefore the accretion rate ) , the poynting flux , and the magnetic field intensity . in figure [ ratio ] , we plot the ratios of @xmath103 , @xmath108 and@xmath93 to @xmath91 . the inter - simulation contrast in the means of these quantities falls by a factor of several from the plot of their absolute values ( fig . [ bndavgplt1 ] ) to the plot of their ratios to @xmath82 ( fig . [ ratio ] ) . thus , we see that their primary dependence is on the magnetic field strength . only in the ratio of the electromagnetic energy flux to magnetic field strength do we see any global variation between the field topologies : @xmath108 in kdpg is consistently a factor of two above qdpa and tdpa . this contrast arises from the strong poynting flux in kdpg associated with the slower - moving , bound portion of the funnel - wall outflow ( as can be seen from figure 10 of * ? ? ? * see [ jets ] ) . the behavior of the fluid - frame stress can be seen in figure [ maxwell ] , which depicts its time - averaged vertically - integrated value as a function of radius . to compute this quantity , we project the stress tensor onto a local tetrad system in each cell and multiply by the fluid - frame cell volume , which is computed by a similar projection @xcite . after integrating over the bound matter on that spherical shell , we normalize the result to the surface area that this finite - thickness shell occupies in the equatorial plane . we present the result in this form in order to compare with the conventional representation of the vertically - integrated stress given by @xcite , which is presented in these terms . the novikov - thorne fluid - frame stress , unlike ours , is _ assumed _ to be zero at and within the isco . in the figure , the novikov - thorne stress is normalized by the time - averaged value of the accretion rate in the simulation . outside the isco , the radial profile of the maxwell stress is similar to the prediction of the novikov - thorne model , but in no case is there a roll - over in the value near the isco . as has been noted previously for the initial dipole field topology @xcite , the @xmath112 stress does not go to zero at the isco , but is continuous down to the event horizon . this is true for all three field topologies , at least in a time - averaged sense , although the amplitude of the stress declines as one goes from dipole to quadrupole to toroidal field topology . in the dipole case , the logarithmic gradient of the stress as a function of radius actually steepens as the accretion flow approaches the horizon . as can be seen from figure [ bndavgplt1 ] , this extra stress does reduce the specific angular momentum carried into the hole by the accretion flow . at the isco the accretion flow in the toroidal field simulation has @xmath113 ( where @xmath114 is the specific angular momentum of the isco ) and the accretion flows in the dipole and quadrupole simulations have @xmath115 ( a change of @xmath116 compared to @xmath117 ) . below the isco , @xmath104 in the quadrupole and toroidal cases remains constant with decreasing @xmath70 . in the dipole simulation , @xmath104 decreases to @xmath118 at @xmath119 , a decrease of @xmath120 compared to @xmath117 . the fall off in the dipole case below the isco is presumably due to the steepening of the maxwell stress in this region . as has been remarked in earlier papers @xcite , @xcite , when @xmath121 , the magnitude of the maxwell stress ( i.e. , the electromagnetic angular momentum flux ) is smaller than the angular momentum flux associated directly with matter ( @xmath122 ) in the inner disk . it follows ( as shown in figure [ ratio ] ) , that the mean accreted angular momentum per unit rest - mass is only slightly below the novikov - thorne value ( i.e. , @xmath123 at the isco ) . nonetheless , this continuity in the stress is important for other reasons : because the dissipation is rather more centrally - concentrated than the stress , this additional stress can be significant in the disk heat budget ; and the relative importance of the additional stress increases sharply for higher black hole spins . thus , we see that the initial magnetic field geometry makes relatively little qualitative difference to the character of the accretion flow . the sense of its ( weak ) influence is that the simulation beginning with a purely toroidal magnetic field results in a saturated field amplitude a factor of two smaller than those resulting from poloidal seed fields . the poynting flux , electromagnetic angular momentum flux , and accretion rate scale in direct proportion to the magnetic field energy density . in their study based on axisymmetric simulations , @xcite also found that certain disk properties are largely independent of the initial field configuration . they found that the shell - averages of magnetic field quantities within the bound portion of the accretion flow ( strictly speaking , where @xmath124 ) follow power - law distributions that are the same regardless of the initial field topology . for example , they found that the field amplitude @xmath125 from the isco out to about @xmath126 . integrating the same magnetic field quantities ( magnetic field strength and the lab - frame projections of the ct - fields ) over the bound portion of the flow in our simulations reveals that these quantities do scale roughly as power - laws in radius , but with shallower gradients than as determined by @xcite . in addition , we see a slow trend with field topology , as well as significant fluctuations in time : @xmath127 in simulation kdpg ( where @xmath128 denotes 1 std . deviation about the time - average ) , @xmath129 in simulation qdpa , and @xmath130 in simulation tdpa . the contrast between our results and @xcite could result from a number of considerations : the initial torus configurations differed in their radial pressure and angular momentum profiles , and our simulations are fully three - dimensional , rather than axisymmetric . these scalings may or may not be universal , but they appear to have little to do with the jet . as we will shortly show in [ jets ] , jet properties differ strongly from one initial magnetic topology to another ; it must be , then , that jets depend on other properties in addition to the current distribution in the disk . @xcite noted that the main disk body is surrounded by a low density region where the magnetic and thermal energies are in approximate equipartition . this region was designated as the coronal envelope . in contrast to the main disk body , where poloidal fluid velocities are dominated by turbulence , the poloidal motions in the coronal envelope are smooth on small scales and create an outgoing , but bound , backflow . figure [ beta ] displays the spatial distribution of the time- and azimuthally - averaged @xmath50 parameter overlaid with contours of the gas density for each of the three different topologies . kdpg shows the familiar strongly - magnetized axial funnel and a corona with relatively large regions of @xmath131 . qdpa shows a lower level of magnetization in the funnel region , while tdpa has much larger regions with @xmath132 overall . to quantify these impressions , we adopt a more precise definition of the corona than that given in @xcite . we define it as those regions of _ bound _ material that lie outside of @xmath133 density - scale heights from the mid - plane , that is those cells that satisfy both @xmath134 and @xmath135 , where @xmath136 is the equatorial density and we measure along curves of constant @xmath70 . using this definition , we integrate over the corona and over time to compute the average total volume and mass fraction in the corona , the volume - averaged @xmath50 parameter , and the mean fluid - frame thermal and magnetic energy per unit mass . these data are given in table [ corona ] . the coronal regions for all three simulations take up half or more of the total simulation volume but contain only @xmath137 of the mass . the volume assigned to the corona for kdpg is a smaller percentage compared to the other models , but kdpg also contains an unbound jet along the funnel axis , which is excluded from the corona by definition . within the coronal volume , the average thermal and magnetic energies per mass are similar , with volume averaged @xmath50 values of @xmath1384 . figure [ angcor ] plots the angular profiles of density @xmath139 , gas pressure @xmath140 , magnetic pressure @xmath141 and @xmath50 @xmath142 at @xmath126 . all three magnetic topology cases have very similar density profiles in the disk body , but appear to differ in the corona . although our averaging period of @xmath77 is 30 orbital periods at the radius shown ( @xmath143 ) , it is not long enough to erase completely fluctuations in the coronal density profile : observe that the quadrupolar case is the most extended in the southern hemisphere " corona while the toroidal case is the most extended in the northern hemisphere " . consistent with the general behavior we have already emphasized , the gas and magnetic pressure profiles in the disk body are very similar in the dipolar and quadrupolar cases , but both pressures are rather smaller in the toroidal field case . nonetheless , all three cases share one important property : the magnetic pressure is almost flat for several scale - heights around the midplane . this finding is consistent with results from shearing - box simulations with much better vertical resolution @xcite . in the corona , the magnetic pressure drops rapidly with height in qdpa and tdpa , but remains nearly constant in kdpg . in all three cases , the magnetic and gas pressures track each other closely , so that the @xmath50 profiles in the three runs are very similar . figure [ angtrptrt ] plots the angular profiles of the electromagnetic contributions to @xmath144 and @xmath145 ( @xmath146 and @xmath147 respectively ) at two different radii , @xmath110 and @xmath109 . at the larger radius , it should come as no surprise that the profile of magnetic pressure shown at that radius in figure [ angcor ] accurately predicts the profile of these two electromagnetic quantities . at the smaller radius , as we have already shown ( fig . [ bndavgplt1 ] ) , the magnetic field is considerably more intense in the dipole case than in the other two , so both the electromagnetic angular momentum flux and energy flux follow suit . to summarize this section , we have found that , just as for the main disk body , the character of disk coronae is only weakly dependent upon the initial field topology . there are modest quantitative contrasts of the same sort as seen in the disk body dipolar and quadrupolar initial fields both create somewhat stronger magnetic fields on average than does initial toroidal field but that is all . the formation of unbound outflows , or jets , is one of the most striking features of the kd models @xcite , all of which were initialized with single - loop dipolar magnetic fields . these outflows generically have two components : a poynting flux jet inside the axial funnel formed by the centrifugal barrier , and a sheath of much denser unbound gas moving outward at sub - relativistic speed along the funnel s outer edge . as described in @xcite , energy for the poynting flux comes from the rotation of the black hole , in a manner closely related to the classical blandford - znajek mechanism . in the case of the schwarzschild hole , the absence of black hole rotation means that the magnetic field in the funnel , although relatively strong , is mostly radial and carries no poynting flux . the funnel wall jet , on the other hand , appears to be powered by the significant gas and magnetic pressure near the isco . the gas retains enough angular momentum to be excluded from the funnel and is squeezed outward along the centrifugal barrier . in contrast , the three - dimensional quadrupole and toroidal field simulations have substantially weaker unbound outflows , if any . figure [ ubndavgplt1 ] shows the time - averaged shell - integrals of the mass outflow rate @xmath148 , magnetic field strength@xmath91 ; along with the electromagnetic energy flux , @xmath92 and angular momentum flux @xmath149 in unbound material between @xmath150 m and @xmath151 m . the values in kdpg are 10100 times larger than those in qdpa , whose outflow quantities are , in turn , an order of magnitude larger than those in tdpa . for example , the unbound mass outflow rate at @xmath152 in kdpg is @xmath153 times that of qdpa , while the mass outflow rate in tdpa is 10 times smaller than in qdpa . the strength of the outflow is directly related to the strength of the funnel field . in kdpg the time - averaged funnel magnetic energy density is 20@xmath154 times greater than in the quadrupole model , while the toroidal model is weaker by another factor of 10 . the mean poynting flux in kdpg is therefore @xmath155 times the poynting flux for qdpa . in other words , unlike the qualitative similarity of the accretion flows produced by these different field topologies , the outflows differ dramatically . to understand better the origin of this sharp contrast , in the next section we focus more closely on exactly how field is introduced into the funnel , and how its magnitude is controlled . we have seen that the strength of the poynting flux jet is very sensitive to the initial magnetic field topology . because the establishment of large - scale field within the funnel appears to be critical to driving a jet with substantial poynting flux , we devote the remainder of this section to a close examination of just how different magnetic topologies lead to different funnel magnetic fields , leaning heavily on high - resolution 2-d axisymmetric simulations . we begin by analyzing the axisymmetric simulation whose initial condition has already been shown to produce a significant jet , the dipole . figure [ dfpevln ] shows a time series of color plots of the @xmath50 parameter overlaid with white contours depicting poloidal field lines . as the radial field is sheared , a toroidal field ( visible in fig . [ dfpevln ] through its effect on @xmath50 ) is created and amplified , leading to poloidal magnetic pressure gradients @xcite . because the initial radial field has opposite sign on opposite sides of the equatorial plane , the toroidal field created by the shear also changes sign across the equator . the inner edge of the torus begins to move inward toward the event horizon , dragging the magnetic field with it . as the gas moves inward , its rotation velocity increases , further strengthening the toroidal field . as the magnetic pressure increases , gas is forced toward the equator where the field is comparatively weak , bringing the regions of oppositely - directed toroidal field closer together ( see , e.g. , the panel illustrating @xmath156 ) . as a result , a strong current sheet forms along the equatorial plane . once the inner edge of the accretion flow arrives at the black hole ( @xmath157 ) , field lines attach to the event horizon , and gas rapidly drains off them . field lines slide toward the poles as they expand outward at relativistic speed , driven by the contrast between the high magnetic pressure near the equatorial plane and the much lower pressure higher up along the axis . this flow moves outward so swiftly because it is almost entirely unencumbered by the inertia of matter . as discussed in @xcite , a magnetic tower forms , similar to the one predicted on analytic grounds @xcite and seen in other simulations @xcite however , in contrast to these models , tightly - wrapped field lines persist only briefly ; the large vertical gradient in magnetic pressure propels the upper loops so rapidly outward that in the steady - state the field lines are better described as loosely helical than as a tightly - wrapped coil . the radial field inside the funnel has the same sign as the vertical component of the initial dipole loop inside the initial pressure maximum ; these field lines close outside of @xmath151 m , reentering the problem volume throughout the corona . a time history of the unbound poynting flux at @xmath152 , normalized by the instantaneous ( coordinate frame ) magnetic field energy within the bound accretion flow , is shown in figure [ 2dtrtdu ] for both the three - dimensional simulation kdpg and the high - resolution axisymmetric model . in both cases , once the jet is established ( a process taking @xmath158 ) , the poynting flux jet persists . it is frequently perturbed , and the amplitude undergoes fluctuations , but its basic properties persist throughout the simulation . excluding initial transients ( i.e. @xmath159 ) , the jet histories of the two simulations are very similar to one another , despite the differences in dimensionality , ( poloidal ) resolution , torus configuration and equation of state : in both cases , the ratio of jet em luminosity to energy density in magnetic energy is @xmath1600.2 per time - unit . this similarity demonstrates that , for this topology at least , the properties of the funnel field depend primarily on the field strength within the initial accretion flow , and that funnel field creation is a primarily axisymmetric process . although the initial magnetic field had structure only on the scale of the disk thickness , the _ essentially laminar _ dynamics of initial inflow lead quickly to the formation of truly large - scale field . on the other hand , because the accretion flow takes longer to be established than the jet does , the magnetic field in the funnel acts in effect as a permanent boundary condition for the accretion flow . in other words , these simulations mimic the long - term behavior of an accretion flow in which there is large - scale net magnetic flux threading the event horizon and filling a cone around the axis , but no large - scale flux piercing the disk proper . as both the three - dimensional and two - dimensional simulations demonstrate , it is entirely possible for long - term quasi - stationary accretion to coexist with a strong poynting flux jet whose foundation is magnetic flux trapped long before . we next consider the evolution of an initial quadrupolar field in high resolution axisymmetry . figure [ qfpevln ] shows a time series of plots illustrating @xmath50 and the poloidal field line structure for this case . initially , the loops of weak poloidal magnetic field enclose two regions , located above and below the pressure maximum ( see figure [ topologies ] , center panel ) . in a similar fashion to the dipole case , these loops are sheared out , creating ever growing toroidal field and producing strong poloidal magnetic pressure gradients . in contrast to the dipole case , however , two current sheets are formed , one above the equatorial plane , one below . the magnetic pressure forces gas toward these current sheets , where the magnetic pressure has a local minimum . as the inner edge of the torus moves radially inward toward the black hole , the current sheets maintain their structure . when the field reaches the black hole , the current sheets are located above and below the equator . as with the dipole , gas drains off the field lines , which anchor themselves in the event horizon , and the field lines between the current sheets and the axis slide toward the poles , where the density and pressure are very low . these field lines expand outward relativistically , filling the funnel region and establishing a large - scale quadrupole field . while this field is in place it supports a poynting - flux jet , but the power in the jet is significantly less than that in the axisymmetric dipolar simulation . averaged from @xmath161 to @xmath162 , the em flux in the dipolar case is twice as strong as in the quadrupolar simulation when measured at @xmath126 , and 10 times stronger measured at @xmath152 ( we will discuss later why the contrast was more than an order of magnitude greater in the 3-d simulations ) . figure [ trtducomp ] compares the dipole jet poynting flux with that of the quadrupole simulation as a function of time at @xmath152 , normalized by the total magnetic field energy in the disk . in terms of this normalized power , the jet in the quadrupole simulation is a factor of 210 weaker than that associated with dipole model ; in other words , much of the actual power contrast between the dipolar and quadrupolar cases can be explained in terms of a reduction in this magnetic efficiency " . unlike the situation for accretion properties , contrasts in jet luminosity from one topology to another can not be described solely in terms of contrasts in mean magnetic field strength . @xcite reported that replacing the dipole with a quadrupole field led to a jet that is weaker by a factor of 23 , but for several reasons one should expect only qualitative , not quantitative agreement with our results . they measured the jet luminosity at @xmath163 and their initial pressure maximum was much closer to the black hole ( @xmath164 ) , so that the radial dynamic range of their genuine accretion flow was considerably smaller . their disk was also geometrically thicker , partly because their code solved a total energy equation , and partly because their initial state was hotter . simple geometry accounts for the qualitative contrast between the jet produced by an initially quadrupolar magnetic geometry and that driven by an initially dipolar field . in the quadrupolar case , relatively small amplitude vertical oscillations in the inner region of the accretion flow can lead to reconnection between magnetic field in the funnel and magnetic field in the disk because vertical field of both signs is available in the same hemisphere . the result is to break the magnetic connection between the black hole horizon and large radius . put another way , the presence of closed field loops on the same side of the equator offers the opportunity for easy collapse of these loops to zero . in the dipole case , however , the sign of the vertical magnetic field passing through the equator is the same everywhere inside @xmath165 . the net result of this geometric contrast is that , while quadrupolar field configurations can pump magnetic field into the outflow funnel and support a jet , they can also equally easily destroy such a jet by reconnection with oppositely - directed field . jets in this case are consequently very unsteady , and exhibit strong variability in the poynting flux , as shown in figure [ trtducomp ] . this variability is a major factor in depressing the time - averaged em luminosity of jets driven by quadrupolar fields . because the code simulates the equations of ideal mhd , reconnection is numerical and occurs at the grid scale . as such , the reconnection rate is resolution dependent ( lower resolution promotes reconnection ) . we can not therefore make specific predictions regarding the variability characteristics of jets in nature , as this is determined ( in our simulations at least ) by competition between processes that are physical in character ( e.g. , orbital dynamics , magnetic forces ) and purely numerical ( the reconnection rate ) . the rate of reconnection is also determined by details of the field geometry , in addition to its basic topology . to investigate this sort of dependence , we have conducted a series of lower resolution ( @xmath166 ) axisymmetric quadrupolar simulations in which we varied the size of the initial torus , and hence the scale of the initial field loops . we kept the inner edge of the torus fixed at @xmath83 , but reduced the angular momentum @xmath27 in order to reduce the vertical thickness of the torus , and consequently reduce the vertical extent of the quadrupolar loops . simulations were carried out for a variety of values of @xmath27 , ranging from @xmath167 to @xmath168 in steps of @xmath169 . for the purposes of our discussion here , we focus on the @xmath170 torus , which serves to illustrate the impact of vertical scale height on the evolution of the quadrupole topology . specifically , setting @xmath171 results in a torus which is initially a factor of two thinner ( in terms of the density scale height , measured along surfaces of constant radius at the location of the pressure maximum ) than the @xmath25 torus previously discussed . the vertical thickness of the resulting quadrupolar field loops is then very similar to that of loops in the sub - keplerian torus used in qdpa . the initial evolution of this system is shown in figure [ qnpevln ] . the evolution proceeds as before : current sheets form above and below the equatorial plane as the inner edge of the torus moves toward the black hole . this time , however , the two current sheets are much closer to the equatorial plane ( contrast fig . [ qnpevln ] with fig . [ qfpevln ] ) . around @xmath172 the current sheets buckle , apparently due to an instability , and the coherent poloidal field structure is destroyed . a similar phenomenon was observed in qdpa , which had a similar scale - height : the current sheets destabilized far from the black hole , destroying the coherent poloidal field structure . as a result , no large scale poloidal field was dragged into the near horizon region by the torus , and the resulting field inside the funnel region is substantially reduced . we note that in three - dimensions the presence of non - axisymmetric mri modes are likely to enhance the rate of reconnection , further decreasing the likelihood that an organized quadrupole configuration will survive to create a significant funnel field . these additional simulations show that the strength of poynting - dominated jets depends on field geometry in two senses : it is topology - dependent and shape - dependent . quadrupolar magnetic topology leads to easier reconnection , and therefore weaker funnel fields and jets , than does dipolar topology , and geometrically thinner quadrupole loops make reconnection still more rapid . we have shown how large dipolar loops can create strong jets , but what happens if the oppositely - directed vertical field on the far side of the loop is accreted ? this did not occur in kdpg , in part because it did not run long enough and in part because the finite mass in our torus meant that the matter outside the pressure maximum moves outward over time , not inward . in a real accretion disk , however , continuing mass re - supply will force the far sides of even large dipolar loops to accrete . to explore this sort of event , we ran a two - dimensional simulation whose initial state contained a series of four narrow dipolar loops ( described in [ numerics ] ) . the initial evolution of this topology is shown in figures [ mfpevln1 ] and [ mfpevln2 ] . the first of these figures shows the descent of the innermost radial loop into the near - horizon region and the establishment of a poynting flux jet . the second figure shows the descent of the second ( moving out in radius ) of these loops and its interaction with the field structures established by the first loop . over the timescale considered , the evolution of this model is determined by the two innermost loops , both initially inside the pressure maximum ( see fig . [ topologies ] ) . as for a single dipole loop , the first loop forms a funnel dipole field by @xmath173 . however , its shape differs significantly from the single loop because the second loop begins its descent toward the inner disk even before the inner edge of the first loop nears the event horizon ( fig . [ mfpevln1 ] ) . as the second loop moves inward , it displaces the field lines of the first loop , forcing the midplane portions of the outer half of the first loop inward , and the off - plane portions of the first loop to higher altitude at their original radius . the result is that the first loop is very nearly transformed into a pair of loops , one on each side of the midplane . after arriving in the inner disk , the field of the second loop expands rapidly toward the funnel region . it pushes on the almost closed sections of the first loop , triggering rapid reconnection where its oppositely - directed segments pass very close to one another at small radius ( fig . [ mfpevln2 ] ) . at this point , the transformation of the first loop into a disconnected quadrupolar loop pair is complete , and the loops fly off to large distance , emptying the funnel of magnetic field . the funnel is then refilled with flux from the second loop . these dramatic events are mirrored in the very large fluctuations in normalized poynting luminosity shown in figure [ trtducomp ] . jumps in output even greater in relative terms than those displayed by the quadrupolar simulation are frequent , and can be associated with these episodes of funnel field establishment and annihilation . note that while the multiple - loop jet is highly variable , similar to the quadrupolar jet , the instantaneous strength of the poynting - flux jet ( when it is strong ) is similar to that of model kdpg when normalized for disk magnetic field strength . in this sense , the efficiency " of these systems in using magnetic field generated in the disk for jets appears to be a function of poloidal field character , declining as the relative importance of higher multipole moments increases . in interpreting this simulation , it is important to bear in mind that , unlike any of the others reported in this paper , its accretion flow never achieves anything resembling a quasi - stationary state . the multiple loop structure also requires high resolution to prevent reconnection from occurring at too rapid a rate . it is for this reason that , unlike the large dipolar loop and quadrupolar caess , we report no 3-d simulation results for this field geometry . thus , we can be less confident than in the other cases that the specific phenomenology we observe may be associated with real systems in a state of continuing accretion . what it _ does _ demonstrate , however , is that accretion of a series of closed dipolar loops can lead to a sequence of field establishment and destruction rather similar to what was seen in the quadrupolar case , and through mechanisms that also resemble those seen when the initial field is quadrupolar . it also supports our qualitative inference that maintenance of a consistent sign of vertical field is essential to the long - term support of a poynting - dominated jet . the lifetime for an episode of strong outflow is roughly the inflow time for the outer anchor point of the loop . in the poloidal cases , magnetic field is injected into the funnel by strong toroidal field pressure in the plunging region . one might expect that this process could act equally efficiently when the magnetic field is toroidal from the start . if the toroidal field has a consistent sign , the radial component created when it expands outward would also have a consistent sign , satisfying the coherent vertical field criterion we have inferred from study of the poloidal simulations . if so , the toroidal field case should also drive a powerful jet . however , it does nothing of the sort . figures [ tdpevln1 ] [ tdpevln2 ] are a series of snapshots of the @xmath50 parameter and the radial field structure in the evolving torus of simulation tdpa . in contrast to the poloidal loop simulations , there is no initial radial field to generate toroidal field through shear . instead , the evolution is governed entirely by the development of the nonaxisymmetric mri . the fastest - growing modes for the mri in this context have large @xmath174 and azimuthal wavelength @xmath175 , which is much shorter than the disk thickness when the field is weak @xcite . the magnetic structures that develop are then predominantly on short lengthscales . note the rapid spatial variation in sign in the radial field in figure [ tdpevln2 ] . with time , the radial field features become somewhat more radially elongated , but field lines do not connect extended regions ; no large - scale coherent poloidal field is generated from the initial toroidal field . magnetic field amplification is less than in the poloidal field initial conditions , and @xmath50 never becomes less than 1 near the hole . matter tends to remain with the field rather than drain either into the hole or toward a localized region such as the equator , as seen in the dipole case . any vertical expansion of the field would therefore have to overcome the inertia of the matter and , as can be seen from figure [ tdpevln2 ] , this does not occur . the absence of a funnel - field in the toroidal topology follows directly from the specific properties of the toroidal field mri : the poloidal fields created by the mri are typically small scale . no radially - extended poloidal field connects the near horizon region to the disk . what radial field there is is accreted faster than it can tap into the differential rotation and generate strong near - hole fields . the main disk - body in the toroidal case remains gas pressure - dominated at all times , and there are no strong magnetic pressure gradients to drive an expansion of the field into the funnel region . the absence of a funnel field means there is no poynting flux jet . it appears , then , that creation of a strong funnel field , and therefore a strong poynting jet when the black hole spins rapidly , depends essentially on the existence of some coherent poloidal field in the initial state . purely toroidal field initial conditions can lead to short coherence - length poloidal loops with rapidly alternating sign , but no large - scale poloidal field components . through a series of numerical studies , we have examined the influence of the initial field topology on the properties of magnetized black hole accretion flows . in particular , we have performed full three - dimensional simulations of three different initial field topologies : a large - scale dipolar loop , a pair of quadrupolar loops separated by the equatorial plane , and a purely toroidal field . apart from the imposed initial field topologies , the initial conditions used in each of these simulations were identical : a @xmath176 hydrostatic torus with a slightly sub - keplerian distribution of angular momentum around a rotating ( @xmath177 ) black hole . the two poloidal field topologies were evolved to @xmath178 m , while the toroidal case was evolved to @xmath179 m . @xcite described the late - time structure of the resulting accretion flow in terms of distinct components : the main disk body , the coronal envelope , and an unbound outflow which itself could be separated into a matter - dominated funnel - wall jet and a poynting - flux jet in the funnel interior . we have found that the late - time averaged properties of the main disk body exhibit only modest quantitative contrasts attributable to different initial field topologies : the dipolar and quadrupolar cases are roughly similar , and both create somewhat stronger magnetic fields than the toroidal case . so long as there is some weak magnetic field present , the mri generates turbulence and drives accretion , although the character of the purely toroidal mri is somewhat different from that produced by poloidal fields . because the maxwell stress controls the surface density required to support a given accretion rate , and the cascade of energy to smaller scales in the mhd turbulence leads ultimately to disk heating , we expect that the radiation emitted by the disk will likewise be only weakly dependent on magnetic topology . it is possible that different magnetic topologies could produce different vertical profiles of dissipation ( which these simulations are not designed to study ) , but the similarity in magnetic pressure profiles ( fig . [ angcor ] ) suggests that even those differences are likely to be minor . in regard to the accretion flow , we found only two dependences on field topology worth noting : for fixed outer torus mass , the initial toroidal field simulation gives a somewhat lower accretion rate . in addition , the magnetic field in the plunging region is , in relative terms , somewhat stronger in the large dipolar loop case than for quadrupolar or toroidal fields , so that stresses near the isco are somewhat stronger when the field topology is dipolar . in other words , the short - wavelength behavior of the mri - driven mhd turbulence in the disk produces near - universal behavior , independent of magnetic field structure . jets are quite different . _ laminar _ flow dynamics can , given the right conditions ( e.g. , the right field topology ) , inflate field with no pre - existing structure on scales larger than a disk scale - height into a truly global field . simulations that begin with a large dipole loop in the initial torus produce a strong poynting - flux jet in the axial funnel and a comparably strong matter - dominated funnel - wall outflow . the inner field lines of the dipolar loop move quickly inward to fill the funnel , and remain there until the opposite side of the loop is accreted , an event which may not happen until quite a long time later . thus , this case mimics a situation in which net magnetic flux has been trapped against the black hole event horizon by previous accretion . when this trapped flux creates a coherent poloidal field in the funnel , the black hole s spin forces nearby field lines into rotation and drives a large outward - directed poynting flux . this sequence of events is not as easily achieved with any of the other field geometries we explored . unlike the dipole configuration , the quadrupole features a pair of current sheets located above and below the equator . these current sheets are subject to instabilities that can lead to reconnection in the magnetic field and loss of extended field coherence . even when the quadrupole is successful in establishing a funnel field , the jet may live only briefly because fluctuations in the accretion flow can generate reconnection events . by the same token , quadrupolar field can also restore field to an empty funnel . consequently , any jet based on quadrupolar field tends to be unstable and episodic , with the timescale and amplitude of changes set both by details of the field geometry and the rate of reconnection . unfortunately , it is very hard to obtain any information about the former in real systems , and in our simulations the latter is controlled by numerical effects and resolution , so it is impossible for us to make predictive statements about the timescale or amplitude of jet variability in this case . all of the models reported on above used a black hole with a spin @xmath180 . to test our conclusions against other black hole spins we also ran two three - dimensional initially quadrupolar simulations with black hole spins @xmath181 and , @xmath182 . the effect of a wide range of black hole spins ( including a counter - rotating hole ) on dipolar models was studied by @xcite . no poynting flux jet is present for a schwarzschild hole regardless of the initial field topology . the energy to drive the poynting flux originates with the spin of the hole . however , the funnel is still filled with radial field , which is considerably stronger in the dipole case compared to the quadrupole . on the other hand , the funnel field strength increases with black hole spin for a given field topology . for the rapidly - spinning hole , the poynting flux jet is stronger than in the @xmath121 simulations for both topologies , but as we have discussed , it is considerably weaker when the initial field is quadrupolar than when it is dipolar . specifically , the ratio between the time - averaged poynting luminosity with dipolar field to the poynting luminosity with quadrupolar field is @xmath183 for @xmath184 and @xmath185 for @xmath121 . thus , the accretion dynamics control what sort of field is brought to the vicinity of the black hole and fill the funnel , and the black hole spin determines how energetic the resulting jet is . the purely toroidal field topology does not generate any jet at all . the only way for toroidal field to create any poloidal field is by initiating non - axisymmetric perturbations of alternating sign . the lengthscales of these perturbations must be relatively short , and the portions of opposite sign must balance in magnitude . fields of this sort can do very little to fill the funnel region because they are even more subject to reconnection than is the quadrupolar field . when the initial field is a single dipolar loop that wraps around the initial pressure maximum of the simulation , all the vertical field brought down to the funnel by the initial accretion has the same direction . in real disks containing closed dipolar loops , oppositely directed vertical field will , sooner or later , arrive in the inner disk , and it may cancel the funnel field . to investigate this possibility , we ran a high - resolution axisymmetric simulation that began with four narrow dipolar loops within the initial torus . its subsequent evolution qualitatively supports this conjecture . the innermost loop produces a funnel field as in the single dipole loop model . when the second loop , oriented oppositely from the first one , descends , it displaces the field established by the first loop . the funnel field established by the first loop reconnects , detaches from the event horizon , and is rapidly expelled from the grid . the second loop then expands poloidally and reestablishes the funnel field . in other words , the accretion of dipolar field loops leads to jet creation , but its structure is highly - variable : the narrower the dipolar loops present in the accretion flow , the more variability we can expect in the jet . when a simulation is initialized with ( possibly multiple ) dipolar field loop(s ) that cross the equatorial plane , the strength of the resulting poynting - flux jet depends on the strength of the magnetic field within the disk itself and on the spin of the hole . poynting - flux jets that are formed from quadrupolar fields are generically much weaker than their dipolar counterparts , even when magnetic field strength in the disk body is taken into account . how much weaker depends on other factors as well , including disk thickness , location of the inner edge of the initial torus , and numerical resolution . in the limited range of numerical experiments reported here , we have seen suppression factors ranging from several to several hundred . real accretion flows are likely to have a mixture , possibly time - varying , of all these field topologies . real jets from black holes will exhibit behavior combining the sorts of events we have seen in this paper : stability over longer timescales when the field is mostly made of large dipolar loops , greater variability and occasional jet quenching when the field has a larger contribution from higher multipoles , jet elimination when the field is mostly toroidal . it is possible that in addition to the sort of coherent funnel field spontaneously created by large closed - loop poloidal structures , there is a large - scale field imposed by external boundary conditions . we plan to investigate the impact of such an externally - imposed field in future work . this study is only one small step in the process of exploring the mechanisms to create large scale field configurations within accretion flows . here the large scale field formation process we have observed results from an interaction between poloidal field components in the accretion flow and the boundary represented by the black hole . the mechanism operates in axisymmetry as well as in three dimensions , so it is not a dynamo if that word is defined as in the antidynamo theorem . many authors have conjectured that an inverse cascade within the three dimensional turbulent disk itself might generate large scale fields ; we see no evidence in these simulations of any disk field with structure on scales more than a few scale - heights , but we caution that the resolution limitations of three dimensional global simulations make this observation less than definitive . the results we have reported also stimulate speculations about whether the contrasts in jet phenomenology from one object , or class of objects , to another is due in part to differences in the mixture of different field topologies , or the interaction of different topologies in the flow . for example , could the jet outbursts sometimes seen in galactic black hole binaries when they cross the jet line " @xcite be due to rapid field annihilation and loop expulsion of the sort observed in the multiple dipolar loop simulation ? perhaps the radio - weakness of most agn ( see e.g. * ? ? ? * ) is due to the predominance of small - scale multipolar field in their accretion flows as a result of accreting from a turbulent interstellar medium ? as we learn more about how it acts , we may find that magnetic topology can , in fact , be inferred from phenomenology . surface normalized to the total magnetic field strength within the disk for simulation kdpg ( black lines ) and the high resolution axisymmetric simulation of the dipole ( magenta line).,scaledwidth=50.0% ] surface normalized by the total magnetic field strength within the disk . the kdpg dipole simulation is shown in black , the quadrupole simulation in blue , and the multiple loop simulation is green.,scaledwidth=50.0% ]

understanding transport of charged solutes across the cell membrane is a problem of paramount importance in biophysics , since it is crucial to regulate many cell functions @xcite . this task is undertaken by transmembrane channels , so the characterization of the permeation of charged particles , ions in particular , is essential . in addition , recently there has been a renewed interest to comprehend transport through biological nanochannels for possible applications in biotechnology @xcite . bacterial porins are macromolecular proteins located at the outer membrane that enable the diffusion of small molecules through the lipid bilayer . as porins are well characterized , both structurally and functionally , they represent model systems to study transport through biological nanochannels . in particular , the ionic transport properties of the ompf porin have been extensively studied since the determination of its x - ray structure @xcite . the simplest description of the complex problem of ionic transport across channels is based on the classical diffusion - advection transport equations : @xmath2 @xmath3 here , @xmath4 is the flux of particles of a given ion of charge @xmath5 and @xmath6 its concentration profile , @xmath0 is a constant diffusion coefficient characterizing the brownian motion ( with a value in principle different from bulk diffusion coefficient ) and @xmath7 is a deterministic force . in the classical nernst - planck equations , the deterministic force arises from the mean - field electrostatic potential , @xmath8 @xcite , whereas in more advanced approaches it may contain steric or entropic forces @xcite . in some cases , integrated solutions of eqs.([cons],[np ] ) under suitable boundary conditions and appropriate assumptions are employed . the goldman - hodkin - katz equation or the henderson equation are well - known examples , being extensively used in the literature to understand the conductance , reversal potential , and other measurable quantities of nanochannels @xcite . in other cases , it is solved self - consistently in combination with poisson s equation of electrostatics , forming the so - called poisson - nernst - planck ( pnp ) equations . this approach allows the analysis of electrically charged nanochannels in a consistent fashion @xcite . in all cases , a continuous mean - field approximation is assumed , i.e. ions are treated not as discrete entities but as continuous charged densities that represent the space - time average of the microscopic motion of the ions . the diffusion coefficient of ions , @xmath0 , characterize their motion within the channel , representing a spatial average over the region considered . in analyzing experiments , diffusion coefficients are often given bulk values when the comparison with experiment is qualitative or are treated as free parameters to adjust by fitting experimental data @xcite . + in recent years , the increase in computer power and the development of new algorithms have permitted the study of ion transport through certain nanochannels employing all - atom molecular dynamics ( md ) simulations @xcite . such analysis captures all the atomic detail of the system ( including the water molecules present ) , which in some cases is necessary to properly account for the interactions of the ions with the interior of the channel . however , these md simulations still remain computationally very expensive , so its use is hindered by limitations in the size of the system ( @xmath9 nm ) and the duration of the processes ( @xmath10 ns ) that such calculations are capable of tackling . nevertheless , even for the analysis of larger systems or longer processes , all - atom md simulations can be used to obtain parameters that coarser ( less detailed ) models employ in their description of the system . in particular , md simulations can be used to obtain diffusion coefficients of ions in the interior of nanochannels , which are central input parameters in many theoretical descriptions of ion permeation ( like , e.g. , brownian dynamics simulations @xcite or approaches based on the nernst - planck equation @xcite ) . tieleman and berendsen @xcite characterized the diffusive dynamics of water molecules in the ompf porin by using the einstein relation for the mean square displacement ( msd ) . they calculated local diffusion coefficients which they averaged out over slices perpendicular to the axis of the pore to obtain the variation of the diffusion coefficient along the axis of the channel . a similar approach was used with the ompf porin @xcite and other channels@xcite to determine the dependence of ionic diffusion coefficients along their axial direction . such approaches , however , do not provide global diffusion coefficients to describe the entire channel , and it is not clear how they can be determined . to circumvent that limitation , hijkoop et al . @xcite , inspired on the work of munakata and kaneko @xcite , used a method based on a first passage time ( fpt ) analysis to compute a global diffusion coefficient of water in the interior of an artificial ( uncharged ) ompf porin . their analysis , however , ignores the influence of free energy barriers and wells on the local movements of water molecules at a specific position along the axis of the pore . in this paper , we use a method based on a first passage time ( fpt ) analysis of the particles trajectories to study the transport of potassium and chloride ions through the ompf porin . this procedure allows us to obtain the effective diffusion coefficients of the ions in the channel from all - atom md simulations @xcite . on the one hand , from the trajectories of the ions provided by the simulations one can calculate several fpt quantities that characterize their passage through the channel . note that such quantities are produced taking into account all the complexity of the all - atom description . on the other hand , it is possible to derive analytical expressions of such fpt quantities for a model based on the 1d diffusion - advection equations with constant parameters , eqs.([cons],[np ] ) . the comparison of the fpt quantities obtained from the md simulations with the analytical formulas give answer to two relevant questions for the characterization of the channel . first , it determines whether the one - dimensional description of the dynamics of the ions in the channel based on eqs.([cons],[np ] ) is valid . second , in case such a description is accurate , we can obtain the diffusion coefficients and effective forces that best define the channel by fitting the fpt quantities calculated from the md simulations with their corresponding analytical expressions . the rest of the paper is organized as follows . in section [ sec : ompf + simulation methods ] we describe the structure of the ompf porin and discuss the model and the details of the md simulations . in section [ sec : first passage approach ] , we define the first passage quantities which will be considered and give the corresponding analytical expressions for the 1d model based on eq.([cons],[np ] ) . in section [ sec : results ] we provide the results of the analysis . we discuss the fpt quantities obtained for the whole channel and also study the fpt properties of different regions within the channel . from this analysis , effective diffusion coefficients and forces are obtained for the different regions . in section [ sec : conclusions ] , the conclusions of the study are given , as well as a table with the obtained diffusion coefficients that characterize the ionic transport of the channel . the biological nanochannel considered here , the outer membrane protein f ( ompf porin ) , is a transmembrane channel located at the outer membrane of the bacteria _ escherichia coli_. the structure of the wt ompf channel is illustrated in figure [ channel ] . the crystallographical structure of the wild type ompf channel is well known from long ago @xcite . it is made of three identical monomerical nanopores , each one assembling into a large 16-stranded antiparallel @xmath11-barrel structure enclosing the transmembrane pore ( figure [ channel ] ) . each aqueous pore has a diameter between 1 - 4 nm , being constricted around half - way through the membrane by a long loop . it is a relatively wide channel in the sense that it allows the simultaneous permeation of both cations and anions ( in _ hydrated _ form ) and also the passive diffusion of larger dipolar molecules . the transport of monovalent ions and molecules across this channel is known to be dominated by electrostatics , as demonstrated by studies considering mutations of the channel ( with different pore sizes and different electrostatic charges ) and different permeating specimens of different sizes and charge distributions @xcite . in this paper we study the transport of potassium and chloride ions through the ompf porin using a fpt approach from results of all - atom molecular dynamics simulations previously published in refs . @xcite . for the sake of completeness , we summarize here the most important features of the simulation . the system is composed of the ompf porin inserted in a popc membrane in contact with a 1 m kcl aqueous ionic solution , see fig.[snapshot ] . the whole system is thermalized at @xmath12k . the coordinates of the ompf porin are obtained from its x - ray structure @xcite . the protonation states employed for the titatrable residues of the protein are the same as those specified in ref . @xcite , which results in an overall charge of @xmath13e per monomer . the molecular dynamics simulations were carried out with namd version 2.6 @xcite . we used the force - field for protein - lipid simulations supplied by namd2 , which is a combination of the charmm22 force field , parameterized to describe protein systems , and the charmm27 force field@xcite , parameterized to describe lipid systems in aqueous media . water was also modeled after the charmm force field @xcite , which describes water using a slightly modified version of the standard tip3p model . within this force field , ions are described as charged lennard - jones spheres . all simulation details can be found in refs.@xcite . to study the ionic transport properties of the channel , the whole system is under the action of a uniform electric field perpendicular to the lipid bilayer ( along the negative @xmath14 axis , see fig.[snapshot ] ) . the magnitude of the field is @xmath15mv / nm , which creates a potential drop of @xmath16 mv across the membrane+channel system . from a computational perspective , the high electrolyte concentration , of approximately 1 m , is convenient , since the large number of ions included in the simulation significantly helps to obtain statistically meaningful averages of the ion properties inside the nanochannel . such high concentrations , however , are not unrealistic . similar concentrations are often used in experiments to measure ion fluxes and selectivity properties of ion channels @xcite . the total simulated time is @xmath17ns , considerably longer than previous md simulations done of the ompf porin @xcite , which also helps to improve the statistical averages of ion properties inside the channel . axis is applied . for clarity , only a fraction of the lipids of the membrane are represented . figure produced using vmd @xcite.,width=321 ] let us consider a diffusing brownian particle under the action of an external force in a one dimensional interval @xmath18 $ ] . to characterize its dynamics in the interval one can compute first passage quantities , which deal with the event of the particle reaching for the first time any of the boundaries ( first passage event ) . the central magnitude that characterizes these events ( from which other first passage quantities can be calculated ) is the distribution of first passage times , given a starting position of the particle within the interval . this is obtained from the probability that the particle reaches for the first time any of the boundaries at a given time @xmath19 . this first passage problem is , thus , approximately equivalent to the problem of a diffusing particle in an interval @xmath18 $ ] with absorbing boundaries @xcite . the analysis based on a first passage approach is well suited to characterize the dynamics of the ions in the channel from the results of molecular dynamics simulations . one can easily calculate the different first - passage quantities characterizing the ions dynamics inside a given interval within the channel from inspection of the individual trajectories provided by the simulations results . a first passage quantity that we employ in our analysis is the _ hitting probability_. it is defined as the probability that a particle which starts at an initial position @xmath20 reaches for the first time either the lower boundary at @xmath21 , @xmath22 , or the upper one at @xmath23 , @xmath24 . the dependence of @xmath25 on @xmath20 is determined by the presence of free energy barriers and external forces acting on the ions . the calculation of the hitting probability from the md trajectories of the ions is simple . at each timestep of the simulation we identify the ions contained within the layer @xmath26 $ ] , with @xmath27 nm ( thiner layers were tested with identical results ) . we then trace all those particles until they reach for the first time any of the virtual boundaries of the interval , moment at which they loose their identity ( although they are not physically removed from the system ) . by keeping track of the number of ions that reach each boundary we can compute , once we have gone through all the simulation , the hitting probabilities of each type of ion for a given initial position @xmath20 . the same procedure is repeated for different starting points @xmath20 ( separated by @xmath28 ) until the whole interval is covered to obtain the hitting probabilities from md simulations , @xmath29 . another magnitude of interest in the application of the fpt approach to the analysis of ionic dynamics in the channel is the _ mean exit time _ ( met ) , @xmath30 . it is defined as the average time taken for a particle to reach for the first time any of the boundaries given that its initial position is @xmath20 . to calculate the met from the md simulations results we divide the system into layers of width @xmath28 as done before , and at each timestep identify the ions within the layer centered at @xmath20 . those particles are tracked until they reach either boundary and the time elapsed is recorded . from the collection of residence times of the particles starting at @xmath20 we compute the average time to reach either boundary . this is repeated for all @xmath20 in the interval to obtain the met from md simulations , @xmath31 . the other fpt quantity that we use in the characterization of ion dynamics in the channel is the _ survival probability _ , @xmath32 . it gives the probability at time @xmath19 of finding a particle within the domain @xmath33 $ ] that will cross the domain from one boundary to the other without crossing the entrance at intermediate times . for the calculation of this quantity , at each timestep of the simulation we identify those ions that enter the considered interval and , for each particle , we record the time of entrance into the interval and a tag which specifies the boundary of entrance . as done in the previous cases , we track those particles until they reach any of the boundaries , where they loose their identity . if the boundary of exit is the same as the boundary of entrance the trajectory is discarded . if the particle exits through the opposite boundary , the exit time is recorded so we can compute the time taken by that particle to cross the interval . from the collection of the crossing times it is then straightforward to calculate the survival probability from md simulations , @xmath34 . in the presence of an external effective force dragging the ions , there are many more events of ions crossing the interval in the direction of the dragging force than in the opposite direction . in order to provide statistically meaningful results , only the survival probability corresponding to the passage of ions in the direction of the dragging force will be considered . the fpt quantities obtained from the md simulations can be used to check the validity and to extract the parameters of a certain theoretical model . based on such a model , one can calculate the corresponding theoretical fpt quantities @xmath35 , where @xmath36 are the parameters of the model . by comparing such theoretical quantities with those computed from md simulations , @xmath37 , we can test the applicability of the model and also , under favorable conditions , determine the parameters @xmath36 that define such model . note that the parameters thus obtained contain in an effective fashion all the information on the geometry and interactions of the all - atomic simulation . here , we consider the simplest diffusion model based on ions characterized by a constant diffusion coefficient @xmath0 moving under the influence of a constant effective force @xmath1 . for such a simple model , the classical transport eqs . ( [ cons],[np ] ) apply and simple analytical expressions for the fpt quantities can be deduced in terms of the parameters @xmath0 and @xmath1 . these analytical expressions are derived by noting that the one - dimensional fpt problem is approximately equivalent to the problem of a diffusing particle in a 1d interval with absorbing boundaries . for detailed derivations , the reader is refereed to standard textbooks such as ref . @xcite . here we limit ourselves to list the results relevant for our analysis . the hitting probabilities are given , in terms of @xmath0 and @xmath1 , by @xcite @xmath38 note that @xmath39 depend solely on one of the parameters of the model , @xmath1 . consequently , the comparison of eqs.([hitprob ] ) with the results from the md simulations , @xmath29 , is well suited to obtain the effective force acting on the ions . within the considered theoretical model , the mean exit time reads @xcite @xmath40 } { 1- e^{\frac{-f_\text{eff}l}{k_bt}}}\,.\ ] ] finally , the analytical expression for the survival probability , @xmath41 , can also be deduced from the diffusion - advection equations , eqs.([cons],[np ] ) , @xcite : @xmath42 with @xmath43 $ ] . at times @xmath44 , @xmath45 is dominated by the first term , with @xmath46 $ ] , which directly depends on the diffusion coefficient . in the analyses done below , based on the comparison of the fpt results from the md simulations with eqs.([hitprob ] ) , ( [ survp ] ) and ( [ met ] ) , the interval length @xmath47 needs to be considered slightly longer than its actual length in the md simulations @xcite . this extra distance , which appears to be dependent on the type of particle and the interval , has been studied and identified before as the _ milne extrapolation length _ @xcite . the necessity for such a ( very small ) correction emerges because the absorbing boundary conditions assumed in @xcite to derive eqs.([hitprob ] ) , ( [ met ] ) and ( [ survp ] ) are not exact for the virtual interval considered in the analysis of the simulations . to check the validity of the method described above , we analyzed first the dynamics of the ions in bulk , far from the channel and membrane , which we know is describable using a 1d nernst - planck equation ( eqs.([cons],[np ] ) ) with constant parameters and whose results can also be obtained from the analysis of the mean square displacement ( msd ) . + for this analysis we have selected a 2nm - thick slab of electrolyte far enough from the channel and membrane , @xmath48nm@xmath49 nm ( see fig.[snapshot ] ) . the delimiting planes of the slab perpendicular to the @xmath14-axis act as virtual absorbing boundaries , while in the transverse directions ( @xmath50 , @xmath51 ) periodic boundary conditions are applied . the protocol that was followed to obtain the parameters of the theoretical model , @xmath0 and @xmath1 , is as follows . the effective force parameter @xmath1 is determined first from fitting the hitting probabilities of reaching the upper boundary of the specified region , @xmath52 , with eqs.([hitprob ] ) . the dependence of these probabilities on the initial position of the particle is determined only by the effective force @xmath1 , being a straight line for @xmath53 and increasing its curvature as the parameter grows ( concave for @xmath54 and convex for @xmath55 ) . second , the diffusion coefficients are obtained from the comparison of the md results for the survival probability , @xmath34 , with eq.([survp ] ) at @xmath44 . the slope of @xmath56 $ ] directly gives the diffusion coefficient , provided that @xmath1 is known . an alternative way to obtain the diffusion coefficients that is used to check the consistency of the results consists of the comparison of the computed mean exit times , @xmath31 , with the analytical expression eq.([met ] ) . , of reaching the upper boundary of the bulk region ( located at @xmath57 nm ) for ( a ) chloride ions ( b ) potassium ions as a function of their initial position , @xmath20 . the linearity of the curves is an evidence supporting that @xmath58.,width=321 ] from the calculation of the hitting probability @xmath52 we have verified that , due to the much higher resistance of the membrane+channel , in the bulk there is a negligible drop of electrostatic potential . the effective force @xmath1 obtained from the comparison of @xmath52 with eq.([hitprob ] ) , which is determined by the value of the electric field , is not distinguishable from @xmath59 ( see fig . ( [ pbulk_hittingprob ] ) ) . the analysis of the survival probability provides the diffusion coefficients for the cl@xmath60 and k@xmath61 ions . from the slope of the fitting straight line represented in figs.([pbulk_survp ] ) we obtain @xmath62nm@xmath63ns and @xmath64nm@xmath63ns with the help of eq.([survp ] ) . these values compare very well with the values obtained from the study of the mean exit time . by fitting the results for the mean exit time computed from the md simulations , @xmath31 , with its corresponding analytical expression , eq.([met ] ) , we obtain @xmath65nm@xmath63ns and @xmath66nm@xmath63ns . such agreement of the parameters , obtained from two independent methods , shows the consistency and robustness of the fpt procedure employed to extract the values of the diffusion coefficient @xmath0 and effective force @xmath1 that characterize the simple 1d model . an extra independent test was done to check the reliability of the results obtained from a first passage time approach . simulations of a 1 m kcl electrolyte in a box with periodic boundary conditions were performed and calculated the diffusion coefficients using einstein s formula for the msd @xcite . the resulting diffusion coefficients are @xmath67nm@xmath63ns and @xmath68nm@xmath63ns , which are in very good agreement with both sets of values obtained from a first passage time approach . the results are also in good agreement with the experimental bulk diffusion coefficients for @xmath69 and @xmath70 in a 1 m kcl concentration at 298k : @xmath71 nm@xmath63ns and @xmath72 nm@xmath63ns @xcite . .,width=321 ] in the following , we present the results for the first passage quantities corresponding to the transport of ions across the whole channel . the channel is defined as the central part of the ompf trimer in which the monomers have a rigid beta barrel structure , leaving out the more flexible loops and turns located at the edges of the protein . in our simulation , the channel approximately extends from @xmath73 nm to @xmath74 nm , see fig . ( [ regions ] ) . the results for the fpt quantities , figs.([pall_hittingprob ] ) and ( [ pall_met ] ) , are obtained from the analysis of the trajectories of the ions of the all - atom md simulation of the system , which contains the exact geometry and considers all interactions within the charmm force field . they provide valuable information on the ion transport inside the channel . from fig.([pall_hittingprob ] ) we can infer the existence of an energy barrier located in the vicinity of @xmath75 nm that prevents the passage of both ions , since the probability of crossing @xmath75 nm is very small . the results on the met represented in fig.([pall_met ] ) provide the characteristic residence times for both ions as a function of their position within the channel . from those values we can obtain the average time of residence in the channel for both ions , @xmath76 ps for chloride and @xmath77 ps for potassium ions . the different position of the maximum of the met for both ions , located at @xmath78 for chloride ions and at @xmath79 for potassium ions ( with @xmath80 being the middle of the channel ) , reflects the overall action of the external electric field ( directed towards negative @xmath14-direction ) , which drags cations along the field and anions in the opposite direction . another interesting outcome of the analysis is that the results for the hitting probabilities and mean exit times shown in figs . ( [ pall_hittingprob ] ) and ( [ pall_met ] ) , contrary to the bulk case , can not be satisfactorily fitted by the analytical expressions eqs.([hitprob ] ) and ( [ met ] ) . our results show that it is not possible to assign constant diffusion coefficients to the ions inside the channel . therefore , this result cast some doubt on the adequacy of one - dimensional models based on the nernst - planck equation with constant diffusion coefficients to describe the ion transport through the whole ompf porin . nm@xmath81 nm . constriction , @xmath82 nm@xmath83 nm . vestibule2 , @xmath84 nm@xmath85 nm . figure produced using vmd @xcite.,width=321 ] in order to reach a better understanding of the ionic transport through the ompf porin , we divided the system into different regions and , through the comparison of the fpt quantities , @xmath37 , with the corresponding analytical expressions , determined the parameters that characterize the ion dynamics ( i.e. , the effective diffusion coefficient @xmath0 and the effective force @xmath1 ) in each region . based on its structure , we split the channel into three different regions : _ constriction _ , defined as the central region between @xmath86 nm and @xmath87 nm ; _ vestibule1 _ , defined by @xmath73 nm and @xmath86 nm ; and _ vestibule2 _ , in between @xmath87 nm and @xmath74 nm ( see fig.([regions ] ) ) . we employed the same protocol used in the analysis of the ions in bulk to study the dynamics of @xmath69 and @xmath70 in the constriction region . although the quantitative agreement between the first passage quantities computed from the simulations results and the corresponding analytical expressions is not accurate , the general functional form of the md results is well captured by the analytical expressions with the appropriate constant parameters ( see figs.([pconstr_hittingprob ] ) , ( [ pconstr_survp ] ) , ( [ pconstr_met ] ) ) . such an agreement ensures the adequacy of the 1d diffusion - advection , eqs.([cons ] , [ np ] ) , to describe the dynamics of the ions inside the constriction . it must be stressed here the reduced number of ions present in the constriction , which results in poor statistics and might contribute to the observed deviations . from the analysis of the hitting probabilities , we obtain the values of the effective force acting on the ions ( see fig.([pconstr_hittingprob ] ) ) . the effective force is @xmath88 k@xmath89t@xmath90/nm and @xmath91 k@xmath89t@xmath90/nm for potassium and chloride ions , respectively . as expected , potassium ions are dragged along the direction of the external electric field , whereas chloride ions are dragged in the opposite direction . the slightly different magnitude of both forces is consistent with the observation that anions and cations follow different pathways in the constriction , as has been reported in previous md studies @xcite and in experiment @xcite . assuming that the section of the channel along the @xmath14-axis is uniform in the constriction ( see fig . ( [ regions ] ) ) , it follows ( see below ) that the effective force acting on the ions is determined solely by the local electric field acting on them . from the values of the effective forces we can estimate the magnitude of the local electric field for potassium ions , @xmath92 v / nm , and for chloride ions , @xmath93 v / nm . the fit of the time dependence of the survival probabilities obtained from the md simulations @xmath34 with eq.([survp ] ) provides the diffusion coefficients @xmath94 nm@xmath95/ns , @xmath96 nm@xmath95/ns . as seen in fig . ( [ pconstr_survp ] ) , the statistics in this case is poor , so the results for the diffusion coefficients are associated with large errors . these values compare well with the values obtained from the analysis of the mean exit time : @xmath97 nm@xmath95/ns , @xmath98 nm@xmath95/ns . also , as seen in fig.([pconstr_met ] ) , the profile of the mean exit time is well reproduced by eq.([met ] ) with such parameters . note that the diffusion coefficients of both ions in the constriction are roughly an order of magnitude smaller than the diffusion coefficients in the bulk . a similar procedure was employed to analyze the dynamics of the ions at the vestibules of the constriction , defined as the regions between @xmath86 nm and @xmath73 nm ( vestibule1 ) and @xmath87 nm and @xmath74 nm ( vestibule2 ) ( see fig . ( [ regions ] ) ) . the first general observation is the remarkable good agreement between the first passage quantities computed from the md simulations results and the corresponding analytical expressions with the appropriate constant parameters ( see figs.([pvest_hittingprob ] ) , ( [ pvest_survp ] ) and ( [ pvest_met ] ) ) . this indicates that the dynamics of the ions in the vestibules is properly described by the simple 1d model based on eqs.([cons],[np ] ) with the appropriate parameters @xmath0 and @xmath1 . for the vestibules , the results for the hitting probabilities @xmath52 give the surprising results shown in figs.([pvest_hittingprob ] ) . in the two regions , the effective foreces acting on both anions and cations have the same sign , being negative in vestibule1 ( @xmath99k@xmath89t@xmath90/nm for chloride ions , and @xmath100k@xmath89t@xmath90/nm for potassium ions ) and positive in vestibule2 ( @xmath101k@xmath89t@xmath90/nm for chloride ions , and @xmath102k@xmath89t@xmath90/nm for potassium ions ) . this indicates that the dominant contribution to the effective force dragging the ions is not charge dependent and its direction must be along the @xmath14-axis ( and of larger magnitude ) in vestibule2 and in the opposite direction in vestibule1 . also , the magnitudes of the effective forces acting on each ion are different . the effective force acting on potassium ions is greater than the force on chloride ions in vestibule1 , but it is smaller in vestibule2 . + such results correspond well with the action of charge independent _ entropic forces _ , which arise when the dynamics of particles through channels is described with one dimensional models . in treating channels of varying cross section , the analysis of the dynamics of ions can be reduced , under appropriate plausible hypothesis , from a three - dimensional problem with a complex boundary to a one - dimensional problem of diffusion along the channel axis @xcite . assuming that the distribution of particles in the planes perpendicular to the axis of the channel is close to its equilibrium distribution , one obtains the modified ficks - jacobs equation . following reguera and rubi s formulation @xcite , in this approximation the flux of particles , @xmath103 , can be expressed as : @xmath104\,.\ ] ] here @xmath6 is the 1d particle concentration ( with dimensions of particles per unit length ) , @xmath105 is a free energy given by the electrostatic energy @xmath106 and the entropy @xmath107 , where @xmath108 is the dimensionless cross section of the channel . this entropic term accounts for the variation of the space accessible for the diffusion of the particles , dragging them towards positions @xmath14 of maximum entropy . note also that eq.([jfj ] ) is a particular case of eq.([cons],[np ] ) with a deterministic force arising from free energy barriers , @xmath109 . from eq.([jfj ] ) we can infer that the effective forces obtained in the vestibules from the analysis of first passage quantities can be decomposed into two additive contributions : @xmath110 here , @xmath111 is the component of the force arising from electrostatic interactions , whose sign depends on the charge of the ion under consideration . @xmath112 accounts for the effective entropic force , which depends solely on the shape of the channel and which points towards the direction that maximizes the available space . hitting probability , , @xmath113 , of reaching the upper boundary for water particles in ( a ) vestibule1 ( b ) vestibule2.,width=321 ] with the help of eqs.([feff ] ) , we can obtain the two contributions , @xmath111 and @xmath112 , from the values of the total effective force computed from the first passage analysis , @xmath114 in vestibule1 , the resultant effective electrostatic force is @xmath115 k@xmath89t@xmath90/nm , and the entropic part is @xmath116k@xmath89t@xmath90/nm . in vestibule2 , the resultant electrostatic component is @xmath117 k@xmath89t@xmath90/nm , and the entropic part is @xmath118 k@xmath89t@xmath90/nm . to confirm the existence of a charge independent entropic force arising from the variation of the available space along the axis of the channel , we have also analyzed the first passage properties of water molecules at both vestibules ( see figs.([pvestaigua_hittingprob ] ) ) . in this case , the only contribution to the effective force is entropic . for vestibule1 we find @xmath119 k@xmath89t@xmath90/nm , and for vestibule2 we find @xmath120 k@xmath89t@xmath90/nm . survival probability of ions in ( a ) vestibule1 ( b ) vestibule2.,width=321 ] at this point we can interpret qualitatively the results obtained from the analysis of the hitting probabilities at the two vestibules . in vestibule1 , the effective section of the channel decreases along the @xmath14-axis , which induces the appearance of an _ entropic force _ in the same direction as the external electric field ( antiparallel to the @xmath14-axis ) . thus , for cations both contributions to the effective force have the same direction whereas for anions they have opposite directions . as the entropic part is dominant , the effective force on both ions is directed towards decreasing @xmath14 values , although its magnitude is larger on potassium ions than on chloride ions . in vestibule2 , on the contrary , the effective section of the channel increases along the @xmath14-axis , so the _ entropic force _ is directed towards increasing @xmath14 values , and the two contributions of the effective force have the same direction for anions and opposite directions for cations . hence , in this case the total effective force is in the direction of the @xmath14-axis and its magnitude is larger for chloride ions than for potassium ions . mean exit time of ions in ( a ) vestibule1 ( b ) vestibule2 as a function of their initial position.,width=321 ] the entropic force can be independently estimated using eqs.([jfj ] ) from the dependence of the effective radius of the channel on the @xmath14-coordinate , as computed from the md simulations . we obtain an average @xmath121 k@xmath89t@xmath90/nm at vestibule1 and an average @xmath122 k@xmath89t@xmath90/nm at vestibule2 . the discrepancies in the precise values obtained for the entropic forces were expected , since the expressions used are only valid for smoother changes of the section of the channel along the @xmath14-coordinate . once the effective forces have been determined we can obtain the values of the diffusion coefficients . the logarithmic fit of the time dependence of the survival probabilities , @xmath34 , provides the diffusion coefficients @xmath123 nm@xmath95/ns and @xmath124 nm@xmath95/ns in vestibule1 , and @xmath125 nm@xmath95/ns , @xmath126 nm@xmath95/ns in vestibule2 ( see figs.([pvest_survp ] ) ) . independently , from the analysis of the mean exit times , we obtain @xmath127 nm@xmath95/ns , @xmath128 nm@xmath95/ns in vestibule1 and @xmath129 nm@xmath95/ns , @xmath130 nm@xmath95/ns in vestibule2 ( see figs.([pvest_met ] ) ) . again , such an agreement of the parameters , obtained from two independent methods , shows the consistency and robustness of the fpt procedure employed to extract the values of the diffusion coefficient @xmath0 . in this paper we have studied the diffusive dynamics of chloride and potassium ions in the interior of the ompf porin under the influence of an external electric field directed along its axis . from the results of extensive all - atom molecular dynamics simulations of the system we computed several first passage time quantities to characterize the dynamics of the ions in the interior of the channel . such quantities , calculated considering all the complexity of the system , provide valuable information on the residence times and presence of free energy barriers for ions in the interior of the ompf porin . these fpt quantities were also used to test the validity of simplified 1d descriptions of the dynamics of the ions in the interior of the channel based on the use of constant diffusion coefficients for the ions . overall , our results show a slow and complex dynamics for the ions inside the channel . the sluggish ionic dynamics inside the channel is characterized by large residence times ( @xmath131 0.8 ns for cl@xmath60 and @xmath132 ns for k@xmath61 ) and very large mean exit times with peaks around @xmath133 ns , see figure [ pall_met ] , which indicate a strong interaction of the channel with the ions . the complexity of the dynamics is demonstrated by the fact that it is not possible to characterize chloride and potassium ions in the interior of the channel with a single diffusion coefficient . however , it is possible to find well - defined diffusion coefficients for some regions of the protein channel . based on its x - ray structure , we have divided the channel into three different regions : constriction , vestibule 1 and vestibule 2 . employing the same fpt analysis we demonstrate that each of such regions is describable by a simple model characterized by a constant diffusion coefficient @xmath0 and a constant deterministic force @xmath1 acting on the ions . the diffusion coefficients @xmath0 which characterize the dynamics of chloride and potassium ions in the different regions , obtained by two independent methods , are given in table [ table : results ] . let us remark that these diffusion coefficients vary by almost an order of magnitude from the bulk to the interior of the constriction . the physical origin of the effective force @xmath1 is also different in the different regions of the channel . the effective forces on the ions at the constriction , as inferred from the fpt results of the md simulations , correspond to the presence of an intense local electric field directed along the applied electric field which drags potassium ions along and chloride ions in the opposite direction . in the vestibules , the effective forces observed on the ions are consistent with the expected effect of entropic forces which arise from the variation of the available space along the axis of the channel ( see eq.([jfj ] ) ) . cc|c|c| & & + & & survival prob . & mean exit time + & & 2.05 & 2.07 + & & 2.11 & 2.08 + & & 0.21 & 0.22 + & & 0.18 & 0.26 + & & 0.66 & 0.59 + & & 0.65 & 0.57 + & & 0.33 & 0.31 + & & 0.49 & 0.44 + we believe that our results have important implications for the modeling of ionic channels . in this field , there is a great interest in having simple , 1d models , with the capability of predicting the essential features of these extremely complex systems . such models ( like , e.g. , the pnp approach ) are based on the classical 1d transport equations @xcite ( see eq.([cons],[np ] ) ) and heavily rely on the existence of a well - defined diffusion coefficient . our work strongly suggests that in general a direct use of a constant diffusion coefficient for the whole channel is not justified . however , it also shows the plausibility of a multi scaling approach , in which these models would play an important role . the idea is to take into account the complexity of the structure of the protein by obtaining diffusion coefficients from detailed md simulations in regions of different geometrical properties . the coefficients thus obtained can be used in the classical 1d descriptions of ionic channels , such as the pnp approach . also , we recall that our work further demonstrates that in such 1d descriptions , entropic forces such as those described in refs.@xcite must be taken into account . work in this direction of a multi - scaling approach is now under way . this work is supported by the spanish government ( grants fis2009 - 13370-c02 - 02 , fis2007 - 60205 and consolider - nanoselect - csd2007 - 00041 ) , generalitat de catalunya ( 2009sgr164 ) and fundaci caixa castell - bancaixa ( p1 - 1a2009 - 13 ) . is supported by the jae doc program of the spanish national research council ( csic ) . jf acknowledges the award of an intramural pie grant from the spanish national research council ( csic ) . the supercomputing resources employed in this work were provided by the cesga supercomputing center , spain . 99 , _ ion channels of excitable membranes _ ( sinaure , sunderland , 2001 ) . , _ nature nanotechnology _ * 4 * , 437 ( 2009 ) . , _ structure _ * 3 * , 1041 ( 1995 ) . , _ diffusion processes _ ( springer , new york , 1967 ) . , _ j. phys , chem . _ * 96 * , 3926 ( 1992 ) . e _ * 64 * , 061106 ( 2001 ) . d. reguera , g. schmid , p. s. burada , j. m. rub , p. reimmann , and p. hnggi , _ phys . lett . _ * 96 * , 130603 ( 2006 ) . , _ soft matter _ * 1 * , 417 ( 2005 ) . , _ biophys . j. _ * 72 * , 97 ( 1997 ) . , _ biophys . j. _ * 76 * , 1346 ( 1999 ) . , _ biophys . j. _ * 72 * , a108 ( 1997 ) . , _ j. mol . biol . _ * 322 * , 851 ( 2002 ) . , _ j. phys . b _ * 113 * , 8745 ( 2009 ) . , _ biophys . j. _ * 88 * , 3745 ( 2005 ) . , _ biophys . j. _ * 97 * , 1898 ( 2009 ) . , _ biophys . j. _ * 99 * , 2107 ( 2010 ) . . _ j. chem . phys . _ * 115 * , 4850 ( 2001 ) . . biol . _ * 322 * , 851 ( 2002 ) . , _ j. mol . * 294 * , 1159 ( 1999 ) . , _ biochemistry _ * 40 * , 6319 ( 2001 ) , _ plos comput . biol . _ * 5 * , 1 ( 2009 ) . , _ biophys . j _ * 74 * , 2786 ( 1998 ) . , _ _ * 319 * , 1177 ( 2002 ) . , _ biophys . j _ * 77 * , 2502 ( 1999 ) . , _ j. chem . _ * 127 * , 085101 ( 2007 ) . , e _ * 47 * , 4076 ( 1993 ) . , _ stochastic precesses in physics and chemistry _ , ( north - holland publishing company , 1981 ) . , _ a guide to first - passage processes _ , ( cambridge university press , 2001 ) . the refined structure of ompf porin from e.coli at 2.4 angstroms resolution . _ protein data bank _ freely available at http://www.pdb.org code 2omf ( doi:10.2210/pdb2omf / pdb ) . , _ biophys . j. _ * 82 * , 160 ( 2002 ) . * 90 * , 1617 ( 2006 ) . , _ softmatter _ , doi : 10.1039/c0sm00904k ( 2010 ) . . graphics _ * 14 * , 33 ( 1996 ) . , _ biophys . j. _ * 90 * , 112 ( 2006 ) . , _ j. comp . chem _ * 26 * , 1781 ( 2005 ) . mackerell , a.d . , brooks , b. , brooks , c.l . , iii , nilsson , l. , roux , b. , won , y. and karplus , m. charmm : the energy function and its parameterization with an overview of the program , in the encyclopedia of computational chemistry , 1 , 271 - 277 , p. v. r. schleyer et al . , editors , ed . john wiley & sons , chichester , 1998 . , _ j. phys . * 20 * , 1345 ( 1987 ) . , _ molecular simulation _ ( accepted ) . preprint available at http://arxiv.org/abs/1005.2857 ( 2010 ) , _ self - diffusion in electrolyte solutions _ , elsevier , amsterdam ( 1989 ) . , _ j. mol . biol . _ * 396 * , 293 ( 2010 ) . , _ j. chem . phys _ * 122 * , 204701 ( 2005 ) . p. kalinay , j. k. percus , _ phys . e _ * 74 * , 041203 ( 2003 ) . e _ * 80 * , 020904 ( 2009 ) .

in the last few years a growing amount of high - quality data concerning white dwarfs ( wds ) , both from space as well as from ground based telescopes , have been produced . in particular , the discovery of wd sequences in globular clusters ( paresce , de marchi , & romaniello 1995 , richer et al . 1997 ) and the observation of their faint ends in old open clusters ( von hippel , gilmore , & jones 1995 , von hippel & gilmore 2000 ) provide new tools for the study of the history of both the halo and the disk of the milky way . the possible use of the observed wd sequences as age indicators for a variety of galactic components is particularly promising . as firstly shown by mestel ( 1952 ) , the luminosity of a wd is largely supplied by its thermal energy content , so that the cooling time scale is inversely proportional to a certain power of the luminosity . thus , comparisons between theoretical estimations of the age - luminosity relation for wds and the observed cooling sequences may be used to derive stellar ages . on the other side , an accurate comparison between the predictions of the theoretical models with specific observations of wds could provide a better comprehension of the physical properties of high density matter . in this context the recent development of the astereoseismology applied to wds allows to check the internal structure of these compact objects ( bradley & winget 1991 , bradley , winget , & wood 1992 , clemens 1993 , metcalfe , winget , & charbonneau 2001 ) . white dwarfs are the final destiny of the evolution of low and intermediate mass stars . the majority of the presently observed wds are post - asymptotic giant branch stars ( agb ) . the core of an agb star is made of ashes of the he - burning , namely a mixture of primary carbon and oxygen , with a minor amount of secondary @xmath2nen , during the h - burning , and , later on during the he - burning , into @xmath2ne via the chain @xmath3n@xmath4f@xmath5o@xmath6ne . thus , the abundance ( in number ) of @xmath2ne in the co core of a wd roughly correspond to the abundance ( in number ) of the original cno . ] . during this evolutionary phase the core mass increases , due to the accretion of fresh material processed by the he - burning shell . in the meantime , a huge mass loss erodes the h - rich envelope until it is reduced down to a critical value ( see e.g. castellani , limongi , & tornamb 1995 ) , and the star departs from the agb . about @xmath7 of the total mass of the resulting wd is the co core builded up during both the central and shell he - burning phases . m@xmath8 is the minimum stellar mass for which a carbon ignition occurs before the onset of the agb and set the upper mass limit for the progenitors of co wds . current stellar evolution theory predicts that the value of m@xmath8 varies between 6.5 and 7.5 m@xmath9 , for metallicity ranging between z=0 and 0.02 ( see cassisi , castellani , & tornamb 1996 and dominguez et al . 1999 for recent reviews of these calculations ) . according to a well established theoretical scenario , single stars having mass lower than m@xmath8 end their life as co wds with masses ranging between about 0.55 and 1.05 m@xmath9 ( see e.g. dominguez et al . the theoretical initial - final mass relation , as derived from stellar evolution models , roughly agrees with the few available measurements of wd masses in nearby open clusters ( weidemann & koester 1983 , weidemann 1987 , 2000 , herwig 1997 ) . however , it should be recalled that stellar rotation could induce the formation of more massive co wd ( dominguez et al . 1996 ) . few wds may have an he - rich core . in principle , stars having initial mass particularly low ( namely m@xmath100.5 m@xmath11 ) never attain temperatures high enough for the he - burning . after the main sequence , they develop a degenerate he core and their final fate should be a low mass he - rich compact remnant . however , owing to the very long evolutionary time scale of these stars ( much longer than the hubble time ) , we can definitely exclude that the present generation of wds could include their remnants . nevertheless , low mass he - rich wds may result from the deviated evolution of more massive progenitors when , under particular circumstances , a complete envelope removal takes place during the first ascent of the red giant branch . for example , a stripping of the envelope may be caused by a close encounter with another star ( livne & tuchman 1988 ) or by roche lobe overflow in close binary systems ( kippenhahn , koll , & weigart 1967 , kippenhahn , thomas , & weigart 1968 , iben & tutukov 1986 ) . it has been also supposed that massive wds could be generated by stars having an initial mass slightly larger than m@xmath8 ( dominguez , tornamb , & isern 1993 , ritossa , garcia - berro , & iben 1996 , 1999 ; garcia - berro , ritossa , & iben 1997 ) . in this case , in fact , an highly degenerate o - ne - mg core is left by the carbon burning . then , after a rather normal agb phase , eventually characterized by an important mass loss , these stars could terminate their life as massive onemg wds ( m@xmath12 m@xmath9 ) . in this paper , we confine our analysis to the co wds having an h - rich envelope , which are the most common type of these compact remnants ( the so called da ) . let us note that the physical evolution of the internal structure of a white dwarf appears rather simple when compared to other stellar evolutionary phases . the residual nuclear burnings provide only a negligible contribution to the total luminosity . the highly degenerate core is almost incompressible , so that only the contraction of the most external layers may still contribute to the release of gravitational energy . only the brightest wds ( logl / l@xmath13 ) suffer a non - negligible energy loss through neutrino production . the cooling time is then determined by the rate of temperature decrease , which depends on the efficiency of the energy transport from the hot core through the thin opaque envelope . a very efficient thermal conduction , due to the highly degenerate electrons , takes place in the whole co core that rapidly becomes almost isothermal . on the contrary , the more external layers of the wd ( i.e. an he - rich mantel and , eventually , an h - rich envelope , where electron are initially only partially or non degenerate ) are efficient insulators and regulate the energy loss rate basically , the time scale of the wd evolution is controlled by the heat capacity of the co core and by the opacity of the external wrap . thus , the main problem of this studies concerns the accurate evaluation of these two physical ingredients and no sophisticated algorithms for the computation of the stellar structure is , in principle , required . nonetheless , we believe that the best approach should be that based on a full evolutionary code calculation , which allows a coherent description of both the progenitor and the wd evolution and may account for all the contributions to the energy production and transport . in this paper we critically re - analyze the influence of the assumed theoretical scheme on the predicted cooling time . in such a way , we try to evaluate the present uncertainty of the cosmo - chronology predictions based on wd models . a rough idea of the present level of uncertainty may be obtained by comparing some recently published theoretical cooling sequences for h - rich atmosphere wd ( see figure 1 ) . one may recognize sizeable differences , in particular near the faint end of the cooling sequence . they are likely due to different evaluations of the basic ingredients of the theoretical cooking : the specific heat in the core , the radiative and the conductive opacities , the treatment of the external convection and the like ; but they could be also partially attributed to differences in the initial models . we argue that the origin of these uncertainties should be identified before using cooling sequences as cosmic clocks . owing to the complexity of the matter , our investigation is far from being exhaustive . many previous works have addressed the question of the reliability of the input physics used in wd models computation . in some case we do not repeat their analysis . for example , we do not address the question of the chemical segregation occurring in the core when the crystallization of the oxygen component takes place , because this argument has been exhaustively discussed in the recent literature ( segretain et al . 1994 , salaris et al . 1997 , montgomery et al . 1999 , isern et al . 1997 , 2000 ) . in some case we extent previous investigations up to the faint end of the cooling sequence , as in the case of the effects of microscopic diffusion . in the following section we illustrate the input physics used to compute the various cooling sequences . in the third section we discuss the uncertainties related to the progenitor evolution that determine the internal stratification of the core and the size of the thin external layers : the he - rich mantel and the h - rich envelope . section four illustrates the dependence of the cooling time scale on some physical ingredients ( eos , opacity and the like ) . a final discussion follows . our wd models have been obtained by means of a full evolutionary code . in particular we have used the franec in the version described by chieffi & straniero ( 1989 ) . the input physics have been completely revised in order to account for the peculiar conditions developed in wd interiors . we have adopted the radiative opacities of opal ( iglesias & rogers 1996 ) for high temperature ( log@xmath14 > 4.0 $ ] ) and the alexander & ferguson ( 1994 ) molecular opacities for the low temperatures ( log@xmath14\leq 4.0 $ ] ) . the conductive opacities have been derived from the work made by itoh and coworkers ( itoh et al . 1983 , mitake , ichimaru , & itoh 1984 , itoh , hayashi , & kohyama 1993 ) . additional models have been obtained by using the hubbard & lampe ( 1969 ) prescriptions and the table provided by a. potekhin ( see potekhin et al . as it is well known ( see e.g. mazzitelli 1994 , dantona & mazzitelli 1990 ) , there is a region in the @xmath15 plane not covered by the opal radiative opacities and where the electron conductivity is not yet dominant . thus one has to extrapolate the radiative opacities in order to fill this gap . the problem is alleviated by the use , for the lower temperature , of the alexander & ferguson ( 1994 ) opacities , which extend at larger densities with respect to the opal . present models have been computed adopting a linear extrapolation of the radiative opacity beyond the upper density provided by opal . we have tested the reliability of this choice by changing the extrapolation method . negligible effects on the cooling sequences have been found . nuclear reaction are almost extinct during the wd evolution , but they play a relevant role in determine the chemical structure of the model at the beginning of the cooling sequence . we have used the rates tabulated by the nacre collaboration ( angulo et al . 1999 ) , except for the @xmath16c@xmath17o . during the he - burning this reaction competes with the @xmath18 and regulates the final c / o ratio in the core ( iben 1972 ) . the great influence on the resulting wd evolution has been deeply investigated ( dantona & mazzitelli 1990 ; salaris et al . we have used two different rates , namely the one reported by caughlan et al . ( 1985 ) and that of caughlan & fowler ( 1988 ) . note that the difference between these two compilation may be roughly considered as representative of the present experimental uncertainty level ( see e.g. buchmann 1996 ) . nuclear reaction rates are corrected for the weak and intermediate electron screening by using the prescriptions of graboske et al . 1973 and de witt , graboske , & cooper 1973 , and for the strong screening by itoh , totsuji , & ichimaru 1977 and itoh et al . 1979 . the efficiency of neutrino emission processes are taken from : haft , raffelt , & weiss 1994 ( plasma neutrinos ) , itoh et al . 1989 ( photo and pair neutrinos ) . as recognized long ago ( salpeter 1961 ) coulomb interactions play a relevant role in wd interior , finally leading to crystallization of the stellar core ( lamb & van horn 1975 ) . thus , a detailed treatment of the thermodynamic behavior of both liquid and solid matter is a necessary physical ingredient to study the evolution of cold wds . the high pressure experienced in the whole core ensures the complete ionization of carbon and oxygen . thus , we have updated and extended the eos for fully ionized matter described by straniero ( 1988 ) . in particular , we have revised the treatment of the electrostatic corrections up to the liquid - solid transition and beyond . the free energy in the fluid phase can be written as @xmath19 where @xmath20 and @xmath21 are respectively the contribution of the ideal gas of ions and electrons . to compute these terms , as in straniero ( 1988 ) , we have assumed that ions follow the boltzmann distribution , while electrons are described by fermi - dirac integrals , for an arbitrary degree of degeneracy and relativistic state . @xmath22 is the excess of ionic free energy due to ion - ion coulomb interactions . for this contribution we adopted the analytical expression by potekhin & chabrier ( 2000 ) that in the region of high ionic coupling parameter ( @xmath23 ) fits the accurate results of recent monte carlo simulations of one component plasma ( de witt & slattery 1999 ) , while for @xmath24 it approaches to the cohen & murphy ( 1969 ) expansion and reproduces the debye - huckel limit for vanishing @xmath25 ( landau & lifshitz 1969 ) . the quantum diffraction correction to ionic free energy has been neglected in the liquid phase . @xmath26 represents the excess electron free energy due to electron - electron interactions . for this term we adopted the non relativistic expression by tanaka , mitake , & ichimaru 1985 . we neglected the relativistic @xmath26 . this is a reasonable approximation since at high density , when degenerate electrons become relativistic , the electron coupling parameter @xmath27 . finally , for @xmath28 , that describes the excess due to ion - electron interactions , we adopted the analytical expression given by potekhin & chabrier ( 2000 ) . as it is well known , the contribution of the ion - ion interactions increases with density , eventually leading to crystallization . according to potekhin & chabrier ( 2000 ) , the liquid - solid phase transition has been assumed at @xmath29 . in the solid phase ( @xmath30 ) the free energy can be written as @xmath31 where the electron terms @xmath21 and @xmath26 are the same as in the fluid phase . for the free energy of the ionic crystal @xmath32 we used the analytical expression by farouki & hamaguchi ( 1993 ) obtained by fitting numerical models of molecular dynamic . the contribution of the ion - electron @xmath28 in the solid phase is from potekhin & chabrier ( 2000 ) . finally , the term @xmath33 represents the quantum correction to the thermodynamic behavior of the ionic coulomb crystal . this contribution is very important in late wd evolutionary phases , in fact , as the wd cools down , the crystallized co core reaches a quantum state ( diffraction parameter @xmath34 , where @xmath35 denotes the ion plasma frequency ) . when this occurs , the ionic contribution to the specific heat decreases as @xmath36 , thus depleting the main thermal reservoir of the star : it is called debye cooling phase . for this term we adopted the analytical expression described by stolzmann & blocker ( 2000 ) , which is based on the free energy of a coulomb crystal studied by chabrier , ashcroft , & de witt ( 1992 ) . the various thermodynamic quantities for pure carbon and pure oxygen have been obtained by analytically deriving the corresponding free energy . finally , an additive volume interpolation is used to calculate the thermodynamic quantities of a co mixturene . ] . to describe the outer layers of partially ionized helium and hydrogen , we adopted the eos of saumon , chabrier , & van horn ( 1995 , scvh ) . this eos requires both high pressure and low pressure extensions . at high pressure ( p@xmath37 dyn/@xmath38 ) , which is in any case large enough to guarantee a full ionization of h and he , we extended the scvh eos following the same procedure described above for the co core . the match between the two eos is generally good . at low pressure , we have used a perfect gas ( including h , h@xmath39 , h@xmath40 , h@xmath41 , he , he@xmath39 and he@xmath42 ) plus radiation . in such a case the classical saha equation has been used to derive the population of the various species . for each temperature , the precise value of the maximum pressure of the perfect gas ( or the minimum of the scvh ) has been varied in order to guarantee a smooth transition between the two eos . tables of this eos are available on the web . in order to fix the external boundary condition of a stellar model , appropriate model atmosphere are needed . in the last decade wd atmosphere theory has significantly improved ( bergeron , waesemael , & fontaine 1991 ; bergeron , saumon , & waesemel 1995 ; bergeron , waesemel , & beauchamp 1995 ; saumon & jacobson 1999 ; bergeron 2000 ) . as it is well known , in the high density / low temperature atmospheres of an old wd the emerging electromagnetic flux significantly departs from the black body spectrum . this is mainly due to the molecular hydrogen recombination and the consequent collision - induced absorption ( cia ) by h@xmath40 - h@xmath40 collisions . cia is the main source of opacity in the infrared and the main cause of increasingly blue color indices for decreasing effective temperature ( t@xmath43 ) in cold wds ( hansen 1998 , 1999 ; saumon & jacobson 1999 ; bergeron 2000 ) . detailed model atmospheres are also crucial ingredient to transform the theoretical quantities ( l , t@xmath43 ) into the observational ones ( magnitude , color ) . in the present computations we have used a solar scaled t-@xmath44 relation ( see chieffi & straniero for details ) up to the onset of the cooling sequence . then , namely when log l / l@xmath45 , we adopt the model atmospheres of bergeron , saumon , & wesemael ( 1995 ) , plus a low temperature extension including the effects of the cia , which has been kindly provided us by p. bergeron . since no metals are included in these models , they are consistent with the usual hypothesis of complete sorting of the external layers ( see next subsection ) . convective boundaries are fixed by using the schwarzschild criterion . concerning the progenitor evolution , the extension ( in mass ) of the convective core during the central he burning have a great influence on the amount of c ( and o ) in the most internal layers of a wd ( imbriani et al . 2001 and dominguez et al . 2001 ) . during this phase , we use the algorithm described by castellani et al . ( 1985 ) to take into account the growth of the convective core induced by the conversion of he ( low opacity ) into c and o ( large opacity ) and the resulting semiconvection . breathing pulses , which occur when the central he mass fraction decreases below about 0.1 , are usually neglected ( but see the discussion in section 3.3 ) . note that any mechanism that increases the size of the well mixed region during the final part of the he - burning ( mechanical overshoot , semiconvection , breathing pulses or rotational induced mixing ) leads to a reduction of the resulting amount of c in the central region of the wd ( see e.g. imbriani et al . 2001 ) . theoretical studies ( fontaine & michaud 1979 , iben & macdonald 1985 , 1986 , althaus et al . 2002 ) , supported by observational evidences ( see e.g. bergeron , ruiz , & legget 1997 ) , indicates that , as a consequence of the gravitational settling of heavy elements , da wds should have a practically pure h envelope and a pure he mantel . in addition , as suggested by salaris et al . ( 1997 ) , the carbon and oxygen profiles left by the he - burning is smoothed by rayleigh - taylor instabilities . in order to account for the element diffusion , when log l / l@xmath45 , we adjust the composition of the envelope and the mantel : all the residual h is putted on the top of a pure he layer . the total mass of h and he is conserved . we also modify the internal c and o profile according the the prescription of salaris et al . ( 1997 ) . owing to the larger @xmath25 , oxygen crystallization occurs when carbon is still liquid . for a certain time , this occurrence produces an unstable stratification of the liquid phase and , in turn , an efficient mixing ( isern et al . 1997 , salaris et al . 1997 ) . this phenomenon is presently not included in our calculations . first of all one should know the physical and the chemical structure of the star at the beginning of the cooling sequence . since our investigation regards da white dwarfs , we need to know the internal stratification of the co core and the extension ( in mass ) of both the he and h - rich layers . the formation of the co core begins during the central he - burning phase and proceeds in the following agb phase . in principle , a lengthy computation through a series of recursive thermal pulses is required , until the mass loss erodes the envelope and the star becomes a wd ( see e.g. iben & renzini 1983 ) . during the thermally pulsing agb phase , the mass of the co core is accreted with the ashes of the he - burning shell , while the mass between the two shells is progressively reduced ( see e.g. straniero et al . thus , it is very important to determine the duration of the agb , which is evidently affected by mass loss : the longer the agb phase , the larger the final mass of the co core and the smaller the mass of the he - rich mantel . agb mass loss rate ranges between 10@xmath46 and 10@xmath47 m@xmath0/yr , but it is still unclear how it is related to the changes of the chemical and physical structure occurring in the cool atmospheres of these giants ( see habing 1996 ) . in addition , if the wd progenitor is a low mass star ( m@xmath481.5 m@xmath0 ) , the mass loss during the first ascent to the tip of the red giant branch ( rgb ) must be also carefully taken into account . the spread in color of the horizontal branch ( hb ) stars in globular clusters is commonly interpreted as an evidence of the different mass loss experienced by similar stars during the rgb phase : bluer hb stars should have a smaller envelope than red hb stars . since the agb phase of blue ( reduced envelope ) hb stars will be significantly shorter than that of a red hb star , the resulting wd will be different . in other words , stars with similar initial mass and chemical composition might produce white dwarfs of different masses and/or different internal stratification , depending on the mass loss history . not only the mass , but also the initial composition of a star determines its evolution and , possibly , its final fate . a detailed descriptions of the internal c and o profiles in wds generated by stars with mass ranging between 1 and 7 m@xmath0 and metallicity 0@xmath10z@xmath100.02 may be found in dominguez et al . ( 2001 ) . this variety of possible progenitor evolutions could frighten the reader . our scope is to quantify the uncertainty on the cooling time of wd , for different galactic components . basing on some numerical experiments , we have evaluated the influence of the pre - wd history on the age calibration . starting from zero age he - burning stellar structures ( i.e. an he core surrounded by an h - rich envelope , in which a suitable mixture of helium and metals is adopted ) , we have computed the evolutionary sequences up to the wd stage . at the beginning of the cooling phase ( @xmath49 ) the envelope has been completely sorted in a pure he buffer and a pure h outer envelope mimicking the effect of the very efficient gravitational settling . table 1 reports the initial parameters of the models and the properties of the resulting white dwarfs , namely : identification label of the model , initial mass ( m@xmath50 ) , metallicity ( z ) , initial he - core mass ( m@xmath51 ) , final mass ( m@xmath52 ) , final amount of he ( @xmath53m@xmath51 ) , final amount of h ( @xmath53m@xmath54 ) , central carbon mass fraction ( x@xmath55 ) , mass of the homogeneous central region ( q = m@xmath56/m@xmath52 , see below for a definition of this parameter ) , wd age when log l / l@xmath0=-5.5 . let us firstly discuss models b , b1 , b2 . the total mass has been maintained constant ( m=0.6 m@xmath9 ) during the whole evolution for both case b and b1 . they differ for the initial mass of the h - rich envelope , namely : @xmath57 and @xmath58 respectively for b and b1 . the third case ( b2 ) has been computed starting from a larger mass ( m=0.64 m@xmath9 ) and imposing a constant mass loss ( @xmath59 m@xmath60 ) since the onset of the thermally pulsing phase . in this case , the initial he - core mass is in agreement with recent evolutionary computation of h - burning models of low mass stars ( see e.g. straniero , chieffi , & limongi 1997 ) . note that the initial total mass has been chosen to obtain a final ( wd ) mass similar to the ( constant ) one of case b and b1 . a low metallicity ( z=@xmath61 ) , adequate for old halo wd progenitors , has been chosen for all these three models . figure 2 illustrates the evolutionary tracks in the hr diagram . the central he - burning phase is similar in the three cases . at the central he exhaustion , model b1 evolves toward the red part of the hr diagram and develops the typical double shell structure of an agb star . when the faster he - burning shell get closer to the h - burning one , it dies down and the star enters in the thermally pulsing phase ( for a more detailed description see e.g. iben & renzini 1983 ) . after two major thermal pulses , the envelope mass is reduced down to about 0.001 m@xmath0 ; then , the star leaves the agb . during the post - agb , a further ( last ) thermal pulse takes place ; the star re - expand and comes back to the asymptotic giant branch . finally , after a second horizontal crossing of the hr diagram , both shells definitely die down and the model settles on the wd cooling sequence . owing to the very small envelope mass , after the central he exhaustion , model b remains in the blue side of the hr diagram , so skipping the classical agb phase . when the outgoing he shell approaches the h - rich layer it dies down . note that the h - burning shell has been practically inactive up to this moment . the following contraction of the star induces some irregular and short h - ignitions . finally this model attains its cooling sequence . in model b2 the duration of the agb and the final wd mass are mainly determined by the assumed mass loss rate , which is , in any case , quite typical for faint tp - agb stars . this models experience 8 thermal pulses ; then , the envelope is reduced down to a certain critical value and the model moves toward the blue side of the hr diagram ( see figure 2 ) . note that the mass loss has been stopped when the effective temperature becomes larger than 7000 k. at variance with case b1 , no more thermal pulses are found in the post - agb phase . the resulting profiles of carbon and oxygen in the core of the three final wd models are shown in figure 3 . the most internal region is produced during the central he - burning and its flatness is a consequence of the convective mixing occurring in the stellar core . as noted by salaris et al . ( 1997 ) , the region immediately outside the flat profile left by the central convection presents a positive gradient of mean molecular weight . the consequent rayleigh - taylor instability induces a mixing and a further little reduction of the central carbon . in column 8 and 9 of table 1 we have reported the central c mass fraction and the mass of the homogeneous central region ( in unit of the total wd mass ) as they should result after the action of this rayleigh - taylor instability ( q = m@xmath56/m@xmath52 ) . the growing c profile in the more external layer is produced during the agb , when the he - burning shell advances in mass toward the stellar surface . the carbon peak at the external border of the co core reflects the incomplete he - burning left by the last thermal pulse . the three chemical structures are rather similar . in particular the amount of carbon ( oxygen ) in the center of the wd and the extension of the inner flat region are practically the same . the most relevant differences is the size ( in mass ) of the he - rich mantel ( see column 6 in table 1 ) . in model b1 the thickness of the mantel is about half of that found in case b. this is the memory of the last thermal pulse experienced by model b1 in the post - agb phase , when the h - burning shell was almost extinct . the amount of hydrogen left in the envelope of the wd is practically the same in the three models . this confirms the fact , already demonstrated in many papers , that for each metallicity and core mass , it exist a maximum mass of the h - rich envelope for a wd ( iben 1982 , fujimoto 1982a , b , castellani , deglinnocenti , & romaniello 1994 , piersanti et al . when the h mass exceeds such a limit , the h - burning is efficiently activated at the base of the envelope . it drives the expansion of the envelope leading to the formation of a giant star rather than to a white dwarf . the cooling sequence obtained starting from the three pre - wd models illustrated in this subsection are shown in figure 4 . they are very similar , with a little difference ( less than 0.5 gyr ) at the faint end . in particular , the cooling of model b1 is a bit faster because the smaller he mantel . note that these three models reach the wd stage through a very different path and they may be considered as representative of the possible changes occurring during the progenitor evolution . in summary , the knowledge of the cooling time scale at the faint end of the wd sequence ( namely at log l / l@xmath0=-5.5 ) is limited by a 3 - 4% error as a consequence of the uncertainty on the leading parameters characterizing the pre - wd history . in many recent studies of cosmochronology ( salaris et al . 1997 , richer et al . 2000 ) wd models based on progenitors having solar chemical composition have been used . this is certainly adequate to study the age of the disk and its components ( open clusters ) . however , when the main goal are the oldest component of our galaxy , halo or globular cluster stars , low metallicity models are more appropriate . in this subsection we would try to evaluate the influence on the cooling sequence of the progenitor chemical composition . for this reason we have computed an additional model producing a wd having m=0.6 m@xmath0 , but with a solar metallicity , namely z=0.02 . the properties of this model are reported in the fourth row of table 1 ( model b3 ) . a comparison with model b ( z=0.0001 ) reveals only small differences . as already discussed by dominguez et al . ( 2001 ) , the central carbon abundance is slightly lower at large metallicity ; they found a maximum 10% variation when the metallicity is varied between 0.0001 and 0.02 . note that the extension of the innermost homogeneous region decreases by increasing the metallicity ( see the values reported in column 9 of table 1 ) . as expected , since the h - burning is more efficient at large metallicity ( because the larger amount of cno ) , the mass of the h - rich envelope is significantly reduced when the star attains the cooling sequence . in spite of this difference , at the faint end model b3 is only 2% younger than model b. the cooling sequences obtained for the two metallicities are compared in figure 5 . the amounts of carbon and oxygen left in the core of a wd have a great influence on its cooling rate ( dantona & mazzitelli 1990 , salaris et al . 1997 ) and plays an important role in determining the observable outcome of a type ia supernova ( dominguez et al . the chemistry of a wd is determined by the competition of the two major nuclear reactions powering the he burning , namely the 3@xmath62 ( the carbon producer ) and @xmath16c@xmath17o ( the carbon destroyer ) . as deeply discussed by imbriani et al . ( 2001 ) , the final c / o not only depends on the rate of these two reactions , but it is significantly influenced by the efficiency of the convective mixing operating near the center of an he burning star . the pre - wd models discussed so far , has been obtained by neglecting any source of extra - mixing , like mechanical overshoot , possibly occurring at the external edge of the convective core . we have also suppressed the so called breathing pulses , which take place when the central mass fraction of he is reduced down to @xmath630.1 ( see castellani et al . 1985 and references therein ) . as shown by imbriani et al . ( 2001 ) , the amount of carbon left in the center is almost insensitive to the occurrence of the convective overshoot , but it is strongly influenced by any additional mixing occurring in the final part of the he - burning , when the carbon previously accumulated by the 3@xmath62 is efficiently converted into oxygen via the @xmath16c@xmath17o reaction . this is exactly the job done by breathing pulses ( bps ) . in this subsection we study the combined effects of a change of the @xmath16c@xmath17o rate and the inclusion of bps . models b to b3 have been obtained by using a high rate for the @xmath16c@xmath17o reaction ( namely , the one reported by caughlan et al . 1985 , hereinafter c85 ) . as it is well known , the adoption of a low rate leads to a significantly larger central abundance of carbon . model b - low has been obtained by adopting the rate reported by caughlan & fowler ( 1988 , hereinafter cf88 ) , which is about 0.4 times the rate of c85 at the relevant temperatures ( around @xmath64 k ) . recent experimental determinations of this rate ( buchmann 1996 and 1997 , kunz et al . 2001 , 2002 ) seem to confirm that the actual value should be intermediate between cf88 and c85 . then , we have also repeated these two calculations by including the effect of the breathing pulses ( model b - bp and b - low - bp ) . the internal chemical profiles of models b - low and b - low - bp are compared in figure 6 , while the influence on the cooling sequence is illustrated in figure 7 . in synthesis , the whole uncertainty ( convection plus nuclear reaction rate ) due to the he - burning phase induces a 9% uncertainty in the estimated cooling time at the faint end of the wd sequence ( see the values reported in the last column of table 1 ) . it is worth noting that this is only a lower limit of the uncertainty caused by the possible variation of the internal composition . in fact , as discussed by isern et al . ( 1997 ) and montgomery et al . ( 1999 ) , the delay of the cooling due to the chemical segregation upon crystallization is larger when @xmath65 . therefore , the occurrence of this phenomenon should hampers the differences between case b - low and case b - bp . a more quantitative estimation may be deduced from the results reported in table 2 of montgomery et al .. note that our b - low model have an internal stratification that roughly correspond to their 50:50 case , while our model b have a composition profile similar to the one of salaris et al . ( 1997 ) . then , according to montgomery et al . , element segregation should increase the age difference between b - low and b - bp of at least 0.6 gyr , thus bringing the whole uncertainty on the age at the faint end of the cooling sequence up to 13 - 14% . we have already discussed the existence of a maximum value for the mass of the h - rich envelope of a wd . however , the actual amount of h at the top of a wd could be substantially lower . in fact , stellar wind may continue during the post - agb phase so that the h - rich envelope could be strongly reduced and even completely lost . it is well known that in an important fraction ( about 10 - 20 @xmath66 ) of the whole wd population the h - rich envelope is practically absent ( db - wds ) . the minimum detectable amount of h is very low ( about @xmath67 m@xmath0 ) . the post - agb mass loss is the easiest explanation of the lack of h in the atmosphere of the db wd . an alternative possibility was advanced by iben & renzini ( 1983 ) . they suggested that the progenitors of the db could have suffered a last thermal pulse during the post - agb , when the h - burning shell is almost extinguished . in this condition , the convective shell generated by the thermal pulse could completely mix the he and the h - rich layers . then , the h dredged down at high temperature is rapidly consumed . we recall that a similar post - agb thermal pulse has been encountered in our model b1 , but no evidence of a deep h mixing has been found ( in this particular case ) . dantona & mazzitelli ( 1989 ) and tassoul et al . ( 1990 ) have investigated the wd properties by changing the amount of h in the envelope . however , due to the lack of reliable opacity and eos , they did nt extent the computation down to the faint end of the cooling sequence . then , we have computed two additional cooling sequences by artificially reducing the h - rich layer of model b , namely : @xmath53 m@xmath68 and @xmath69 for model b - lh and b - vlh , respectively ( row 5 and 6 of table 1 ) . the comparison between these two models and model b is shown in figure 8 . as expected , the cooling is faster in models with a smaller envelope . the age reduction at log l / l@xmath70 would be as large as 1 gyr if the residual mass of the h - rich envelope was of the order of @xmath67 m@xmath0 . in this section we address the uncertainty caused by some input physics directly affecting the wd evolution . degenerate electrons dominates the heat conduction in white dwarf interiors . two different tabulations of the thermal conductivity by electrons are commonly used in the computation of wd models , namely : the one presented by hubbard & lampe ( 1969 , hereinafter hl69 ) and those by itoh and coworkers ( itoh et al . 1983 , mitake et al . 1984 , itoh et al . 1993 , hereinafter i93 ) . hl69 considered partially degenerate electrons scattered by ions . they also included the effect of the electron - electron interaction , which is negligible in extreme degeneracy regime ( due to the pauli exclusion principle ) , but becomes important if degeneracy is not strong , as in the external region of a wd . the main limitation of this calculation is due to the assumption of non - relativistic electrons . in practice , the hubbard and lampe conductive opacities are only adequate for densities lower than @xmath71 g/@xmath72 . actually , a stronger limitation of this calculation is often ignored . in fact , in order to smooth out the discontinuity in the conductive opacity occurring at the liquid / solid phase transition , the conductive opacities were artificially and progressively increased for @xmath7310 . let us recall that at the end of the sixties it was believed that the phase transition from liquid to solid would occur for rather low values of the coulomb coupling parameter ( @xmath74 ) . this explains the choice of a particularly low @xmath25 to manage the discontinuity occurring at the phase transition . nevertheless , the most recent studies find that the crystallization begins at @xmath75 ( we adopt 175 ) , in any case well above @xmath76 . the work made by itoh and coworkers is a substantial improvement of the calculation of electron conduction . they treat inter - ionic correlations and electron - ion interactions by means of an improved theory , based on one - component monte carlo calculation . however , since they assume fully degenerate electrons , their opacities are only accurate for large value of the degeneracy parameter . at lower density ( or at larger temperature ) , i93 probably underestimate the effects of the electron - electron interactions . in figure 9 we illustrate the situation in the case of the core of our reference model ( b ) . the empty area in the left panel shows the allowed region in the case of hl69 , while the right panel refers to i93 . the two ( almost vertical ) series of arrows shows the evolution of the physical conditions at the center and at the external boundary of the core . this figure demonstrate that hl69 is not adequate to describe the heat conduction in the major part of the wd structure , even if the density is somewhat lower than the relativistic limit . on the contrary the condition of extreme degeneracy is well satisfied and the i93 assumption seems appropriated for the whole co core . concerning the more external layers , figure 10 illustrate the case for the he - rich mantel . as in figure 9 , the left and the right panel show the region of validity for hl69 and i93 , respectively . the two series of arrows marks the boundaries of the mantel . hl69 should be preferred in the first part of the cooling sequence , since electrons may be partially degenerate while @xmath25 is generally rather small . the situations change toward the faint end of the cooling sequence , during which the i93 should be more appropriate . note that there exist a thin region in the t-@xmath77 plane in which both calculation are , in principle , valid . it happens , for example in the case of an he - rich mixture , for @xmath78 . the thick line in figure 10 marks this condition . however , a comparison along the @xmath79=0.1 line shows that the hl69 opacities are generally larger ( 20 - 40 % ) than the i93 ones . such a mismatch reveals a more profound difference between the two calculations , which can not be easily understood . in figure 11 we report the cooling sequences obtained by using the hl69 and i93 ( case b - hl69 and case b , respectively ) . in the former case , we have generated the conductive opacities by means of the original code provided by hubbard . as expected , the lower cooling rate obtained with hubbard & lampe reflects the larger value of the conductive opacities . note that the major discrepancies between the two cooling sequences occurs at about log(l / l@xmath80=-4 , exactly when electron conduction becomes the dominant mechanism of energy transportation at the base of the convective envelope . as it is well known ( see e.g. tassoul et al . 1990 ) , this occurrence produces a temporary reduction of the cooling rate . since many authors use a combination of the hl69 and i93 , in figure 11 we have also reported an additional model obtained by using i93 for @xmath81 and hl69 elsewhere ( case b - comb ) . this test shows that the discrepancy mainly concerns the conductive opacity in the weakly - degenerate regime . in conclusion , no ones of the prescriptions widely adopted in the computation of models of wds appear adequate to describe the whole structure of these cool stars . unfortunately , the substantial disagreement , at temperature and density for which both computations should be correct , suggests extreme caution in using combined tables . since the differences of these two theoretical prescriptions imply a 17% variation in the estimated wd age ( at the faint end of the cooling sequence , see the last column in table 1 ) , we conclude that the calculation of the conductive opacities ( especially in partially degenerate regime ) deserves much attention in view of a reliable calibration of the wd cooling time . in this context a recent paper by potekhin et al . ( 1999 , see also potekhin 1999 ) address this problem , extending the computation of electron conductivity to the weakly - degenerate regime . other improvements , as , in particular , refined ion structure factors and multi - phonon scattering , have been included . the conductive opacity obtained by potekhin and coworkers are intermediate between those of hl69 and i93 . nevertheless , as shown in figure 11 , the resulting cooling sequence does not substantially differs from the one obtained by using the i93 . on the contrary , the cooling time is significantly smaller then those obtained in the model b - hl69 and b - comb . some properties of the model obtained by means of the electron conductivity calculated by potekhin et al . ( 1999 ) are reported in table 1 ( model b - pot ) . chabrier et al . 2000 compare wd cooling sequences obtained by changing the prescriptions for the contribution of the ion - electron interactions to the free energy . in particular they found a significant decrease of the cooling time ( about 10% ) when the prescription of potekhin & chabrier ( 2000 ) are used in the solid phase instead of the result given by yakovlev & shalybkov ( 1989 ) . as pointed out by potekhin and chabrier , the ion - electron interactions are generally negligible , except in the solid state at very high density . for carbon the ion - electron contribution to the heat capacity becomes very important at t=@xmath82 k and @xmath83 g @xmath84 but at t=@xmath71 k , @xmath77 must be larger then @xmath85 g @xmath84 , a density never attained in co white dwarfs ( see figure 8 of potekhin & chabrier ) . we have investigated the overall effect of this contribution to the heat capacity by computing an additional model ( model b - noie ) by fully neglecting the ion - electron interactions . we recall that our `` standard '' computation are based on the potekhin & chabrier prescription . the result is shown in figure 12 ( see also table 1 ) . this test clearly demonstrates that the contribution of the ion - electron interaction derived by potekhin & chabrier produces a negligible effect on the cooling of co white dwarfs . many studies have shown that the time - scale of diffusion is short enough to sort the external layers of a wd , thus explaining the observed mono - elemental feature of wd spectra ( fontaine & michaud 1979 , muchmore 1984 , paquette et al . many authors have discussed the effects of diffusion on the wd structure and evolution ( see in particular iben & macdonald 1985 and 1986 , althaus et al . all the models presented here have been obtained by assuming a complete sorting of the h and he layers at the beginning of the cooling sequence ( @xmath86 ) . since we are mainly interested in quantify the overall variation of the cooling time , we have computed a new model in which the chemical stratification of both the envelope and the mantel is the one modeled by the nuclear burning and , eventually , by convection ( model b - nodif ) . we note that the adopted model atmospheres does not include the effect of metals . since the assumed metallicity is quite low ( z=0.0001 ) , the consequent inconsistency should not dramatically affect our result . the comparison between the properties of this model and those of the _ standard _ one ( model b ) is illustrated in figure 13 and in table 1 . in the standard case , the larger radiative opacity of the pure h induces the formation of a deeper convective envelope . as a consequence , the slow cooling phase , which takes place when the internal border of the convective layer reaches the region dominated by electron conduction , is anticipated . in conclusion , at the faint end of the cooling sequence , the model in which the effect of microscopic diffusion are neglected is a 11% younger than the standard model . note that this is the younger model we found . as it is well known the cooling time depends , among the other things , on the wd mass . figure 14 shows models in the mass range 0.5 - 0.9 @xmath87 computed under the same prescriptions adopted for the case b. in table 2 we have summarized the properties of these models . isochrones and luminosity functions based on these models are available on the web . in the last decade cosmochronology based on wds has become a promising tool of scientific inquiry thanks to both the availability of a large amount of high quality data for cool and faint wds and the improvement of the understanding of the high density plasma physics . however , as demonstrated by the rather large discrepancies among the recently published theoretical cooling sequences , a firm calibration of the age - luminosity relation is still not available . we have analyzed some of the main sources of uncertainty affecting the theoretical cooling time . physical and chemical parameters characterizing the white dwarfs and the progenitors evolutions have been revised . concerning progenitors , we found that the larger uncertainty is due to the combined action of convective mixing a nuclear reactions operating during the central he - burning phase . both these processes are largely affected by theoretical and experimental uncertainties . they determine the amount of c ( and o ) left in the core of the wd and , in turn , have a great influence on the predicted cooling rate . a conservative analysis allow us to conclude that the overall impact of the uncertainties due to the progenitors evolution on the estimated wd ages at logl / l@xmath88 is of the order of 2 gyr . we hope that this uncertainty will be significantly reduced in the next future thanks to the renewed effort of nuclear physicists in measuring the @xmath16c@xmath17o reaction rate at low energy ( see gialanella et al . 2001 and kunz et al . 2001 ) . concerning wd physics we emphasize the relevance of a reliable description of the electron conductivity which is the main mechanism of energy transport in wd interiors . actually , a consistent part of the large discrepancies found in the comparisons of published cooling sequences may be due to the adopted conductive opacity . the old hubbard and lampe ( 1969 ) conductive opacity are probably overestimated , even in the weakly degenerate regime . the latest computations by itoh and coworkers and potekhin et al . ( 1999 ) imply a substantial reduction of the age . at log@xmath89 , we obtain models @xmath90 gyr younger than that obtained with hl69 . let us conclude by noting that most of the uncertainties discussed in this paper mainly affect old wds , but have negligible influence on the age estimated for young wds , whose luminosity is larger than @xmath61 l@xmath0 . this implies that the presently available theoretical scenario may be safely used to estimate the age of young stellar systems as , for example , the intermediate age open clusters ( t@xmath483 gyr ) . we wish to thank p. bergeron for kindly providing us the models atmosphere and s. deglinnocenti , luciano piersanti and jordy isern for the many helpful comments . we thank also s. shore for the careful reading of the manuscript and the several useful discussions . we are deeply indebted to v. castellani for continuous encouragements and for a critical review of the manuscript . this work has been supported by the miur italian grant cofin2000 . alexander d.r . , & ferguson j.w . 1994 , , 437 , 879 althaus , l.g . , serenelli , a.m. , corsico , a.h . , & benvenuto , o.g . 2002 , , 330 , 685 angulo , c. et al . 1999 , a , 656 , 3 benvenuto , o.g . , & althaus , l.g . 1999 , , 303 , 30 bergeron , p. 2000 , private comunication bergeron , p. , ruiz , m.t . , & leggett , s.k . 1997 , , 108 , 339 bergeron , p. , saumon , d. , & wesemael , f. 1995 , , 443 , 764 bergeron , p. , wesemael , f. , & beauchamp , a. 1995 , pasp , 107 , 1047 bergeron , p. , wesemael , f. , & fontaine , g. 1991 , , 367 , 253 bradley , p.a . , & winget , d.e . 1991 , , 75 , 463 bradley , p.a . , winget , d.e . , & wood , m.a . 1992 , , 391 , l33 buchmann , l. 1996 , , 468l , 127 buchmann , l. 1997 , , 479l , 153 cassisi , s. , castellani , v. , & tornamb , a. 1996 , , 459 , 298 castellani , m. , limongi , m. , & tornamb , a. 1995 , , 450 , 275 castellani , v. , chieffi , a. , tornamb , a. , & pulone , l. 1985 , , 296 , 204 castellani , v. , deglinnocenti , s. , & romaniello , m. 1994 , , 423 , 266 caughlan , g.r . , & fowler , w.a . 1988 , atomic data nucl . data tables , 40 , 283 caughlan , g.r . , fowler , w.a . , harris , m.j . , & zimmermann , b.a . 1985 , atomic data nucl . data tables , 32 , 197 chabrier , g. , ashcroft , n.w . , & de witt , h.e . 1992 , nat . , 360 , 48 chabrier , g. , brassard , p. , fontaine , g. , & saumon , d. 2000 , , 543 , 216 chieffi a. , & straniero o. 1989 , , 71 , 47 clemens , j.c . 1993 , baltic . astron . , 2 , 407 cohen , e.g.d . , & murphy , t.j . 1969 , fluids , 12 , 1404 dantona , f. , & mazzitelli , i. 1989 , , 347 , 934 dantona , f. , & mazzitelli , i. 1990 , ara&a , 28 , 139 de witt , h.e . , graboske , h.c . , & cooper , m.s . 1973 , , 181 , 439 de witt , h. , & slattery , w. 1999 , contrib . plasma phys . , 39 , 97 dominguez , i. , chieffi , a. , limongi , m. , & straniero , o. 1999 , , 524 , 226 dominguez , i. , hflich , p. , & straniero , o. 2001 , , 557 , 279 dominguez , i. , straniero , o. , tornamb , a. , & isern , j. 1996 , , 472 , 783 dominguez , i. , tornamb , a. , & isern , j. 1993 , , 419 , 268 farouki , r.t . , & hamaguchi , s. 1993 , phys . e. , 47 , 4330 fontaine , g. , & michaud , g. 1979 , , 231 , 826 fujimoto , m.y . 1982a , , 257 , 767 fujimoto , m.y . 1982b , , 257 , 752 garcia - berro , e. , ritossa , c. , & iben , i.jr . 1997 , , 485 , 765 gialanella et al . 2001 , eur . j.a . , 11 , 357 graboske , h. , de witt , h. , grossman , a. , & cooper m. 1973 , , 181 , 457 haft , m. , raffelt , g. , & weiss a. 1994 , , 425 , 222 habing , h.j . 1996 , a&a rev , 7 , 97 hansen , b.m . 1998 , nat . , 394 , 860 hansen , b.m . 1999 , , 520 , 680 herwig , f. 1997 , in proceedings of the 10th european workshop on white dwarfs , held in blanes , spain , 17 - 21 june 1996 . j. isern , m. hernanz , and e. gracia - berro . publisher : dordrecht : kluwer academic publishers , 1997 hubbard , w.b . , & lampe , m. 1969 , , 18 , 297 iben , i.j . , 1972 , , 178 , 433 iben , jr.i . 1982 , , 259 , 244 iben , jr.i . , & macdonald , j. 1985 , , 296 , 540 iben , jr.i . , & macdonald , j. 1986 , , 301 , 164 iben , jr.i . , & renzini , a. 1983 , ara&a , 21 , 271 iben , jr.i . , & tutukov , a.v . 1986 , , 311 , 742 iglesias , a. , & rogers , f.j . 1996 , , 464 , 943 imbriani , g. , limongi , m. , gialanella , l. , terrasi , f. , straniero , o. , & chieffi , a. 2001 , , 558 , 903 isern , j. , garcia - berro , e. , hernanz , m. , & chabrier , g. 2000 , , 528 , 397 isern , j. , mochkovitch , r. , garcia - berro , e. , & hernanz , m. 1997 , , 485 , 308 itoh , n. , adachi , t. , nakagawa , m. , kohyama , y. , & munakata , h. 1989 , , 339 , 354 itoh n. , hayashi , h. , & kohyama , y. 1993 , , 418 , 405 itoh n. , mitake s. , iyetomi h. , & ichimaru s. , 1983 , , 273 , 774 itoh , n. , totsuji , h. , & ichimaru , s. 1977 , , 218 , 477 itoh , n. , totsuji , h. , ichimaru , s. , & de witt , h. 1979 , , 234 , 1079 kippenhahn , r. , kohl , k. , & weigert , a. 1967 , zeitschrift fur astrophysik , 66 , 58 kippenhahn , r. , thomas , h.c . , & weigert , a. 1968 , zeitschrift fur astrophysik , 69 , 265 kunz r. , jaeger f.m . , mayer a. , & hammer j.w . 2001 , phys . lett . , 86 , 3244 kunz r. , jaeger f.m . , mayer a. , hammer j.w . , staudt , g. , harissopulos , s. , & paradellis , t. 2002 , , 567 , 643 lamb , d. q. , & van horn , h. m. 1975 , , 200 , 306 landau , l.d . , & lifshitz , e.m . 1969 , `` statistical physics '' , pergamon , oxford livne , e. , & tuchman , y. 1988 , , 332 , 271 mazzitelli , i. 1994 , `` the equation of state in astrophysics '' , proceedings of iau colloquium no . 147 , saint - malo , france , 14 - 18 june 1993 , cambridge : cambridge university press , 1994 , ed . g. chabrier and e. schatzman , p.144 mestel , l. 1952 , , 112 , 583 metcalfe , t.s . , winget , d.e . , & charbonneau , p. 2001 , , 557 , 1021 mitake , s. , ichimaru , s. , & itoh , n. 1984 , , 277 , 375 montgomery , m.h . , klumpe , e.w . , winget , d. e. , & wood , m.a . 1999 , , 525 , 482 muchmore , d. 1984 , , 278 , 769 paquette , c. , pelletier , c. , fontaine , g. , & michaud , g. 1986 , , 61 , 197 paresce , f. , de marchi , g. , & romaniello , m. 1995 , , 440 , 216 piersanti , l. , cassisi , s. , iben , i.jr . , & tornamb , a. 2000 , , 535 , 932 potekhin , a.y . 1999 , a&a , 351 , 787 potekhin , a.y . , baiko , d.a . , haensel , p. , & yakovlev , d.g . 1999 , a&a , 346 , 345 potekhin , a.y . , & chabrier , g. 2000 , phys . e. , 62 , 8554 richer , h.b . 1997 , , 484 , 741 richer , h.b . 2000 , , 529 , 318 ritossa , c. , garcia - berro , e. , & iben , jr.i 1996 , , 460 , 489 ritossa , c. , garcia - berro , e. , & iben , jr.i 1999 , , 515 , 381 salaris , m. , dominguez , i. , garcia - berro , e. , hernanz , m. , isern , j. , & mochkovitch , r. 1997 , , 486 , 413 salaris , m. , garcia - berro , e. , hernanz , m. , isern , j. , & saumon , d. 2000 , , 544 , 1036 salpeter , e.e . 1961 , , 134 , 669 saumon , d. , chabrier , g. , & van horn , h.m . 1995 , , 99 , 713 saumon , d. , & jacobson , s.b . 1999 , , 511 , l107 segretain , l. , chabrier , g. , hernanz , m. , garcia - berro , e. , isern , j. , & mochkovitch , r. 1994 , , 434 , 641 stolzmann , w. , & blocker , t. 2000 , a&a , 361 , 115 straniero , o. 1988 , a&a , 76 , 157 straniero , o. , chieffi , a. , & limongi , m. 1997 , , 490 , 425 straniero , o. , chieffi , a. , limongi , m. , busso , m. , gallino , r. , & arlandini , c. 1997 , , 478 , 332 tanaka , s. , mitake , s. , & ichimaru , s. 1985 , phys . a , 32 , 1896 tassoul , m. , fontaine , g. , & winget , d.e . 1990 , , 72 , 335 von hippel , t. , gilmore , g. , & jones , d.h.p . 1995 , , 273l , 39 von hippel , t. , & gilmore , g. 2000 , , 120 , 1384 weidemann , v. 1987 , a&a , 188 , 74 weidemann , v. 2000 , a&a , 363 , 647 weidemann , v. , & koester , d. 1983 , a&a , 121 , 77 wood , m.a . 1995 , in `` white dwarfs '' , springer verlag , ed : d. koester & k. werner , 41 yakovlev , d. , & shalybkov , d. 1989 , astron . space phys . , 7 , 311 cccccccccc b & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & @xmath93 + b1 & 0.600 & 0.0001 & 0.580 & 0.600 & @xmath94 & @xmath95 & 0.258 & 0.542 & @xmath96 + b2 & 0.640 & 0.0001 & 0.515 & 0.601 & @xmath97 & @xmath98 & 0.249 & 0.516 & @xmath99 + b3 & 0.600 & 0.0200 & 0.599 & 0.600 & @xmath100 & @xmath101 & 0.248 & 0.509 & @xmath102 + b - lh & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath103 & 0.263 & 0.573 & @xmath104 + b - vlh & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath105 & 0.263 & 0.573 & @xmath106 + b - bp & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath107 & @xmath108 & 0.151 & 0.316 & @xmath109 + b - low & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath110 & @xmath111 & 0.511 & 0.483 & @xmath112 + b - low - bp & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath113 & @xmath114 & 0.381 & 0.434 & @xmath115 + b - hl69 & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & @xmath116 + b - comb & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & @xmath117 + b - pot & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & @xmath118 + b - noie & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & @xmath119 + b - nodif & 0.600 & 0.0001 & 0.599 & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & @xmath120 + ccccccc a & 0.500 & @xmath121 & @xmath122 & 0.221 & 0.571 & 15.00 + b & 0.600 & @xmath91 & @xmath92 & 0.263 & 0.573 & 14.20 + c & 0.700 & @xmath123 & @xmath124 & 0.294 & 0.560 & 12.90 + d & 0.800 & @xmath125 & @xmath126 & 0.302 & 0.529 & 12.31 + e & 0.900 & @xmath127 & @xmath128 & 0.309 & 0.492 & 11.32 +

in this paper we address the realization and identification problem for a subclass of piecewise - affine hybrid systems . * contribution of the paper * we define the class of _ piecewise - linear systems _ ( abbreviated by * pwl * ) . * pwl*s are a subclass of piecewise - affine hybrid systems . the continuous dynamics of * pwl * is determined by a finite collection of affine subsystems . however , in contrast to traditional piecewise - affine systems , we allow any change of the continuous state during a discrete - state transition , as long as the new state belongs to the set of designated initial states of the affine subsystem associated with the new discrete state . in addition , we do not impose any specific mechanism for triggering discrete - state transitions . we formulate the realization problem for this system class and partially solve it by providing necessary and sufficient conditions for existence of a realization . we also present conditions for minimality . we show that the outputs of any * pwl * can also be described by a switched ar model . the main conclusion is that any system can be transformed to a minimal system _ with one discrete state _ while preserving input - output behavior . this means that without further restrictions , the identification problem for such systems is not necessarily interesting . we discuss a number of restrictions on the system structure which avoid this problem . note that the conclusion above is not valid for other classes of hybrid systems . for hybrid systems from @xcite , there examples of input - output maps which provenly can not be realized by systems with one discrete state . in addition , we present an identification algorithm for systems with full observations . this algorithm is illustrated by example of physical and biological relevance . * motivation * the motivation for studying realization theory for * pwl*s is that it provides the theoretical foundations for systems identification . the motivation for investigating identification of * pwl*s is the following . * first , several systems of interest can be modeled by * pwl*s and * pwl*s tend to be convenient for analysis . however the parameters of * pwl * models are often not directly available and hence they have to be estimated from measurements . * the problem of estimating the network dynamics of complex biological systems can be reduced to the identification problem for * pwl*s . * identification of * pwl*s and biological networks * below we will elaborate on the relationship between identification of * pwl*s and the estimation of dynamic interactions of complex dynamical systems . numerous vital processes in nature involve complex signaling networks , varying from gene - protein interaction networks @xcite and complex communication in microbes @xcite to the synchronization in the heart of higher animals @xcite . in most cases , these signaling networks do not have a fixed topology , but vary their structure according to their internal states and certain external conditions @xcite . this flexibility allows the organism to exhibit complex behavior and flexible responses to various environmental conditions . the underlying biological problem is to describe these flexible dynamical networks , based on ( partial ) observations of their internal states . examples of potential observations are time series of gene expressions , protein densities , oxygen stress and sugar concentrations . in almost all cases , there is no explicit mathematical model available that even remotely describes the network dynamics . this is caused by the sheer size of the problem ; the many thousands of genes , rnas , proteins and other types of molecules involved , the complexity and idiosyncrasy of their interactions , and the large amount of noise in biological systems . basically , many agents involved in the network are unknown let alone their mechanism of interaction . mathematically , the processes above can be viewed as nonlinear dynamical systems with unknown parameters . such systems often have several equilibria , and hence their behavior can be seen as a mixture of certain affine systems , where each affine system is obtained by linearizing the original system around one of the equilibria . that is , such systems can be approximated by a * pwl*. in addition to their simplicity , the advantage of * pwl * approximations is that they neatly capture the interaction among various state components around various equilibria . more precisely , interaction between the @xmath0th and @xmath1th state components can be viewed as the property that the @xmath2th entry of the jacobian of the system around equilibrium is non - zero . since we are often interested exactly in interactions rather than the detailed dynamics , it makes sense to recast the problem of identifying such interactions into the problem of identifying * pwl*s approximations . while the paper does not directly address identification of interactions , we believe that the presented results represent a step towards the solution of that problem . * related work * identification of piecewise - linear ( -affine ) hybrid systems has been subject of intensive research , @xcite . realization theory for hybrid systems was investigated in @xcite . the application of hybrid systems to modelling complex dynamical systems , originating from biology , and their identification has been a subject of intensive research @xcite . the results of this paper are new , to the best of our knowledge . what distinguishes the contribution of this paper from the existing work is * ( a ) * the system class considered , in particular , the emphasis on continuous - time systems , * ( b ) * the emphasis on realization theory and algorithms , * ( c ) * the details of the identification algorithm . * outline of the paper * [ sect : pwl ] presents the formal definition of the system class of interest . [ sect : approx ] discusses the relationship between nonlinear systems with complex dynamics and piecewise - affine hybrid systems . [ sect : real ] presents the results on realization theory , and [ sect : ident ] presents the identification algorithm and the results of the numerical experiments . * notation * we use the standard notation . we denote by @xmath3 the time - axis . we denote by @xmath4 the @xmath5 identity matrix . we denote the set of natural numbers including zero by @xmath6 . the aim of the section is to define the class of piecewise - linear systems formally . a _ piecewise - linear system _ ( abbreviated as pwl ) is a dynamical system determined by @xmath7 here @xmath8 is the finite set of discrete modes , @xmath9 is the output space , @xmath10 is the state - space of the system in mode @xmath11 , for each @xmath12 @xmath13 , @xmath14 , @xmath15 , and @xmath16 are the parameters of the affine system in mode @xmath11 , @xmath17 is the set of initial states of the affine system in mode @xmath11 . the _ state space _ @xmath18 of @xmath19 is @xmath20 we call @xmath19 _ linear _ , if @xmath21 and @xmath22 for all @xmath11 , otherwise @xmath19 is called _ affine_. we will use the following short - hand notation @xmath23 informally , the evolution of @xmath19 takes place as follows . as long as the value of the discrete state @xmath24 at time @xmath25 does not change , the continuous state and the continuous output an time @xmath25 change according to the affine system @xmath26 and @xmath27 . the discrete state can change at any time , however , we do not allow consecutive changes of discrete states immediately one after the other . if the discrete state changes to @xmath28 at time @xmath25 , then the new continuous state should satisfy @xmath29 , i.e. it should belong to the set of the designated initial states of the new discrete state @xmath28 . note that we do not specify the mechanism which triggers the change of discrete state . we also do not specify the initial discrete state , it is chosen by an unspecified mechanism . hence , the _ description above allows for several state- and output - trajectories . _ note that the definition of * pwl*s presented above differs from the one used in the literature . in the standard definition it is assumed that after a discrete - state transition , the new continuous state depends on the previous one . in contrast , here we only require that the new continuous state belongs to the designated set of initial states of the affine system associated with the new discrete state . this implies that _ the behavior of the system in discrete state @xmath30 depends only on the affine system associated with that discrete state and it does not depend on the discrete modes which were visited in the past . _ another consequence of the definition is that the model is not predictive . that is , the knowledge of the system parameters and the inputs does not uniquely determine the state- and output - trajectory of the system . in order to define the evolution of * pwl*s formally , we need to introduce the following notation and terminology . a time event sequence is a strictly monotone sequence @xmath31 such that @xmath32 , @xmath33 and for all @xmath34 , @xmath35 . if @xmath36 then we require that @xmath37 . if @xmath38 , then by convention @xmath39 . the role of time event sequences is to formalize the time instances at which discrete events occur . the restrictions formulated in the definition imply that the set of switching times does not have an accumulation point , i.e. no zeno - behavior can take place . a state - trajectory of @xmath19 is a map @xmath40 such that there exists a time event sequence @xmath41 and a sequence of discrete modes @xmath42 such that for all @xmath43 , @xmath44 , it holds that for all @xmath45 , @xmath46 and @xmath47 the time event sequence @xmath48 is called the sequence of _ switching times _ of the state - trajectory @xmath49 . we _ denote by @xmath50 the set of all state - trajectories of @xmath19_. an output - trajectory of @xmath19 is a map @xmath51 such that the following holds . there exists a state - trajectory @xmath49 of @xmath19 , such that @xmath52 for all @xmath53 . here , @xmath54 is the _ readout - map _ of @xmath19 , defined as @xmath55 we _ denote by @xmath56 the set of all output - trajectories of @xmath19_. in the sequel , unless stated otherwise , _ @xmath57 denotes a function @xmath58_. the definition above implies that the external behavior of a * pwl * is exactly a function of this type . the function @xmath57 is said to be realized by * pwl * @xmath19 , if @xmath59 is an output - trajectory of @xmath19 , i.e. if @xmath60 . in this case @xmath19 is called a realization of @xmath57 . the dimension of a * pwl * @xmath19 is a tuple @xmath61 , where @xmath62 if @xmath19 is linear , and @xmath63 , if @xmath19 is affine . that is , the first element of @xmath64 is the number of discrete states , the second element is the number of continuous state components . the additional dimension in the case of affine * pwl*s stems from the need to store the vectors @xmath65 , @xmath66 , @xmath11 . we use the following partial order relation on @xmath67 . we say that @xmath68 is smaller than or equal @xmath69 , denoted by @xmath70 , if @xmath71 and @xmath72 . note that the order relation @xmath73 in @xmath67 is indeed a partial order , it is not possible to compare all elements of @xmath67 . a * pwl * @xmath19 is said to be a minimal realization of @xmath57 , if for any * pwl * @xmath74 which is a realization of @xmath57 , @xmath75 . the definition above implies that if @xmath19 is minimal , then it is possible to compare its dimension to the dimension of any other realization of @xmath57 . since we work with a partial order on dimensions , the existence of a minimal system is not at all obvious . find conditions for existence of a * pwl * realization of @xmath57 . characterize minimal * pwl*s realizations of @xmath57 . find algorithms for computing a * pwl*s realization from finite data . as it was indicated before , our motivation for studying the realization problem for * pwl*s is to lay the theoretical foundations for identification of * pwl*s . we present below the formulation of the identification problem . assume that the value of @xmath57 and its derivatives up to order @xmath76 , @xmath77 is measured at time instances @xmath78 . based on the ( possibly noisy ) data @xmath79 , find a * pwl * realization of @xmath57 . our main motivation for studying identification of * pwl*s is the following . the problem of identifying interaction networks of complex dynamical systems can be reduced to the identification problem of * pwl*s . indeed , consider a partially observed non - linear system of the form @xmath80 where @xmath81 and @xmath82 are sufficiently smooth functions . we are interested in the dynamics of the network structure of . by _ the network structure of at time @xmath25 _ we mean the directed graph with nodes numbered by @xmath83 and where an edge goes from node @xmath0 to node @xmath1 if the @xmath0th component of @xmath84 influences the @xmath1th component of @xmath85 . as the values of @xmath84 changes , so does @xmath86 . hence the network structure described above depends on time . the basic question of interest is how to derive the dynamics of the interaction network based on the observed output @xmath87 . one obvious approach is to estimate @xmath57 and @xmath88 from @xmath87 , but this is feasible only if special restrictions are put on @xmath57 and @xmath88 . notice , however , that if we look at local linearizations of @xmath57 around an equilibrium and the interactions are strong enough , then @xmath89 influences @xmath90 , if the @xmath2th entry of the jacobian of @xmath57 is non - zero . prompted by the considerations above we propose the following approach to modelling and identification of the network structure of systems of the form . we assume that has finitely many equilibrium points . we assume that under influence of noise the system jumps between neighborhoods of these equilibrium points . we assume that the transition from the neighborhood of one equilibrium point to the neighborhood of another equilibrium point takes place very fast . hence , the behavior of the system during the transition can be ignored . we associate a * pwl * @xmath19 of the form with as follows . let @xmath91 be the number of equilibrium points . for each @xmath11 , let @xmath92 be the corresponding equilibrium point and define @xmath93 here @xmath94 and @xmath95 denote the jacobian of @xmath57 and @xmath88 respectively . let @xmath96 be a suitably chosen neighborhood of @xmath92 . it then follows that if the state @xmath84 of is close enough to @xmath92 , we can approximate as follows @xmath97 as it was noted above , the relevant information regarding the interactions of state variables is already included in @xmath98 . hence , _ for the purposes of network reconstructions , identification of the * pwl*s approximation is sufficient_. below we present some basic results on realization theory of * pwl*s . first we present a characterization of minimality , after that we present conditions for existence of a realization and a realization algorithm . throughout the section , _ @xmath57 denotes a function @xmath58_. in order to present our first result , we introduce the notion of a _ linear system with state - jumps_. a linear system with state jumps , abbreviated as ( * lssj * ) is a linear * pwl * @xmath19 of the form with one discrete state , i.e. @xmath99 . we will identify the * lssj * @xmath19 with the collection of data @xmath100 , where @xmath101 and @xmath102 . the reason we call a * pwl * with one discrete state a linear system with state jumps is that the system behaves as a linear system , with the exception that its state occasionally jumps back to one of the initial states . for * lssj*s , we can easily define the concepts of observability and span - reachability . a * lssj * @xmath103 is called _ observable _ , if @xmath104 is an observable pair , and @xmath19 is called _ span - reachable _ , if @xmath105 . [ lssj : min ] note that @xmath106 is an @xmath107 invariant subspace containing @xmath108 , hence by restricting @xmath107 and @xmath109 to @xmath110 we can transform @xmath19 to a span - reachable * lssj * @xmath111 while preserving output trajectories , i.e. @xmath112 . by applying linear observability reduction , we can transform the * lssj * @xmath111 to a * lssj * @xmath113 which is span - reachable , observable and has the same output trajectories as @xmath19 , i.e. @xmath114 . [ phas : min ] assume that @xmath57 admits a realization by a * pwl*. then @xmath57 admits a minimal * pwl * realization @xmath19 such that @xmath19 is a span - reachable and observable * lssj*. a * lssj * realization of @xmath57 is minimal if and only if it is span - reachable and observable . the theorem above says that in general , the external behavior of any * pwl*s can be represented by a linear system with several initial states . that is , without further restrictions , the realization and identification problems for * pwl*s are equivalent to that of * lssjs*. the proof of theorem [ phas : min ] relies on the following transformations . [ pahs2linpahs ] define the _ linear _ * pwls * @xmath115 associated with an affine @xmath19 as follows . @xmath116 where @xmath117 , @xmath118 , @xmath119 and @xmath120 the output trajectories of @xmath19 and @xmath115 coincide , i.e. @xmath121 , and @xmath122 . a state trajectory @xmath49 of @xmath115 is always a map of the form @xmath123 , @xmath53 such that @xmath124 , @xmath53 is a state - trajectory of @xmath19 . since @xmath125 , the statement of the proposition follows . let @xmath19 be a linear * pwl * of the form . define the * lssj * @xmath126 associated with @xmath19 as follows . @xmath127 where @xmath62 and @xmath128 the output trajectories of @xmath19 and @xmath126 are the same , i.e. @xmath129 , moreover , @xmath130 where @xmath131 . for each @xmath11 , let @xmath132 be the subset of @xmath133 of the form @xmath134 , @xmath135 . for each @xmath136 of the above form , let @xmath137 . it then follows that @xmath138 is a state - trajectory of @xmath126 if and only if @xmath139 is a state - trajectory of @xmath19 , where @xmath140 for some @xmath11 if and only if @xmath141 . since @xmath142 for all @xmath143 , the statement of the proposition follows . using the two transformations defined above , and remark [ lssj : min ] , we can present the proof of theorem [ phas : min ] . assume that @xmath19 is a * pwl * realization of @xmath57 . if @xmath19 is affine , then replace @xmath19 with the associated linear * pwl * @xmath115 , which is also a realization of @xmath57 . hence , we can assume that @xmath19 is already a linear * pwl*. construct then the * lssj * @xmath126 associated with @xmath19 and apply remark [ lssj : min ] to @xmath126 to obtain a span - reachable and observable * lssj * @xmath113 such that @xmath144 . it then follows that @xmath145 . hence , since @xmath19 was arbitrary , it is enough to look for minimal realizations among span - reachable and observable * lssj*s . notice that for any two * lssj*s @xmath146 and @xmath147 , either @xmath148 or @xmath149 , hence , among all the possible * lssj*s realizations of @xmath57 , there must exist a minimal one . the conditions for existence of a realization will be formulated using the rank of the hankel - matrix of @xmath57 . in order to define the hankel - matrix of @xmath57 , we have to define the notion of markov - parameters . to that end , we need the notion of _ piecewise - analytic functions_. the map @xmath57 is called _ piecewise - analytic _ , if there exist a finite or infinite number of time instances , @xmath150 , @xmath151 , @xmath152 , @xmath44 for some @xmath153 , such that the following holds . if @xmath154 , let @xmath155 . then we require that @xmath33 and @xmath156 and for each @xmath44 , @xmath152 , @xmath57 is analytic on @xmath157 , but @xmath57 is not analytic on any neighborhood of @xmath158 in @xmath159 . we call the points @xmath160 the points of _ non - analyticity_. we define the set @xmath161 of indices of points of non - analyticity . the intuition behind the definition is as follows . if @xmath57 has a realization by a * pwl * @xmath19 , then the only points where @xmath57 is not analytic are the points where the corresponding state - trajectory of @xmath19 switches from one discrete mode to another . in fact , one can show that there always exists a state - trajectory of @xmath19 which yields @xmath57 as output trajectory and which switches only at time instances at which @xmath57 is not analytic . hence , if @xmath57 has a realization by a * pwl * , then @xmath57 is piecewise - analytic and the points of non - analyticity tell us the switching times of a * pwl * realization of @xmath57 . in the sequel , _ @xmath57 is assumed to be piecewise - analytic_. assume that @xmath57 is piecewise - analytic and let @xmath160 be the points of non - analyticity of @xmath57 . for each @xmath162 , define the @xmath0th _ markov - parameter _ @xmath163 of @xmath57 as a sequence @xmath164 @xmath165 it is easy to see that the collection of markov - parameters @xmath166 determines the map @xmath57 uniquely . we use the markov - parameters to define the hankel - matrix of @xmath57 . we define the _ hankel - matrix _ @xmath167 of @xmath57 as the infinite matrix , rows of which are indexed by @xmath168 , and columns of which are indexed by @xmath169 . the entry of @xmath167 indexed by row index @xmath170 and by column index @xmath171 equals @xmath172_{(i , r),(j , l)}=(\mathbf{m}_{l}^f(i+j))_{r}\ ] ] where @xmath173 denotes the @xmath76th entry of markov - parameter @xmath174 . the rank of @xmath167 , denoted by @xmath175 , is the dimension of the linear space spanned by the columns of @xmath167 . [ theo : ex ] the map @xmath57 can be realized by a * pwl * if and only if it is piecewise - analytic and @xmath176 . moreover , a minimal * lssj * realization of @xmath57 of dimension @xmath177 can be constructed from @xmath167 . * only if * assume that @xmath57 can be realized by a * pwl*. then it can be realized by a * lssj * @xmath103 . then @xmath178 @xmath179 for some @xmath180 , i.e. @xmath181 are the markov - parameters of the linear system @xmath104 from some initial condition . it then follows from linear systems theory that @xmath182 . * if * assume that @xmath183 and fix a basis in the linear span of the columns of @xmath167 . define @xmath109 as the matrix in this basis of the linear map which maps a column to the vector formed by the rows of that column indexed by @xmath184 , in this order . define the matrix @xmath107 as the matrix in this basis of the linear map which maps the column indexed by @xmath171 to the column indexed by @xmath185 . finally , let @xmath108 be the set of coordinate vectors of the columns of @xmath167 indexed by @xmath186 , @xmath187 . then @xmath188 is a * lssj * realization of @xmath57 . if @xmath19 is another * lssj * realization of @xmath57 , then from the * only if * part it follows that @xmath189 , i.e. @xmath190 is a minimal realization of @xmath57 . in algorithm [ alg0 ] we present a kalman - ho - like realization algorithm for * lssj*s and hence * pwl*s . to this end , we need the following definition . [ def : part ] fix integers + @xmath191 . define the set @xmath192 define the submatrix @xmath193 of @xmath167 as the matrix which is formed by the intersection of the rows of @xmath167 indexed by the elements of @xmath194 and the columns of @xmath167 indexed by the elements of @xmath195 . compute the decomposition @xmath196 such that @xmath197 and @xmath198 and @xmath199 . define @xmath200 as the matrix formed by the columns of @xmath201 indexed by elements of @xmath202 . let @xmath203 the matrix such that the column of @xmath204 indexed by @xmath171 equals the column of @xmath201 indexed by @xmath185 . define @xmath205 as follows . * the @xmath0th row of @xmath109 equals the row of @xmath206 indexed by @xmath207 , @xmath208 . * the matrix @xmath107 is the solution of the equation @xmath209 * the set @xmath108 consists of the columns of @xmath201 indexed by indices of the form @xmath186 , @xmath210 . [ alg0:theo ] if @xmath211 , and @xmath212 , then for some @xmath213 , the * lssj * returned by algorithm [ alg0 ] is a minimal realization of @xmath57 . the conditions above hold if @xmath57 has a realization by a * lssj * of dimension at most @xmath214 and with a finite set of initial states . the results on realization theory of * pwl*s indicate that the realization and identification problems for * pwl*s are in general ill - posed in the following sense . realizability by a * pwl*s is equivalent to realizability by a minimal * pwl*s with one single discrete state . hence , there seems to be no intrinsic way to choose discrete states based on the data . since the original intention was use the combination of the dynamics in various discrete modes to explain complex dynamics , the above conclusion is an unpleasant one . note that the main problem is that one can increase the dimension of the continuous state - space to encode discrete states . one way to remedy this is to place restriction on the dimension . to this end , we introduce the notion of @xmath215 realization . a * pwl * @xmath19 is said to be a @xmath215 * pwl * if @xmath19 is of the form and @xmath216 and for all @xmath11 , @xmath217 . the map @xmath57 is said to admit a @xmath215 realization , if there exists a @xmath215 * pwl * which is a realization of @xmath57 . the motivation for using @xmath215 realizations is that by choosing an appropriate @xmath218 , we can make sure that the growth in the number of continuous states can not be used to replace discrete states . below we present conditions for existence of a @xmath215 realization of @xmath57 . to this end , we have to introduce the following concept . consider the partitioning @xmath219 of the set of @xmath220 . for each @xmath221 , let @xmath222 be the sub - matrix of hankel - matrix @xmath167 which is formed by the columns of @xmath167 indexed by indices of the form @xmath171 , @xmath223 and @xmath224 . [ theo : pos1 ] the map @xmath57 has a realization by a linear @xmath215 * pwl * if and only if there exists a partitioning @xmath225 of @xmath226 with @xmath227 such that for all @xmath228 , @xmath229 . if the latter condition holds , then @xmath19 can be computed from @xmath167 with @xmath8 and with @xmath230 for all @xmath11 . theorem [ theo : pos1 ] above says that @xmath57 can be realized by a @xmath215 realization if the columns of the hankel - matrix @xmath167 can be divided in @xmath231 clusters such that for each fixed @xmath232 , all columns indexed by @xmath233 end up in the same cluster . moreover , the linear span of the elements of each cluster should have dimension at most @xmath218 . in order to prove theorem [ theo : pos1 ] , we need the following transformation of a * lssj * to a @xmath215 realization . [ lss2pwls ] consider a partition @xmath234 of @xmath220 for some @xmath235 . consider an observable * lssj * @xmath236 such that @xmath237 is a realization of @xmath57 . then there exists a unique state trajectory @xmath49 of @xmath237 , such that the switching times of @xmath49 coincide with the points of non - analyticity @xmath238 of @xmath57 and @xmath239 for all @xmath53 . define the * pwl * realization @xmath240 of @xmath57 _ associated with @xmath237 and @xmath241 _ as follows . the * pwl * @xmath240 is of the form , such that @xmath242 , and * for all @xmath11 , let @xmath243 be the set of states @xmath244 such that for some @xmath224 , @xmath245 . define @xmath246 set @xmath247 and choose a basis of @xmath132 . * for each @xmath11 , define @xmath98 as the matrix in the basis of @xmath132 of the linear map @xmath248 obtained by restricting @xmath107 to @xmath132 . * let @xmath96 be the set of vector representations in the basis of @xmath132 of the elements of @xmath243 . * let @xmath249 be the matrix representation in the basis of @xmath132 of the linear map @xmath250 obtained by restricting @xmath109 to @xmath132 . if @xmath237 is a * lssj * realization of @xmath57 , then the * pwl * @xmath240 is a realization of @xmath57 . * only if * assume that @xmath19 is a * pwl * realization of @xmath57 of the form , such that @xmath216 and @xmath217 for all @xmath11 . then there exists a state trajectory @xmath49 of @xmath19 such that @xmath251 and the switching times of @xmath49 contain the points of non - analyticity @xmath238 of @xmath57 . define @xmath252 . it then follows that @xmath253 is a partitioning of @xmath220 . moreover , for any @xmath254 , @xmath255 is the markov - parameter of the linear system @xmath256 from some initial condition . hence , by using linear systems theory we can show that @xmath257 . * if * if the conditions of the theorem hold , then @xmath258 . construct the * lssj * realization @xmath190 of @xmath57 , as in the proof of theorem [ theo : ex ] . apply definition [ lss2pwls ] to obtain a linear * pwl * @xmath259 . then @xmath19 has @xmath227 discrete states . notice that the space @xmath132 from definition [ theo : pos1 ] is then isomorphic to the column space of @xmath222 , @xmath260 . hence @xmath261 . similarly to the general case , in algorithm [ alg05 ] we state an algorithm for computing a @xmath215 realization of @xmath57 . apply algorithm [ alg0 ] to @xmath262 and denote the result by @xmath263 . [ alg05:clust ] compute a partitioning @xmath225 of @xmath264 such that @xmath227 and for all @xmath228 , @xmath265 . here @xmath266 is the sub - matrix of @xmath193 formed by the columns @xmath171 where @xmath267 and @xmath268 . return the * pwl * @xmath269 . [ alg05:theo ] if @xmath57 has a @xmath215 realization of dimension at most @xmath270 , and and with a finite set of initial states , then for some @xmath213 , algorithm [ alg0 ] returns a minimal realization of @xmath57 . below we show that @xmath57 is realizable by a * pwl * if and only if @xmath57 satisfies a switched ar model . a _ switched ar system _ ( abbreviated as * sars * ) is a tuple @xmath271 where @xmath8 , @xmath272 , @xmath273 and @xmath274 for all @xmath275 . the function @xmath57 is said to satisfy the * sars * model @xmath276 , if the following holds . let @xmath238 be the points of non - analiticity of @xmath57 . for any @xmath162 , and for any @xmath277 , denote by @xmath278 the @xmath279th order right - hand derivative @xmath280 . then we require that for any @xmath281 there exist @xmath282 , such that @xmath283 , @xmath284 the set @xmath285 is called the set of discrete modes of @xmath276 . notice that if @xmath99 , then the * sars * @xmath276 corresponds to an ar system , and @xmath57 satisfies @xmath276 if it satisfies an ar equation . note , however , that in contrast to trajectories of an ar systems , @xmath57 is not smooth . [ sars : theo ] the function @xmath57 has a realization by a * pwl * with @xmath91 discrete states if and only if there exists a * sars * @xmath276 with @xmath91 discrete modes such that @xmath57 satisfies @xmath276 . * only if * let @xmath19 be a realization of @xmath57 . without loss of generality we can assume that @xmath19 is linear , it is of the form and @xmath286 for all @xmath11 . then for each @xmath162 , there exists @xmath282 such that on @xmath157 , @xmath57 is an output trajectory of the linear system @xmath287 . from classical theory we then know that there exists matrices @xmath288 such that holds . * if * assume @xmath57 satisfies a * sars * @xmath276 of the form . define the linear * pwl * @xmath289 of the form such that @xmath289 is a realization of @xmath57 as follows . the sets of discrete modes of @xmath289 and @xmath276 coincide , @xmath286 , @xmath290 and the matrices @xmath249 and @xmath291 , @xmath11 are the matrices of the linear system which corresponds to the ar @xmath292 . here we consider the linear system , state vector of which is the regressor @xmath293 . theorem [ phas : min ] and theorem [ sars : theo ] yield the following . the function @xmath57 has a realization by a * pwl * if and only if it satisfies an ar system , i.e. for some @xmath294 , @xmath295 , @xmath296 below we present an algorithm which computes a realization with full observations based on measurements at finitely many time instances . more precisely , the algorithm solves the following problem . [ ident : full_obs ] fix integers @xmath272 and @xmath295 . consider a piecewise - analytic function @xmath297 and a finite sequence of time instances @xmath298 and assume that @xmath299 are known , i.e. we know that value of @xmath57 and its derivative at time instances @xmath300 . find a * pwl * @xmath19 of the form with full - observations such that @xmath19 realizes @xmath57 , and @xmath301 , @xmath21 , and @xmath286 for all @xmath11 . * motivation of the identification problem * the motivation for considering the identification problem with full observations is the following . 1 . notice that the identification problem with full observations and the identification problem for * sars*s are equivalent . indeed , a * pwl * with full observation can be considered as a * sars*s with @xmath302 . conversely , for a * sars * @xmath276 , the associated * pwl * @xmath289 from the proof of theorem [ sars : theo ] can be viewed as a * pwl * with full observations , if the high - order derivatives of @xmath57 can be measured . hence , by theorem [ sars : theo ] , _ the identification problem for * pwl*s is equivalent to the identification problem for * pwl * with full observations _ , if we assume that high - order derivatives of the output can be measured too . the problem of identification with full observations is still a non - trivial problem . even in this case we have problems with identifiability , see example [ example1 ] . 3 . the solution of problem [ ident : full_obs ] enables us to prove experimentally the feasibility of approximating complex systems by * pwl*s . this is done by applying the identification algorithm which solves problem [ ident : full_obs ] to simulated trajectories of several well - known complex systems . [ example1 ] consider the * lssj*s @xmath303 and @xmath304 where @xmath305 is the identity matrix and @xmath306 and the remaining parameters are as follows ; @xmath307 and @xmath308 . it then follows that @xmath309 , @xmath53 can be realized both by @xmath146 and @xmath147 , but clearly @xmath310 . in other words , _ realizations of @xmath57 with full observation are not identifiable . _ the reason for this phenomenon is that the state of both @xmath146 and @xmath147 live in the subspace @xmath311 , @xmath312 , hence the difference in @xmath313 and @xmath314 is not visible . there is a wealth of results on identification of * sars * , mostly addressing the discrete - time siso case . many of these results , in particular , the algebraic approach @xcite could probably be adapted to the system class of this paper . investigating such adaptations remains a topic of future research . note that it is not at all clear that the convergence results for the discrete - time case have meaningful counterparts in the continuous - time case . in algorithm [ alg2 ] we present a solution for solving problem [ ident : full_obs ] . algorithm [ alg2 ] is an iteration consisting of the following steps . first , the algorithm starts with a random initialization of @xmath315 . subsequently , in each iteration step , the current estimates @xmath316 are updated by applying first the step [ alg : pwl-1 ] for estimating the updated weights with the fixed system parameters @xmath317 , and then applying step [ alg : pwl-2 ] with the previously updated weights @xmath318 for updating the estimates of the system parameters @xmath317 . the interpretation of the weights is as follows : @xmath319 is one if in @xmath158 the discrete state @xmath11 is active , and it is zero otherwise . the algorithm terminates when an absolute criterion @xmath320 falls below a pre - specified threshold @xmath321 . the iteration fails if after a pre - defined maximum number of iterations @xmath322 the criterion @xmath320 has not yet reached the lower threshold @xmath321 . unfortunately , the conditions under which the algorithm terminates and returns a correct realizations are not known yet . the criterion @xmath320 of the algorithm always decreases or stays the same during the iterations , and the same holds for the cost functions in step [ alg : pwl-1][alg : pwl-2 ] . hence , the criterion @xmath320 and the cost functions in step [ alg : pwl-1][alg : pwl-2 ] will converge to a fixed value . whether this value is a global optimum remains a topic of future research . note that the true parameters of a realization of @xmath57 render @xmath320 zero , but there might be many of them , see example [ example1 ] . initialize the estimates of @xmath323 , @xmath324 , @xmath325 . [ alg : pwl-1 ] solve the optimization for the weights @xmath326 \end{split}\ ] ] where for @xmath327 , @xmath328 , @xmath329 , @xmath11 , @xmath330 note that the optimization problem above is a linear programming problem , hence the optimal values of @xmath331 take values in the set @xmath332 . [ alg : pwl-2 ] using the weights @xmath331 recalculate the optimal values of @xmath333 , @xmath11 by solving the following minimization problem . @xmath334 the optimization problem is equivalent to the following linear least squares problem . collect the unknowns @xmath333 , @xmath11 into the matrix @xmath335 @xmath336 define the matrices @xmath337 and @xmath338 @xmath339 then the solution of can be read off from the entries of @xmath335 , where @xmath340 update @xmath341 using the new values . return @xmath19 of the form with @xmath286 , @xmath301 , @xmath21 and @xmath342 as computed in the last iteration , for all @xmath11 , and @xmath343 for all @xmath11 . we conducted several numerical experiments . they were all performed on a pc with intel pentium 4 cpu 2.8 ghz processor and 1.21 gb ram memory under windows xp , using matlab 7.5 including the optimization toolbox . in line with the definitions above , we use @xmath231 and @xmath218 to quantify the dimension of the * pwl * of interest and @xmath344 to denote the number of data points . dynamical data @xmath299 was sampled from a given dynamical model @xmath345 at regular time intervals @xmath346 , @xmath347 for some @xmath348 . several initial states were chosen randomly , and a sampled trajectory was obtained for each of them . we then concatenated these sampled trajectories into one time series . notice that concatenation of finite components of two output trajectories of a * pwl * is itself a finite component of a valid output trajectory . hence , the combined data can be viewed as originating from the measurements of one valid output trajectory of a * pwl*. gaussian noise with zero mean and variance @xmath349 was added to the obtained data . finally , algorithm [ alg2 ] was applied to the resulting data . in the numerical experiments we first tested the performance of the algorithm on artificial * pwl * systems . furthermore , we studied the algorithm on lorenz systems and on the tyson - novak model for the cell cycle of budding yeast . we generate an artificial * pwl * system containing @xmath231 affine subsystems , each of the same dimension @xmath350 . the state switching is obtained by partitioning the state space in @xmath231 subsets @xmath351 . the switching is generated by using the following switching law : the discrete mode @xmath0 is active , if the continuous state belongs to @xmath352 . in the absence of noise , and if the subsystems are sufficiently sampled , the estimation algorithm is usually able to find the system parameters , though occasionally it gets stuck in a local minimum . the amount of empirical data required to reconstruct the system parameters , within the bounds set by the noise , increases proportionally with @xmath218 and @xmath231 . figure [ n1 ] below shows the result of the algorithm for @xmath353 , @xmath354 , @xmath355 data points . the voronoi partitioning of state space of the original dataset ( black ) and the reconstructed dataset ( red ) are identical . the reconstructed system parameters and the original system parameters are identical as well . ] further tests reveal that algorithm [ alg2 ] is able to find the correct parameters even if the regions @xmath351 are disconnected . when applying zero - mean gaussian noise @xmath356 , the accuracy of the reconstruction decreases as @xmath357 increases . figure [ n3 ] below shows the result on a dataset of dimension @xmath353 with @xmath354 subsystems and @xmath355 data points with a snr ( signal - to - noise ratio ) of @xmath358 . two subsystems could not be reconstructed , but the correlation between the three reconstructed systems with the best fitting original subsystems is @xmath359 ] we applied algorithm [ alg2 ] to lorenz systems @xcite . this example provides insight how the identification algorithm performs on systems that are not piecewise affine by nature . the lorenz attractor can be visualized approximately as consisting of two linear subsystems ` left ' and ` right ' , with two different central attractors . a third region can be defined that involves the transition between the left and right subsystem . in figure [ n4 ] the application of algorithm [ alg2 ] to lorenz systems with @xmath360 data points is presented for @xmath361 . for @xmath362 the two main parts of the lorenz attractor are found . for @xmath363 also the transition region from the left to the right attractor is found . applying more than @xmath364 subsystems does not improve the performance . , title="fig : " ] , title="fig : " ] we applied algorithm [ alg2 ] to data simulated using the tyson - novak model @xcite for the yeast cell cycle . the cell cycle is the succession of events whereby a cell grows and divides into two daughter cells . the simulated data was used to construct a * pwl * approximation of the system using algorithm [ alg2 ] . this resulted in a clear partition of the cell cycle dynamics as a function of the selected number of classes @xmath231 . the best qualitative results were obtained for @xmath362 , see figure [ n6-a ] . moreover , for @xmath365 the results did not improve . it was found that the obtained two discrete modes correspond to the phases _ s _ ( dna synthesis ) and _ m _ ( mitosis ) of the cell cycle . the reconstruction remains valid under noise with snr at most @xmath366 . ] we presented some basic results on realization theory of * pwl*s and a practical identification algorithm . analysis of the correctness and convergence of the presented algorithm remains a topic of future research . bako , l. , mercre , g. , vidal , r. , and lecoeuche . , s. ( 2009 ) . identification of switched linear state space models without minimum dwell time . in _ ifac symposium on system identification , saint malo , france_. ma , y. and vidal , r. ( 2005 ) . identification of deterministic switched arx systems via identification of algebraic varieties . in _ hybrid systems : computation and control _ , volume 3414 of _ lncs _ , 449 465 .

in the last three decades transport through molecular junctions @xcite has attracted much attention because of fundamental aspects of the processes involved , as well as of potential practical prospects . in particular , studies in this direction have intensified since the discovery of new forms of carbon allotropes , starting with the fullerenes @xcite and carbon nanotubes @xcite ( cnts ) in the 1980s and 1990s respectively , and the isolation of graphene @xcite in 2004 . the above carbon allotropes differ in shape ( curvature ) , topology , and dimensionality , with the fullerenes being zero - dimensional ( 0d ) with spherical or prolate shape , the cnts being one - dimensional ( 1d ) cylinders , and graphene being a two - dimensional ( 2d ) plane ( or 1d planar ribbons @xcite ) . in the fullerenes the carbon - atom network is made of ( non - adjacent ) hexagons and pentagons , whereas the cnts and graphene are entirely hexagonal lattices ( curved in cnts ) described in terms of a unit cell with a two - atom basis with the two carbon atoms occupying two sublattices ( @xmath0 and @xmath1 , also mapped into the up and down pseudospin states @xcite ) . in the absence of defects , in - plane ( @xmath2 ) bonding occurs through @xmath3 hybrid orbitals and out - of - plane bonding ( @xmath4 ) involves the @xmath5 orbital ; in the following only the physics of @xmath4-states is considered . the hexagonal network topology of graphene gives rise to relativistic behavior of the low - energy excitations which is captured by the ultra - relativistic massless dirac - weyl ( dw ) equation , with the fermi velocity of graphene ( @xmath6 ) replacing the velocity of light . @xcite among the many manifestations of the relativistic behavior in graphene is the klein paradox , that is `` ... unimpeded penetration of relativistic particles through high and wide potential barriers is one of the most exotic and counterintuitive consequences of quantum electrodynamics.''@xcite the surprising relativistic behavior in graphene has indeed been recognized in the 2010 nobel award in physics to a. geim and k. novosolev . another carbon - based system that was the subject of an earlier ( 2000 ) chemistry nobel award ( to a.j . heeger , a.g . macdiarmid , and h. shirakawa ) is conductive polymers , with polyacetelyne ( pac ) being a representative example . @xcite pac is a 1d chain of carbon atoms forming a conjugated polymer with @xmath7 hybridization that leads to one unpaired electron per carbon atom ( half - filled @xmath4-band ) as in graphene . linearizing the spectrum of this 1d equally - spaced carbon chain at the fermi level ( that is at the dirac points , i.e. , the zero - energy points of the energy vs momentum dispersion relation ) results in a 1d massless dirac - equation description of the low energy excitations of the system @xcite however , the equally - spaced 1d system is unstable and consequently it distorts ( structurally ) spontaneously ( peierls distortion @xcite ) yielding a modulation ( alternation ) of the spacing between successive sites that results in dimerization of successive atoms along the chain and the opening of a gap in the electronic spectrum . this dimerization can occur in two energetically degenerate , but spatially distinct , patterns , termed as equivalent domain structures . either of these domains is a realization of the 1d dirac equation with a constant mass term , @xmath8 . disruption of the dimerization pattern ( e.g. , by transforming from one domain to another along the chain ) creates a domain wall which can be described in the 1d dirac formulation through the use of a position - dependent mass term of alternating sign ( with the spatial mass - profile connecting @xmath9 with @xmath10 ) . the solution of this generalized 1d dirac equation is a soliton characterized by having zero energy and by being localized at the domain wall . in this paper , the dirac equation with position - dependent mass will be used in investigations of the electronic structure , transport , and magnetic - field - induced phenomena ( aharonov - bohm ) in graphene nanostructures such as nanoribbon junctions and rings . @xcite from the above we conclude that in the two extreme size - domains , that is a 2d infinite graphene sheet and a 1d carbon chain ( pac ) , the systems exhibit behavior that is relativistic in nature . in an attempt to bridge between these two size - domains , we briefly discuss in the following systems of successively larger width , starting from the polyacene ( a quasi-1d chain of fused benzene rings ) which may be regarded as the narrowest graphene nanoribbon ( gnr ) with zigzag edge terminations . polyacene was investigated @xcite first in 1983 . it was found that , as in graphene and pac ( without distortion ) , the valence and conduction bands of undistorted polyacene touch at the edge of the brillouin zone . however , unlike 2d graphene and pac , the dispersion relation about the touching point is quadratic , conferring a non - relativistic ( schrdinger equation ) character . we show in this paper that this surprising finding persists for sufficiently narrow gnrs with zigzag termination ( zgnrs ) . however , narrow armchair - terminated gnrs ( agnrs ) are found here to maintain relativistic behavior , with metallic ones being massless and semiconducting ones being massive ( both classes obeying the einstein energy relation ) . it this paper , we discuss mainly manifestations of relativistic and/or nonrelativistic quantum behavior explored through theoretical considerations of transport measurements in segmented graphene nanoribbons of variable width , and spectral and topological effects in graphene nanorings in the presence of magnetic fields . transport through narrow graphene channels @xmath11 particularly bottom - up fabricated and atomically - precise graphene nanoribbons @xcite @xmath11 is expected to offer ingress to unique behavior of dirac electrons in graphene nanostructures . in particular , the wave nature of elementary particles ( e.g. , electrons and photons ) is commonly manifested and demonstrated in transport processes . because of an exceptionally high electron mobility and a long mean - free path , it has been anticipated that graphene @xcite devices hold the promise for the realization , measurement , and possible utilization of fundamental aspects of coherent and ballistic transport behavior , which to date have been observed , with varying degrees of success , mainly at semiconductor interfaces @xcite , quantum point contacts @xcite , metallic wires @xcite , and carbon nanotubes @xcite . another manifestation of coherent ballistic transport are interference phenomena , reflecting the wave nature of the transporting physical object , and associated most often with optical ( electromagnetic waves , photons ) systems or with analogies to such systems ( that is , the behavior of massless particles , as in graphene sheets ) . measurements of interference patterns are commonly made with the use of interferometers , most familiar among them the multi - pass optical fabry - prot ( ofp ) interferometer @xcite . the advent of 2d forms of carbon allotropes has motivated the study of optical - like interference phenomena associated with relativistic massless electrons , as in the case of metallic carbon nanotubes @xcite and graphene 2d @xmath12-@xmath13 junctions @xcite . ( we note that the hallmark of the ofp is that the energy separation between successive maxima of the interference pattern varies as the inverse of the cavity length @xmath14 . ) for gnrs with segments of different widths , our investigations reveal diverse fabry - prot transport modes beyond the ofp case , with conductance quantization steps ( @xmath15 , with @xmath16 ) found only for uniform gnrs . in particular , three distinct categories of fabry - prot interference patterns are identified : 1 . _ fp - a : _ an _ optical _ fp pattern corresponding to _ massless _ graphene electrons exhibiting _ equal spacing _ between neighboring peaks . this pattern is associated with metallic armchair nanoribbon central segments . this category is subdivided further to _ fp - a1 _ and _ fp - a2 _ depending on whether a valence - to - conduction gap is absent ( _ fp - a1 _ , associated with metallic armchair leads ) , or present ( _ fp - a2 _ , corresponding to semiconducting armchair leads ) . 2 . _ fp - b : _ a _ massive relativistic _ fp pattern exhibiting a shift in the conduction onset due to the valence - to - conduction gap and _ unequal peak spacings_. this pattern is associated with semiconducting armchair nanoribbon central segments , irrespective of whether the leads are metallic armchair , semiconducting armchair , or zigzag . _ fp - c : _ a _ massive non - relativistic _ fp pattern with @xmath17 _ peak spacings _ , but with a vanishing valence - to - conduction gap , @xmath14 being the length of the central segment . this pattern is the one expected for usual semiconductors described by the ( nonrelativistic ) schrdinger equation , and it is associated with zigzag nanoribbon central segments , irrespective of whether zigzag or metallic armchair leads are used . we report in this paper on the unique apects of transport through segmented gnrs obtained from tight - binding non - equilibrium green s function @xcite ( tb - negf ) calculations in conjunction with an analysis based on a one - dimensional ( 1d ) relativistic dirac continuum model . this continuum model goes beyond the physics of the massless dirac - weyl ( dw ) electron , familiar from two - dimensional ( 2d ) honeycomb carbon sheets @xcite , and it is referred to by us as the dirac - fabry - prot ( dfp ) theory ( see below for the choice of name ) . in particular , it is shown that the valence - to - conduction energy gap in armchair gnr ( agnr ) segments , as well as the barriers at the interfaces between nanoribbon segments , can be incorporated in an effective position - dependent mass term in the dirac hamiltonian ; the transport solutions associated with this hamiltonian exhibit conductance patterns comparable to those obtained from the microscopic negf calculations . for zigzag graphene nanoribbon ( zgnr ) segments , the valence - to - conduction energy gap vanishes , and the mass term is consonant with excitations corresponding to massive nonrelativistic schrdinger - type carriers . the faithful reproduction of these unique tb - negf conductance patterns by the dfp theory , including mixed armchair - zigzag configurations ( where the carriers transit from a relativistic to a nonrelativistic regime ) , provides a unifying framework for analysis of coherent transport phenomena and for interpretation of experiments targeting fundamental understanding of transport in gnrs and the future development of graphene nanoelectronics . to demonstrate the aforementioned soliton formation due to structural topological effects ( discussed by us above in the context of polyacetylene ) , we explore with numerical tight - binding calculations and a dirac - kronig - penney ( dkp ) approach , soliton formation and charge fractionization in graphene rhombic rings ; this dkp approach is based on a generalized dirac equation with alternating - sign position - dependent masses . before leaving the introduction , we mention that , due to their importance as fundamental phenomena , aharonov - bohm - type effects in graphene - nanoribbons systems have attracted ( in addition to refs . ) considerable theoretical attention . @xcite these latter theoretical papers , however , based their analysis exclusively on tight - binding and/or dft calculations@xcite , or they used in addition a two - dimensional dirac equation with infinite - mass boundary conditions.@xcite transcending the level of current understanding which explores direct similaritiess with the aharonov - bohm physics in semiconductor and metallic mesoscopic rings , our work here analyzes the tb results in conjunction with a continuum 1d generalized dirac equation ( that incorporates a position - dependent mass term ) , and thus it enables investigations of until - now unexplored topological aspects and relativistic quantum - field analogies of the ab effect in graphene nanosystems . furthermore we note that oscillations in the conductance of graphene nanoribbons in the presence of magnetic barriers were found in a theoretical study @xcite ( using exclusively a tb - negf approach ) , as well as in an experimental investigation of high quality bilayer nanoribbons , @xcite and they were attributed to fabry - prot - type interference . in the absence of a continuum dirac analysis , the precise relation of such oscillatory patterns to our fabry - prot theory ( based on the incorporation of a position - dependent mass term in the dirac equation at zero magnetic field ) warrants further investigation . finally , gap engineering in graphene ribbons under strong external fields was studied in ref . . we stress again that one of the main results in this paper is the appearance of `` forbidden '' solitonic states inside the energy gap in the context of the low - magnetic - field aharonov - bohm spectra of graphene nanorings ; see the part titled `` aharonov - bohm spectra of rhombic graphene rings '' in the results section . the energy of a fermion ( with one - dimensional momentum @xmath18 ) is given by the einstein relativistic relation @xmath19 , where @xmath8 is the rest mass and @xmath20 is the fermi velocity of graphene . in a gapped uniform graphene nanoribbon , the mass parameter is related to the particle - hole energy gap , @xmath21 , as @xmath22 . following the relativisitic quantum - field lagrangian formalism , the mass @xmath8 is replaced by a position - dependent scalar higgs field @xmath23 , to which the relativistic fermionic field @xmath24 couples through the yukawa lagrangian @xcite @xmath25 ( @xmath26 being a pauli matrix ) . for @xmath27 ( constant ) @xmath28 , and the massive fermion dirac theory is recovered . the dirac equation is generalized as ( here we do not consider applied electric or magnetic fields ) @xmath29 \psi + i \hbar v_f \alpha \frac{\partial \psi}{\partial x } - \beta \phi(x ) \psi=0 . \label{direq}\ ] ] in one dimension , the fermion field is a two - component spinor @xmath30 ; @xmath31 and @xmath32 stand , respectively , for the upper and lower component and @xmath33 and @xmath26 can be any two of the three pauli matrices . note that the higgs field enters in the last term of eq . ( [ direq ] ) . @xmath34 in the first term is the usual electrostatic potential , which is inoperative due to the klein phenomenon and thus is set to zero for the case of the armchair nanoribbons ( where the excitations are relativistic ) . the generalized dirac eq . ( [ direq ] ) is used in the construction of the transfer matrices of the dirac - fabry - prot model described below . the building block of the dfp model is a 2@xmath352 wave - function matrix @xmath36 formed by the components of two independent spinor solutions ( at a point @xmath37 ) of the onedimensional , first - order generalized dirac equation [ see eq . ( 3 ) in the main paper ] . @xmath36 plays @xcite the role of the wronskian matrix @xmath38 used in the second - order nonrelativistic kronig - penney model . following ref . and generalizing to @xmath39 regions , we use the simple form of @xmath36 in the dirac representation ( @xmath40 , @xmath41 ) , namely @xmath42 where @xmath43 the transfer matrix for a given region ( extending between two matching points @xmath44 and @xmath45 specifying the potential steps @xmath46 ) is the product @xmath47 ; this latter matrix depends only on the width @xmath48 of the region , and not separately on @xmath44 or @xmath45 . the transfer matrix corresponding to a series of @xmath39 regions can be formed @xcite as the product @xmath49 where @xmath50 is the width of the @xmath51th region [ with @xmath52 . the transfer matrix associated with the transmission of a free fermion of mass @xmath8 ( incoming from the right ) through the multiple mass barriers is the product @xmath53 with @xmath54 , @xmath55 ; for armchair leads @xmath56 , while for zigzag leads @xmath57 . naturally , in the case of metallic armchair leads , @xmath58 , since @xmath59 . then the transmission coefficient @xmath60 is @xmath61 while the reflection coefficient is given by @xmath62 at zero temperature , the conductance is given by @xmath63 ; @xmath60 is the transmission coefficient in eq . ( [ trans ] ) . + the transfer matrix corresponding to either half of the rhombus can be formed @xcite as the product @xmath64 with @xmath14 being the length of half of the perimeter of the rhombus ; @xmath65 , with @xmath66 and @xmath67 . the transfer matrix associated with the complete unit cell ( encircling the rhombic ring ) is the product @xmath68 following refs . , we consider the superlattice generated from the virtual periodic translation of the unit cell as a result of the application of a magnetic field @xmath1 perpendicular to the ring . then the aharonov - bohm energy spectra are given as solutions of the dispersion relation @xmath69 = \operatorname{tr}[{\bf t}(e)]/2 , \label{disrel}\ ] ] where we have explicitly denoted the dependence of the r.h.s . on the energy @xmath70 ; @xmath71 for a rhombus with type - i corners and @xmath72 for a rhombus with type - ii corners . the energy spectra and single - particle densities do not depend on a specific representation . however , the wave functions ( upper and lower spinor components of the fermionic field @xmath73 ) do depend on the representation used . to transform the initial dkp wave functions to the ( @xmath74 , @xmath75 ) representation , which corresponds to the natural separation of the tight - binding amplitudes into the @xmath0 and @xmath1 sublattices , we apply successively the unitary transformations @xmath76 and @xmath77 . to describe the properties of graphene nanostructures in the tight - binding approximation , we use the hamiltonian @xmath78 with @xmath79 indicating summation over the nearest - neighbor sites @xmath80 . @xmath81 ev is the hopping parameter of two - dimensional graphene . to calculate the tb - negf transmission coefficients , the hamiltonian ( [ htb ] ) is employed in conjunction with the well known transport formalism which is based on the nonequilibrium green s functions @xcite . according to the landauer theory , the linear conductance is @xmath82 , where the transmission coefficient is calculated as @xmath83 $ ] . the green s function @xmath84 is given by @xmath85 with @xmath86 being the hamiltonian of the isolated device ( junction without the leads ) . the self - energies @xmath87 are given by @xmath88^{-1}\tau_{l(r)}^\dagger$ ] , where the hopping matrices @xmath89 describe the left ( right ) device - to - lead coupling , and @xmath90 is the hamiltonian of the semi - infinite left ( right ) lead . the broadening matrices are given by @xmath91 $ ] . + + + + + our results for a 3-segment ( @xmath92 , where @xmath93 is the lead width and @xmath94 is the width of the central segment ) _ all - semiconducting _ agnr , aaa ( sss ) , are portrayed in fig . [ fig1 ] [ see schematic lattice diagrams in figs . [ fig1](i ) and [ fig1](ii ) ] . a uniform semiconducting armchair gnr [ see fig . [ fig1](i ) ] exhibits ballistic quantized - conductance steps [ see fig . [ fig1](a ) ] . in contrast , conductance quantization is absent for a nonuniform 3-segment ( @xmath95 ) agnr ; see figs . [ fig1](b ) @xmath11 [ fig1](d ) . here oscillations appear instead of quantized steps . the first oscillation appears at an energy @xmath96 [ fig . [ fig1](b ) ] , which reflects the intrinsic gap @xmath97 of the semiconducting central segment belonging to the class ii of agnrs , specified @xcite by a width @xmath98 , @xmath99 . we recall that as a function of their width , @xmath100 , the armchair graphene nanoribbons fall into three classes : ( i ) @xmath101 ( semiconducting , @xmath102 ) , ( ii ) @xmath103 ( semiconducting , @xmath102 ) , and ( iii ) @xmath104 ( metallic , @xmath105 ) , @xmath99 . that the leads are semiconducting does not have any major effect . this is due to the fact that @xmath106 , and as a result the energy gap @xmath107 of the central segment is larger than the energy gap @xmath108 of the semiconducting leads [ see schematic in fig . [ fig1](iii ) ] . in the opposite case ( central segment wider than the leads ) , the energy gap of the semiconducting leads would have determined the onset of the conductance oscillations . the armchair gnr case with interchanged widths ( i.e. , @xmath109 instead of @xmath95 ) is portrayed in figs . [ fig1](e ) @xmath11 [ fig1](f ) . in this case the energy gap of the semiconducting leads ( being the largest ) determines the onset of the conductance oscillations . it is a testimonial of the consistency of our dfp method that it can reproduce [ see fig . [ fig1](d ) and fig . [ fig1](f ) ] both the @xmath95 and @xmath109 tb - negf conductances ; this is achieved with very similar sets of parameters taking into consideration the central - segment - leads interchange . we note that the larger spacing between peaks ( and also the smaller number of peaks ) in the @xmath109 case is due to the smaller mass of the central segment ( @xmath110 instead of @xmath111 ) . further insight can be gained by an analysis of the discrete energies associated with the humps of the conductance oscillations in fig . [ fig1](c ) and the resonant spikes in fig . [ fig1](e ) . indeed a simplified approximation for the electron confinement in the continuum model consists in considering the graphene electrons as being trapped within a 1d infinite - mass square well ( imsw ) of length @xmath112 ( the mass terms are infinite outside the interval @xmath112 and the coupling to the leads vanishes ) . the discrete spectrum of the electrons in this case is given @xcite by @xmath113 where the wave numbers @xmath114 are solutions of the transcendental equation @xmath115 when @xmath59 [ massless dirac - weyl ( dw ) electrons ] , one finds for the spectrum of the imsw model : @xmath116 with @xmath117 . @xmath118 which is twice the energy @xmath119 of the lowest state . as is well known , a constant energy separation of the intensity peaks , inversely proportional to the length of the resonating cavity [ here @xmath112 , see eqs . ( [ edis ] ) and ( [ eoff ] ) above ] is the hallmark of the optical fabry - prot , reflecting the linear energy dispersion of the photon in optics @xcite or a massless dw electron in graphene structures.@xcite most revealing is the energy offset away form zero of the first conductance peak , which equals exactly one - half of the constant energy separation between the peaks . in one - dimension , this is the hallmark of a massless fermion subject to an infinite - mass - barrier confinement @xcite . naturally , in the case of a semiconducting segment ( see below ) , this equidistant behavior and @xmath120-offset of the conductance peaks do not apply ; this case is accounted for by the dirac - fabry - prot model presented in methods , and it is more general than the optical fabry - prot theory associated with a photonic cavity . @xcite in the nonrelativistic limit , i.e. , when @xmath121 , one gets @xmath122 which yields the well known relations @xmath123 and @xmath124 for a massive relativistic electron , as is the case with the semiconducting agnrs in this paper , one has to numerically solve eq . ( [ eimsw ] ) and then substitute the corresponding value of @xmath114 in einstein s energy relation given by eq . ( [ erel ] ) . from an inspection of fig . [ fig1 ] , one can conclude that the physics associated with the all - semiconducting aaa junction is that of multiple reflections of a _ massive _ relativistic dirac fermion bouncing back and forth from the edges of a `` quantum box '' created by a double - mass barrier [ see the schematic of the double - mass barrier in figs . [ fig1](iii ) and [ fig1](v ) ] . in particular , to a good approximation the energies of the conductance oscillation peaks are given by the imsw eq.([eimsw ] ) with @xmath125 ( @xmath95 ) or @xmath126 ( @xmath109 ) . in this respect , the separation energy between successive peaks in figs . [ fig1](b ) , [ fig1](c ) , [ fig1](e ) , and [ fig1](f ) is not a constant , unlike the case of an all - metallic junction @xcite ( or a photonic cavity @xcite ) . the conductance patterns in figs . [ fig1](d ) and [ fig1](f ) correspond to the category _ fp - b _ ( see the introductory section ) . these patterns can not be accounted for by the optical fabry - prot theory , but they are well reproduced by the generalized dirac - fabry - prot model introduced by us in the methods section . our results for a 3-segment ( @xmath127 ) _ semiconducting - metallic - semiconducting _ agnr , aaa ( sms ) , are portrayed in fig . [ fig2 ] [ left , see schematic lattice diagram in fig . [ fig2](i ) ] . the first fp oscillation in the tb - negf conductance displayed in fig . [ fig2](a ) appears at an energy @xmath96 , which reflects the intrinsic gap @xmath97 of the semiconducting leads ( with @xmath128 ) . the energy spacing between the peaks in fig . [ fig2](a ) is constant in agreement with the metallic ( massless dw electrons ) character of the central segment with @xmath129 . the tb - negf pattern in fig . [ fig2](a ) corresponds to the fabry - prot category _ fp - a2_. as seen from fig . [ fig2](b ) , our generalized dirac - fabry - prot theory is again capable of faithfully reproducing this behavior . a deeper understanding of the aaa ( sms ) case can be gained via an inspection of the density of states ( dos ) plotted in fig . [ fig2](c ) for the total segmented agnr ( central segment plus leads ) and in fig . [ fig2](d ) for the the isolated leads . in fig . [ fig2](c ) , nine equidistant resonance lines are seen . their energies are close to those resulting from the imsw eq . ( [ imsw0 ] ) ( with @xmath130 , see the caption of fig . [ fig2 ] ) for a massless dw electron . out of these nine resonances , the first five do not conduct [ compare figs . [ fig2](a ) and [ fig2](c ) ] because their energies are lower than the minimum energy ( i.e. , @xmath131 ) of the incoming electrons in the leads [ see the onset of the first band ( marked by an arrow ) in the dos curve displayed in fig . [ fig2](d ) ] . as shown in refs . , a detailed description of hygrogen passivation requires that the hopping parameters @xmath132 for the nearest - neighbor c - c bonds at the armchair edges be given by @xmath133 . taking this modification into account , our results for a 3-segment _ semiconducting - metallic - semiconducting _ agnr are portrayed in fig . [ fig2 ] [ right , see schematic lattice diagram in fig . [ fig2](iii ) ] ; this lattice configuration is denoted as `` aaa ( sms ) h - passivation . '' the first fp oscillation in the tb - negf conductance displayed in fig . [ fig2](e ) appears at an energy @xmath134 , which reflects the intrinsic gap @xmath97 of the properly passivated semiconducting leads ( with @xmath128 ) . the energy spacing between the peaks in fig . [ fig2](e ) is slightly away from being constant in agreement with the small mass @xmath135 acquired by the central segment with @xmath129 , due to taking @xmath133 . as seen from fig . [ fig2](f ) , our generalized dirac - fabry - prot theory is again capable of faithfully reproducing this behavior . a deeper understanding of the aaa ( sms)-h - passivation case can be gained via an inspection of the dos plotted in fig . [ fig2](g ) for the total segmented agnr ( central segment plus leads ) and in fig . [ fig2](h ) for the isolated leads . in fig . [ fig2](g ) , eight ( almost , but not exactly , equidistant ) resonance lines are seen . their energies are close to those resulting from the imsw eq . ( [ eimsw ] ) ( with @xmath136 and @xmath135 ; see the caption of fig . [ fig2 ] ) for a dirac electron with a small mass . out of these eight resonances , the first six do not conduct [ compare figs . [ fig2](e ) and [ fig2](g ) ] because their energies are lower than the minimum energy ( i.e. , @xmath137 ) of the incoming electrons in the leads [ see the onset of the first band ( marked by an arrow ) in the dos curve displayed in fig . [ fig2](h ) ] . from the above we conclude that hydrogen passivation of the agnr resulted in a small shift of the location of the states , and opening of a small gap for the central metallic narrower ( with a width of @xmath138 ) segment , but did not modify the conductance record in any qualitative way . moreover , the passivation effect can be faithfully captured by the dirac fp model by a small readjustment of the model parameters . , with @xmath139 ( @xmath140 ) being the number of carbon atoms specifying the width of the ribbon segments . the armchair leads in the aza junction are metallic ( @xmath141 , class iii agnr ) . ( a)-(b ) tb - negf conductance for the zzz and aza junction , respectively . ( c ) dfp conductance reproducing the tb - negf result in ( a ) for the zzz junction . ( d ) dfp conductance reproducing the tb - negf result in ( b ) for the aza junction . in spite of the different edge morphology , the fabry - prot patterns in ( a ) and ( b ) are very similar . the central zigzag segment controls the fabry - prot patterns . according to the continuum dfp analysis , the physics underlying such patterns is that of a massive nonrelativistic schrdinger fermionic carrier performing multiple reflections within a cavity defined by a double - mass barrier [ see diagram in ( iii ) ] , but with the additional feature that @xmath142 and @xmath143 are also considered for segments or leads with zigzag edge terminations ( see text for details ) . the mass and @xmath144 parameters used in the dfp calculations were @xmath145 , @xmath146 , with @xmath147 , @xmath142 , @xmath148 , @xmath149 , @xmath150 , @xmath151 , @xmath143 in ( c ) and @xmath152 , @xmath153 , with @xmath154 , @xmath142 , @xmath155 , @xmath156 , @xmath157 , @xmath158 , @xmath159 in ( d ) . @xmath160 and @xmath161 denote parameters of the leads . @xmath70 is the energy in units of @xmath162 . @xmath163 nm is the graphene lattice constant ; @xmath81 ev is the hopping parameter.,width=434 ] it is interesting to investigate the sensitivity of the interference features on the edge morphology . we show in this section that the relativistic transport treatment applied to segmented armchaie gnrs does not maintain for the case of a nanoribbon segment with zigzag edge terminations . in fact zigzag gnr ( zgnr ) segments exhibit properties akin to the well - known transport in usual semiconductors , i.e. , their excitations are governed by the nonrelativistic schrdinger equation . before discussing segmented gnrs with zigzag edge terminations , we remark that such gnrs with uniform width exhibit stepwise quantization of the conductance , similar to the case of a uniform armchair - edge - terminated gnr [ see fig . [ fig1](a ) ] . in fig . [ fig3](a ) , we display the conductance in a three - segment junction [ see lattice schematic in fig . [ fig3](i ) ] when all three segments have zigzag edge terminations ( denoted as zzz ) , but the central one is narrower than the lead segments . the main finding is that the central segment behaves again as a resonant cavity that yields an oscillatory conductance pattern where the peak spacings are unequal [ fig . [ fig3](a ) ] . this feature , which deviates from the optical fabry - prot behavior , appeared also in the dfp patterns for a three - segment armchair junction whose central segment was semiconducting , albeit with a different dependence on @xmath14 [ see figs . [ fig1 ] and eq . ( [ eimsw ] ) ] . moreover , from a set of systematic calculations ( not shown ) employing different lengths and widths , we found that the energy of the resonant levels in zgnr segments varies on the average as @xmath164 , where the integer @xmath13 counts the resonances and @xmath14 indicates the length of the central segment . however , a determining difference with the armchair gnr case in fig . [ fig1 ] is the vanishing of the valence - to - conductance gap in the zigzag case of fig . [ fig3](a ) . it is well known that the above features are associated with resonant transport of electronic excitations that obey the nonrelativistic second - order schrdinger equation . naturally , one could formulate a continuum transport theory based on transfer matrices ( see methods ) that use the 1d schrdinger equation instead of the generalized dirac eq . ( [ direq ] ) . such a schrdinger - equation continuum approach , however , is unable to describe mixed armchair - zigzag interfaces ( see below ) , where the electron transits between two extreme regimes , i.e. , an ultrarelativistic ( i.e. , including the limit of vanishing carrier mass ) dirac regime ( armchair segment ) and a nonrelativistic schrdinger regime ( zigzag segment ) . we have thus been led to adopt the same dirac - type transfer - matrix approach as with the armchair gnrs , but with nonvanishing potentials @xmath57 . this amounts to shifts ( in opposite senses ) of the energy scales for particle and hole excitations , respectively , and it yields the desired vanishing value for the valence - to - conduction gap of zigzag gnrs . the calculated dfp conductance that reproduces well the tb - negf result for the zzz junction [ fig.[fig3](a ) ] is displayed in fig . [ fig3](c ) ; the parameters used in the dfp calculation are given in the caption of fig . we note that the carrier mass ( @xmath165 ) in the central zigzag segment exhibits an energy dependence . this is similar to a well known effect ( due to nonparabolicity in the @xmath166 dispersion ) in the transport theory of usual semiconductors @xcite . we further note that the average mass associated with a zigzag segment is an order of magnitude larger than that found for semiconducting armchair segments of similar width ( see caption in fig . [ fig1 ] ) , and this yields energy levels @xmath164 close to the nonrelativistic limit [ see eq . ( [ ennr ] ) ] . we note that the fp pattern of the zzz junction belongs to the category _ fp - c_. fig . [ fig3 ] ( right column ) presents an example of a mixed armchair - zigzag - armchair ( aza ) junction , where the central segment has again zigzag edge terminations [ see lattice schematic in fig . [ fig3](ii ) ] . the corresponding tb - negf conductance is displayed in fig . [ fig3](b ) . in spite of the different morphology of the edges between the leads ( armchair ) and the central segment ( zigzag ) , the conductance profile of the aza junction [ fig . [ fig3](b ) ] is very similar to that of the zzz junction [ fig . [ fig3](a ) ] . this means that the characteristics of the transport are determined mainly by the central segment , with the left and right leads , whether zigzag or armchair , acting as reservoirs supplying the impinging electrons . the dfp result reproducing the tb - negf conductance is displayed in fig . [ fig3](d ) , and the parameters used are given in the caption . we stress that the mixed aza junction represents a rather unusual physical regime , where an ultrarelativistic dirac - weyl massless charge carrier ( due to the metallic armchair gnrs in the leads ) transits to a nonrelativistic massive schrdinger electron in the central segment . we note that the fp pattern of the aza junction belongs to the _ fp - c _ caregory . . ( b ) tb spectrum for a nanoring with type - ii corners and the same width @xmath167 . these armchair graphene rings are semiconducting ( type - i ) and metallic ( type - ii ) . the three ( four ) lowest - in - energy two - membered bands are shown . the hole states ( with @xmath168 , not shown ) are symmetric to the particle states ( with @xmath169 ) . ( c , d ) dkp spectra reproducing the tb ones in ( a ) and ( b ) , respectively . insets in ( c ) and ( d ) : schematics of the higgs fields ( position - dependent mass ) @xmath170 employed in the dkp modeling . @xmath170 is approximated by steplike functions @xmath46 ; @xmath51 counts the three regions of each half of the rhombus ( @xmath171 and @xmath172 ) , and @xmath13 ( @xmath173 ) . the non - zero ( constant ) variable - mass values of @xmath170 are indicated by yellow ( red ) color when positive ( negative ) . the parameters used in the dkp modeling are : ( c ) @xmath174 , @xmath175 , @xmath176 , @xmath177 [ see corresponding schematic inset in ( c ) ] and ( d ) @xmath178 , @xmath179 , @xmath176 , @xmath180 with @xmath181 [ see schematic inset in ( d ) ] . note the two - membered braided bands and the `` forbidden '' band [ within the gap , in ( b ) and ( d ) ] . the twofold forbidden band with @xmath182 appears as a straight line due to the very small amplitude oscillations of its two members . @xmath163 nm is the graphene lattice constant and @xmath81 ev is the hopping parameter . the edge terminations of both the inside and outside sides of the ring are armchair.,width=510 ] the energy of a particle ( with onedimensional momentum @xmath18 ) is given by the einstein relativistic relation @xmath19 , where @xmath8 is the rest mass . as aforementioned , in armchair graphene ribbons , the mass parameter is related to the particle - hole energy gap , @xmath21 , as @xmath22 . in relativistic quantum field theory , the mass of elementary particles is imparted through interaction with a scalar field known as the higgs field . accordingly , the mass @xmath8 is replaced by a position - dependent higgs field @xmath183 , to which the relativistic fermionic field @xmath24 couples through the yukawa lagrangian @xcite @xmath25 ( @xmath26 being a pauli matrix ) . in the elementary - particles standard model , @xcite such coupling is responsible for the masses of quarks and leptons . for @xmath27 ( constant ) @xmath28 , and the massive fermion dirac theory is recovered . we exploit the generalized dirac physics governed by a total lagrangian density @xmath184 , where the fermionic part is given by @xmath185 and the scalar - field part has the form @xmath186 with the potential @xmath187 ( second term ) assumed to have a double - well @xmath188 form ; @xmath189 and @xmath190 are constants . henceforth , the dirac equation ( see methods ) is generalized as @xmath191 in one dimension , the fermion field is a two - component spinor @xmath30 ; @xmath31 and @xmath32 stand , respectively , for the upper and lower component and @xmath33 and @xmath26 can be any two of the three pauli matrices . -sublattice ( red ) and @xmath1-sublattice ( blue ) components of the tb state with energy @xmath192 at @xmath193 , belonging to the forbidden solitonic band of the type - ii nanoring with @xmath167 [ see fig . [ fig44](b ) ] . ( b ) upper ( red ) and lower ( blue ) spinor components for the corresponding state ( forbidden band ) according to the dkp spectrum [ see fig . [ fig44](d ) ] , reproducing the tb behavior of the type - ii nanoring with @xmath167 ( @xmath181 ) . the tb and dkp wave functions for all states of the solitonic band are similar to those displayed here . the wave functions here represent a pair of solitons . for contrast , see fig . 10 in ref . which portrays schematically the spinor @xmath194 for a _ single _ fermionic soliton attached to a higgs field with a smooth kink - soliton analytic shape @xmath195 . @xmath196 is a solution @xcite of the lagrangian in eq . ( [ lagrphi ] ) . dkp densities in units of @xmath197 . , width=472 ] a graphene polygonal ring can be viewed as made of connected graphene - nanoribbon fragments ( here we consider agnrs ) . the excitations of an infinite agnr are described by the 1d massive dirac equation , see eq . ( [ direq1 ] ) with @xmath74 , @xmath75 , and @xmath198 . the two ( in general ) unequal hopping parameters @xmath199 and @xmath200 are associated with an effective 1d tight - binding problem ) and are given @xcite by @xmath201 $ ] , @xmath202 and @xmath203 ; @xmath100 is the number of carbon atoms specifying the width of the nanoribbon and @xmath81 ev is the hopping parameter for 2d graphene . the effective @xcite tb hamiltonian of an agnr has a form similar to that used in _ trans_-polyacetylene ( a single chain of carbon atoms ) . in _ trans_-polyacetylene , the inequality of @xmath199 and @xmath200 ( referred to as dimerization ) is a consequence of the aforementioned peierls distortion induced by the electron - phonon coupling . for an armchair graphene ring , this inequality is a topological effect associated with the geometry of the edge and the width of the ribbon . we recall that as a function of their width , @xmath100 , the armchair graphene nanoribbons fall into three classes : ( i ) @xmath204 ( semiconducting , @xmath205 ) , ( ii ) @xmath98 ( semiconducting , @xmath205 ) , and ( iii ) @xmath206 ( metallic @xmath207 ) , @xmath99 . we adapt the `` crystal '' approach @xcite to the aharonov - bohm ( ab ) effect , and introduce a virtual dirac - kronig - penney @xcite ( dkp ) relativistic superlattice ( see methods ) . charged fermions in a perpendicular magnetic field circulating around the ring behave like electrons in a spatially periodic structure ( period @xmath208 ) with the magnetic flux @xmath209 ( @xmath210 ) playing the role of the bloch wave vector @xmath211 , i.e. , @xmath212 [ see the cosine term in eq . ( [ disrel ] ) ] . naturally , nanorings with arms made of nanoribbon segments belonging to the semiconducting classes may be expected to exhibit a particle - hole gap ( particle - antiparticle gap in rqf theory ) . indeed this is found for a rhombic armchair graphene ring ( agr ) [ see gap @xmath21 in fig . [ fig44](a ) ] with a width of @xmath167 carbon atoms having type - i corners . suprisingly , a rhombic armchair graphene nanoring of the same width @xmath167 , but having corners of type - ii , demonstrates a different behavior , showing a `` forbidden '' band ( with @xmath182 ) in the middle of the gap region [ see fig . [ fig44](b ) ] . this behavior of rhombic armchair graphene rings with type - ii corners can be explained through analogies with rqf theoretical models , describing single zero - energy fermionic solitons with fractional charge @xcite or their modifications when forming soliton / anti - soliton systems . @xcite ( a solution of the equation of motion corresponding to eq . ( [ lagrphi ] ) , is a @xmath213 kink soliton , @xmath196 . the solution of eq . ( [ direq1 ] ) with @xmath214 is the fermionic soliton . ) we model the rhombic ring with the use of a continuous 1d kronig - penney @xcite model ( see methods ) based on the generalized dirac equation ( [ direq1 ] ) , allowing variation of the scalar field @xmath170 along the ring s arms . we find that the dkp model reproduces [ see fig . [ fig44](d ) ] the spectrum of the type - ii rhombic ring ( including the forbidden band ) when considering alternating masses @xmath215 associated with each half of the ring [ see inset in fig . [ fig44](d ) ] . in analogy with the physics of _ trans_-polyacetylene ( see remarks in the introductory section ) , the positive and negative masses correspond to two degenerate domains associated with the two possible dimerization patterns @xcite @xmath216 and @xmath217 , which are possible in a single - atom chain . the transition zones between the two domains ( here two of the four corners of the rhombic ring ) are referred to as the domain walls . for a single soliton , a ( precise ) zero - energy fermionic excitation emerges , localized at the domain wall . in the case of soliton - antisoliton pairs , paired energy levels with small positive and negative values appear within the gap . the tb spectrum in fig . [ fig44](b ) exhibits a forbidden band of two paired @xmath218 levels , a property fully reproduced by the dkp model that employs two alternating mass domains [ fig . [ fig44](d ) . the twofold forbidden band with @xmath182 appears as a straight line due to the very small amplitude oscillations of its two members . @xmath163 nm is the graphene lattice constant and @xmath81 ev is the hopping parameter . the strong localization of a fraction of a fermion at the domain walls ( two of the rhombus corners ) , characteristic of fermionic solitons @xcite and of soliton / anti - soliton pairs , @xcite is clearly seen in the tb density distributions ( modulus of single - particle wave functions ) displayed in fig . [ fig55](a ) . the tb @xmath0 ( @xmath1 ) sublattice component of the tight - binding wave functions localizes at the odd numbered corners . these alternating localization patterns are faithfully reproduced [ see fig . [ fig55](b ) ] by the upper , @xmath219 , and lower , @xmath220 , spinor components of the continuum dkp model . the soliton - antisoliton pair in fig . [ fig55](b ) generates an @xmath221 charge fractionization at each of the odd - numbered corners , which is similar to the @xmath221 fractionization familiar from polyacetylene . the absence of a forbidden band ( i.e. , solitonic excitations within the gap ) in the spectrum of the type - i rhombic nanorings [ see fig . [ fig44](a ) ] indicates that the corners in this case do not induce an alternation between the two equivalent dimerized domains ( represented by @xmath215 in the dkp model ) . here the corners do not act as topological domain walls . nevertheless , direct correspondence between the tb and dkp spectra is achieved here too by using a variable higgs field defined as @xmath222 with @xmath223 and @xmath177 [ see the schematic inset in fig . [ fig44](c ) and the dkp spectrum plotted in the same figure ] . in this paper we focused on manifestations of relativistic and/or nonrelativistic quantum behavior explored through theoretical considerations of transport in graphene nanostructures and spectral and topological effects in graphene nanorings in the presence of magnetic fields . in particular , we investigated the emergence of new behavior of electrons in atomically precise segmented graphene nanoribbons ( gnrs ) with different edge terminations ( armchair , zigzag and mixed ones ) , and in graphene rings . to these aims we have employed tight - binding calculations of electronic states with and without applied magnetic fields , the non - equilibrium green s function transport theory , and a newly developed dirac continuum model that absorbs the valence - to - conductance energy gaps as position - dependent masses , including topological - in - origin mass - barriers at the contacts between segments . the electronic conductance has been found to exhibit fabry - prot oscillations , or resonant tunneling , associated with partial confinement and formation of a quantum box ( resonant cavity ) in the junction . along with the familiar optical fp oscillations , exhibiting equal spacing between neighboring peaks , that we find for massless electrons in gnrs with metallic armchair central segments , we find other fp categories that differ from the optical one . in particular , our calculations reveal : ( a ) a massive relativistic fp pattern exhibiting a valence - to - conduction gap and unequal peak spacings . this pattern is associated with semiconducting armchair nanoribbon central segments , irrespective of whether the armchair leads are matallic or semiconducting . ( b ) a massive non - relativistic fp pattern with @xmath17 peak spacings , but with a vanishing valence - to - conduction gap . this pattern is the one expected for the carriers in usual semiconductors described by the ( nonrelativistic ) schrdinger equation , and it is associated with zigzag nanoribbon central segments , regardless of whether zigzag or metallic armchair leads are used . perfect quantized - conductance flat steps were found only for uniform gnrs . in the absence of extraneous factors , like disorder , in our theoretical model , the deviations from the perfect quantized - conductance steps were unexpected . however , this aforementioned behavior obtained through tb - negf calculations is well accounted for by a 1d contimuum fermionic dirac - fabry - prot interference theory ( see methods ) . this approach employs an effective position - dependent mass term in the dirac hamiltonian to absorb the finite - width ( valence - to - conduction ) gap in armchair nanoribbon segments , as well as the barriers at the interfaces between nanoribbon segments forming a junction . for zigzag nanoribbon segments the mass term in the dirac equation reflects the nonrelativistic schrodinger - type behavior of the excitations . the carrier mass in zigzag - terminated gnr segments is much larger than the particle mass in semiconducting armchair - terminated gnr segments . furthermore in the zigzag gnr segments ( which are always characterized by a vanishing valence - to - conduction energy gap ) , the mass corresponds simply to the carrier mass . in the armchair gnr segments , the carrier mass endows ( in addition ) the segment with a valence - to - conduction energy gap , according to einstein s relativistic energy relation [ see eq . ( 1 ) ] . we concluded with a brief discussion of the physics of electrons in segmented polygonal rings , which may be regarded as constructed by connecting gnr segments . evaluation of the electronic states in a rhombic graphene nanoring under the influence of an applied magnetic field in the aharonov - bohm regime , and their analysis with the use of a relativistic quantum - field theoretical model , unveils development of a topological - in - origin zero - energy soliton state and charge fractionization . the above findings point to a most fundamental underlying physics , namely that the topology of disruptions of the regular honeycomb lattice ( e.g. , variable width segments , corners , edges ) generates a scalar - potential field ( position - dependent mass , identified @xcite also as a higgs - type field ) , which when integrated into a generalized dirac equation for the electrons provides a unifying framework for the analysis of transport processes through graphene segmented junctions and the nature of electronic states in graphene nanorings . with growing activities and further improvements in the areas of bottom - up fabrication and manipulation of atomically precise @xcite graphene nanostructures and the anticipated measurement of conductance through them , the above findings could serve as impetus and implements aiding the design and interpretation of future experiments . 999 nitzan , a. ; ratner , m. a. electron transport in molecular wire junctions . _ science _ * 2003 * , _ 300 _ , 1384 - 1389 . joachim , c. ; ratner , m. a. molecular electronics : some views on transport junctions and beyond . _ proc . ( usa ) _ * 2005 * , _ 102 _ , 8801 - 8808 . lindsay , s. m. ; ratner , m. a. molecular transport junctions : clearing mists . . maters . _ * 2007 * , _ 19 _ , 23 - 31 . hod , o. ; rabani , e. ; baer , r. magnetoresistance of nanoscale molecular devices . _ * 2006 * , _ 39 _ , 109 - 117 . bergfield , j. p. ; ratner , m. a. forty years of molecular electronics : non - equilibrium heat and charge transport at the nanoscale . _ phys . solidi b _ * 2013 * , _ 250 _ , 2249 - 2266 . rai , d. ; hod , o. ; nitzan , a. magnetic field control of the current through molecular ring junctions . _ j. phys . lett . _ * 2011 * , _ 2 _ , 2118 - 2124 . kroto , h. w. ; heath , j. r. ; obrien , s. c. ; curl r. f. ; smalley , r. e. c60 : buckminsterfullerine . _ nature _ * 1985 * , _ 318 _ , 162 - 163 . iijima , s. helical microtubules of graphitic carbon . _ nature _ * 1991 * , _ 354 _ , 56 - 58 . novoselov , k. s. ; geim , a. k. ; morozov , s. v. ; jiang , d. ; zhang , y. ; dubonos , s. v. ; grigorieva , i. v. ; firsov , a. a. electric field effect in atomically thin carbon films . _ science _ * 2004 * , _ 306 _ , 666 - 669 . nakada , k. ; fujita , m. ; dresselhaus , g. ; dresselhaus , m. s. edge state in graphene ribbons : nanometer size effect and edge shape dependence . b : condens . matter mater . phys . _ * 1996 * , _ 54 _ , 17954 - 17961 . son , y. w. ; cohen , m. l. ; louie , s. g. energy gaps in graphene nanoribbons . _ * 2006 * , _ 97 _ , 216803 . wang , z. f. ; li , q. ; zheng , h. ; ren , h. ; su , h. ; shi , q. w. ; chen , j. tuning the electronic structure of graphene nanoribbons through chemical edge modification : a theoretical study . _ phys . rev . b : condens . matter mater . _ * 2007 * , _ 75 _ , 113406 . wakabayashi , k. ; sasaki , k. ; nakanishi , t ; enoki , t. electronic states of graphene nanoribbons and analytical solutions . technol . adv . mater . _ * 2010 * , _ 11 _ , 054504 , and references therein . castro neto , a. h. ; guinea , f. ; peres , n. m. r. ; novoselov , k. s. ; geim , a. k. the electronic properties of graphene . phys . _ * 2009 * , _ 81 _ , 109 - 162 . katsnelson , m. i. ; novoselov , k. s. ; geim , a. k. chiral tunnelling and the klein paradox in graphene . _ nature phys . _ * 2006 * , _ 2 _ , 620 - 625 . heeger , a. j. ; kivelson , s. ; schrieffer , j. r. ; su , w .- p . solitons in conducting polymers . phys . _ * 1988 * , _ 60 _ , 781 - 847 . jackiw , r. ; schrieffer , j. r. solitons with fermion number @xmath120 in condensed matter and relativistic field theories . b _ * 1981 * , _ 190 [ fs3 ] _ , 253 - 265 . jackiw , r. fractional and majorana fermions : the physics of zero - energy modes . _ phys . scr . _ * 2012 * , _ t146 _ , 014005 . peierls , r. f. _ quantum theory of solids _ ; clarendon press : oxford , * 1955*. romanovsky , i. ; yannouleas , c. ; landman , u. patterns of the aharonov - bohm oscillations in graphene nanorings . _ b : condens . matter mater . * 2012 * , _ 85 _ , 165434 . romanovsky , i. ; yannouleas , c. ; landman , u. topological effects and particle physics analogies beyond the massless dirac - weyl fermion in graphene nanorings . b : condens . matter mater . phys . _ * 2013 * , _ 87 _ , 165431 . yannouleas , c. ; romanovsky , i. ; landman , u. beyond the constant - mass dirac physics : solitons , charge fractionization , and the emergence of topological insulators in graphene rings . b : condens . matter mater . * 2014 * , _ 89 _ , 035432 . kivelson , s. ; chapman , o. i. polyacene and a new class of quasi - one - dimensional conductors . b : condens . matter mater . phys . _ * 1983 * , _ 28 _ , 7236 - 7243 . cai , j. m. ; ruffieux , p. ; jaafar , r. ; bieri , m. ; braun , t. ; blankenburg , s. ; muoth , m. ; seitsonen , a. p. ; saleh , m. ; feng , x .- l . ; et al . atomically precise bottom - up fabrication of graphene nanoribbons . _ nature _ * 2010 * , _ 466 _ , 470 - 473 . m. s. graphene : ribbons piece - by - piece . mat . _ * 2010 * , _ 9 _ , 611 - 612 . huang , h. ; wei , d. ; sun , j. ; wong , s. l. ; feng , y. p. ; castro neto , a. h. ; wee , a. t. s. spatially resolved electronic structures of atomically precise armchair graphene nanoribbons . _ rep . _ * 2012 * , _ 2 _ , 983 . van der lit , j. ; boneschanscher , m. p. ; vanmaekelbergh , d. ; ijs , m. ; uppstu , a. ; ervasti , m. ; harju , a. ; liljeroth , p. ; swart , i. suppression of electron - vibron coupling in graphene nanoribbons contacted via a single atom . commun . _ * 2013 * , _ 4 _ , 2023 . narita , a. ; feng , x .- ; hernandez , y. ; jensen , s. a. ; bonn , m. ; yang , h. f. ; verzhbitskiy , i. a. ; casiraghi , c. ; hansen , m. r. ; koch , a. h. r. ; et al . synthesis of structurally well - defined and liquid - phase - processable graphene nanoribbons . chem . _ * 2014 * , _ 6 _ , 126 - 132 . hartley , c. s. graphene synthesis : nanoribbons from the bottom - up . chem . _ * 2014 * , _ 6 _ , 91 - 92 . vo , t. h. ; shekhirev , m. ; kunkel , d. a. ; morton , m. d. ; berglund , e. ; kong , l. ; wilson , p. m. ; dowben , p. a. ; enders , a. ; sinitskii , a. large - scale solution synthesis of narrow graphene nanoribbons . commun . _ * 2014 * , _ 5 _ , 3189 . blankenburg , s. ; cai j. m. ; ruffieux , p. ; jaafar , r. ; passerone , d. ; feng , x .- l . ; mllen , k. ; fasel , r. ; pignedoli , c. a. intraribbon heterojunction formation in ultranarrow graphene nanoribbons . _ acs nano _ _ 2012 _ , _ 6 _ , 2020 - 2025 . van houten , h. ; beenakker , c. w. j. ; williamson , j. g. ; broekaart , m. e. i. ; van loosdrecht , p. h. m. coherent electron focusing with quantum point contacts in a two - dimensional electron gas . b : condens . matter mater . _ * 1989 * , _ 39 _ , 8556 - 8575 . ji , y. ; chung , y. c. ; sprinzak , d. ; heiblum , m. ; mahalu , d. ; shtrikman , h. an electronic mach - zehnder interferometer . _ nature _ * 2003 * , _ 422 _ , 415 - 418 . vanwees , b. j. ; van houten , h. ; beenakker , c. w. j. ; williamson , j. g. ; kouwenhoven , l. p. ; van der marel , d. ; foxon , c. t. quantized conductance of point contacts in a two - dimensional electron gas . lett . _ * 1988 * , _ 60 _ , 848 - 850 . pascual , j. i. ; mndez , j. ; gmez - herrero , j. ; bar , a. m. ; garcia , n. ; landman , u. ; luedtke , w. d. ; bogachek , e. n. ; cheng , h .- p . properties of metallic nanowires : from conductance quantization to localization . _ science _ * 1995 * , _ 267 _ , 1793 - 1795 . liang , w. ; bockrath , m. ; bozovic , d. ; hafner , j. h. ; tinkham , m. ; park , h. fabry - prot interference in a nanotube electron waveguide . _ nature _ * 2001 * , _ 411 _ , 665 - 669 . vaughan , j.m . ; _ the fabry - prot interferometer _ ; hilger : bristol , * 1989*. lipson , s. g. ; lipson , h. ; tannhauser , d. s. _ optical physics _ ; cambridge univ . press : london , * 1995 * , 3rd ed . ; rickhaus , p. ; maurand , r. ; liu , m .- h . ; weiss , m. ; richter , k. ; schnenberger , ch . ballistic interferences in suspended graphene . commun . _ * 2013 * , _ 4 _ , 2342 . datta , s. ; _ quantum transport : atom to transistor _ ; cambridge university press : cambridge , * 2005*. lake , r. ; datta , s. nonequilibrium greens - function method applied to double - barrier resonant - tunneling diodes . b : condens . matter mater . phys . _ * 1992 * , _ 45 _ , 6670 - 6685 . ezawa , m. peculiar band gap structure of graphene nanoribbons . ( c ) _ * 2007 * , _ 4 _ , 489 - 492 . recher , p. ; trauzettel , b. ; rycerz , a. ; blanter , ya . m. ; beenakker , c. w. j. ; morpurgo , a. f. aharonov - bohm effect and broken valley degeneracy in graphene rings . b : condens . matter mater . phys . _ * 2007 * , _ 76 _ , 235404 . bahamon , d. a. ; pereira , a. l. c. ; schulz , p. a. inner and outer edge states in graphene rings : a numerical investigation . b : condens . matter mater . phys . _ * 2009 * , _ 79 _ , 125414 . wurm , j. ; wimmer , m. ; baranger , h. u. ; richter , k. graphene rings in magnetic fields : aharonov - bohm effect and valley splitting . _ semicond . technol . _ * 2010 * , _ 25 _ , 034003 . hung nguyen , v. ; niquet , y. m. ; dollfus , p. aharonov - bohm effect and giant magnetoresistance in graphene nanoribbon rings . b : condens . matter mater . phys . _ * 2013 * , _ 88 _ , 035408 . hung nguyen , v. ; niquet , y. m. ; dollfus , p. the interplay between the aharonov - bohm interference and parity selective tunneling in graphene nanoribbon rings . _ j. phys . : condens . matter _ * 2014 * , _ 26 _ , 205301 . xu , h. ; heinzel , t. ; evaldsson , m. ; zozoulenko , i. v. magnetic barriers in graphene nanoribbons : theoretical study of transport properties . _ j. phys . : condens . matter _ * 2008 * , _ 77 _ , 245401 . jiao , l. ; wang , x. ; diankov , g. ; wang , h. ; dai , h. facile synthesis of high - quality graphene nanoribbons . nanotech . _ * 2010 * , _ 5 _ , 321 - 325 . ritter , c. ; makler , s. s. ; latg , a. energy - gap modulations of graphene ribbons under external fields : a theoretical study . _ j. phys . : condens . matter _ * 2008 * , _ 77 _ , 195443 . mckellar , b. h. j. ; stephenson , jr . , g. j. relativistic quarks in one - dimensional periodic structures . c _ * 1987 * , _ 35 _ , 2262 - 2271 . bttiker , m. ; imry , y. ; landauer , r. josephson behavior in small normal one - dimensional rings . * 1983 * , _ 96 _ , 365 - 367 . alberto , p. ; fiolhais , c. ; gil , v. m. s. relativistic particle in a box . j. phys . _ * 1996 * , _ 17 _ , 19 - 24 . lpez - villanueva , j. a. ; melchor , i. ; cartujo , p. ; carceller , j. e. modified schrdinger - equation including nonparabolicity for the study of a two - dimensional electron gas . b : condens . matter mater . phys . _ * 1993 * , _ 48 _ , 1626 - 1631 . mackenzie , r. ; palmer , w. f. bags via bosonization . d _ * 1990 * , _ 42 _ , 701 - 709 . bednyakov , v. a. ; giokaris , n. d. ; bednyakov , a. v. on higgs mass generation mechanism in the standard model . nucl . _ * 2008 * , _ 39 _ , 13 - 36 ; arxiv : hep - ph/0703280 . griffith , d. ; _ introduction to elementary particles _ ; wiley - vch : weinheim , germany , * 2008 * ; 2nd edition . zheng , h. x. ; wang , z. f. ; luo , t. ; shi , q. w. ; chen , j. analytical study of electronic structure in armchair graphene nanoribbons . _ phys . rev . b : condens . matter mater . phys . _ * 2007 * , _ 75 _ , 165414 . jackiw , r. ; rebbi , c. solitons with fermion number @xmath120 . d _ * 1976 * , _ 13 _ , 3398 - 3409 . jackiw , r. ; kerman , a. k. ; klebanov , i. ; semenoff , g. fluctuations of fractional charge in soliton anti - soliton systems . b _ * 1983 * , _ 225 _ , 233 - 246 . + * acknowledgements : * this work was supported by a grant from the office of basic energy sciences of the us department of energy under contract no . fg05 - 86er45234 . computations were made at the gatech center for computational materials science . + + * author contributions : * i.r . & c.y . performed the computations . c.y . , i.r . & u.l . analyzed the results . & u.l . wrote the manuscript . + + * competing financial interests : * the authors declare no competing financial interests . + + * correspondence * should be addressed to u.l . : ( uzi.landman@physics.gatech.edu ) .